1 //====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE 1 & 2 Instructions Classes
19 //===----------------------------------------------------------------------===//
21 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
22 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
23 RegisterClass RC, X86MemOperand x86memop,
25 let isCommutable = 1 in {
26 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
28 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
29 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
30 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
32 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
34 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
35 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
36 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
39 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
40 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
41 string asm, string SSEVer, string FPSizeStr,
42 Operand memopr, ComplexPattern mem_cpat,
44 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
46 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
47 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
48 [(set RC:$dst, (!cast<Intrinsic>(
49 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
50 RC:$src1, RC:$src2))]>;
51 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
53 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
54 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
55 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
56 SSEVer, "_", OpcodeStr, FPSizeStr))
57 RC:$src1, mem_cpat:$src2))]>;
60 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
61 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
62 RegisterClass RC, ValueType vt,
63 X86MemOperand x86memop, PatFrag mem_frag,
64 Domain d, bit Is2Addr = 1> {
65 let isCommutable = 1 in
66 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
68 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
69 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
70 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], d>;
72 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
74 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
75 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
76 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], d>;
79 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
80 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
81 string OpcodeStr, X86MemOperand x86memop,
82 list<dag> pat_rr, list<dag> pat_rm,
84 bit rr_hasSideEffects = 0> {
85 let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
86 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
88 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
89 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
91 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
93 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
94 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
98 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
99 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
100 string asm, string SSEVer, string FPSizeStr,
101 X86MemOperand x86memop, PatFrag mem_frag,
102 Domain d, bit Is2Addr = 1> {
103 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
105 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
106 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
107 [(set RC:$dst, (!cast<Intrinsic>(
108 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
109 RC:$src1, RC:$src2))], d>;
110 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
112 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
113 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
114 [(set RC:$dst, (!cast<Intrinsic>(
115 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
116 RC:$src1, (mem_frag addr:$src2)))], d>;
119 //===----------------------------------------------------------------------===//
120 // Non-instruction patterns
121 //===----------------------------------------------------------------------===//
123 // A vector extract of the first f32/f64 position is a subregister copy
124 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
125 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
126 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
127 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
129 // A 128-bit subvector extract from the first 256-bit vector position
130 // is a subregister copy that needs no instruction.
131 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
132 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
133 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
134 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
136 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
137 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
138 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
139 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
141 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
142 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
143 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
144 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
146 // A 128-bit subvector insert to the first 256-bit vector position
147 // is a subregister copy that needs no instruction.
148 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
149 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
150 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
151 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
152 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
153 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
154 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
155 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
156 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
157 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
158 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
159 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
161 // Implicitly promote a 32-bit scalar to a vector.
162 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
163 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
164 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
165 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
166 // Implicitly promote a 64-bit scalar to a vector.
167 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
168 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
169 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
170 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
172 // Bitcasts between 128-bit vector types. Return the original type since
173 // no instruction is needed for the conversion
174 let Predicates = [HasSSE2] in {
175 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
176 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
177 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
178 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
179 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
180 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
181 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
182 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
183 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
184 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
185 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
186 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
187 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
188 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
189 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
190 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
191 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
192 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
193 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
194 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
195 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
196 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
197 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
198 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
199 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
200 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
201 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
202 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
203 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
204 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
207 // Bitcasts between 256-bit vector types. Return the original type since
208 // no instruction is needed for the conversion
209 let Predicates = [HasAVX] in {
210 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
211 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
212 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
213 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
214 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
215 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
216 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
217 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
218 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
219 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
220 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
221 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
222 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
223 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
224 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
225 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
226 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
227 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
228 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
229 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
230 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
231 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
232 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
233 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
234 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
235 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
236 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
237 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
238 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
239 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
242 // Alias instructions that map fld0 to pxor for sse.
243 // This is expanded by ExpandPostRAPseudos.
244 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
246 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
247 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
248 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
249 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
252 //===----------------------------------------------------------------------===//
253 // AVX & SSE - Zero/One Vectors
254 //===----------------------------------------------------------------------===//
256 // Alias instruction that maps zero vector to pxor / xorp* for sse.
257 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
258 // swizzled by ExecutionDepsFix to pxor.
259 // We set canFoldAsLoad because this can be converted to a constant-pool
260 // load of an all-zeros value if folding it would be beneficial.
261 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
262 isPseudo = 1, neverHasSideEffects = 1 in {
263 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
266 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
267 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
268 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
269 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
270 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
271 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
274 // The same as done above but for AVX. The 256-bit ISA does not support PI,
275 // and doesn't need it because on sandy bridge the register is set to zero
276 // at the rename stage without using any execution unit, so SET0PSY
277 // and SET0PDY can be used for vector int instructions without penalty
278 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
279 // JIT implementatioan, it does not expand the instructions below like
280 // X86MCInstLower does.
281 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
282 isCodeGenOnly = 1 in {
283 let Predicates = [HasAVX] in {
284 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
285 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
286 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
287 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
289 let Predicates = [HasAVX2], neverHasSideEffects = 1 in
290 def AVX2_SET0 : PDI<0xef, MRMInitReg, (outs VR256:$dst), (ins), "",
294 let Predicates = [HasAVX2], AddedComplexity = 5 in {
295 def : Pat<(v4i64 immAllZerosV), (AVX2_SET0)>;
296 def : Pat<(v8i32 immAllZerosV), (AVX2_SET0)>;
297 def : Pat<(v16i16 immAllZerosV), (AVX2_SET0)>;
298 def : Pat<(v32i8 immAllZerosV), (AVX2_SET0)>;
301 // AVX has no support for 256-bit integer instructions, but since the 128-bit
302 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
303 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
304 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
305 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
307 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
308 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
309 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
311 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
312 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
313 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
315 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
316 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
317 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
319 // We set canFoldAsLoad because this can be converted to a constant-pool
320 // load of an all-ones value if folding it would be beneficial.
321 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
322 // JIT implementation, it does not expand the instructions below like
323 // X86MCInstLower does.
324 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
325 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
326 let Predicates = [HasAVX] in
327 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
328 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
329 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
330 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
331 let Predicates = [HasAVX2] in
332 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
333 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
337 //===----------------------------------------------------------------------===//
338 // SSE 1 & 2 - Move FP Scalar Instructions
340 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
341 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
342 // is used instead. Register-to-register movss/movsd is not modeled as an
343 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
344 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
345 //===----------------------------------------------------------------------===//
347 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
348 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
349 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
351 // Loading from memory automatically zeroing upper bits.
352 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
353 PatFrag mem_pat, string OpcodeStr> :
354 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
355 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
356 [(set RC:$dst, (mem_pat addr:$src))]>;
359 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
360 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
362 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
363 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
366 // For the disassembler
367 let isCodeGenOnly = 1 in {
368 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
369 (ins VR128:$src1, FR32:$src2),
370 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
372 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
373 (ins VR128:$src1, FR64:$src2),
374 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
378 let canFoldAsLoad = 1, isReMaterializable = 1 in {
379 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
381 let AddedComplexity = 20 in
382 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
386 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
387 "movss\t{$src, $dst|$dst, $src}",
388 [(store FR32:$src, addr:$dst)]>, XS, VEX, VEX_LIG;
389 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
390 "movsd\t{$src, $dst|$dst, $src}",
391 [(store FR64:$src, addr:$dst)]>, XD, VEX, VEX_LIG;
394 let Constraints = "$src1 = $dst" in {
395 def MOVSSrr : sse12_move_rr<FR32, v4f32,
396 "movss\t{$src2, $dst|$dst, $src2}">, XS;
397 def MOVSDrr : sse12_move_rr<FR64, v2f64,
398 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
400 // For the disassembler
401 let isCodeGenOnly = 1 in {
402 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
403 (ins VR128:$src1, FR32:$src2),
404 "movss\t{$src2, $dst|$dst, $src2}", []>, XS;
405 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
406 (ins VR128:$src1, FR64:$src2),
407 "movsd\t{$src2, $dst|$dst, $src2}", []>, XD;
411 let canFoldAsLoad = 1, isReMaterializable = 1 in {
412 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
414 let AddedComplexity = 20 in
415 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
418 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
419 "movss\t{$src, $dst|$dst, $src}",
420 [(store FR32:$src, addr:$dst)]>;
421 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
422 "movsd\t{$src, $dst|$dst, $src}",
423 [(store FR64:$src, addr:$dst)]>;
426 let Predicates = [HasAVX] in {
427 let AddedComplexity = 15 in {
428 // Extract the low 32-bit value from one vector and insert it into another.
429 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
430 (VMOVSSrr (v4f32 VR128:$src1),
431 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
432 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
433 (VMOVSSrr (v4i32 VR128:$src1),
434 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
436 // Extract the low 64-bit value from one vector and insert it into another.
437 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
438 (VMOVSDrr (v2f64 VR128:$src1),
439 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
440 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
441 (VMOVSDrr (v2i64 VR128:$src1),
442 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
444 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
445 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
446 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
447 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
448 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
450 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
451 // MOVS{S,D} to the lower bits.
452 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
453 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
454 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
455 (VMOVSSrr (v4f32 (V_SET0)),
456 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
457 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
458 (VMOVSSrr (v4i32 (V_SET0)),
459 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
460 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
461 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
463 // Move low f32 and clear high bits.
464 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
465 (SUBREG_TO_REG (i32 0),
466 (VMOVSSrr (v4f32 (V_SET0)),
467 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
468 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
469 (SUBREG_TO_REG (i32 0),
470 (VMOVSSrr (v4i32 (V_SET0)),
471 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
474 let AddedComplexity = 20 in {
475 // MOVSSrm zeros the high parts of the register; represent this
476 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
477 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
478 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
479 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
480 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
481 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
482 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
484 // MOVSDrm zeros the high parts of the register; represent this
485 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
486 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
487 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
488 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
489 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
490 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
491 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
492 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
493 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
494 def : Pat<(v2f64 (X86vzload addr:$src)),
495 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
497 // Represent the same patterns above but in the form they appear for
499 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
500 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
501 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
502 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
503 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
504 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
505 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
506 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
507 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
509 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
510 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
511 (SUBREG_TO_REG (i32 0),
512 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
514 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
515 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
516 (SUBREG_TO_REG (i64 0),
517 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
519 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
520 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
521 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
523 // Move low f64 and clear high bits.
524 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
525 (SUBREG_TO_REG (i32 0),
526 (VMOVSDrr (v2f64 (V_SET0)),
527 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
529 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
530 (SUBREG_TO_REG (i32 0),
531 (VMOVSDrr (v2i64 (V_SET0)),
532 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_sd)), sub_xmm)>;
534 // Extract and store.
535 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
538 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
539 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
542 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
544 // Shuffle with VMOVSS
545 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
546 (VMOVSSrr VR128:$src1, FR32:$src2)>;
547 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
548 (VMOVSSrr (v4i32 VR128:$src1),
549 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
550 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
551 (VMOVSSrr (v4f32 VR128:$src1),
552 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
555 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
556 (SUBREG_TO_REG (i32 0),
557 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
558 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
559 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
560 (SUBREG_TO_REG (i32 0),
561 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
562 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
564 // Shuffle with VMOVSD
565 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
566 (VMOVSDrr VR128:$src1, FR64:$src2)>;
567 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
568 (VMOVSDrr (v2i64 VR128:$src1),
569 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
570 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
571 (VMOVSDrr (v2f64 VR128:$src1),
572 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
573 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
574 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
576 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
577 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
581 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
582 (SUBREG_TO_REG (i32 0),
583 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
584 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
585 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
586 (SUBREG_TO_REG (i32 0),
587 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
588 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
591 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
592 // is during lowering, where it's not possible to recognize the fold cause
593 // it has two uses through a bitcast. One use disappears at isel time and the
594 // fold opportunity reappears.
595 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
596 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
598 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
599 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
601 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
602 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
604 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
605 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
609 let Predicates = [HasSSE1] in {
610 let AddedComplexity = 15 in {
611 // Extract the low 32-bit value from one vector and insert it into another.
612 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
613 (MOVSSrr (v4f32 VR128:$src1),
614 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
615 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
616 (MOVSSrr (v4i32 VR128:$src1),
617 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
619 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
620 // MOVSS to the lower bits.
621 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
622 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
623 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
624 (MOVSSrr (v4f32 (V_SET0)),
625 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
626 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
627 (MOVSSrr (v4i32 (V_SET0)),
628 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
631 let AddedComplexity = 20 in {
632 // MOVSSrm zeros the high parts of the register; represent this
633 // with SUBREG_TO_REG.
634 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
635 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
636 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
637 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
638 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
639 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
642 // Extract and store.
643 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
646 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
648 // Shuffle with MOVSS
649 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
650 (MOVSSrr VR128:$src1, FR32:$src2)>;
651 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
652 (MOVSSrr (v4i32 VR128:$src1),
653 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
654 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
655 (MOVSSrr (v4f32 VR128:$src1),
656 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
659 let Predicates = [HasSSE2] in {
660 let AddedComplexity = 15 in {
661 // Extract the low 64-bit value from one vector and insert it into another.
662 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
663 (MOVSDrr (v2f64 VR128:$src1),
664 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
665 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
666 (MOVSDrr (v2i64 VR128:$src1),
667 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
669 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
670 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
671 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
672 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
673 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
675 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
676 // MOVSD to the lower bits.
677 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
678 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
681 let AddedComplexity = 20 in {
682 // MOVSDrm zeros the high parts of the register; represent this
683 // with SUBREG_TO_REG.
684 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
685 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
686 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
687 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
688 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
689 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
690 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
691 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
692 def : Pat<(v2f64 (X86vzload addr:$src)),
693 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
696 // Extract and store.
697 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
700 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
702 // Shuffle with MOVSD
703 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
704 (MOVSDrr VR128:$src1, FR64:$src2)>;
705 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
706 (MOVSDrr (v2i64 VR128:$src1),
707 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
708 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
709 (MOVSDrr (v2f64 VR128:$src1),
710 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
711 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
712 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
713 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
714 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
716 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
717 // is during lowering, where it's not possible to recognize the fold cause
718 // it has two uses through a bitcast. One use disappears at isel time and the
719 // fold opportunity reappears.
720 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
721 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
722 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
723 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
724 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
725 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
726 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
727 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
730 //===----------------------------------------------------------------------===//
731 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
732 //===----------------------------------------------------------------------===//
734 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
735 X86MemOperand x86memop, PatFrag ld_frag,
736 string asm, Domain d,
737 bit IsReMaterializable = 1> {
738 let neverHasSideEffects = 1 in
739 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
740 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
741 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
742 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
743 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
744 [(set RC:$dst, (ld_frag addr:$src))], d>;
747 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
748 "movaps", SSEPackedSingle>, TB, VEX;
749 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
750 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
751 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
752 "movups", SSEPackedSingle>, TB, VEX;
753 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
754 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
756 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
757 "movaps", SSEPackedSingle>, TB, VEX;
758 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
759 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
760 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
761 "movups", SSEPackedSingle>, TB, VEX;
762 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
763 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
764 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
765 "movaps", SSEPackedSingle>, TB;
766 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
767 "movapd", SSEPackedDouble>, TB, OpSize;
768 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
769 "movups", SSEPackedSingle>, TB;
770 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
771 "movupd", SSEPackedDouble, 0>, TB, OpSize;
773 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
774 "movaps\t{$src, $dst|$dst, $src}",
775 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
776 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
777 "movapd\t{$src, $dst|$dst, $src}",
778 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
779 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
780 "movups\t{$src, $dst|$dst, $src}",
781 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
782 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
783 "movupd\t{$src, $dst|$dst, $src}",
784 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
785 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
786 "movaps\t{$src, $dst|$dst, $src}",
787 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)]>, VEX;
788 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
789 "movapd\t{$src, $dst|$dst, $src}",
790 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)]>, VEX;
791 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
792 "movups\t{$src, $dst|$dst, $src}",
793 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
794 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
795 "movupd\t{$src, $dst|$dst, $src}",
796 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
799 let isCodeGenOnly = 1 in {
800 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
802 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
803 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
805 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
806 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
808 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
809 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
811 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
812 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
814 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
815 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
817 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
818 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
820 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
821 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
823 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
826 let Predicates = [HasAVX] in {
827 def : Pat<(v8i32 (X86vzmovl
828 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
829 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
830 def : Pat<(v4i64 (X86vzmovl
831 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
832 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
833 def : Pat<(v8f32 (X86vzmovl
834 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
835 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
836 def : Pat<(v4f64 (X86vzmovl
837 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
838 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
842 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
843 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
844 (VMOVUPSYmr addr:$dst, VR256:$src)>;
846 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
847 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
848 (VMOVUPDYmr addr:$dst, VR256:$src)>;
850 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
851 "movaps\t{$src, $dst|$dst, $src}",
852 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
853 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
854 "movapd\t{$src, $dst|$dst, $src}",
855 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
856 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
857 "movups\t{$src, $dst|$dst, $src}",
858 [(store (v4f32 VR128:$src), addr:$dst)]>;
859 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
860 "movupd\t{$src, $dst|$dst, $src}",
861 [(store (v2f64 VR128:$src), addr:$dst)]>;
864 let isCodeGenOnly = 1 in {
865 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
866 "movaps\t{$src, $dst|$dst, $src}", []>;
867 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
868 "movapd\t{$src, $dst|$dst, $src}", []>;
869 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
870 "movups\t{$src, $dst|$dst, $src}", []>;
871 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
872 "movupd\t{$src, $dst|$dst, $src}", []>;
875 let Predicates = [HasAVX] in {
876 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
877 (VMOVUPSmr addr:$dst, VR128:$src)>;
878 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
879 (VMOVUPDmr addr:$dst, VR128:$src)>;
882 let Predicates = [HasSSE1] in
883 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
884 (MOVUPSmr addr:$dst, VR128:$src)>;
885 let Predicates = [HasSSE2] in
886 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
887 (MOVUPDmr addr:$dst, VR128:$src)>;
889 // Use vmovaps/vmovups for AVX integer load/store.
890 let Predicates = [HasAVX] in {
891 // 128-bit load/store
892 def : Pat<(alignedloadv2i64 addr:$src),
893 (VMOVAPSrm addr:$src)>;
894 def : Pat<(loadv2i64 addr:$src),
895 (VMOVUPSrm addr:$src)>;
897 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
898 (VMOVAPSmr addr:$dst, VR128:$src)>;
899 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
900 (VMOVAPSmr addr:$dst, VR128:$src)>;
901 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
902 (VMOVAPSmr addr:$dst, VR128:$src)>;
903 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
904 (VMOVAPSmr addr:$dst, VR128:$src)>;
905 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
906 (VMOVUPSmr addr:$dst, VR128:$src)>;
907 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
908 (VMOVUPSmr addr:$dst, VR128:$src)>;
909 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
910 (VMOVUPSmr addr:$dst, VR128:$src)>;
911 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
912 (VMOVUPSmr addr:$dst, VR128:$src)>;
914 // 256-bit load/store
915 def : Pat<(alignedloadv4i64 addr:$src),
916 (VMOVAPSYrm addr:$src)>;
917 def : Pat<(loadv4i64 addr:$src),
918 (VMOVUPSYrm addr:$src)>;
919 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
920 (VMOVAPSYmr addr:$dst, VR256:$src)>;
921 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
922 (VMOVAPSYmr addr:$dst, VR256:$src)>;
923 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
924 (VMOVAPSYmr addr:$dst, VR256:$src)>;
925 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
926 (VMOVAPSYmr addr:$dst, VR256:$src)>;
927 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
928 (VMOVUPSYmr addr:$dst, VR256:$src)>;
929 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
930 (VMOVUPSYmr addr:$dst, VR256:$src)>;
931 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
932 (VMOVUPSYmr addr:$dst, VR256:$src)>;
933 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
934 (VMOVUPSYmr addr:$dst, VR256:$src)>;
937 // Use movaps / movups for SSE integer load / store (one byte shorter).
938 // The instructions selected below are then converted to MOVDQA/MOVDQU
939 // during the SSE domain pass.
940 let Predicates = [HasSSE1] in {
941 def : Pat<(alignedloadv2i64 addr:$src),
942 (MOVAPSrm addr:$src)>;
943 def : Pat<(loadv2i64 addr:$src),
944 (MOVUPSrm addr:$src)>;
946 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
947 (MOVAPSmr addr:$dst, VR128:$src)>;
948 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
949 (MOVAPSmr addr:$dst, VR128:$src)>;
950 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
951 (MOVAPSmr addr:$dst, VR128:$src)>;
952 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
953 (MOVAPSmr addr:$dst, VR128:$src)>;
954 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
955 (MOVUPSmr addr:$dst, VR128:$src)>;
956 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
957 (MOVUPSmr addr:$dst, VR128:$src)>;
958 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
959 (MOVUPSmr addr:$dst, VR128:$src)>;
960 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
961 (MOVUPSmr addr:$dst, VR128:$src)>;
964 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
965 // bits are disregarded. FIXME: Set encoding to pseudo!
966 let neverHasSideEffects = 1 in {
967 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
968 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
969 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
970 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
971 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
972 "movaps\t{$src, $dst|$dst, $src}", []>;
973 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
974 "movapd\t{$src, $dst|$dst, $src}", []>;
977 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
978 // bits are disregarded. FIXME: Set encoding to pseudo!
979 let canFoldAsLoad = 1, isReMaterializable = 1 in {
980 let isCodeGenOnly = 1 in {
981 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
982 "movaps\t{$src, $dst|$dst, $src}",
983 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
984 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
985 "movapd\t{$src, $dst|$dst, $src}",
986 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
988 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
989 "movaps\t{$src, $dst|$dst, $src}",
990 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
991 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
992 "movapd\t{$src, $dst|$dst, $src}",
993 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
996 //===----------------------------------------------------------------------===//
997 // SSE 1 & 2 - Move Low packed FP Instructions
998 //===----------------------------------------------------------------------===//
1000 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
1001 PatFrag mov_frag, string base_opc,
1003 def PSrm : PI<opc, MRMSrcMem,
1004 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1005 !strconcat(base_opc, "s", asm_opr),
1008 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1009 SSEPackedSingle>, TB;
1011 def PDrm : PI<opc, MRMSrcMem,
1012 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1013 !strconcat(base_opc, "d", asm_opr),
1014 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
1015 (scalar_to_vector (loadf64 addr:$src2)))))],
1016 SSEPackedDouble>, TB, OpSize;
1019 let AddedComplexity = 20 in {
1020 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1021 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1023 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1024 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1025 "\t{$src2, $dst|$dst, $src2}">;
1028 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1029 "movlps\t{$src, $dst|$dst, $src}",
1030 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1031 (iPTR 0))), addr:$dst)]>, VEX;
1032 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1033 "movlpd\t{$src, $dst|$dst, $src}",
1034 [(store (f64 (vector_extract (v2f64 VR128:$src),
1035 (iPTR 0))), addr:$dst)]>, VEX;
1036 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1037 "movlps\t{$src, $dst|$dst, $src}",
1038 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1039 (iPTR 0))), addr:$dst)]>;
1040 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1041 "movlpd\t{$src, $dst|$dst, $src}",
1042 [(store (f64 (vector_extract (v2f64 VR128:$src),
1043 (iPTR 0))), addr:$dst)]>;
1045 let Predicates = [HasAVX] in {
1046 let AddedComplexity = 20 in {
1047 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1048 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1049 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1050 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1051 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1052 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1053 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1054 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1055 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1056 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1059 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1060 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1061 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1062 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1063 VR128:$src2)), addr:$src1),
1064 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1066 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1067 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1068 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1069 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1070 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1072 // Shuffle with VMOVLPS
1073 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1074 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1075 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1076 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1077 def : Pat<(X86Movlps VR128:$src1,
1078 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1079 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1081 // Shuffle with VMOVLPD
1082 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1083 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1084 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1085 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1086 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1087 (scalar_to_vector (loadf64 addr:$src2)))),
1088 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1091 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1093 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1094 def : Pat<(store (v4i32 (X86Movlps
1095 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1096 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1097 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1099 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1100 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1102 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1105 let Predicates = [HasSSE1] in {
1106 let AddedComplexity = 20 in {
1107 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1108 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1109 (MOVLPSrm VR128:$src1, addr:$src2)>;
1110 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1111 (MOVLPSrm VR128:$src1, addr:$src2)>;
1114 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1115 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1116 (iPTR 0))), addr:$src1),
1117 (MOVLPSmr addr:$src1, VR128:$src2)>;
1118 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1119 (MOVLPSmr addr:$src1, VR128:$src2)>;
1120 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1121 VR128:$src2)), addr:$src1),
1122 (MOVLPSmr addr:$src1, VR128:$src2)>;
1124 // Shuffle with MOVLPS
1125 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1126 (MOVLPSrm VR128:$src1, addr:$src2)>;
1127 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1128 (MOVLPSrm VR128:$src1, addr:$src2)>;
1129 def : Pat<(X86Movlps VR128:$src1,
1130 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1131 (MOVLPSrm VR128:$src1, addr:$src2)>;
1132 def : Pat<(X86Movlps VR128:$src1,
1133 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1134 (MOVLPSrm VR128:$src1, addr:$src2)>;
1137 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1139 (MOVLPSmr addr:$src1, VR128:$src2)>;
1140 def : Pat<(store (v4i32 (X86Movlps
1141 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1143 (MOVLPSmr addr:$src1, VR128:$src2)>;
1146 let Predicates = [HasSSE2] in {
1147 let AddedComplexity = 20 in {
1148 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1149 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1150 (MOVLPDrm VR128:$src1, addr:$src2)>;
1151 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1152 (MOVLPDrm VR128:$src1, addr:$src2)>;
1155 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1156 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1157 (MOVLPDmr addr:$src1, VR128:$src2)>;
1158 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1159 (MOVLPDmr addr:$src1, VR128:$src2)>;
1161 // Shuffle with MOVLPD
1162 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1163 (MOVLPDrm VR128:$src1, addr:$src2)>;
1164 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1165 (MOVLPDrm VR128:$src1, addr:$src2)>;
1166 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1167 (scalar_to_vector (loadf64 addr:$src2)))),
1168 (MOVLPDrm VR128:$src1, addr:$src2)>;
1171 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1173 (MOVLPDmr addr:$src1, VR128:$src2)>;
1174 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1176 (MOVLPDmr addr:$src1, VR128:$src2)>;
1179 //===----------------------------------------------------------------------===//
1180 // SSE 1 & 2 - Move Hi packed FP Instructions
1181 //===----------------------------------------------------------------------===//
1183 let AddedComplexity = 20 in {
1184 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1185 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1187 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1188 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1189 "\t{$src2, $dst|$dst, $src2}">;
1192 // v2f64 extract element 1 is always custom lowered to unpack high to low
1193 // and extract element 0 so the non-store version isn't too horrible.
1194 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1195 "movhps\t{$src, $dst|$dst, $src}",
1196 [(store (f64 (vector_extract
1197 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1198 (undef)), (iPTR 0))), addr:$dst)]>,
1200 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1201 "movhpd\t{$src, $dst|$dst, $src}",
1202 [(store (f64 (vector_extract
1203 (v2f64 (unpckh VR128:$src, (undef))),
1204 (iPTR 0))), addr:$dst)]>,
1206 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1207 "movhps\t{$src, $dst|$dst, $src}",
1208 [(store (f64 (vector_extract
1209 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1210 (undef)), (iPTR 0))), addr:$dst)]>;
1211 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1212 "movhpd\t{$src, $dst|$dst, $src}",
1213 [(store (f64 (vector_extract
1214 (v2f64 (unpckh VR128:$src, (undef))),
1215 (iPTR 0))), addr:$dst)]>;
1217 let Predicates = [HasAVX] in {
1219 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1220 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1221 def : Pat<(X86Movlhps VR128:$src1,
1222 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1223 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1224 def : Pat<(X86Movlhps VR128:$src1,
1225 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1226 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1227 def : Pat<(X86Movlhps VR128:$src1,
1228 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1229 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1231 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1232 // is during lowering, where it's not possible to recognize the load fold
1233 // cause it has two uses through a bitcast. One use disappears at isel time
1234 // and the fold opportunity reappears.
1235 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1236 (scalar_to_vector (loadf64 addr:$src2)))),
1237 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1239 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1240 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1241 (scalar_to_vector (loadf64 addr:$src2)))),
1242 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1245 def : Pat<(store (f64 (vector_extract
1246 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1247 (bc_v2f64 (v4f32 VR128:$src))), (iPTR 0))), addr:$dst),
1248 (VMOVHPSmr addr:$dst, VR128:$src)>;
1249 def : Pat<(store (f64 (vector_extract
1250 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))), addr:$dst),
1251 (VMOVHPDmr addr:$dst, VR128:$src)>;
1254 let Predicates = [HasSSE1] in {
1256 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1257 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1258 def : Pat<(X86Movlhps VR128:$src1,
1259 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1260 (MOVHPSrm VR128:$src1, addr:$src2)>;
1261 def : Pat<(X86Movlhps VR128:$src1,
1262 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1263 (MOVHPSrm VR128:$src1, addr:$src2)>;
1264 def : Pat<(X86Movlhps VR128:$src1,
1265 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1266 (MOVHPSrm VR128:$src1, addr:$src2)>;
1269 def : Pat<(store (f64 (vector_extract
1270 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1271 (bc_v2f64 (v4f32 VR128:$src))), (iPTR 0))), addr:$dst),
1272 (MOVHPSmr addr:$dst, VR128:$src)>;
1275 let Predicates = [HasSSE2] in {
1276 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1277 // is during lowering, where it's not possible to recognize the load fold
1278 // cause it has two uses through a bitcast. One use disappears at isel time
1279 // and the fold opportunity reappears.
1280 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1281 (scalar_to_vector (loadf64 addr:$src2)))),
1282 (MOVHPDrm VR128:$src1, addr:$src2)>;
1284 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1285 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1286 (scalar_to_vector (loadf64 addr:$src2)))),
1287 (MOVHPDrm VR128:$src1, addr:$src2)>;
1290 def : Pat<(store (f64 (vector_extract
1291 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))),addr:$dst),
1292 (MOVHPDmr addr:$dst, VR128:$src)>;
1295 //===----------------------------------------------------------------------===//
1296 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1297 //===----------------------------------------------------------------------===//
1299 let AddedComplexity = 20 in {
1300 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1301 (ins VR128:$src1, VR128:$src2),
1302 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1304 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1306 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1307 (ins VR128:$src1, VR128:$src2),
1308 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1310 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1313 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1314 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1315 (ins VR128:$src1, VR128:$src2),
1316 "movlhps\t{$src2, $dst|$dst, $src2}",
1318 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1319 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1320 (ins VR128:$src1, VR128:$src2),
1321 "movhlps\t{$src2, $dst|$dst, $src2}",
1323 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1326 let Predicates = [HasAVX] in {
1328 let AddedComplexity = 20 in {
1329 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1330 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1331 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1332 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1334 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1335 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1336 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1338 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1339 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1340 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1341 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1342 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1343 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1346 let AddedComplexity = 20 in {
1347 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1348 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1349 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1351 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1352 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1353 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1354 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1355 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1358 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1359 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1360 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1361 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1364 let Predicates = [HasSSE1] in {
1366 let AddedComplexity = 20 in {
1367 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1368 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1369 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1370 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1372 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1373 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1374 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1376 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1377 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1378 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1379 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1380 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1381 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1384 let AddedComplexity = 20 in {
1385 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1386 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1387 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1389 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1390 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1391 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1392 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1393 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1396 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1397 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1398 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1399 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1402 //===----------------------------------------------------------------------===//
1403 // SSE 1 & 2 - Conversion Instructions
1404 //===----------------------------------------------------------------------===//
1406 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1407 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1409 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1410 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1411 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1412 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1415 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1416 X86MemOperand x86memop, string asm> {
1417 let neverHasSideEffects = 1 in {
1418 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, []>;
1420 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, []>;
1421 } // neverHasSideEffects = 1
1424 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1425 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1426 string asm, Domain d> {
1427 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1428 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1429 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1430 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1433 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1434 X86MemOperand x86memop, string asm> {
1435 let neverHasSideEffects = 1 in {
1436 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1437 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1439 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1440 (ins DstRC:$src1, x86memop:$src),
1441 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1442 } // neverHasSideEffects = 1
1445 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1446 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1448 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1449 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1451 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1452 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1454 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1455 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1456 VEX, VEX_W, VEX_LIG;
1458 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1459 // register, but the same isn't true when only using memory operands,
1460 // provide other assembly "l" and "q" forms to address this explicitly
1461 // where appropriate to do so.
1462 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1464 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1465 VEX_4V, VEX_W, VEX_LIG;
1466 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1468 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1470 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1471 VEX_4V, VEX_W, VEX_LIG;
1473 let Predicates = [HasAVX], AddedComplexity = 1 in {
1474 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1475 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1476 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1477 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1478 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1479 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1480 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1481 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1483 def : Pat<(f32 (sint_to_fp GR32:$src)),
1484 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1485 def : Pat<(f32 (sint_to_fp GR64:$src)),
1486 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1487 def : Pat<(f64 (sint_to_fp GR32:$src)),
1488 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1489 def : Pat<(f64 (sint_to_fp GR64:$src)),
1490 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1493 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1494 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1495 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1496 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1497 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1498 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1499 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1500 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1501 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1502 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1503 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1504 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1505 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1506 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1507 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1508 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1510 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1511 // and/or XMM operand(s).
1513 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1514 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1516 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1517 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1518 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1519 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1520 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1521 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1524 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1525 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1526 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1527 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1529 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1530 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1531 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1532 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1533 (ins DstRC:$src1, x86memop:$src2),
1535 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1536 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1537 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1540 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1541 f128mem, load, "cvtsd2si">, XD, VEX;
1542 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1543 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1546 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1547 // Get rid of this hack or rename the intrinsics, there are several
1548 // intructions that only match with the intrinsic form, why create duplicates
1549 // to let them be recognized by the assembler?
1550 defm VCVTSD2SI : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
1551 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_LIG;
1552 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
1553 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W,
1556 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1557 f128mem, load, "cvtsd2si{l}">, XD;
1558 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1559 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1562 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1563 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1564 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1565 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1567 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1568 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1569 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1570 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1573 let Constraints = "$src1 = $dst" in {
1574 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1575 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1577 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1578 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1579 "cvtsi2ss{q}">, XS, REX_W;
1580 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1581 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1583 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1584 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1585 "cvtsi2sd">, XD, REX_W;
1590 // Aliases for intrinsics
1591 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1592 f32mem, load, "cvttss2si">, XS, VEX;
1593 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1594 int_x86_sse_cvttss2si64, f32mem, load,
1595 "cvttss2si">, XS, VEX, VEX_W;
1596 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1597 f128mem, load, "cvttsd2si">, XD, VEX;
1598 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1599 int_x86_sse2_cvttsd2si64, f128mem, load,
1600 "cvttsd2si">, XD, VEX, VEX_W;
1601 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1602 f32mem, load, "cvttss2si">, XS;
1603 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1604 int_x86_sse_cvttss2si64, f32mem, load,
1605 "cvttss2si{q}">, XS, REX_W;
1606 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1607 f128mem, load, "cvttsd2si">, XD;
1608 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1609 int_x86_sse2_cvttsd2si64, f128mem, load,
1610 "cvttsd2si{q}">, XD, REX_W;
1612 let Pattern = []<dag> in {
1613 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1614 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1616 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1617 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1619 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1620 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1621 SSEPackedSingle>, TB, VEX;
1622 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1623 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1624 SSEPackedSingle>, TB, VEX;
1627 let Pattern = []<dag> in {
1628 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1629 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1630 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1631 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1632 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1633 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1634 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1637 let Predicates = [HasAVX] in {
1638 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1639 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1640 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1641 (VCVTSS2SIrm addr:$src)>;
1642 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1643 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1644 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1645 (VCVTSS2SI64rm addr:$src)>;
1648 let Predicates = [HasSSE1] in {
1649 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1650 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1651 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1652 (CVTSS2SIrm addr:$src)>;
1653 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1654 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1655 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1656 (CVTSS2SI64rm addr:$src)>;
1661 // Convert scalar double to scalar single
1662 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1663 (ins FR64:$src1, FR64:$src2),
1664 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1667 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1668 (ins FR64:$src1, f64mem:$src2),
1669 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1670 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1672 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1675 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1676 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1677 [(set FR32:$dst, (fround FR64:$src))]>;
1678 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1679 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1680 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1681 Requires<[HasSSE2, OptForSize]>;
1683 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1684 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1686 let Constraints = "$src1 = $dst" in
1687 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1688 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1690 // Convert scalar single to scalar double
1691 // SSE2 instructions with XS prefix
1692 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1693 (ins FR32:$src1, FR32:$src2),
1694 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1695 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1697 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1698 (ins FR32:$src1, f32mem:$src2),
1699 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1700 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1702 let Predicates = [HasAVX] in {
1703 def : Pat<(f64 (fextend FR32:$src)),
1704 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1705 def : Pat<(fextend (loadf32 addr:$src)),
1706 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1707 def : Pat<(extloadf32 addr:$src),
1708 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1711 def : Pat<(extloadf32 addr:$src),
1712 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1713 Requires<[HasAVX, OptForSpeed]>;
1715 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1716 "cvtss2sd\t{$src, $dst|$dst, $src}",
1717 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1718 Requires<[HasSSE2]>;
1719 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1720 "cvtss2sd\t{$src, $dst|$dst, $src}",
1721 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1722 Requires<[HasSSE2, OptForSize]>;
1724 // extload f32 -> f64. This matches load+fextend because we have a hack in
1725 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1727 // Since these loads aren't folded into the fextend, we have to match it
1729 def : Pat<(fextend (loadf32 addr:$src)),
1730 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1731 def : Pat<(extloadf32 addr:$src),
1732 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1734 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1735 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1736 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1737 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1738 VR128:$src2))]>, XS, VEX_4V,
1740 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1741 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1742 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1743 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1744 (load addr:$src2)))]>, XS, VEX_4V,
1746 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1747 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1748 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1749 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1750 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1751 VR128:$src2))]>, XS,
1752 Requires<[HasSSE2]>;
1753 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1754 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1755 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1756 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1757 (load addr:$src2)))]>, XS,
1758 Requires<[HasSSE2]>;
1761 // Convert doubleword to packed single/double fp
1762 // SSE2 instructions without OpSize prefix
1763 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1764 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1765 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1766 TB, VEX, Requires<[HasAVX]>;
1767 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1768 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1769 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1770 (bitconvert (memopv2i64 addr:$src))))]>,
1771 TB, VEX, Requires<[HasAVX]>;
1772 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1773 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1774 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1775 TB, Requires<[HasSSE2]>;
1776 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1777 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1778 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1779 (bitconvert (memopv2i64 addr:$src))))]>,
1780 TB, Requires<[HasSSE2]>;
1782 // FIXME: why the non-intrinsic version is described as SSE3?
1783 // SSE2 instructions with XS prefix
1784 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1785 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1786 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1787 XS, VEX, Requires<[HasAVX]>;
1788 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1789 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1790 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1791 (bitconvert (memopv2i64 addr:$src))))]>,
1792 XS, VEX, Requires<[HasAVX]>;
1793 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1794 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1795 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1796 XS, Requires<[HasSSE2]>;
1797 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1798 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1799 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1800 (bitconvert (memopv2i64 addr:$src))))]>,
1801 XS, Requires<[HasSSE2]>;
1804 // Convert packed single/double fp to doubleword
1805 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1806 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1807 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1808 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1809 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1810 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1811 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1812 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1813 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1814 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1815 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1816 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1818 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1819 "cvtps2dq\t{$src, $dst|$dst, $src}",
1820 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1822 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1824 "cvtps2dq\t{$src, $dst|$dst, $src}",
1825 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1826 (memop addr:$src)))]>, VEX;
1827 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1828 "cvtps2dq\t{$src, $dst|$dst, $src}",
1829 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1830 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1831 "cvtps2dq\t{$src, $dst|$dst, $src}",
1832 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1833 (memop addr:$src)))]>;
1835 // SSE2 packed instructions with XD prefix
1836 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1837 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1838 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1839 XD, VEX, Requires<[HasAVX]>;
1840 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1841 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1842 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1843 (memop addr:$src)))]>,
1844 XD, VEX, Requires<[HasAVX]>;
1845 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1846 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1847 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1848 XD, Requires<[HasSSE2]>;
1849 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1850 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1851 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1852 (memop addr:$src)))]>,
1853 XD, Requires<[HasSSE2]>;
1856 // Convert with truncation packed single/double fp to doubleword
1857 // SSE2 packed instructions with XS prefix
1858 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1859 "cvttps2dq\t{$src, $dst|$dst, $src}",
1861 (int_x86_sse2_cvttps2dq VR128:$src))]>, VEX;
1862 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1863 "cvttps2dq\t{$src, $dst|$dst, $src}",
1864 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1865 (memop addr:$src)))]>, VEX;
1866 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1867 "cvttps2dq\t{$src, $dst|$dst, $src}",
1869 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))]>, VEX;
1870 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1871 "cvttps2dq\t{$src, $dst|$dst, $src}",
1872 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1873 (memopv8f32 addr:$src)))]>, VEX;
1875 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1876 "cvttps2dq\t{$src, $dst|$dst, $src}",
1878 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1879 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1880 "cvttps2dq\t{$src, $dst|$dst, $src}",
1882 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1884 let Predicates = [HasAVX] in {
1885 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1886 (Int_VCVTDQ2PSrr VR128:$src)>;
1887 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1888 (Int_VCVTDQ2PSrm addr:$src)>;
1890 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1891 (VCVTTPS2DQrr VR128:$src)>;
1892 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1893 (VCVTTPS2DQrm addr:$src)>;
1895 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1896 (VCVTDQ2PSYrr VR256:$src)>;
1897 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1898 (VCVTDQ2PSYrm addr:$src)>;
1900 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1901 (VCVTTPS2DQYrr VR256:$src)>;
1902 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1903 (VCVTTPS2DQYrm addr:$src)>;
1906 let Predicates = [HasSSE2] in {
1907 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1908 (Int_CVTDQ2PSrr VR128:$src)>;
1909 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1910 (Int_CVTDQ2PSrm addr:$src)>;
1912 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1913 (CVTTPS2DQrr VR128:$src)>;
1914 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1915 (CVTTPS2DQrm addr:$src)>;
1918 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1919 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1921 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1922 let isCodeGenOnly = 1 in
1923 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1924 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1925 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1926 (memop addr:$src)))]>, VEX;
1927 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1928 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1929 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1930 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1931 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1932 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1933 (memop addr:$src)))]>;
1935 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1936 // register, but the same isn't true when using memory operands instead.
1937 // Provide other assembly rr and rm forms to address this explicitly.
1938 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1939 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1942 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1943 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1944 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1945 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1948 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1949 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1950 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1951 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1953 // Convert packed single to packed double
1954 let Predicates = [HasAVX] in {
1955 // SSE2 instructions without OpSize prefix
1956 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1957 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1958 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1959 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1960 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1961 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1962 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1963 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1965 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1966 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1967 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1968 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1970 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1971 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1972 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1973 TB, VEX, Requires<[HasAVX]>;
1974 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1975 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1976 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1977 (load addr:$src)))]>,
1978 TB, VEX, Requires<[HasAVX]>;
1979 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1980 "cvtps2pd\t{$src, $dst|$dst, $src}",
1981 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1982 TB, Requires<[HasSSE2]>;
1983 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1984 "cvtps2pd\t{$src, $dst|$dst, $src}",
1985 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1986 (load addr:$src)))]>,
1987 TB, Requires<[HasSSE2]>;
1989 // Convert packed double to packed single
1990 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1991 // register, but the same isn't true when using memory operands instead.
1992 // Provide other assembly rr and rm forms to address this explicitly.
1993 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1994 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1995 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1996 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1999 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2000 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
2001 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2002 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
2005 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2006 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
2007 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2008 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
2009 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2010 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
2011 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2012 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
2015 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2016 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2017 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2018 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
2020 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2021 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2022 (memop addr:$src)))]>;
2023 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2024 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2025 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2026 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2027 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2028 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2029 (memop addr:$src)))]>;
2031 // AVX 256-bit register conversion intrinsics
2032 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2033 // whenever possible to avoid declaring two versions of each one.
2034 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2035 (VCVTDQ2PSYrr VR256:$src)>;
2036 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2037 (VCVTDQ2PSYrm addr:$src)>;
2039 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2040 (VCVTPD2PSYrr VR256:$src)>;
2041 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2042 (VCVTPD2PSYrm addr:$src)>;
2044 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2045 (VCVTPS2DQYrr VR256:$src)>;
2046 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2047 (VCVTPS2DQYrm addr:$src)>;
2049 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2050 (VCVTPS2PDYrr VR128:$src)>;
2051 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2052 (VCVTPS2PDYrm addr:$src)>;
2054 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2055 (VCVTTPD2DQYrr VR256:$src)>;
2056 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2057 (VCVTTPD2DQYrm addr:$src)>;
2059 // Match fround and fextend for 128/256-bit conversions
2060 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2061 (VCVTPD2PSYrr VR256:$src)>;
2062 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2063 (VCVTPD2PSYrm addr:$src)>;
2065 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2066 (VCVTPS2PDYrr VR128:$src)>;
2067 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2068 (VCVTPS2PDYrm addr:$src)>;
2070 //===----------------------------------------------------------------------===//
2071 // SSE 1 & 2 - Compare Instructions
2072 //===----------------------------------------------------------------------===//
2074 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2075 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2076 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2077 string asm, string asm_alt> {
2078 def rr : SIi8<0xC2, MRMSrcReg,
2079 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2080 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2081 def rm : SIi8<0xC2, MRMSrcMem,
2082 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2083 [(set RC:$dst, (OpNode (VT RC:$src1),
2084 (ld_frag addr:$src2), imm:$cc))]>;
2086 // Accept explicit immediate argument form instead of comparison code.
2087 let neverHasSideEffects = 1 in {
2088 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2089 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2091 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2092 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2096 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2097 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2098 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2099 XS, VEX_4V, VEX_LIG;
2100 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2101 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2102 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2103 XD, VEX_4V, VEX_LIG;
2105 let Constraints = "$src1 = $dst" in {
2106 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2107 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2108 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2110 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2111 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2112 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2116 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2117 Intrinsic Int, string asm> {
2118 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2119 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2120 [(set VR128:$dst, (Int VR128:$src1,
2121 VR128:$src, imm:$cc))]>;
2122 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2123 (ins VR128:$src1, x86memop:$src, SSECC:$cc), asm,
2124 [(set VR128:$dst, (Int VR128:$src1,
2125 (load addr:$src), imm:$cc))]>;
2128 // Aliases to match intrinsics which expect XMM operand(s).
2129 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2130 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2132 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2133 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2135 let Constraints = "$src1 = $dst" in {
2136 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2137 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2138 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2139 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2143 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2144 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2145 ValueType vt, X86MemOperand x86memop,
2146 PatFrag ld_frag, string OpcodeStr, Domain d> {
2147 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2148 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2149 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2150 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2151 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2152 [(set EFLAGS, (OpNode (vt RC:$src1),
2153 (ld_frag addr:$src2)))], d>;
2156 let Defs = [EFLAGS] in {
2157 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2158 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2159 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2160 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2162 let Pattern = []<dag> in {
2163 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2164 "comiss", SSEPackedSingle>, TB, VEX,
2166 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2167 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2171 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2172 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2173 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2174 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2176 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2177 load, "comiss", SSEPackedSingle>, TB, VEX;
2178 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2179 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2180 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2181 "ucomiss", SSEPackedSingle>, TB;
2182 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2183 "ucomisd", SSEPackedDouble>, TB, OpSize;
2185 let Pattern = []<dag> in {
2186 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2187 "comiss", SSEPackedSingle>, TB;
2188 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2189 "comisd", SSEPackedDouble>, TB, OpSize;
2192 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2193 load, "ucomiss", SSEPackedSingle>, TB;
2194 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2195 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2197 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2198 "comiss", SSEPackedSingle>, TB;
2199 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2200 "comisd", SSEPackedDouble>, TB, OpSize;
2201 } // Defs = [EFLAGS]
2203 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2204 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2205 Intrinsic Int, string asm, string asm_alt,
2207 let isAsmParserOnly = 1 in {
2208 def rri : PIi8<0xC2, MRMSrcReg,
2209 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2210 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2211 def rmi : PIi8<0xC2, MRMSrcMem,
2212 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2213 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2216 // Accept explicit immediate argument form instead of comparison code.
2217 def rri_alt : PIi8<0xC2, MRMSrcReg,
2218 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2220 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2221 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2225 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2226 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2227 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2228 SSEPackedSingle>, TB, VEX_4V;
2229 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2230 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2231 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2232 SSEPackedDouble>, TB, OpSize, VEX_4V;
2233 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2234 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2235 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2236 SSEPackedSingle>, TB, VEX_4V;
2237 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2238 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2239 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2240 SSEPackedDouble>, TB, OpSize, VEX_4V;
2241 let Constraints = "$src1 = $dst" in {
2242 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2243 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2244 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2245 SSEPackedSingle>, TB;
2246 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2247 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2248 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2249 SSEPackedDouble>, TB, OpSize;
2252 let Predicates = [HasAVX] in {
2253 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2254 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2255 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2256 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2257 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2258 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2259 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2260 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2262 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2263 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2264 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2265 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2266 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2267 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2268 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2269 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2272 let Predicates = [HasSSE1] in {
2273 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2274 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2275 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2276 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2279 let Predicates = [HasSSE2] in {
2280 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2281 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2282 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2283 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2286 //===----------------------------------------------------------------------===//
2287 // SSE 1 & 2 - Shuffle Instructions
2288 //===----------------------------------------------------------------------===//
2290 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2291 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2292 ValueType vt, string asm, PatFrag mem_frag,
2293 Domain d, bit IsConvertibleToThreeAddress = 0> {
2294 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2295 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2296 [(set RC:$dst, (vt (shufp:$src3
2297 RC:$src1, (mem_frag addr:$src2))))], d>;
2298 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2299 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2300 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2302 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2305 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2306 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2307 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2308 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2309 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2310 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2311 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2312 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2313 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2314 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2315 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2316 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2318 let Constraints = "$src1 = $dst" in {
2319 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2320 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2321 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2323 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2324 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2325 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2329 let Predicates = [HasAVX] in {
2330 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2331 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2332 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2333 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2334 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2335 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2336 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2337 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2338 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2339 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2340 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2341 // fall back to this for SSE1)
2342 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2343 (VSHUFPSrri VR128:$src2, VR128:$src1,
2344 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2345 // Special unary SHUFPSrri case.
2346 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2347 (VSHUFPSrri VR128:$src1, VR128:$src1,
2348 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2349 // Special binary v4i32 shuffle cases with SHUFPS.
2350 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2351 (VSHUFPSrri VR128:$src1, VR128:$src2,
2352 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2353 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2354 (bc_v4i32 (memopv2i64 addr:$src2)))),
2355 (VSHUFPSrmi VR128:$src1, addr:$src2,
2356 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2357 // Special unary SHUFPDrri cases.
2358 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2359 (VSHUFPDrri VR128:$src1, VR128:$src1,
2360 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2361 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2362 (VSHUFPDrri VR128:$src1, VR128:$src1,
2363 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2364 // Special binary v2i64 shuffle cases using SHUFPDrri.
2365 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2366 (VSHUFPDrri VR128:$src1, VR128:$src2,
2367 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2369 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2370 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2371 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2372 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2373 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2374 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2375 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2376 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2377 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2378 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2381 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2382 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2383 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2384 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2385 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2387 def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2388 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2389 def : Pat<(v8f32 (X86Shufp VR256:$src1,
2390 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2391 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2393 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2394 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2395 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2396 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2397 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2399 def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2400 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2401 def : Pat<(v4f64 (X86Shufp VR256:$src1,
2402 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2403 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2406 let Predicates = [HasSSE1] in {
2407 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2408 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2409 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2410 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2411 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2412 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2413 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2414 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2415 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2416 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2417 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2418 // fall back to this for SSE1)
2419 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2420 (SHUFPSrri VR128:$src2, VR128:$src1,
2421 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2422 // Special unary SHUFPSrri case.
2423 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2424 (SHUFPSrri VR128:$src1, VR128:$src1,
2425 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2428 let Predicates = [HasSSE2] in {
2429 // Special binary v4i32 shuffle cases with SHUFPS.
2430 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2431 (SHUFPSrri VR128:$src1, VR128:$src2,
2432 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2433 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2434 (bc_v4i32 (memopv2i64 addr:$src2)))),
2435 (SHUFPSrmi VR128:$src1, addr:$src2,
2436 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2437 // Special unary SHUFPDrri cases.
2438 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2439 (SHUFPDrri VR128:$src1, VR128:$src1,
2440 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2441 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2442 (SHUFPDrri VR128:$src1, VR128:$src1,
2443 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2444 // Special binary v2i64 shuffle cases using SHUFPDrri.
2445 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2446 (SHUFPDrri VR128:$src1, VR128:$src2,
2447 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2448 // Generic SHUFPD patterns
2449 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2450 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2451 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2452 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2453 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2454 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2455 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2456 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2457 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2458 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2461 //===----------------------------------------------------------------------===//
2462 // SSE 1 & 2 - Unpack Instructions
2463 //===----------------------------------------------------------------------===//
2465 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2466 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2467 PatFrag mem_frag, RegisterClass RC,
2468 X86MemOperand x86memop, string asm,
2470 def rr : PI<opc, MRMSrcReg,
2471 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2473 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2474 def rm : PI<opc, MRMSrcMem,
2475 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2477 (vt (OpNode RC:$src1,
2478 (mem_frag addr:$src2))))], d>;
2481 let AddedComplexity = 10 in {
2482 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2483 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2484 SSEPackedSingle>, TB, VEX_4V;
2485 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2486 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2487 SSEPackedDouble>, TB, OpSize, VEX_4V;
2488 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2489 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2490 SSEPackedSingle>, TB, VEX_4V;
2491 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2492 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2493 SSEPackedDouble>, TB, OpSize, VEX_4V;
2495 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2496 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2497 SSEPackedSingle>, TB, VEX_4V;
2498 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2499 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2500 SSEPackedDouble>, TB, OpSize, VEX_4V;
2501 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2502 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2503 SSEPackedSingle>, TB, VEX_4V;
2504 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2505 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2506 SSEPackedDouble>, TB, OpSize, VEX_4V;
2508 let Constraints = "$src1 = $dst" in {
2509 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2510 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2511 SSEPackedSingle>, TB;
2512 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2513 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2514 SSEPackedDouble>, TB, OpSize;
2515 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2516 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2517 SSEPackedSingle>, TB;
2518 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2519 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2520 SSEPackedDouble>, TB, OpSize;
2521 } // Constraints = "$src1 = $dst"
2522 } // AddedComplexity
2524 let Predicates = [HasAVX], AddedComplexity = 1 in {
2525 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2526 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2527 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2528 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2529 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2530 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2531 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2532 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2534 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2535 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2536 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2537 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2538 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2539 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2540 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2541 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2543 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2544 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2545 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2546 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2547 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2548 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2549 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2550 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2552 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2553 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2554 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2555 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2556 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2557 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2558 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2559 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2561 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2562 // problem is during lowering, where it's not possible to recognize the load
2563 // fold cause it has two uses through a bitcast. One use disappears at isel
2564 // time and the fold opportunity reappears.
2565 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2566 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2567 let AddedComplexity = 10 in
2568 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2569 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2572 let Predicates = [HasSSE1] in {
2573 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2574 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2575 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2576 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2577 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2578 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2579 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2580 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2583 let Predicates = [HasSSE2] in {
2584 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2585 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2586 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2587 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2588 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2589 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2590 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2591 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2593 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2594 // problem is during lowering, where it's not possible to recognize the load
2595 // fold cause it has two uses through a bitcast. One use disappears at isel
2596 // time and the fold opportunity reappears.
2597 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2598 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2600 let AddedComplexity = 10 in
2601 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2602 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2605 //===----------------------------------------------------------------------===//
2606 // SSE 1 & 2 - Extract Floating-Point Sign mask
2607 //===----------------------------------------------------------------------===//
2609 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2610 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2612 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2613 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2614 [(set GR32:$dst, (Int RC:$src))], d>;
2615 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2616 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2619 let Predicates = [HasAVX] in {
2620 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2621 "movmskps", SSEPackedSingle>, TB, VEX;
2622 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2623 "movmskpd", SSEPackedDouble>, TB,
2625 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2626 "movmskps", SSEPackedSingle>, TB, VEX;
2627 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2628 "movmskpd", SSEPackedDouble>, TB,
2631 def : Pat<(i32 (X86fgetsign FR32:$src)),
2632 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2634 def : Pat<(i64 (X86fgetsign FR32:$src)),
2635 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2637 def : Pat<(i32 (X86fgetsign FR64:$src)),
2638 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2640 def : Pat<(i64 (X86fgetsign FR64:$src)),
2641 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2645 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2646 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2647 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2648 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2650 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2651 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2652 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2653 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2657 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2658 SSEPackedSingle>, TB;
2659 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2660 SSEPackedDouble>, TB, OpSize;
2662 def : Pat<(i32 (X86fgetsign FR32:$src)),
2663 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2664 sub_ss))>, Requires<[HasSSE1]>;
2665 def : Pat<(i64 (X86fgetsign FR32:$src)),
2666 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2667 sub_ss))>, Requires<[HasSSE1]>;
2668 def : Pat<(i32 (X86fgetsign FR64:$src)),
2669 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2670 sub_sd))>, Requires<[HasSSE2]>;
2671 def : Pat<(i64 (X86fgetsign FR64:$src)),
2672 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2673 sub_sd))>, Requires<[HasSSE2]>;
2675 //===---------------------------------------------------------------------===//
2676 // SSE2 - Packed Integer Logical Instructions
2677 //===---------------------------------------------------------------------===//
2679 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2681 /// PDI_binop_rm - Simple SSE2 binary operator.
2682 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2683 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2684 X86MemOperand x86memop, bit IsCommutable = 0,
2686 let isCommutable = IsCommutable in
2687 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2688 (ins RC:$src1, RC:$src2),
2690 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2691 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2692 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2693 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2694 (ins RC:$src1, x86memop:$src2),
2696 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2697 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2698 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2699 (bitconvert (memop_frag addr:$src2)))))]>;
2701 } // ExeDomain = SSEPackedInt
2703 // These are ordered here for pattern ordering requirements with the fp versions
2705 let Predicates = [HasAVX] in {
2706 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2707 i128mem, 1, 0>, VEX_4V;
2708 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2709 i128mem, 1, 0>, VEX_4V;
2710 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2711 i128mem, 1, 0>, VEX_4V;
2712 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2713 i128mem, 0, 0>, VEX_4V;
2716 let Constraints = "$src1 = $dst" in {
2717 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2719 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2721 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2723 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2725 } // Constraints = "$src1 = $dst"
2727 let Predicates = [HasAVX2] in {
2728 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2729 i256mem, 1, 0>, VEX_4V;
2730 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2731 i256mem, 1, 0>, VEX_4V;
2732 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2733 i256mem, 1, 0>, VEX_4V;
2734 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2735 i256mem, 0, 0>, VEX_4V;
2738 //===----------------------------------------------------------------------===//
2739 // SSE 1 & 2 - Logical Instructions
2740 //===----------------------------------------------------------------------===//
2742 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2744 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2746 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2747 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2749 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2750 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2752 let Constraints = "$src1 = $dst" in {
2753 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2754 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2756 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2757 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2761 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2762 let mayLoad = 0 in {
2763 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2764 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2765 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2768 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2769 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2771 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2773 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2775 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2776 // are all promoted to v2i64, and the patterns are covered by the int
2777 // version. This is needed in SSE only, because v2i64 isn't supported on
2778 // SSE1, but only on SSE2.
2779 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2780 !strconcat(OpcodeStr, "ps"), f128mem, [],
2781 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2782 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2784 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2785 !strconcat(OpcodeStr, "pd"), f128mem,
2786 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2787 (bc_v2i64 (v2f64 VR128:$src2))))],
2788 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2789 (memopv2i64 addr:$src2)))], 0>,
2791 let Constraints = "$src1 = $dst" in {
2792 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2793 !strconcat(OpcodeStr, "ps"), f128mem,
2794 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2795 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2796 (memopv2i64 addr:$src2)))]>, TB;
2798 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2799 !strconcat(OpcodeStr, "pd"), f128mem,
2800 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2801 (bc_v2i64 (v2f64 VR128:$src2))))],
2802 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2803 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2807 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2809 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2811 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2812 !strconcat(OpcodeStr, "ps"), f256mem,
2813 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2814 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2815 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2817 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2818 !strconcat(OpcodeStr, "pd"), f256mem,
2819 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2820 (bc_v4i64 (v4f64 VR256:$src2))))],
2821 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2822 (memopv4i64 addr:$src2)))], 0>,
2826 // AVX 256-bit packed logical ops forms
2827 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2828 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2829 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2830 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2832 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2833 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2834 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2835 let isCommutable = 0 in
2836 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2838 //===----------------------------------------------------------------------===//
2839 // SSE 1 & 2 - Arithmetic Instructions
2840 //===----------------------------------------------------------------------===//
2842 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2845 /// In addition, we also have a special variant of the scalar form here to
2846 /// represent the associated intrinsic operation. This form is unlike the
2847 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2848 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2850 /// These three forms can each be reg+reg or reg+mem.
2853 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2855 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2857 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2858 OpNode, FR32, f32mem, Is2Addr>, XS;
2859 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2860 OpNode, FR64, f64mem, Is2Addr>, XD;
2863 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2865 let mayLoad = 0 in {
2866 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2867 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2868 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2869 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2873 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2875 let mayLoad = 0 in {
2876 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2877 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2878 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2879 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2883 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2885 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2886 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2887 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2888 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2891 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2893 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2894 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2895 SSEPackedSingle, Is2Addr>, TB;
2897 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2898 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2899 SSEPackedDouble, Is2Addr>, TB, OpSize;
2902 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2903 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2904 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2905 SSEPackedSingle, 0>, TB;
2907 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2908 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2909 SSEPackedDouble, 0>, TB, OpSize;
2912 // Binary Arithmetic instructions
2913 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2914 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2915 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2916 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2917 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2918 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2919 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2920 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2922 let isCommutable = 0 in {
2923 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2924 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2925 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2926 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2927 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2928 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2929 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2930 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2931 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2932 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2933 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2934 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2935 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2936 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2937 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2938 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2939 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2940 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2941 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2942 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2945 let Constraints = "$src1 = $dst" in {
2946 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2947 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2948 basic_sse12_fp_binop_s_int<0x58, "add">;
2949 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2950 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2951 basic_sse12_fp_binop_s_int<0x59, "mul">;
2953 let isCommutable = 0 in {
2954 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2955 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2956 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2957 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2958 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2959 basic_sse12_fp_binop_s_int<0x5E, "div">;
2960 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2961 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2962 basic_sse12_fp_binop_s_int<0x5F, "max">,
2963 basic_sse12_fp_binop_p_int<0x5F, "max">;
2964 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2965 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2966 basic_sse12_fp_binop_s_int<0x5D, "min">,
2967 basic_sse12_fp_binop_p_int<0x5D, "min">;
2972 /// In addition, we also have a special variant of the scalar form here to
2973 /// represent the associated intrinsic operation. This form is unlike the
2974 /// plain scalar form, in that it takes an entire vector (instead of a
2975 /// scalar) and leaves the top elements undefined.
2977 /// And, we have a special variant form for a full-vector intrinsic form.
2979 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2980 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2981 SDNode OpNode, Intrinsic F32Int> {
2982 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2983 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2984 [(set FR32:$dst, (OpNode FR32:$src))]>;
2985 // For scalar unary operations, fold a load into the operation
2986 // only in OptForSize mode. It eliminates an instruction, but it also
2987 // eliminates a whole-register clobber (the load), so it introduces a
2988 // partial register update condition.
2989 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2990 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2991 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2992 Requires<[HasSSE1, OptForSize]>;
2993 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2994 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2995 [(set VR128:$dst, (F32Int VR128:$src))]>;
2996 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2997 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2998 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
3001 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
3002 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3003 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
3004 !strconcat(OpcodeStr,
3005 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3007 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
3008 !strconcat(OpcodeStr,
3009 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3010 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3011 (ins VR128:$src1, ssmem:$src2),
3012 !strconcat(OpcodeStr,
3013 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3016 /// sse1_fp_unop_p - SSE1 unops in packed form.
3017 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3018 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3019 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3020 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
3021 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3022 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3023 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
3026 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3027 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3028 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3029 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3030 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
3031 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3032 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3033 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
3036 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3037 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3038 Intrinsic V4F32Int> {
3039 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3040 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3041 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
3042 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3043 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3044 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
3047 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3048 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3049 Intrinsic V4F32Int> {
3050 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3051 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3052 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
3053 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3054 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3055 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
3058 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3059 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3060 SDNode OpNode, Intrinsic F64Int> {
3061 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3062 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3063 [(set FR64:$dst, (OpNode FR64:$src))]>;
3064 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3065 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3066 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3067 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
3068 Requires<[HasSSE2, OptForSize]>;
3069 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3070 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3071 [(set VR128:$dst, (F64Int VR128:$src))]>;
3072 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3073 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3074 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
3077 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3078 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3079 let neverHasSideEffects = 1 in {
3080 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3081 !strconcat(OpcodeStr,
3082 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3084 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3085 !strconcat(OpcodeStr,
3086 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3088 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3089 (ins VR128:$src1, sdmem:$src2),
3090 !strconcat(OpcodeStr,
3091 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3094 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3095 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3097 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3098 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3099 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3100 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3101 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3102 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3105 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3106 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3107 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3108 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3109 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3110 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3111 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3112 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3115 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3116 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3117 Intrinsic V2F64Int> {
3118 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3119 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3120 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3121 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3122 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3123 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3126 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3127 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3128 Intrinsic V2F64Int> {
3129 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3130 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3131 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3132 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3133 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3134 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3137 let Predicates = [HasAVX] in {
3139 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3140 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3142 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3143 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3144 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3145 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3146 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3147 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3148 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3149 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3152 // Reciprocal approximations. Note that these typically require refinement
3153 // in order to obtain suitable precision.
3154 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3155 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3156 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3157 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3158 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3160 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3161 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3162 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3163 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3164 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3167 let AddedComplexity = 1 in {
3168 def : Pat<(f32 (fsqrt FR32:$src)),
3169 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3170 def : Pat<(f32 (fsqrt (load addr:$src))),
3171 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3172 Requires<[HasAVX, OptForSize]>;
3173 def : Pat<(f64 (fsqrt FR64:$src)),
3174 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3175 def : Pat<(f64 (fsqrt (load addr:$src))),
3176 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3177 Requires<[HasAVX, OptForSize]>;
3179 def : Pat<(f32 (X86frsqrt FR32:$src)),
3180 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3181 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3182 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3183 Requires<[HasAVX, OptForSize]>;
3185 def : Pat<(f32 (X86frcp FR32:$src)),
3186 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3187 def : Pat<(f32 (X86frcp (load addr:$src))),
3188 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3189 Requires<[HasAVX, OptForSize]>;
3192 let Predicates = [HasAVX], AddedComplexity = 1 in {
3193 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3194 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3195 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3196 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3198 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3199 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3201 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3202 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3203 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3204 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3206 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3207 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3209 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3210 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3211 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3212 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3214 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3215 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3217 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3218 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3219 (VRCPSSr (f32 (IMPLICIT_DEF)),
3220 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3222 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3223 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3227 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3228 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3229 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3230 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3231 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3232 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3234 // Reciprocal approximations. Note that these typically require refinement
3235 // in order to obtain suitable precision.
3236 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3237 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3238 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3239 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3240 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3241 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3243 // There is no f64 version of the reciprocal approximation instructions.
3245 //===----------------------------------------------------------------------===//
3246 // SSE 1 & 2 - Non-temporal stores
3247 //===----------------------------------------------------------------------===//
3249 let AddedComplexity = 400 in { // Prefer non-temporal versions
3250 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3251 (ins f128mem:$dst, VR128:$src),
3252 "movntps\t{$src, $dst|$dst, $src}",
3253 [(alignednontemporalstore (v4f32 VR128:$src),
3255 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3256 (ins f128mem:$dst, VR128:$src),
3257 "movntpd\t{$src, $dst|$dst, $src}",
3258 [(alignednontemporalstore (v2f64 VR128:$src),
3261 let ExeDomain = SSEPackedInt in
3262 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3263 (ins f128mem:$dst, VR128:$src),
3264 "movntdq\t{$src, $dst|$dst, $src}",
3265 [(alignednontemporalstore (v2i64 VR128:$src),
3268 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3269 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3271 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3272 (ins f256mem:$dst, VR256:$src),
3273 "movntps\t{$src, $dst|$dst, $src}",
3274 [(alignednontemporalstore (v8f32 VR256:$src),
3276 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3277 (ins f256mem:$dst, VR256:$src),
3278 "movntpd\t{$src, $dst|$dst, $src}",
3279 [(alignednontemporalstore (v4f64 VR256:$src),
3281 let ExeDomain = SSEPackedInt in
3282 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3283 (ins f256mem:$dst, VR256:$src),
3284 "movntdq\t{$src, $dst|$dst, $src}",
3285 [(alignednontemporalstore (v4i64 VR256:$src),
3289 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3290 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3291 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3292 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3293 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3294 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3296 let AddedComplexity = 400 in { // Prefer non-temporal versions
3297 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3298 "movntps\t{$src, $dst|$dst, $src}",
3299 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3300 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3301 "movntpd\t{$src, $dst|$dst, $src}",
3302 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3304 let ExeDomain = SSEPackedInt in
3305 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3306 "movntdq\t{$src, $dst|$dst, $src}",
3307 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)]>;
3309 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3310 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3312 // There is no AVX form for instructions below this point
3313 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3314 "movnti{l}\t{$src, $dst|$dst, $src}",
3315 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3316 TB, Requires<[HasSSE2]>;
3317 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3318 "movnti{q}\t{$src, $dst|$dst, $src}",
3319 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3320 TB, Requires<[HasSSE2]>;
3323 //===----------------------------------------------------------------------===//
3324 // SSE 1 & 2 - Prefetch and memory fence
3325 //===----------------------------------------------------------------------===//
3327 // Prefetch intrinsic.
3328 let Predicates = [HasSSE1] in {
3329 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3330 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3331 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3332 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3333 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3334 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3335 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3336 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3340 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3341 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3342 TB, Requires<[HasSSE2]>;
3344 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3345 // was introduced with SSE2, it's backward compatible.
3346 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3348 // Load, store, and memory fence
3349 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3350 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3351 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3352 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3353 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3354 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3356 def : Pat<(X86SFence), (SFENCE)>;
3357 def : Pat<(X86LFence), (LFENCE)>;
3358 def : Pat<(X86MFence), (MFENCE)>;
3360 //===----------------------------------------------------------------------===//
3361 // SSE 1 & 2 - Load/Store XCSR register
3362 //===----------------------------------------------------------------------===//
3364 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3365 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3366 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3367 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3369 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3370 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3371 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3372 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3374 //===---------------------------------------------------------------------===//
3375 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3376 //===---------------------------------------------------------------------===//
3378 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3380 let neverHasSideEffects = 1 in {
3381 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3382 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3383 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3384 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3386 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3387 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3388 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3389 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3392 let isCodeGenOnly = 1 in {
3393 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3394 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3395 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3396 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3397 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3398 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3399 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3400 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3403 let canFoldAsLoad = 1, mayLoad = 1 in {
3404 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3405 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3406 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3407 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3408 let Predicates = [HasAVX] in {
3409 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3410 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3411 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3412 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3416 let mayStore = 1 in {
3417 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3418 (ins i128mem:$dst, VR128:$src),
3419 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3420 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3421 (ins i256mem:$dst, VR256:$src),
3422 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3423 let Predicates = [HasAVX] in {
3424 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3425 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3426 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3427 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3431 let neverHasSideEffects = 1 in
3432 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3433 "movdqa\t{$src, $dst|$dst, $src}", []>;
3435 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3436 "movdqu\t{$src, $dst|$dst, $src}",
3437 []>, XS, Requires<[HasSSE2]>;
3440 let isCodeGenOnly = 1 in {
3441 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3442 "movdqa\t{$src, $dst|$dst, $src}", []>;
3444 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3445 "movdqu\t{$src, $dst|$dst, $src}",
3446 []>, XS, Requires<[HasSSE2]>;
3449 let canFoldAsLoad = 1, mayLoad = 1 in {
3450 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3451 "movdqa\t{$src, $dst|$dst, $src}",
3452 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3453 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3454 "movdqu\t{$src, $dst|$dst, $src}",
3455 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3456 XS, Requires<[HasSSE2]>;
3459 let mayStore = 1 in {
3460 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3461 "movdqa\t{$src, $dst|$dst, $src}",
3462 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3463 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3464 "movdqu\t{$src, $dst|$dst, $src}",
3465 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3466 XS, Requires<[HasSSE2]>;
3469 // Intrinsic forms of MOVDQU load and store
3470 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3471 "vmovdqu\t{$src, $dst|$dst, $src}",
3472 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3473 XS, VEX, Requires<[HasAVX]>;
3475 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3476 "movdqu\t{$src, $dst|$dst, $src}",
3477 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3478 XS, Requires<[HasSSE2]>;
3480 } // ExeDomain = SSEPackedInt
3482 let Predicates = [HasAVX] in {
3483 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3484 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3485 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3488 //===---------------------------------------------------------------------===//
3489 // SSE2 - Packed Integer Arithmetic Instructions
3490 //===---------------------------------------------------------------------===//
3492 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3494 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3495 RegisterClass RC, PatFrag memop_frag,
3496 X86MemOperand x86memop, bit IsCommutable = 0,
3498 let isCommutable = IsCommutable in
3499 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3500 (ins RC:$src1, RC:$src2),
3502 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3503 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3504 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3505 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3506 (ins RC:$src1, x86memop:$src2),
3508 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3509 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3510 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3513 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3514 string OpcodeStr, Intrinsic IntId,
3515 Intrinsic IntId2, RegisterClass RC,
3517 // src2 is always 128-bit
3518 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3519 (ins RC:$src1, VR128:$src2),
3521 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3522 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3523 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3524 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3525 (ins RC:$src1, i128mem:$src2),
3527 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3528 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3529 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3530 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3531 (ins RC:$src1, i32i8imm:$src2),
3533 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3534 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3535 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3538 } // ExeDomain = SSEPackedInt
3540 // 128-bit Integer Arithmetic
3542 let Predicates = [HasAVX] in {
3543 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3544 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3545 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3546 i128mem, 1, 0>, VEX_4V;
3547 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3548 i128mem, 1, 0>, VEX_4V;
3549 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3550 i128mem, 1, 0>, VEX_4V;
3551 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3552 i128mem, 1, 0>, VEX_4V;
3553 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3554 i128mem, 0, 0>, VEX_4V;
3555 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3556 i128mem, 0, 0>, VEX_4V;
3557 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3558 i128mem, 0, 0>, VEX_4V;
3559 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3560 i128mem, 0, 0>, VEX_4V;
3563 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3564 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3565 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3566 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3567 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3568 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3569 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3570 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3571 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3572 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3573 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3574 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3575 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3576 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3577 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3578 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3579 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3580 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3581 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3582 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3583 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3584 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3585 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3586 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3587 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3588 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3589 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3590 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3591 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3592 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3593 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3594 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3595 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3596 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3597 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3598 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3599 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3600 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3603 let Predicates = [HasAVX2] in {
3604 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3605 i256mem, 1, 0>, VEX_4V;
3606 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3607 i256mem, 1, 0>, VEX_4V;
3608 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3609 i256mem, 1, 0>, VEX_4V;
3610 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3611 i256mem, 1, 0>, VEX_4V;
3612 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3613 i256mem, 1, 0>, VEX_4V;
3614 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3615 i256mem, 0, 0>, VEX_4V;
3616 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3617 i256mem, 0, 0>, VEX_4V;
3618 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3619 i256mem, 0, 0>, VEX_4V;
3620 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3621 i256mem, 0, 0>, VEX_4V;
3624 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3625 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3626 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3627 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3628 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3629 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3630 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3631 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3632 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3633 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3634 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3635 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3636 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3637 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3638 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3639 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3640 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3641 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3642 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3643 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3644 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3645 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3646 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3647 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3648 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3649 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3650 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3651 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3652 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3653 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3654 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3655 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3656 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3657 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3658 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3659 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3660 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3661 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3664 let Constraints = "$src1 = $dst" in {
3665 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3667 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3669 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3671 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3673 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3675 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3677 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3679 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3681 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3685 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3686 VR128, memopv2i64, i128mem>;
3687 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3688 VR128, memopv2i64, i128mem>;
3689 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3690 VR128, memopv2i64, i128mem>;
3691 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3692 VR128, memopv2i64, i128mem>;
3693 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3694 VR128, memopv2i64, i128mem, 1>;
3695 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3696 VR128, memopv2i64, i128mem, 1>;
3697 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3698 VR128, memopv2i64, i128mem, 1>;
3699 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3700 VR128, memopv2i64, i128mem, 1>;
3701 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3702 VR128, memopv2i64, i128mem, 1>;
3703 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3704 VR128, memopv2i64, i128mem, 1>;
3705 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3706 VR128, memopv2i64, i128mem, 1>;
3707 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3708 VR128, memopv2i64, i128mem, 1>;
3709 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3710 VR128, memopv2i64, i128mem, 1>;
3711 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3712 VR128, memopv2i64, i128mem, 1>;
3713 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3714 VR128, memopv2i64, i128mem, 1>;
3715 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3716 VR128, memopv2i64, i128mem, 1>;
3717 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3718 VR128, memopv2i64, i128mem, 1>;
3719 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3720 VR128, memopv2i64, i128mem, 1>;
3721 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3722 VR128, memopv2i64, i128mem, 1>;
3724 } // Constraints = "$src1 = $dst"
3726 //===---------------------------------------------------------------------===//
3727 // SSE2 - Packed Integer Logical Instructions
3728 //===---------------------------------------------------------------------===//
3730 let Predicates = [HasAVX] in {
3731 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3732 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3734 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3735 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3737 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3738 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3741 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3742 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3744 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3745 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3747 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3748 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3751 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3752 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3754 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3755 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3758 let ExeDomain = SSEPackedInt in {
3759 // 128-bit logical shifts.
3760 def VPSLLDQri : PDIi8<0x73, MRM7r,
3761 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3762 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3764 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3766 def VPSRLDQri : PDIi8<0x73, MRM3r,
3767 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3768 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3770 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3772 // PSRADQri doesn't exist in SSE[1-3].
3774 } // Predicates = [HasAVX]
3776 let Predicates = [HasAVX2] in {
3777 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3778 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3780 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3781 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3783 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3784 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3787 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3788 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3790 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3791 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3793 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3794 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3797 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3798 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3800 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3801 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3804 let ExeDomain = SSEPackedInt in {
3805 // 256-bit logical shifts.
3806 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3807 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3808 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3810 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3812 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3813 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3814 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3816 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3818 // PSRADQYri doesn't exist in SSE[1-3].
3820 } // Predicates = [HasAVX2]
3822 let Constraints = "$src1 = $dst" in {
3823 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3824 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3826 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3827 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3829 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3830 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3833 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3834 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3836 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3837 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3839 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3840 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3843 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3844 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3846 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3847 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3850 let ExeDomain = SSEPackedInt in {
3851 // 128-bit logical shifts.
3852 def PSLLDQri : PDIi8<0x73, MRM7r,
3853 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3854 "pslldq\t{$src2, $dst|$dst, $src2}",
3856 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3857 def PSRLDQri : PDIi8<0x73, MRM3r,
3858 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3859 "psrldq\t{$src2, $dst|$dst, $src2}",
3861 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3862 // PSRADQri doesn't exist in SSE[1-3].
3864 } // Constraints = "$src1 = $dst"
3866 let Predicates = [HasAVX] in {
3867 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3868 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3869 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3870 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3871 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3872 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3874 // Shift up / down and insert zero's.
3875 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3876 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3877 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3878 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3880 def : Pat<(v8i16 (X86vshli VR128:$src1, (i32 imm:$src2))),
3881 (VPSLLWri VR128:$src1, imm:$src2)>;
3882 def : Pat<(v4i32 (X86vshli VR128:$src1, (i32 imm:$src2))),
3883 (VPSLLDri VR128:$src1, imm:$src2)>;
3884 def : Pat<(v2i64 (X86vshli VR128:$src1, (i32 imm:$src2))),
3885 (VPSLLQri VR128:$src1, imm:$src2)>;
3887 def : Pat<(v8i16 (X86vsrli VR128:$src1, (i32 imm:$src2))),
3888 (VPSRLWri VR128:$src1, imm:$src2)>;
3889 def : Pat<(v4i32 (X86vsrli VR128:$src1, (i32 imm:$src2))),
3890 (VPSRLDri VR128:$src1, imm:$src2)>;
3891 def : Pat<(v2i64 (X86vsrli VR128:$src1, (i32 imm:$src2))),
3892 (VPSRLQri VR128:$src1, imm:$src2)>;
3894 def : Pat<(v8i16 (X86vsrai VR128:$src1, (i32 imm:$src2))),
3895 (VPSRAWri VR128:$src1, imm:$src2)>;
3896 def : Pat<(v4i32 (X86vsrai VR128:$src1, (i32 imm:$src2))),
3897 (VPSRADri VR128:$src1, imm:$src2)>;
3899 def : Pat<(v8i16 (X86vshl VR128:$src1, (v8i16 VR128:$src2))),
3900 (VPSLLWrr VR128:$src1, VR128:$src2)>;
3901 def : Pat<(v8i16 (X86vshl VR128:$src1, (bc_v8i16 (memopv2i64 addr:$src2)))),
3902 (VPSLLWrm VR128:$src1, addr:$src2)>;
3903 def : Pat<(v4i32 (X86vshl VR128:$src1, (v4i32 VR128:$src2))),
3904 (VPSLLDrr VR128:$src1, VR128:$src2)>;
3905 def : Pat<(v4i32 (X86vshl VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3906 (VPSLLDrm VR128:$src1, addr:$src2)>;
3907 def : Pat<(v2i64 (X86vshl VR128:$src1, (v2i64 VR128:$src2))),
3908 (VPSLLQrr VR128:$src1, VR128:$src2)>;
3909 def : Pat<(v2i64 (X86vshl VR128:$src1, (memopv2i64 addr:$src2))),
3910 (VPSLLQrm VR128:$src1, addr:$src2)>;
3912 def : Pat<(v8i16 (X86vsrl VR128:$src1, (v8i16 VR128:$src2))),
3913 (VPSRLWrr VR128:$src1, VR128:$src2)>;
3914 def : Pat<(v8i16 (X86vsrl VR128:$src1, (bc_v8i16 (memopv2i64 addr:$src2)))),
3915 (VPSRLWrm VR128:$src1, addr:$src2)>;
3916 def : Pat<(v4i32 (X86vsrl VR128:$src1, (v4i32 VR128:$src2))),
3917 (VPSRLDrr VR128:$src1, VR128:$src2)>;
3918 def : Pat<(v4i32 (X86vsrl VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3919 (VPSRLDrm VR128:$src1, addr:$src2)>;
3920 def : Pat<(v2i64 (X86vsrl VR128:$src1, (v2i64 VR128:$src2))),
3921 (VPSRLQrr VR128:$src1, VR128:$src2)>;
3922 def : Pat<(v2i64 (X86vsrl VR128:$src1, (memopv2i64 addr:$src2))),
3923 (VPSRLQrm VR128:$src1, addr:$src2)>;
3925 def : Pat<(v8i16 (X86vsra VR128:$src1, (v8i16 VR128:$src2))),
3926 (VPSRAWrr VR128:$src1, VR128:$src2)>;
3927 def : Pat<(v8i16 (X86vsra VR128:$src1, (bc_v8i16 (memopv2i64 addr:$src2)))),
3928 (VPSRAWrm VR128:$src1, addr:$src2)>;
3929 def : Pat<(v4i32 (X86vsra VR128:$src1, (v4i32 VR128:$src2))),
3930 (VPSRADrr VR128:$src1, VR128:$src2)>;
3931 def : Pat<(v4i32 (X86vsra VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3932 (VPSRADrm VR128:$src1, addr:$src2)>;
3935 let Predicates = [HasAVX2] in {
3936 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3937 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3938 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3939 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3941 def : Pat<(v16i16 (X86vshli VR256:$src1, (i32 imm:$src2))),
3942 (VPSLLWYri VR256:$src1, imm:$src2)>;
3943 def : Pat<(v8i32 (X86vshli VR256:$src1, (i32 imm:$src2))),
3944 (VPSLLDYri VR256:$src1, imm:$src2)>;
3945 def : Pat<(v4i64 (X86vshli VR256:$src1, (i32 imm:$src2))),
3946 (VPSLLQYri VR256:$src1, imm:$src2)>;
3948 def : Pat<(v16i16 (X86vsrli VR256:$src1, (i32 imm:$src2))),
3949 (VPSRLWYri VR256:$src1, imm:$src2)>;
3950 def : Pat<(v8i32 (X86vsrli VR256:$src1, (i32 imm:$src2))),
3951 (VPSRLDYri VR256:$src1, imm:$src2)>;
3952 def : Pat<(v4i64 (X86vsrli VR256:$src1, (i32 imm:$src2))),
3953 (VPSRLQYri VR256:$src1, imm:$src2)>;
3955 def : Pat<(v16i16 (X86vsrai VR256:$src1, (i32 imm:$src2))),
3956 (VPSRAWYri VR256:$src1, imm:$src2)>;
3957 def : Pat<(v8i32 (X86vsrai VR256:$src1, (i32 imm:$src2))),
3958 (VPSRADYri VR256:$src1, imm:$src2)>;
3960 def : Pat<(v16i16 (X86vshl VR256:$src1, (v8i16 VR128:$src2))),
3961 (VPSLLWYrr VR256:$src1, VR128:$src2)>;
3962 def : Pat<(v16i16 (X86vshl VR256:$src1, (bc_v8i16 (memopv2i64 addr:$src2)))),
3963 (VPSLLWYrm VR256:$src1, addr:$src2)>;
3964 def : Pat<(v8i32 (X86vshl VR256:$src1, (v4i32 VR128:$src2))),
3965 (VPSLLDYrr VR256:$src1, VR128:$src2)>;
3966 def : Pat<(v8i32 (X86vshl VR256:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3967 (VPSLLDYrm VR256:$src1, addr:$src2)>;
3968 def : Pat<(v4i64 (X86vshl VR256:$src1, (v2i64 VR128:$src2))),
3969 (VPSLLQYrr VR256:$src1, VR128:$src2)>;
3970 def : Pat<(v4i64 (X86vshl VR256:$src1, (memopv2i64 addr:$src2))),
3971 (VPSLLQYrm VR256:$src1, addr:$src2)>;
3973 def : Pat<(v16i16 (X86vsrl VR256:$src1, (v8i16 VR128:$src2))),
3974 (VPSRLWYrr VR256:$src1, VR128:$src2)>;
3975 def : Pat<(v16i16 (X86vsrl VR256:$src1, (bc_v8i16 (memopv2i64 addr:$src2)))),
3976 (VPSRLWYrm VR256:$src1, addr:$src2)>;
3977 def : Pat<(v8i32 (X86vsrl VR256:$src1, (v4i32 VR128:$src2))),
3978 (VPSRLDYrr VR256:$src1, VR128:$src2)>;
3979 def : Pat<(v8i32 (X86vsrl VR256:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3980 (VPSRLDYrm VR256:$src1, addr:$src2)>;
3981 def : Pat<(v4i64 (X86vsrl VR256:$src1, (v2i64 VR128:$src2))),
3982 (VPSRLQYrr VR256:$src1, VR128:$src2)>;
3983 def : Pat<(v4i64 (X86vsrl VR256:$src1, (memopv2i64 addr:$src2))),
3984 (VPSRLQYrm VR256:$src1, addr:$src2)>;
3986 def : Pat<(v16i16 (X86vsra VR256:$src1, (v8i16 VR128:$src2))),
3987 (VPSRAWYrr VR256:$src1, VR128:$src2)>;
3988 def : Pat<(v16i16 (X86vsra VR256:$src1, (bc_v8i16 (memopv2i64 addr:$src2)))),
3989 (VPSRAWYrm VR256:$src1, addr:$src2)>;
3990 def : Pat<(v8i32 (X86vsra VR256:$src1, (v4i32 VR128:$src2))),
3991 (VPSRADYrr VR256:$src1, VR128:$src2)>;
3992 def : Pat<(v8i32 (X86vsra VR256:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3993 (VPSRADYrm VR256:$src1, addr:$src2)>;
3996 let Predicates = [HasSSE2] in {
3997 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3998 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3999 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4000 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4001 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4002 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4004 // Shift up / down and insert zero's.
4005 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4006 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4007 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4008 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4010 def : Pat<(v8i16 (X86vshli VR128:$src1, (i32 imm:$src2))),
4011 (PSLLWri VR128:$src1, imm:$src2)>;
4012 def : Pat<(v4i32 (X86vshli VR128:$src1, (i32 imm:$src2))),
4013 (PSLLDri VR128:$src1, imm:$src2)>;
4014 def : Pat<(v2i64 (X86vshli VR128:$src1, (i32 imm:$src2))),
4015 (PSLLQri VR128:$src1, imm:$src2)>;
4017 def : Pat<(v8i16 (X86vsrli VR128:$src1, (i32 imm:$src2))),
4018 (PSRLWri VR128:$src1, imm:$src2)>;
4019 def : Pat<(v4i32 (X86vsrli VR128:$src1, (i32 imm:$src2))),
4020 (PSRLDri VR128:$src1, imm:$src2)>;
4021 def : Pat<(v2i64 (X86vsrli VR128:$src1, (i32 imm:$src2))),
4022 (PSRLQri VR128:$src1, imm:$src2)>;
4024 def : Pat<(v8i16 (X86vsrai VR128:$src1, (i32 imm:$src2))),
4025 (PSRAWri VR128:$src1, imm:$src2)>;
4026 def : Pat<(v4i32 (X86vsrai VR128:$src1, (i32 imm:$src2))),
4027 (PSRADri VR128:$src1, imm:$src2)>;
4029 def : Pat<(v8i16 (X86vshl VR128:$src1, (v8i16 VR128:$src2))),
4030 (PSLLWrr VR128:$src1, VR128:$src2)>;
4031 def : Pat<(v8i16 (X86vshl VR128:$src1, (bc_v8i16 (memopv2i64 addr:$src2)))),
4032 (PSLLWrm VR128:$src1, addr:$src2)>;
4033 def : Pat<(v4i32 (X86vshl VR128:$src1, (v4i32 VR128:$src2))),
4034 (PSLLDrr VR128:$src1, VR128:$src2)>;
4035 def : Pat<(v4i32 (X86vshl VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
4036 (PSLLDrm VR128:$src1, addr:$src2)>;
4037 def : Pat<(v2i64 (X86vshl VR128:$src1, (v2i64 VR128:$src2))),
4038 (PSLLQrr VR128:$src1, VR128:$src2)>;
4039 def : Pat<(v2i64 (X86vshl VR128:$src1, (memopv2i64 addr:$src2))),
4040 (PSLLQrm VR128:$src1, addr:$src2)>;
4042 def : Pat<(v8i16 (X86vsrl VR128:$src1, (v8i16 VR128:$src2))),
4043 (PSRLWrr VR128:$src1, VR128:$src2)>;
4044 def : Pat<(v8i16 (X86vsrl VR128:$src1, (bc_v8i16 (memopv2i64 addr:$src2)))),
4045 (PSRLWrm VR128:$src1, addr:$src2)>;
4046 def : Pat<(v4i32 (X86vsrl VR128:$src1, (v4i32 VR128:$src2))),
4047 (PSRLDrr VR128:$src1, VR128:$src2)>;
4048 def : Pat<(v4i32 (X86vsrl VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
4049 (PSRLDrm VR128:$src1, addr:$src2)>;
4050 def : Pat<(v2i64 (X86vsrl VR128:$src1, (v2i64 VR128:$src2))),
4051 (PSRLQrr VR128:$src1, VR128:$src2)>;
4052 def : Pat<(v2i64 (X86vsrl VR128:$src1, (memopv2i64 addr:$src2))),
4053 (PSRLQrm VR128:$src1, addr:$src2)>;
4055 def : Pat<(v8i16 (X86vsra VR128:$src1, (v8i16 VR128:$src2))),
4056 (PSRAWrr VR128:$src1, VR128:$src2)>;
4057 def : Pat<(v8i16 (X86vsra VR128:$src1, (bc_v8i16 (memopv2i64 addr:$src2)))),
4058 (PSRAWrm VR128:$src1, addr:$src2)>;
4059 def : Pat<(v4i32 (X86vsra VR128:$src1, (v4i32 VR128:$src2))),
4060 (PSRADrr VR128:$src1, VR128:$src2)>;
4061 def : Pat<(v4i32 (X86vsra VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
4062 (PSRADrm VR128:$src1, addr:$src2)>;
4065 //===---------------------------------------------------------------------===//
4066 // SSE2 - Packed Integer Comparison Instructions
4067 //===---------------------------------------------------------------------===//
4069 let Predicates = [HasAVX] in {
4070 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
4071 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
4072 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
4073 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
4074 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
4075 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
4076 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
4077 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4078 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
4079 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4080 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
4081 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4083 def : Pat<(v16i8 (X86pcmpeq VR128:$src1, VR128:$src2)),
4084 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
4085 def : Pat<(v16i8 (X86pcmpeq VR128:$src1,
4086 (bc_v16i8 (memopv2i64 addr:$src2)))),
4087 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
4088 def : Pat<(v8i16 (X86pcmpeq VR128:$src1, VR128:$src2)),
4089 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
4090 def : Pat<(v8i16 (X86pcmpeq VR128:$src1,
4091 (bc_v8i16 (memopv2i64 addr:$src2)))),
4092 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
4093 def : Pat<(v4i32 (X86pcmpeq VR128:$src1, VR128:$src2)),
4094 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
4095 def : Pat<(v4i32 (X86pcmpeq VR128:$src1,
4096 (bc_v4i32 (memopv2i64 addr:$src2)))),
4097 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
4099 def : Pat<(v16i8 (X86pcmpgt VR128:$src1, VR128:$src2)),
4100 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
4101 def : Pat<(v16i8 (X86pcmpgt VR128:$src1,
4102 (bc_v16i8 (memopv2i64 addr:$src2)))),
4103 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
4104 def : Pat<(v8i16 (X86pcmpgt VR128:$src1, VR128:$src2)),
4105 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
4106 def : Pat<(v8i16 (X86pcmpgt VR128:$src1,
4107 (bc_v8i16 (memopv2i64 addr:$src2)))),
4108 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
4109 def : Pat<(v4i32 (X86pcmpgt VR128:$src1, VR128:$src2)),
4110 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
4111 def : Pat<(v4i32 (X86pcmpgt VR128:$src1,
4112 (bc_v4i32 (memopv2i64 addr:$src2)))),
4113 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
4116 let Predicates = [HasAVX2] in {
4117 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
4118 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
4119 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
4120 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
4121 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
4122 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
4123 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
4124 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4125 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
4126 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4127 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
4128 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4130 def : Pat<(v32i8 (X86pcmpeq VR256:$src1, VR256:$src2)),
4131 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
4132 def : Pat<(v32i8 (X86pcmpeq VR256:$src1,
4133 (bc_v32i8 (memopv4i64 addr:$src2)))),
4134 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
4135 def : Pat<(v16i16 (X86pcmpeq VR256:$src1, VR256:$src2)),
4136 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
4137 def : Pat<(v16i16 (X86pcmpeq VR256:$src1,
4138 (bc_v16i16 (memopv4i64 addr:$src2)))),
4139 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
4140 def : Pat<(v8i32 (X86pcmpeq VR256:$src1, VR256:$src2)),
4141 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
4142 def : Pat<(v8i32 (X86pcmpeq VR256:$src1,
4143 (bc_v8i32 (memopv4i64 addr:$src2)))),
4144 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
4146 def : Pat<(v32i8 (X86pcmpgt VR256:$src1, VR256:$src2)),
4147 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
4148 def : Pat<(v32i8 (X86pcmpgt VR256:$src1,
4149 (bc_v32i8 (memopv4i64 addr:$src2)))),
4150 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
4151 def : Pat<(v16i16 (X86pcmpgt VR256:$src1, VR256:$src2)),
4152 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
4153 def : Pat<(v16i16 (X86pcmpgt VR256:$src1,
4154 (bc_v16i16 (memopv4i64 addr:$src2)))),
4155 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
4156 def : Pat<(v8i32 (X86pcmpgt VR256:$src1, VR256:$src2)),
4157 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
4158 def : Pat<(v8i32 (X86pcmpgt VR256:$src1,
4159 (bc_v8i32 (memopv4i64 addr:$src2)))),
4160 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
4163 let Constraints = "$src1 = $dst" in {
4164 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
4165 VR128, memopv2i64, i128mem, 1>;
4166 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
4167 VR128, memopv2i64, i128mem, 1>;
4168 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
4169 VR128, memopv2i64, i128mem, 1>;
4170 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
4171 VR128, memopv2i64, i128mem>;
4172 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
4173 VR128, memopv2i64, i128mem>;
4174 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
4175 VR128, memopv2i64, i128mem>;
4176 } // Constraints = "$src1 = $dst"
4178 let Predicates = [HasSSE2] in {
4179 def : Pat<(v16i8 (X86pcmpeq VR128:$src1, VR128:$src2)),
4180 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
4181 def : Pat<(v16i8 (X86pcmpeq VR128:$src1,
4182 (bc_v16i8 (memopv2i64 addr:$src2)))),
4183 (PCMPEQBrm VR128:$src1, addr:$src2)>;
4184 def : Pat<(v8i16 (X86pcmpeq VR128:$src1, VR128:$src2)),
4185 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
4186 def : Pat<(v8i16 (X86pcmpeq VR128:$src1,
4187 (bc_v8i16 (memopv2i64 addr:$src2)))),
4188 (PCMPEQWrm VR128:$src1, addr:$src2)>;
4189 def : Pat<(v4i32 (X86pcmpeq VR128:$src1, VR128:$src2)),
4190 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
4191 def : Pat<(v4i32 (X86pcmpeq VR128:$src1,
4192 (bc_v4i32 (memopv2i64 addr:$src2)))),
4193 (PCMPEQDrm VR128:$src1, addr:$src2)>;
4195 def : Pat<(v16i8 (X86pcmpgt VR128:$src1, VR128:$src2)),
4196 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
4197 def : Pat<(v16i8 (X86pcmpgt VR128:$src1,
4198 (bc_v16i8 (memopv2i64 addr:$src2)))),
4199 (PCMPGTBrm VR128:$src1, addr:$src2)>;
4200 def : Pat<(v8i16 (X86pcmpgt VR128:$src1, VR128:$src2)),
4201 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
4202 def : Pat<(v8i16 (X86pcmpgt VR128:$src1,
4203 (bc_v8i16 (memopv2i64 addr:$src2)))),
4204 (PCMPGTWrm VR128:$src1, addr:$src2)>;
4205 def : Pat<(v4i32 (X86pcmpgt VR128:$src1, VR128:$src2)),
4206 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
4207 def : Pat<(v4i32 (X86pcmpgt VR128:$src1,
4208 (bc_v4i32 (memopv2i64 addr:$src2)))),
4209 (PCMPGTDrm VR128:$src1, addr:$src2)>;
4212 //===---------------------------------------------------------------------===//
4213 // SSE2 - Packed Integer Pack Instructions
4214 //===---------------------------------------------------------------------===//
4216 let Predicates = [HasAVX] in {
4217 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4218 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4219 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4220 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4221 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4222 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4225 let Predicates = [HasAVX2] in {
4226 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4227 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4228 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4229 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4230 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4231 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4234 let Constraints = "$src1 = $dst" in {
4235 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4236 VR128, memopv2i64, i128mem>;
4237 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4238 VR128, memopv2i64, i128mem>;
4239 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4240 VR128, memopv2i64, i128mem>;
4241 } // Constraints = "$src1 = $dst"
4243 //===---------------------------------------------------------------------===//
4244 // SSE2 - Packed Integer Shuffle Instructions
4245 //===---------------------------------------------------------------------===//
4247 let ExeDomain = SSEPackedInt in {
4248 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4250 def ri : Ii8<0x70, MRMSrcReg,
4251 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4252 !strconcat(OpcodeStr,
4253 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4254 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4256 def mi : Ii8<0x70, MRMSrcMem,
4257 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4258 !strconcat(OpcodeStr,
4259 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4260 [(set VR128:$dst, (vt (pshuf_frag:$src2
4261 (bc_frag (memopv2i64 addr:$src1)),
4265 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4267 def Yri : Ii8<0x70, MRMSrcReg,
4268 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4269 !strconcat(OpcodeStr,
4270 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4271 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4273 def Ymi : Ii8<0x70, MRMSrcMem,
4274 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4275 !strconcat(OpcodeStr,
4276 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4277 [(set VR256:$dst, (vt (pshuf_frag:$src2
4278 (bc_frag (memopv4i64 addr:$src1)),
4281 } // ExeDomain = SSEPackedInt
4283 let Predicates = [HasAVX] in {
4284 let AddedComplexity = 5 in
4285 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4288 // SSE2 with ImmT == Imm8 and XS prefix.
4289 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4292 // SSE2 with ImmT == Imm8 and XD prefix.
4293 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4296 let AddedComplexity = 5 in
4297 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4298 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4299 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4300 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4301 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4303 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4305 (VPSHUFDmi addr:$src1, imm:$imm)>;
4306 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4308 (VPSHUFDmi addr:$src1, imm:$imm)>;
4309 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4310 (VPSHUFDri VR128:$src1, imm:$imm)>;
4311 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4312 (VPSHUFDri VR128:$src1, imm:$imm)>;
4313 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4314 (VPSHUFHWri VR128:$src, imm:$imm)>;
4315 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4317 (VPSHUFHWmi addr:$src, imm:$imm)>;
4318 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4319 (VPSHUFLWri VR128:$src, imm:$imm)>;
4320 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4322 (VPSHUFLWmi addr:$src, imm:$imm)>;
4325 let Predicates = [HasAVX2] in {
4326 let AddedComplexity = 5 in
4327 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4330 // SSE2 with ImmT == Imm8 and XS prefix.
4331 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4334 // SSE2 with ImmT == Imm8 and XD prefix.
4335 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4339 let Predicates = [HasSSE2] in {
4340 let AddedComplexity = 5 in
4341 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4343 // SSE2 with ImmT == Imm8 and XS prefix.
4344 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4346 // SSE2 with ImmT == Imm8 and XD prefix.
4347 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4349 let AddedComplexity = 5 in
4350 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4351 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4352 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4353 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4354 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4356 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4358 (PSHUFDmi addr:$src1, imm:$imm)>;
4359 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4361 (PSHUFDmi addr:$src1, imm:$imm)>;
4362 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4363 (PSHUFDri VR128:$src1, imm:$imm)>;
4364 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4365 (PSHUFDri VR128:$src1, imm:$imm)>;
4366 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4367 (PSHUFHWri VR128:$src, imm:$imm)>;
4368 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4370 (PSHUFHWmi addr:$src, imm:$imm)>;
4371 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4372 (PSHUFLWri VR128:$src, imm:$imm)>;
4373 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4375 (PSHUFLWmi addr:$src, imm:$imm)>;
4378 //===---------------------------------------------------------------------===//
4379 // SSE2 - Packed Integer Unpack Instructions
4380 //===---------------------------------------------------------------------===//
4382 let ExeDomain = SSEPackedInt in {
4383 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4384 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4385 def rr : PDI<opc, MRMSrcReg,
4386 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4388 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4389 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4390 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4391 def rm : PDI<opc, MRMSrcMem,
4392 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4394 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4395 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4396 [(set VR128:$dst, (OpNode VR128:$src1,
4397 (bc_frag (memopv2i64
4401 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4402 SDNode OpNode, PatFrag bc_frag> {
4403 def Yrr : PDI<opc, MRMSrcReg,
4404 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4405 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4406 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4407 def Yrm : PDI<opc, MRMSrcMem,
4408 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4409 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4410 [(set VR256:$dst, (OpNode VR256:$src1,
4411 (bc_frag (memopv4i64 addr:$src2))))]>;
4414 let Predicates = [HasAVX] in {
4415 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4416 bc_v16i8, 0>, VEX_4V;
4417 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4418 bc_v8i16, 0>, VEX_4V;
4419 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4420 bc_v4i32, 0>, VEX_4V;
4421 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4422 bc_v2i64, 0>, VEX_4V;
4424 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4425 bc_v16i8, 0>, VEX_4V;
4426 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4427 bc_v8i16, 0>, VEX_4V;
4428 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4429 bc_v4i32, 0>, VEX_4V;
4430 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4431 bc_v2i64, 0>, VEX_4V;
4434 let Predicates = [HasAVX2] in {
4435 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4437 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4439 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4441 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4444 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4446 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4448 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4450 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4454 let Constraints = "$src1 = $dst" in {
4455 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4457 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4459 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4461 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4464 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4466 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4468 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4470 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4473 } // ExeDomain = SSEPackedInt
4475 // Patterns for using AVX1 instructions with integer vectors
4476 // Here to give AVX2 priority
4477 let Predicates = [HasAVX] in {
4478 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4479 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4480 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4481 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4482 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4483 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4484 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4485 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4487 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4488 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4489 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4490 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4491 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4492 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4493 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4494 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4497 // Splat v2f64 / v2i64
4498 let AddedComplexity = 10 in {
4499 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4500 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4501 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4502 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4505 //===---------------------------------------------------------------------===//
4506 // SSE2 - Packed Integer Extract and Insert
4507 //===---------------------------------------------------------------------===//
4509 let ExeDomain = SSEPackedInt in {
4510 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4511 def rri : Ii8<0xC4, MRMSrcReg,
4512 (outs VR128:$dst), (ins VR128:$src1,
4513 GR32:$src2, i32i8imm:$src3),
4515 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4516 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4518 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4519 def rmi : Ii8<0xC4, MRMSrcMem,
4520 (outs VR128:$dst), (ins VR128:$src1,
4521 i16mem:$src2, i32i8imm:$src3),
4523 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4524 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4526 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4531 let Predicates = [HasAVX] in
4532 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4533 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4534 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4535 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4536 imm:$src2))]>, TB, OpSize, VEX;
4537 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4538 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4539 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4540 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4544 let Predicates = [HasAVX] in {
4545 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4546 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4547 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4548 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4549 []>, TB, OpSize, VEX_4V;
4552 let Constraints = "$src1 = $dst" in
4553 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4555 } // ExeDomain = SSEPackedInt
4557 //===---------------------------------------------------------------------===//
4558 // SSE2 - Packed Mask Creation
4559 //===---------------------------------------------------------------------===//
4561 let ExeDomain = SSEPackedInt in {
4563 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4564 "pmovmskb\t{$src, $dst|$dst, $src}",
4565 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4566 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4567 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4569 let Predicates = [HasAVX2] in {
4570 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4571 "pmovmskb\t{$src, $dst|$dst, $src}",
4572 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4573 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4574 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4577 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4578 "pmovmskb\t{$src, $dst|$dst, $src}",
4579 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4581 } // ExeDomain = SSEPackedInt
4583 //===---------------------------------------------------------------------===//
4584 // SSE2 - Conditional Store
4585 //===---------------------------------------------------------------------===//
4587 let ExeDomain = SSEPackedInt in {
4590 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4591 (ins VR128:$src, VR128:$mask),
4592 "maskmovdqu\t{$mask, $src|$src, $mask}",
4593 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4595 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4596 (ins VR128:$src, VR128:$mask),
4597 "maskmovdqu\t{$mask, $src|$src, $mask}",
4598 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4601 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4602 "maskmovdqu\t{$mask, $src|$src, $mask}",
4603 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4605 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4606 "maskmovdqu\t{$mask, $src|$src, $mask}",
4607 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4609 } // ExeDomain = SSEPackedInt
4611 //===---------------------------------------------------------------------===//
4612 // SSE2 - Move Doubleword
4613 //===---------------------------------------------------------------------===//
4615 //===---------------------------------------------------------------------===//
4616 // Move Int Doubleword to Packed Double Int
4618 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4619 "movd\t{$src, $dst|$dst, $src}",
4621 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4622 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4623 "movd\t{$src, $dst|$dst, $src}",
4625 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4627 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4628 "mov{d|q}\t{$src, $dst|$dst, $src}",
4630 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4631 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4632 "mov{d|q}\t{$src, $dst|$dst, $src}",
4633 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4635 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4636 "movd\t{$src, $dst|$dst, $src}",
4638 (v4i32 (scalar_to_vector GR32:$src)))]>;
4639 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4640 "movd\t{$src, $dst|$dst, $src}",
4642 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4643 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4644 "mov{d|q}\t{$src, $dst|$dst, $src}",
4646 (v2i64 (scalar_to_vector GR64:$src)))]>;
4647 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4648 "mov{d|q}\t{$src, $dst|$dst, $src}",
4649 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4651 //===---------------------------------------------------------------------===//
4652 // Move Int Doubleword to Single Scalar
4654 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4655 "movd\t{$src, $dst|$dst, $src}",
4656 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4658 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4659 "movd\t{$src, $dst|$dst, $src}",
4660 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4662 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4663 "movd\t{$src, $dst|$dst, $src}",
4664 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4666 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4667 "movd\t{$src, $dst|$dst, $src}",
4668 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4670 //===---------------------------------------------------------------------===//
4671 // Move Packed Doubleword Int to Packed Double Int
4673 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4674 "movd\t{$src, $dst|$dst, $src}",
4675 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4677 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4678 (ins i32mem:$dst, VR128:$src),
4679 "movd\t{$src, $dst|$dst, $src}",
4680 [(store (i32 (vector_extract (v4i32 VR128:$src),
4681 (iPTR 0))), addr:$dst)]>, VEX;
4682 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4683 "movd\t{$src, $dst|$dst, $src}",
4684 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4686 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4687 "movd\t{$src, $dst|$dst, $src}",
4688 [(store (i32 (vector_extract (v4i32 VR128:$src),
4689 (iPTR 0))), addr:$dst)]>;
4691 //===---------------------------------------------------------------------===//
4692 // Move Packed Doubleword Int first element to Doubleword Int
4694 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4695 "mov{d|q}\t{$src, $dst|$dst, $src}",
4696 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4698 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4700 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4701 "mov{d|q}\t{$src, $dst|$dst, $src}",
4702 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4705 //===---------------------------------------------------------------------===//
4706 // Bitcast FR64 <-> GR64
4708 let Predicates = [HasAVX] in
4709 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4710 "vmovq\t{$src, $dst|$dst, $src}",
4711 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4713 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4714 "mov{d|q}\t{$src, $dst|$dst, $src}",
4715 [(set GR64:$dst, (bitconvert FR64:$src))]>, VEX;
4716 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4717 "movq\t{$src, $dst|$dst, $src}",
4718 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>,
4721 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4722 "movq\t{$src, $dst|$dst, $src}",
4723 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4724 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4725 "mov{d|q}\t{$src, $dst|$dst, $src}",
4726 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4727 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4728 "movq\t{$src, $dst|$dst, $src}",
4729 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4731 //===---------------------------------------------------------------------===//
4732 // Move Scalar Single to Double Int
4734 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4735 "movd\t{$src, $dst|$dst, $src}",
4736 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4737 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4738 "movd\t{$src, $dst|$dst, $src}",
4739 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4740 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4741 "movd\t{$src, $dst|$dst, $src}",
4742 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4743 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4744 "movd\t{$src, $dst|$dst, $src}",
4745 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4747 //===---------------------------------------------------------------------===//
4748 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4750 let AddedComplexity = 15 in {
4751 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4752 "movd\t{$src, $dst|$dst, $src}",
4753 [(set VR128:$dst, (v4i32 (X86vzmovl
4754 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4756 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4757 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4758 [(set VR128:$dst, (v2i64 (X86vzmovl
4759 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4762 let AddedComplexity = 15 in {
4763 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4764 "movd\t{$src, $dst|$dst, $src}",
4765 [(set VR128:$dst, (v4i32 (X86vzmovl
4766 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4767 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4768 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4769 [(set VR128:$dst, (v2i64 (X86vzmovl
4770 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4773 let AddedComplexity = 20 in {
4774 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4775 "movd\t{$src, $dst|$dst, $src}",
4777 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4778 (loadi32 addr:$src))))))]>,
4780 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4781 "movd\t{$src, $dst|$dst, $src}",
4783 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4784 (loadi32 addr:$src))))))]>;
4787 let Predicates = [HasAVX] in {
4788 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4789 let AddedComplexity = 20 in {
4790 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4791 (VMOVZDI2PDIrm addr:$src)>;
4792 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4793 (VMOVZDI2PDIrm addr:$src)>;
4795 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4796 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4797 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4798 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4799 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4800 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4801 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4804 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4805 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4806 (MOVZDI2PDIrm addr:$src)>;
4807 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4808 (MOVZDI2PDIrm addr:$src)>;
4811 // These are the correct encodings of the instructions so that we know how to
4812 // read correct assembly, even though we continue to emit the wrong ones for
4813 // compatibility with Darwin's buggy assembler.
4814 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4815 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4816 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4817 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4818 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4819 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4820 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4821 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4822 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4823 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4824 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4825 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4827 //===---------------------------------------------------------------------===//
4828 // SSE2 - Move Quadword
4829 //===---------------------------------------------------------------------===//
4831 //===---------------------------------------------------------------------===//
4832 // Move Quadword Int to Packed Quadword Int
4834 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4835 "vmovq\t{$src, $dst|$dst, $src}",
4837 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4838 VEX, Requires<[HasAVX]>;
4839 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4840 "movq\t{$src, $dst|$dst, $src}",
4842 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4843 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4845 //===---------------------------------------------------------------------===//
4846 // Move Packed Quadword Int to Quadword Int
4848 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4849 "movq\t{$src, $dst|$dst, $src}",
4850 [(store (i64 (vector_extract (v2i64 VR128:$src),
4851 (iPTR 0))), addr:$dst)]>, VEX;
4852 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4853 "movq\t{$src, $dst|$dst, $src}",
4854 [(store (i64 (vector_extract (v2i64 VR128:$src),
4855 (iPTR 0))), addr:$dst)]>;
4857 //===---------------------------------------------------------------------===//
4858 // Store / copy lower 64-bits of a XMM register.
4860 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4861 "movq\t{$src, $dst|$dst, $src}",
4862 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4863 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4864 "movq\t{$src, $dst|$dst, $src}",
4865 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4867 let AddedComplexity = 20 in
4868 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4869 "vmovq\t{$src, $dst|$dst, $src}",
4871 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4872 (loadi64 addr:$src))))))]>,
4873 XS, VEX, Requires<[HasAVX]>;
4875 let AddedComplexity = 20 in
4876 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4877 "movq\t{$src, $dst|$dst, $src}",
4879 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4880 (loadi64 addr:$src))))))]>,
4881 XS, Requires<[HasSSE2]>;
4883 let Predicates = [HasAVX], AddedComplexity = 20 in {
4884 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4885 (VMOVZQI2PQIrm addr:$src)>;
4886 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4887 (VMOVZQI2PQIrm addr:$src)>;
4888 def : Pat<(v2i64 (X86vzload addr:$src)),
4889 (VMOVZQI2PQIrm addr:$src)>;
4892 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4893 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4894 (MOVZQI2PQIrm addr:$src)>;
4895 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4896 (MOVZQI2PQIrm addr:$src)>;
4897 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4900 let Predicates = [HasAVX] in {
4901 def : Pat<(v4i64 (X86vzload addr:$src)),
4902 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4905 //===---------------------------------------------------------------------===//
4906 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4907 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4909 let AddedComplexity = 15 in
4910 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4911 "vmovq\t{$src, $dst|$dst, $src}",
4912 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4913 XS, VEX, Requires<[HasAVX]>;
4914 let AddedComplexity = 15 in
4915 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4916 "movq\t{$src, $dst|$dst, $src}",
4917 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4918 XS, Requires<[HasSSE2]>;
4920 let AddedComplexity = 20 in
4921 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4922 "vmovq\t{$src, $dst|$dst, $src}",
4923 [(set VR128:$dst, (v2i64 (X86vzmovl
4924 (loadv2i64 addr:$src))))]>,
4925 XS, VEX, Requires<[HasAVX]>;
4926 let AddedComplexity = 20 in {
4927 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4928 "movq\t{$src, $dst|$dst, $src}",
4929 [(set VR128:$dst, (v2i64 (X86vzmovl
4930 (loadv2i64 addr:$src))))]>,
4931 XS, Requires<[HasSSE2]>;
4934 let AddedComplexity = 20 in {
4935 let Predicates = [HasAVX] in {
4936 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4937 (VMOVZPQILo2PQIrm addr:$src)>;
4938 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4939 (VMOVZPQILo2PQIrr VR128:$src)>;
4941 let Predicates = [HasSSE2] in {
4942 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4943 (MOVZPQILo2PQIrm addr:$src)>;
4944 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4945 (MOVZPQILo2PQIrr VR128:$src)>;
4949 // Instructions to match in the assembler
4950 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4951 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4952 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4953 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4954 // Recognize "movd" with GR64 destination, but encode as a "movq"
4955 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4956 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4958 // Instructions for the disassembler
4959 // xr = XMM register
4962 let Predicates = [HasAVX] in
4963 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4964 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4965 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4966 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4968 //===---------------------------------------------------------------------===//
4969 // SSE3 - Conversion Instructions
4970 //===---------------------------------------------------------------------===//
4972 // Convert Packed Double FP to Packed DW Integers
4973 let Predicates = [HasAVX] in {
4974 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4975 // register, but the same isn't true when using memory operands instead.
4976 // Provide other assembly rr and rm forms to address this explicitly.
4977 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4978 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4979 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4980 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4983 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4984 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4985 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4986 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4989 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4990 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4991 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4992 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4995 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4996 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4997 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4998 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
5000 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
5001 (VCVTPD2DQYrr VR256:$src)>;
5002 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
5003 (VCVTPD2DQYrm addr:$src)>;
5005 // Convert Packed DW Integers to Packed Double FP
5006 let Predicates = [HasAVX] in {
5007 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
5008 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
5009 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5010 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
5011 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
5012 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
5013 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5014 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
5017 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
5018 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
5019 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5020 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
5022 // AVX 256-bit register conversion intrinsics
5023 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
5024 (VCVTDQ2PDYrr VR128:$src)>;
5025 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
5026 (VCVTDQ2PDYrm addr:$src)>;
5028 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
5029 (VCVTPD2DQYrr VR256:$src)>;
5030 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
5031 (VCVTPD2DQYrm addr:$src)>;
5033 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
5034 (VCVTDQ2PDYrr VR128:$src)>;
5035 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
5036 (VCVTDQ2PDYrm addr:$src)>;
5038 //===---------------------------------------------------------------------===//
5039 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
5040 //===---------------------------------------------------------------------===//
5041 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
5042 ValueType vt, RegisterClass RC, PatFrag mem_frag,
5043 X86MemOperand x86memop> {
5044 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
5045 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5046 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
5047 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5048 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5049 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
5052 let Predicates = [HasAVX] in {
5053 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
5054 v4f32, VR128, memopv4f32, f128mem>, VEX;
5055 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
5056 v4f32, VR128, memopv4f32, f128mem>, VEX;
5057 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
5058 v8f32, VR256, memopv8f32, f256mem>, VEX;
5059 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
5060 v8f32, VR256, memopv8f32, f256mem>, VEX;
5062 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
5063 memopv4f32, f128mem>;
5064 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
5065 memopv4f32, f128mem>;
5067 let Predicates = [HasAVX] in {
5068 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5069 (VMOVSHDUPrr VR128:$src)>;
5070 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
5071 (VMOVSHDUPrm addr:$src)>;
5072 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5073 (VMOVSLDUPrr VR128:$src)>;
5074 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
5075 (VMOVSLDUPrm addr:$src)>;
5076 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
5077 (VMOVSHDUPYrr VR256:$src)>;
5078 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
5079 (VMOVSHDUPYrm addr:$src)>;
5080 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
5081 (VMOVSLDUPYrr VR256:$src)>;
5082 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
5083 (VMOVSLDUPYrm addr:$src)>;
5086 let Predicates = [HasSSE3] in {
5087 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5088 (MOVSHDUPrr VR128:$src)>;
5089 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
5090 (MOVSHDUPrm addr:$src)>;
5091 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5092 (MOVSLDUPrr VR128:$src)>;
5093 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
5094 (MOVSLDUPrm addr:$src)>;
5097 //===---------------------------------------------------------------------===//
5098 // SSE3 - Replicate Double FP - MOVDDUP
5099 //===---------------------------------------------------------------------===//
5101 multiclass sse3_replicate_dfp<string OpcodeStr> {
5102 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5103 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5104 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
5105 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
5106 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5108 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
5112 // FIXME: Merge with above classe when there're patterns for the ymm version
5113 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
5114 let Predicates = [HasAVX] in {
5115 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
5116 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5118 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
5119 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5124 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
5125 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
5126 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
5128 let Predicates = [HasAVX] in {
5129 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
5131 (VMOVDDUPrm addr:$src)>;
5132 let AddedComplexity = 5 in {
5133 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
5134 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
5135 (VMOVDDUPrm addr:$src)>;
5136 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
5137 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
5138 (VMOVDDUPrm addr:$src)>;
5140 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5141 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5142 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5143 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5144 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5145 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5146 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5147 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5148 def : Pat<(X86Movddup (bc_v2f64
5149 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5150 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5153 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
5154 (VMOVDDUPYrm addr:$src)>;
5155 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
5156 (VMOVDDUPYrm addr:$src)>;
5157 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
5158 (VMOVDDUPYrm addr:$src)>;
5159 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
5160 (VMOVDDUPYrm addr:$src)>;
5161 def : Pat<(X86Movddup (v4f64 VR256:$src)),
5162 (VMOVDDUPYrr VR256:$src)>;
5163 def : Pat<(X86Movddup (v4i64 VR256:$src)),
5164 (VMOVDDUPYrr VR256:$src)>;
5167 let Predicates = [HasSSE3] in {
5168 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
5170 (MOVDDUPrm addr:$src)>;
5171 let AddedComplexity = 5 in {
5172 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
5173 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
5174 (MOVDDUPrm addr:$src)>;
5175 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
5176 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
5177 (MOVDDUPrm addr:$src)>;
5179 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5180 (MOVDDUPrm addr:$src)>;
5181 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5182 (MOVDDUPrm addr:$src)>;
5183 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5184 (MOVDDUPrm addr:$src)>;
5185 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5186 (MOVDDUPrm addr:$src)>;
5187 def : Pat<(X86Movddup (bc_v2f64
5188 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5189 (MOVDDUPrm addr:$src)>;
5192 //===---------------------------------------------------------------------===//
5193 // SSE3 - Move Unaligned Integer
5194 //===---------------------------------------------------------------------===//
5196 let Predicates = [HasAVX] in {
5197 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5198 "vlddqu\t{$src, $dst|$dst, $src}",
5199 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5200 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5201 "vlddqu\t{$src, $dst|$dst, $src}",
5202 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5204 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5205 "lddqu\t{$src, $dst|$dst, $src}",
5206 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
5208 //===---------------------------------------------------------------------===//
5209 // SSE3 - Arithmetic
5210 //===---------------------------------------------------------------------===//
5212 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5213 X86MemOperand x86memop, bit Is2Addr = 1> {
5214 def rr : I<0xD0, MRMSrcReg,
5215 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5217 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5218 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5219 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
5220 def rm : I<0xD0, MRMSrcMem,
5221 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5223 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5224 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5225 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
5228 let Predicates = [HasAVX] in {
5229 let ExeDomain = SSEPackedSingle in {
5230 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5231 f128mem, 0>, TB, XD, VEX_4V;
5232 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5233 f256mem, 0>, TB, XD, VEX_4V;
5235 let ExeDomain = SSEPackedDouble in {
5236 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5237 f128mem, 0>, TB, OpSize, VEX_4V;
5238 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5239 f256mem, 0>, TB, OpSize, VEX_4V;
5242 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5243 let ExeDomain = SSEPackedSingle in
5244 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5246 let ExeDomain = SSEPackedDouble in
5247 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5248 f128mem>, TB, OpSize;
5251 //===---------------------------------------------------------------------===//
5252 // SSE3 Instructions
5253 //===---------------------------------------------------------------------===//
5256 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5257 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5258 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5260 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5261 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5262 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5264 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5266 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5267 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5268 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5270 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5271 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5272 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5274 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5275 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5276 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5278 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5280 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5281 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5282 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5285 let Predicates = [HasAVX] in {
5286 let ExeDomain = SSEPackedSingle in {
5287 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5288 X86fhadd, 0>, VEX_4V;
5289 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5290 X86fhsub, 0>, VEX_4V;
5291 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5292 X86fhadd, 0>, VEX_4V;
5293 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5294 X86fhsub, 0>, VEX_4V;
5296 let ExeDomain = SSEPackedDouble in {
5297 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5298 X86fhadd, 0>, VEX_4V;
5299 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5300 X86fhsub, 0>, VEX_4V;
5301 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5302 X86fhadd, 0>, VEX_4V;
5303 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5304 X86fhsub, 0>, VEX_4V;
5308 let Constraints = "$src1 = $dst" in {
5309 let ExeDomain = SSEPackedSingle in {
5310 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5311 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5313 let ExeDomain = SSEPackedDouble in {
5314 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5315 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5319 //===---------------------------------------------------------------------===//
5320 // SSSE3 - Packed Absolute Instructions
5321 //===---------------------------------------------------------------------===//
5324 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5325 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5326 Intrinsic IntId128> {
5327 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5329 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5330 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5333 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5335 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5338 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5341 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5342 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5343 Intrinsic IntId256> {
5344 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5346 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5347 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5350 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5352 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5355 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5358 let Predicates = [HasAVX] in {
5359 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5360 int_x86_ssse3_pabs_b_128>, VEX;
5361 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5362 int_x86_ssse3_pabs_w_128>, VEX;
5363 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5364 int_x86_ssse3_pabs_d_128>, VEX;
5367 let Predicates = [HasAVX2] in {
5368 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5369 int_x86_avx2_pabs_b>, VEX;
5370 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5371 int_x86_avx2_pabs_w>, VEX;
5372 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5373 int_x86_avx2_pabs_d>, VEX;
5376 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5377 int_x86_ssse3_pabs_b_128>;
5378 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5379 int_x86_ssse3_pabs_w_128>;
5380 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5381 int_x86_ssse3_pabs_d_128>;
5383 //===---------------------------------------------------------------------===//
5384 // SSSE3 - Packed Binary Operator Instructions
5385 //===---------------------------------------------------------------------===//
5387 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5388 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5389 Intrinsic IntId128, bit Is2Addr = 1> {
5390 let isCommutable = 1 in
5391 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5392 (ins VR128:$src1, VR128:$src2),
5394 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5395 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5396 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5398 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5399 (ins VR128:$src1, i128mem:$src2),
5401 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5402 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5404 (IntId128 VR128:$src1,
5405 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5408 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5409 Intrinsic IntId256> {
5410 let isCommutable = 1 in
5411 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5412 (ins VR256:$src1, VR256:$src2),
5413 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5414 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5416 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5417 (ins VR256:$src1, i256mem:$src2),
5418 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5420 (IntId256 VR256:$src1,
5421 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5424 let ImmT = NoImm, Predicates = [HasAVX] in {
5425 let isCommutable = 0 in {
5426 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw",
5427 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5428 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd",
5429 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5430 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5431 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5432 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw",
5433 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5434 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd",
5435 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5436 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5437 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5438 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5439 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5440 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb",
5441 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5442 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb",
5443 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5444 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw",
5445 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5446 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd",
5447 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5449 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5450 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5453 let ImmT = NoImm, Predicates = [HasAVX2] in {
5454 let isCommutable = 0 in {
5455 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw",
5456 int_x86_avx2_phadd_w>, VEX_4V;
5457 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd",
5458 int_x86_avx2_phadd_d>, VEX_4V;
5459 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5460 int_x86_avx2_phadd_sw>, VEX_4V;
5461 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw",
5462 int_x86_avx2_phsub_w>, VEX_4V;
5463 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd",
5464 int_x86_avx2_phsub_d>, VEX_4V;
5465 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5466 int_x86_avx2_phsub_sw>, VEX_4V;
5467 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5468 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5469 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb",
5470 int_x86_avx2_pshuf_b>, VEX_4V;
5471 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb",
5472 int_x86_avx2_psign_b>, VEX_4V;
5473 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw",
5474 int_x86_avx2_psign_w>, VEX_4V;
5475 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd",
5476 int_x86_avx2_psign_d>, VEX_4V;
5478 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5479 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5482 // None of these have i8 immediate fields.
5483 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5484 let isCommutable = 0 in {
5485 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw",
5486 int_x86_ssse3_phadd_w_128>;
5487 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd",
5488 int_x86_ssse3_phadd_d_128>;
5489 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5490 int_x86_ssse3_phadd_sw_128>;
5491 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw",
5492 int_x86_ssse3_phsub_w_128>;
5493 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd",
5494 int_x86_ssse3_phsub_d_128>;
5495 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5496 int_x86_ssse3_phsub_sw_128>;
5497 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5498 int_x86_ssse3_pmadd_ub_sw_128>;
5499 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb",
5500 int_x86_ssse3_pshuf_b_128>;
5501 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb",
5502 int_x86_ssse3_psign_b_128>;
5503 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw",
5504 int_x86_ssse3_psign_w_128>;
5505 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd",
5506 int_x86_ssse3_psign_d_128>;
5508 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5509 int_x86_ssse3_pmul_hr_sw_128>;
5512 let Predicates = [HasAVX] in {
5513 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5514 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5515 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5516 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5518 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5519 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5520 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5521 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5522 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5523 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5525 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5526 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5527 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5528 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5529 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5530 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5531 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5532 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5535 let Predicates = [HasAVX2] in {
5536 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5537 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5538 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5539 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5540 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5541 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5543 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5544 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5545 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5546 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5547 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5548 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5549 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5550 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5553 let Predicates = [HasSSSE3] in {
5554 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5555 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5556 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5557 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5559 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5560 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5561 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5562 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5563 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5564 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5566 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5567 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5568 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5569 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5570 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5571 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5572 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5573 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5576 //===---------------------------------------------------------------------===//
5577 // SSSE3 - Packed Align Instruction Patterns
5578 //===---------------------------------------------------------------------===//
5580 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5581 let neverHasSideEffects = 1 in {
5582 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5583 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5585 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5587 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5590 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5591 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5593 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5595 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5600 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5601 let neverHasSideEffects = 1 in {
5602 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5603 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5605 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5608 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5609 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5611 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5616 let Predicates = [HasAVX] in
5617 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5618 let Predicates = [HasAVX2] in
5619 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5620 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5621 defm PALIGN : ssse3_palign<"palignr">;
5623 let Predicates = [HasAVX2] in {
5624 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5625 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5626 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5627 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5628 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5629 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5630 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5631 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5634 let Predicates = [HasAVX] in {
5635 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5636 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5637 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5638 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5639 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5640 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5641 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5642 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5645 let Predicates = [HasSSSE3] in {
5646 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5647 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5648 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5649 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5650 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5651 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5652 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5653 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5656 //===---------------------------------------------------------------------===//
5657 // SSSE3 - Thread synchronization
5658 //===---------------------------------------------------------------------===//
5660 let usesCustomInserter = 1 in {
5661 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5662 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5663 Requires<[HasSSE3]>;
5664 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5665 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5666 Requires<[HasSSE3]>;
5669 let Uses = [EAX, ECX, EDX] in
5670 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5671 Requires<[HasSSE3]>;
5672 let Uses = [ECX, EAX] in
5673 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5674 Requires<[HasSSE3]>;
5676 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5677 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5679 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5680 Requires<[In32BitMode]>;
5681 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5682 Requires<[In64BitMode]>;
5684 //===----------------------------------------------------------------------===//
5685 // SSE4.1 - Packed Move with Sign/Zero Extend
5686 //===----------------------------------------------------------------------===//
5688 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5689 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5690 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5691 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5693 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5694 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5696 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5700 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5702 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5703 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5704 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5706 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5707 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5708 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5711 let Predicates = [HasAVX] in {
5712 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5714 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5716 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5718 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5720 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5722 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5726 let Predicates = [HasAVX2] in {
5727 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5728 int_x86_avx2_pmovsxbw>, VEX;
5729 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5730 int_x86_avx2_pmovsxwd>, VEX;
5731 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5732 int_x86_avx2_pmovsxdq>, VEX;
5733 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5734 int_x86_avx2_pmovzxbw>, VEX;
5735 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5736 int_x86_avx2_pmovzxwd>, VEX;
5737 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5738 int_x86_avx2_pmovzxdq>, VEX;
5741 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5742 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5743 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5744 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5745 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5746 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5748 let Predicates = [HasAVX] in {
5749 // Common patterns involving scalar load.
5750 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5751 (VPMOVSXBWrm addr:$src)>;
5752 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5753 (VPMOVSXBWrm addr:$src)>;
5755 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5756 (VPMOVSXWDrm addr:$src)>;
5757 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5758 (VPMOVSXWDrm addr:$src)>;
5760 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5761 (VPMOVSXDQrm addr:$src)>;
5762 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5763 (VPMOVSXDQrm addr:$src)>;
5765 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5766 (VPMOVZXBWrm addr:$src)>;
5767 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5768 (VPMOVZXBWrm addr:$src)>;
5770 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5771 (VPMOVZXWDrm addr:$src)>;
5772 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5773 (VPMOVZXWDrm addr:$src)>;
5775 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5776 (VPMOVZXDQrm addr:$src)>;
5777 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5778 (VPMOVZXDQrm addr:$src)>;
5781 let Predicates = [HasSSE41] in {
5782 // Common patterns involving scalar load.
5783 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5784 (PMOVSXBWrm addr:$src)>;
5785 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5786 (PMOVSXBWrm addr:$src)>;
5788 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5789 (PMOVSXWDrm addr:$src)>;
5790 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5791 (PMOVSXWDrm addr:$src)>;
5793 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5794 (PMOVSXDQrm addr:$src)>;
5795 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5796 (PMOVSXDQrm addr:$src)>;
5798 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5799 (PMOVZXBWrm addr:$src)>;
5800 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5801 (PMOVZXBWrm addr:$src)>;
5803 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5804 (PMOVZXWDrm addr:$src)>;
5805 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5806 (PMOVZXWDrm addr:$src)>;
5808 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5809 (PMOVZXDQrm addr:$src)>;
5810 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5811 (PMOVZXDQrm addr:$src)>;
5815 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5816 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5817 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5818 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5820 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5821 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5823 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5827 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5829 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5830 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5831 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5833 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5834 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5836 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5840 let Predicates = [HasAVX] in {
5841 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5843 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5845 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5847 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5851 let Predicates = [HasAVX2] in {
5852 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5853 int_x86_avx2_pmovsxbd>, VEX;
5854 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5855 int_x86_avx2_pmovsxwq>, VEX;
5856 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5857 int_x86_avx2_pmovzxbd>, VEX;
5858 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5859 int_x86_avx2_pmovzxwq>, VEX;
5862 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5863 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5864 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5865 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5867 let Predicates = [HasAVX] in {
5868 // Common patterns involving scalar load
5869 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5870 (VPMOVSXBDrm addr:$src)>;
5871 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5872 (VPMOVSXWQrm addr:$src)>;
5874 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5875 (VPMOVZXBDrm addr:$src)>;
5876 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5877 (VPMOVZXWQrm addr:$src)>;
5880 let Predicates = [HasSSE41] in {
5881 // Common patterns involving scalar load
5882 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5883 (PMOVSXBDrm addr:$src)>;
5884 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5885 (PMOVSXWQrm addr:$src)>;
5887 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5888 (PMOVZXBDrm addr:$src)>;
5889 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5890 (PMOVZXWQrm addr:$src)>;
5893 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5894 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5895 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5896 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5898 // Expecting a i16 load any extended to i32 value.
5899 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5900 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5901 [(set VR128:$dst, (IntId (bitconvert
5902 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5906 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5908 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5909 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5910 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5912 // Expecting a i16 load any extended to i32 value.
5913 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5914 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5915 [(set VR256:$dst, (IntId (bitconvert
5916 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5920 let Predicates = [HasAVX] in {
5921 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5923 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5926 let Predicates = [HasAVX2] in {
5927 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5928 int_x86_avx2_pmovsxbq>, VEX;
5929 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5930 int_x86_avx2_pmovzxbq>, VEX;
5932 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5933 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5935 let Predicates = [HasAVX] in {
5936 // Common patterns involving scalar load
5937 def : Pat<(int_x86_sse41_pmovsxbq
5938 (bitconvert (v4i32 (X86vzmovl
5939 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5940 (VPMOVSXBQrm addr:$src)>;
5942 def : Pat<(int_x86_sse41_pmovzxbq
5943 (bitconvert (v4i32 (X86vzmovl
5944 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5945 (VPMOVZXBQrm addr:$src)>;
5948 let Predicates = [HasSSE41] in {
5949 // Common patterns involving scalar load
5950 def : Pat<(int_x86_sse41_pmovsxbq
5951 (bitconvert (v4i32 (X86vzmovl
5952 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5953 (PMOVSXBQrm addr:$src)>;
5955 def : Pat<(int_x86_sse41_pmovzxbq
5956 (bitconvert (v4i32 (X86vzmovl
5957 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5958 (PMOVZXBQrm addr:$src)>;
5961 //===----------------------------------------------------------------------===//
5962 // SSE4.1 - Extract Instructions
5963 //===----------------------------------------------------------------------===//
5965 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5966 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5967 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5968 (ins VR128:$src1, i32i8imm:$src2),
5969 !strconcat(OpcodeStr,
5970 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5971 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5973 let neverHasSideEffects = 1, mayStore = 1 in
5974 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5975 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5976 !strconcat(OpcodeStr,
5977 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5980 // There's an AssertZext in the way of writing the store pattern
5981 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5984 let Predicates = [HasAVX] in {
5985 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5986 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5987 (ins VR128:$src1, i32i8imm:$src2),
5988 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5991 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5994 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5995 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5996 let neverHasSideEffects = 1, mayStore = 1 in
5997 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5998 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5999 !strconcat(OpcodeStr,
6000 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6003 // There's an AssertZext in the way of writing the store pattern
6004 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
6007 let Predicates = [HasAVX] in
6008 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
6010 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
6013 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6014 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
6015 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6016 (ins VR128:$src1, i32i8imm:$src2),
6017 !strconcat(OpcodeStr,
6018 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6020 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
6021 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6022 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
6023 !strconcat(OpcodeStr,
6024 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6025 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
6026 addr:$dst)]>, OpSize;
6029 let Predicates = [HasAVX] in
6030 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
6032 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
6034 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
6035 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
6036 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
6037 (ins VR128:$src1, i32i8imm:$src2),
6038 !strconcat(OpcodeStr,
6039 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6041 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
6042 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6043 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
6044 !strconcat(OpcodeStr,
6045 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6046 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
6047 addr:$dst)]>, OpSize, REX_W;
6050 let Predicates = [HasAVX] in
6051 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
6053 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
6055 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
6057 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
6058 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
6059 (ins VR128:$src1, i32i8imm:$src2),
6060 !strconcat(OpcodeStr,
6061 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6063 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
6065 def mr : SS4AIi8<opc, MRMDestMem, (outs),
6066 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
6067 !strconcat(OpcodeStr,
6068 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6069 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
6070 addr:$dst)]>, OpSize;
6073 let ExeDomain = SSEPackedSingle in {
6074 let Predicates = [HasAVX] in {
6075 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
6076 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
6077 (ins VR128:$src1, i32i8imm:$src2),
6078 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
6081 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
6084 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
6085 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6088 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6090 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6093 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6094 Requires<[HasSSE41]>;
6096 //===----------------------------------------------------------------------===//
6097 // SSE4.1 - Insert Instructions
6098 //===----------------------------------------------------------------------===//
6100 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
6101 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6102 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6104 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6106 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6108 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
6109 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6110 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
6112 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6114 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6116 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
6117 imm:$src3))]>, OpSize;
6120 let Predicates = [HasAVX] in
6121 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
6122 let Constraints = "$src1 = $dst" in
6123 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6125 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6126 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6127 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6129 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6131 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6133 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6135 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6136 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6138 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6140 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6142 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6143 imm:$src3)))]>, OpSize;
6146 let Predicates = [HasAVX] in
6147 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6148 let Constraints = "$src1 = $dst" in
6149 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6151 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6152 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6153 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6155 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6157 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6159 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6161 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6162 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6164 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6166 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6168 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6169 imm:$src3)))]>, OpSize;
6172 let Predicates = [HasAVX] in
6173 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6174 let Constraints = "$src1 = $dst" in
6175 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6177 // insertps has a few different modes, there's the first two here below which
6178 // are optimized inserts that won't zero arbitrary elements in the destination
6179 // vector. The next one matches the intrinsic and could zero arbitrary elements
6180 // in the target vector.
6181 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6182 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6183 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6185 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6187 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6189 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6191 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6192 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6194 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6196 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6198 (X86insrtps VR128:$src1,
6199 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6200 imm:$src3))]>, OpSize;
6203 let ExeDomain = SSEPackedSingle in {
6204 let Predicates = [HasAVX] in
6205 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6206 let Constraints = "$src1 = $dst" in
6207 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6210 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6211 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6213 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6214 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6215 Requires<[HasSSE41]>;
6217 //===----------------------------------------------------------------------===//
6218 // SSE4.1 - Round Instructions
6219 //===----------------------------------------------------------------------===//
6221 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6222 X86MemOperand x86memop, RegisterClass RC,
6223 PatFrag mem_frag32, PatFrag mem_frag64,
6224 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6225 let ExeDomain = SSEPackedSingle in {
6226 // Intrinsic operation, reg.
6227 // Vector intrinsic operation, reg
6228 def PSr : SS4AIi8<opcps, MRMSrcReg,
6229 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6230 !strconcat(OpcodeStr,
6231 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6232 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6235 // Vector intrinsic operation, mem
6236 def PSm : SS4AIi8<opcps, MRMSrcMem,
6237 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6238 !strconcat(OpcodeStr,
6239 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6241 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6243 } // ExeDomain = SSEPackedSingle
6245 let ExeDomain = SSEPackedDouble in {
6246 // Vector intrinsic operation, reg
6247 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6248 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6249 !strconcat(OpcodeStr,
6250 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6251 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6254 // Vector intrinsic operation, mem
6255 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6256 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6257 !strconcat(OpcodeStr,
6258 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6260 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6262 } // ExeDomain = SSEPackedDouble
6265 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6268 Intrinsic F64Int, bit Is2Addr = 1> {
6269 let ExeDomain = GenericDomain in {
6271 def SSr : SS4AIi8<opcss, MRMSrcReg,
6272 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6274 !strconcat(OpcodeStr,
6275 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6276 !strconcat(OpcodeStr,
6277 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6280 // Intrinsic operation, reg.
6281 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6282 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6284 !strconcat(OpcodeStr,
6285 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6286 !strconcat(OpcodeStr,
6287 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6288 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6291 // Intrinsic operation, mem.
6292 def SSm : SS4AIi8<opcss, MRMSrcMem,
6293 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6295 !strconcat(OpcodeStr,
6296 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6297 !strconcat(OpcodeStr,
6298 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6300 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6304 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6305 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6307 !strconcat(OpcodeStr,
6308 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6309 !strconcat(OpcodeStr,
6310 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6313 // Intrinsic operation, reg.
6314 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6315 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6317 !strconcat(OpcodeStr,
6318 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6319 !strconcat(OpcodeStr,
6320 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6321 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6324 // Intrinsic operation, mem.
6325 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6326 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6328 !strconcat(OpcodeStr,
6329 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6330 !strconcat(OpcodeStr,
6331 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6333 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6335 } // ExeDomain = GenericDomain
6338 // FP round - roundss, roundps, roundsd, roundpd
6339 let Predicates = [HasAVX] in {
6341 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6342 memopv4f32, memopv2f64,
6343 int_x86_sse41_round_ps,
6344 int_x86_sse41_round_pd>, VEX;
6345 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6346 memopv8f32, memopv4f64,
6347 int_x86_avx_round_ps_256,
6348 int_x86_avx_round_pd_256>, VEX;
6349 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6350 int_x86_sse41_round_ss,
6351 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6353 def : Pat<(ffloor FR32:$src),
6354 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6355 def : Pat<(f64 (ffloor FR64:$src)),
6356 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6357 def : Pat<(f32 (fnearbyint FR32:$src)),
6358 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6359 def : Pat<(f64 (fnearbyint FR64:$src)),
6360 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6361 def : Pat<(f32 (fceil FR32:$src)),
6362 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6363 def : Pat<(f64 (fceil FR64:$src)),
6364 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6365 def : Pat<(f32 (frint FR32:$src)),
6366 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6367 def : Pat<(f64 (frint FR64:$src)),
6368 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6369 def : Pat<(f32 (ftrunc FR32:$src)),
6370 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6371 def : Pat<(f64 (ftrunc FR64:$src)),
6372 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6375 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6376 memopv4f32, memopv2f64,
6377 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6378 let Constraints = "$src1 = $dst" in
6379 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6380 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6382 def : Pat<(ffloor FR32:$src),
6383 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6384 def : Pat<(f64 (ffloor FR64:$src)),
6385 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6386 def : Pat<(f32 (fnearbyint FR32:$src)),
6387 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6388 def : Pat<(f64 (fnearbyint FR64:$src)),
6389 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6390 def : Pat<(f32 (fceil FR32:$src)),
6391 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6392 def : Pat<(f64 (fceil FR64:$src)),
6393 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6394 def : Pat<(f32 (frint FR32:$src)),
6395 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6396 def : Pat<(f64 (frint FR64:$src)),
6397 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6398 def : Pat<(f32 (ftrunc FR32:$src)),
6399 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6400 def : Pat<(f64 (ftrunc FR64:$src)),
6401 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6403 //===----------------------------------------------------------------------===//
6404 // SSE4.1 - Packed Bit Test
6405 //===----------------------------------------------------------------------===//
6407 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6408 // the intel intrinsic that corresponds to this.
6409 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6410 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6411 "vptest\t{$src2, $src1|$src1, $src2}",
6412 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6414 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6415 "vptest\t{$src2, $src1|$src1, $src2}",
6416 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6419 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6420 "vptest\t{$src2, $src1|$src1, $src2}",
6421 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6423 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6424 "vptest\t{$src2, $src1|$src1, $src2}",
6425 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6429 let Defs = [EFLAGS] in {
6430 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6431 "ptest\t{$src2, $src1|$src1, $src2}",
6432 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6434 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6435 "ptest\t{$src2, $src1|$src1, $src2}",
6436 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6440 // The bit test instructions below are AVX only
6441 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6442 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6443 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6444 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6445 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6446 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6447 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6448 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6452 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6453 let ExeDomain = SSEPackedSingle in {
6454 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6455 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6457 let ExeDomain = SSEPackedDouble in {
6458 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6459 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6463 //===----------------------------------------------------------------------===//
6464 // SSE4.1 - Misc Instructions
6465 //===----------------------------------------------------------------------===//
6467 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6468 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6469 "popcnt{w}\t{$src, $dst|$dst, $src}",
6470 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6472 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6473 "popcnt{w}\t{$src, $dst|$dst, $src}",
6474 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6475 (implicit EFLAGS)]>, OpSize, XS;
6477 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6478 "popcnt{l}\t{$src, $dst|$dst, $src}",
6479 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6481 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6482 "popcnt{l}\t{$src, $dst|$dst, $src}",
6483 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6484 (implicit EFLAGS)]>, XS;
6486 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6487 "popcnt{q}\t{$src, $dst|$dst, $src}",
6488 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6490 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6491 "popcnt{q}\t{$src, $dst|$dst, $src}",
6492 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6493 (implicit EFLAGS)]>, XS;
6498 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6499 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6500 Intrinsic IntId128> {
6501 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6503 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6504 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6505 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6507 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6510 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6513 let Predicates = [HasAVX] in
6514 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6515 int_x86_sse41_phminposuw>, VEX;
6516 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6517 int_x86_sse41_phminposuw>;
6519 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6520 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6521 Intrinsic IntId128, bit Is2Addr = 1> {
6522 let isCommutable = 1 in
6523 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6524 (ins VR128:$src1, VR128:$src2),
6526 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6527 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6528 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6529 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6530 (ins VR128:$src1, i128mem:$src2),
6532 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6533 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6535 (IntId128 VR128:$src1,
6536 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6539 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6540 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6541 Intrinsic IntId256> {
6542 let isCommutable = 1 in
6543 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6544 (ins VR256:$src1, VR256:$src2),
6545 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6546 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6547 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6548 (ins VR256:$src1, i256mem:$src2),
6549 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6551 (IntId256 VR256:$src1,
6552 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6555 let Predicates = [HasAVX] in {
6556 let isCommutable = 0 in
6557 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6559 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6561 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6563 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6565 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6567 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6569 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6571 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6573 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6575 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6577 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6580 def : Pat<(v2i64 (X86pcmpeq VR128:$src1, VR128:$src2)),
6581 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6582 def : Pat<(v2i64 (X86pcmpeq VR128:$src1, (memop addr:$src2))),
6583 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6586 let Predicates = [HasAVX2] in {
6587 let isCommutable = 0 in
6588 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6589 int_x86_avx2_packusdw>, VEX_4V;
6590 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6591 int_x86_avx2_pcmpeq_q>, VEX_4V;
6592 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6593 int_x86_avx2_pmins_b>, VEX_4V;
6594 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6595 int_x86_avx2_pmins_d>, VEX_4V;
6596 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6597 int_x86_avx2_pminu_d>, VEX_4V;
6598 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6599 int_x86_avx2_pminu_w>, VEX_4V;
6600 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6601 int_x86_avx2_pmaxs_b>, VEX_4V;
6602 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6603 int_x86_avx2_pmaxs_d>, VEX_4V;
6604 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6605 int_x86_avx2_pmaxu_d>, VEX_4V;
6606 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6607 int_x86_avx2_pmaxu_w>, VEX_4V;
6608 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6609 int_x86_avx2_pmul_dq>, VEX_4V;
6611 def : Pat<(v4i64 (X86pcmpeq VR256:$src1, VR256:$src2)),
6612 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6613 def : Pat<(v4i64 (X86pcmpeq VR256:$src1, (memop addr:$src2))),
6614 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6617 let Constraints = "$src1 = $dst" in {
6618 let isCommutable = 0 in
6619 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6620 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6621 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6622 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6623 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6624 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6625 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6626 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6627 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6628 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6629 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6632 let Predicates = [HasSSE41] in {
6633 def : Pat<(v2i64 (X86pcmpeq VR128:$src1, VR128:$src2)),
6634 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6635 def : Pat<(v2i64 (X86pcmpeq VR128:$src1, (memop addr:$src2))),
6636 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6639 /// SS48I_binop_rm - Simple SSE41 binary operator.
6640 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6641 ValueType OpVT, bit Is2Addr = 1> {
6642 let isCommutable = 1 in
6643 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6644 (ins VR128:$src1, VR128:$src2),
6646 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6647 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6648 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6650 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6651 (ins VR128:$src1, i128mem:$src2),
6653 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6654 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6655 [(set VR128:$dst, (OpNode VR128:$src1,
6656 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6660 /// SS48I_binop_rm - Simple SSE41 binary operator.
6661 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6663 let isCommutable = 1 in
6664 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6665 (ins VR256:$src1, VR256:$src2),
6666 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6667 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6669 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6670 (ins VR256:$src1, i256mem:$src2),
6671 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6672 [(set VR256:$dst, (OpNode VR256:$src1,
6673 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6677 let Predicates = [HasAVX] in
6678 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6679 let Predicates = [HasAVX2] in
6680 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6681 let Constraints = "$src1 = $dst" in
6682 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6684 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6685 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6686 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6687 X86MemOperand x86memop, bit Is2Addr = 1> {
6688 let isCommutable = 1 in
6689 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6690 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6692 !strconcat(OpcodeStr,
6693 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6694 !strconcat(OpcodeStr,
6695 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6696 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6698 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6699 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6701 !strconcat(OpcodeStr,
6702 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6703 !strconcat(OpcodeStr,
6704 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6707 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6711 let Predicates = [HasAVX] in {
6712 let isCommutable = 0 in {
6713 let ExeDomain = SSEPackedSingle in {
6714 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6715 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6716 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6717 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6719 let ExeDomain = SSEPackedDouble in {
6720 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6721 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6722 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6723 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6725 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6726 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6727 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6728 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6730 let ExeDomain = SSEPackedSingle in
6731 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6732 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6733 let ExeDomain = SSEPackedDouble in
6734 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6735 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6736 let ExeDomain = SSEPackedSingle in
6737 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6738 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6741 let Predicates = [HasAVX2] in {
6742 let isCommutable = 0 in {
6743 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6744 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6745 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6746 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6750 let Constraints = "$src1 = $dst" in {
6751 let isCommutable = 0 in {
6752 let ExeDomain = SSEPackedSingle in
6753 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6754 VR128, memopv4f32, i128mem>;
6755 let ExeDomain = SSEPackedDouble in
6756 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6757 VR128, memopv2f64, i128mem>;
6758 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6759 VR128, memopv2i64, i128mem>;
6760 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6761 VR128, memopv2i64, i128mem>;
6763 let ExeDomain = SSEPackedSingle in
6764 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6765 VR128, memopv4f32, i128mem>;
6766 let ExeDomain = SSEPackedDouble in
6767 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6768 VR128, memopv2f64, i128mem>;
6771 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6772 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6773 RegisterClass RC, X86MemOperand x86memop,
6774 PatFrag mem_frag, Intrinsic IntId> {
6775 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6776 (ins RC:$src1, RC:$src2, RC:$src3),
6777 !strconcat(OpcodeStr,
6778 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6779 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6780 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6782 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6783 (ins RC:$src1, x86memop:$src2, RC:$src3),
6784 !strconcat(OpcodeStr,
6785 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6787 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6789 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6792 let Predicates = [HasAVX] in {
6793 let ExeDomain = SSEPackedDouble in {
6794 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6795 memopv2f64, int_x86_sse41_blendvpd>;
6796 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6797 memopv4f64, int_x86_avx_blendv_pd_256>;
6798 } // ExeDomain = SSEPackedDouble
6799 let ExeDomain = SSEPackedSingle in {
6800 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6801 memopv4f32, int_x86_sse41_blendvps>;
6802 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6803 memopv8f32, int_x86_avx_blendv_ps_256>;
6804 } // ExeDomain = SSEPackedSingle
6805 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6806 memopv2i64, int_x86_sse41_pblendvb>;
6809 let Predicates = [HasAVX2] in {
6810 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6811 memopv4i64, int_x86_avx2_pblendvb>;
6814 let Predicates = [HasAVX] in {
6815 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6816 (v16i8 VR128:$src2))),
6817 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6818 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6819 (v4i32 VR128:$src2))),
6820 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6821 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6822 (v4f32 VR128:$src2))),
6823 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6824 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6825 (v2i64 VR128:$src2))),
6826 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6827 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6828 (v2f64 VR128:$src2))),
6829 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6830 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6831 (v8i32 VR256:$src2))),
6832 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6833 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6834 (v8f32 VR256:$src2))),
6835 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6836 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6837 (v4i64 VR256:$src2))),
6838 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6839 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6840 (v4f64 VR256:$src2))),
6841 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6844 let Predicates = [HasAVX2] in {
6845 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6846 (v32i8 VR256:$src2))),
6847 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6850 /// SS41I_ternary_int - SSE 4.1 ternary operator
6851 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6852 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6854 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6855 (ins VR128:$src1, VR128:$src2),
6856 !strconcat(OpcodeStr,
6857 "\t{$src2, $dst|$dst, $src2}"),
6858 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6861 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6862 (ins VR128:$src1, i128mem:$src2),
6863 !strconcat(OpcodeStr,
6864 "\t{$src2, $dst|$dst, $src2}"),
6867 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6871 let ExeDomain = SSEPackedDouble in
6872 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6873 int_x86_sse41_blendvpd>;
6874 let ExeDomain = SSEPackedSingle in
6875 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6876 int_x86_sse41_blendvps>;
6877 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6878 int_x86_sse41_pblendvb>;
6880 let Predicates = [HasSSE41] in {
6881 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6882 (v16i8 VR128:$src2))),
6883 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6884 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6885 (v4i32 VR128:$src2))),
6886 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6887 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6888 (v4f32 VR128:$src2))),
6889 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6890 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6891 (v2i64 VR128:$src2))),
6892 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6893 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6894 (v2f64 VR128:$src2))),
6895 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6898 let Predicates = [HasAVX] in
6899 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6900 "vmovntdqa\t{$src, $dst|$dst, $src}",
6901 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6903 let Predicates = [HasAVX2] in
6904 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6905 "vmovntdqa\t{$src, $dst|$dst, $src}",
6906 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6908 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6909 "movntdqa\t{$src, $dst|$dst, $src}",
6910 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6913 //===----------------------------------------------------------------------===//
6914 // SSE4.2 - Compare Instructions
6915 //===----------------------------------------------------------------------===//
6917 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6918 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6919 Intrinsic IntId128, bit Is2Addr = 1> {
6920 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6921 (ins VR128:$src1, VR128:$src2),
6923 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6924 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6925 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6927 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6928 (ins VR128:$src1, i128mem:$src2),
6930 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6931 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6933 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6936 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6937 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6938 Intrinsic IntId256> {
6939 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6940 (ins VR256:$src1, VR256:$src2),
6941 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6942 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6944 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6945 (ins VR256:$src1, i256mem:$src2),
6946 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6948 (IntId256 VR256:$src1, (memopv4i64 addr:$src2)))]>, OpSize;
6951 let Predicates = [HasAVX] in {
6952 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6955 def : Pat<(v2i64 (X86pcmpgt VR128:$src1, VR128:$src2)),
6956 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6957 def : Pat<(v2i64 (X86pcmpgt VR128:$src1, (memop addr:$src2))),
6958 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6961 let Predicates = [HasAVX2] in {
6962 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6965 def : Pat<(v4i64 (X86pcmpgt VR256:$src1, VR256:$src2)),
6966 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6967 def : Pat<(v4i64 (X86pcmpgt VR256:$src1, (memop addr:$src2))),
6968 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6971 let Constraints = "$src1 = $dst" in
6972 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6974 let Predicates = [HasSSE42] in {
6975 def : Pat<(v2i64 (X86pcmpgt VR128:$src1, VR128:$src2)),
6976 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6977 def : Pat<(v2i64 (X86pcmpgt VR128:$src1, (memop addr:$src2))),
6978 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6981 //===----------------------------------------------------------------------===//
6982 // SSE4.2 - String/text Processing Instructions
6983 //===----------------------------------------------------------------------===//
6985 // Packed Compare Implicit Length Strings, Return Mask
6986 multiclass pseudo_pcmpistrm<string asm> {
6987 def REG : PseudoI<(outs VR128:$dst),
6988 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6989 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6991 def MEM : PseudoI<(outs VR128:$dst),
6992 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6993 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6994 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6997 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6998 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6999 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
7002 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
7003 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
7004 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7005 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
7007 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
7008 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7009 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
7012 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
7013 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
7014 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7015 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
7017 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
7018 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7019 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
7022 // Packed Compare Explicit Length Strings, Return Mask
7023 multiclass pseudo_pcmpestrm<string asm> {
7024 def REG : PseudoI<(outs VR128:$dst),
7025 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7026 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
7027 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7028 def MEM : PseudoI<(outs VR128:$dst),
7029 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7030 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
7031 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
7034 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7035 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
7036 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
7039 let Predicates = [HasAVX],
7040 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7041 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
7042 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7043 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
7045 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
7046 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7047 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
7050 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7051 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
7052 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7053 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
7055 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
7056 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7057 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
7060 // Packed Compare Implicit Length Strings, Return Index
7061 let Defs = [ECX, EFLAGS] in {
7062 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
7063 def rr : SS42AI<0x63, MRMSrcReg, (outs),
7064 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7065 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7066 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
7067 (implicit EFLAGS)]>, OpSize;
7068 def rm : SS42AI<0x63, MRMSrcMem, (outs),
7069 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7070 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7071 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
7072 (implicit EFLAGS)]>, OpSize;
7076 let Predicates = [HasAVX] in {
7077 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
7079 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
7081 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
7083 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
7085 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
7087 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
7091 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
7092 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
7093 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
7094 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
7095 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
7096 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
7098 // Packed Compare Explicit Length Strings, Return Index
7099 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
7100 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
7101 def rr : SS42AI<0x61, MRMSrcReg, (outs),
7102 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7103 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7104 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
7105 (implicit EFLAGS)]>, OpSize;
7106 def rm : SS42AI<0x61, MRMSrcMem, (outs),
7107 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7108 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7110 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
7111 (implicit EFLAGS)]>, OpSize;
7115 let Predicates = [HasAVX] in {
7116 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
7118 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
7120 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
7122 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
7124 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
7126 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
7130 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
7131 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
7132 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
7133 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
7134 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
7135 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
7137 //===----------------------------------------------------------------------===//
7138 // SSE4.2 - CRC Instructions
7139 //===----------------------------------------------------------------------===//
7141 // No CRC instructions have AVX equivalents
7143 // crc intrinsic instruction
7144 // This set of instructions are only rm, the only difference is the size
7146 let Constraints = "$src1 = $dst" in {
7147 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
7148 (ins GR32:$src1, i8mem:$src2),
7149 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7151 (int_x86_sse42_crc32_32_8 GR32:$src1,
7152 (load addr:$src2)))]>;
7153 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
7154 (ins GR32:$src1, GR8:$src2),
7155 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7157 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
7158 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7159 (ins GR32:$src1, i16mem:$src2),
7160 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7162 (int_x86_sse42_crc32_32_16 GR32:$src1,
7163 (load addr:$src2)))]>,
7165 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7166 (ins GR32:$src1, GR16:$src2),
7167 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7169 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7171 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7172 (ins GR32:$src1, i32mem:$src2),
7173 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7175 (int_x86_sse42_crc32_32_32 GR32:$src1,
7176 (load addr:$src2)))]>;
7177 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7178 (ins GR32:$src1, GR32:$src2),
7179 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7181 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7182 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7183 (ins GR64:$src1, i8mem:$src2),
7184 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7186 (int_x86_sse42_crc32_64_8 GR64:$src1,
7187 (load addr:$src2)))]>,
7189 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7190 (ins GR64:$src1, GR8:$src2),
7191 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7193 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7195 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7196 (ins GR64:$src1, i64mem:$src2),
7197 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7199 (int_x86_sse42_crc32_64_64 GR64:$src1,
7200 (load addr:$src2)))]>,
7202 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7203 (ins GR64:$src1, GR64:$src2),
7204 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7206 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7210 //===----------------------------------------------------------------------===//
7211 // AES-NI Instructions
7212 //===----------------------------------------------------------------------===//
7214 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7215 Intrinsic IntId128, bit Is2Addr = 1> {
7216 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7217 (ins VR128:$src1, VR128:$src2),
7219 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7220 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7221 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7223 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7224 (ins VR128:$src1, i128mem:$src2),
7226 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7227 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7229 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7232 // Perform One Round of an AES Encryption/Decryption Flow
7233 let Predicates = [HasAVX, HasAES] in {
7234 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7235 int_x86_aesni_aesenc, 0>, VEX_4V;
7236 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7237 int_x86_aesni_aesenclast, 0>, VEX_4V;
7238 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7239 int_x86_aesni_aesdec, 0>, VEX_4V;
7240 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7241 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7244 let Constraints = "$src1 = $dst" in {
7245 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7246 int_x86_aesni_aesenc>;
7247 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7248 int_x86_aesni_aesenclast>;
7249 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7250 int_x86_aesni_aesdec>;
7251 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7252 int_x86_aesni_aesdeclast>;
7255 // Perform the AES InvMixColumn Transformation
7256 let Predicates = [HasAVX, HasAES] in {
7257 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7259 "vaesimc\t{$src1, $dst|$dst, $src1}",
7261 (int_x86_aesni_aesimc VR128:$src1))]>,
7263 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7264 (ins i128mem:$src1),
7265 "vaesimc\t{$src1, $dst|$dst, $src1}",
7266 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7269 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7271 "aesimc\t{$src1, $dst|$dst, $src1}",
7273 (int_x86_aesni_aesimc VR128:$src1))]>,
7275 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7276 (ins i128mem:$src1),
7277 "aesimc\t{$src1, $dst|$dst, $src1}",
7278 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7281 // AES Round Key Generation Assist
7282 let Predicates = [HasAVX, HasAES] in {
7283 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7284 (ins VR128:$src1, i8imm:$src2),
7285 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7287 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7289 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7290 (ins i128mem:$src1, i8imm:$src2),
7291 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7293 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7296 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7297 (ins VR128:$src1, i8imm:$src2),
7298 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7300 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7302 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7303 (ins i128mem:$src1, i8imm:$src2),
7304 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7306 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7309 //===----------------------------------------------------------------------===//
7310 // CLMUL Instructions
7311 //===----------------------------------------------------------------------===//
7313 // Carry-less Multiplication instructions
7314 let neverHasSideEffects = 1 in {
7315 // AVX carry-less Multiplication instructions
7316 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7317 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7318 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7322 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7323 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7324 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7327 let Constraints = "$src1 = $dst" in {
7328 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7329 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7330 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7334 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7335 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7336 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7338 } // Constraints = "$src1 = $dst"
7339 } // neverHasSideEffects = 1
7342 multiclass pclmul_alias<string asm, int immop> {
7343 def : InstAlias<!strconcat("pclmul", asm,
7344 "dq {$src, $dst|$dst, $src}"),
7345 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7347 def : InstAlias<!strconcat("pclmul", asm,
7348 "dq {$src, $dst|$dst, $src}"),
7349 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7351 def : InstAlias<!strconcat("vpclmul", asm,
7352 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7353 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7355 def : InstAlias<!strconcat("vpclmul", asm,
7356 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7357 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7359 defm : pclmul_alias<"hqhq", 0x11>;
7360 defm : pclmul_alias<"hqlq", 0x01>;
7361 defm : pclmul_alias<"lqhq", 0x10>;
7362 defm : pclmul_alias<"lqlq", 0x00>;
7364 //===----------------------------------------------------------------------===//
7366 //===----------------------------------------------------------------------===//
7368 //===----------------------------------------------------------------------===//
7369 // VBROADCAST - Load from memory and broadcast to all elements of the
7370 // destination operand
7372 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7373 X86MemOperand x86memop, Intrinsic Int> :
7374 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7375 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7376 [(set RC:$dst, (Int addr:$src))]>, VEX;
7378 // AVX2 adds register forms
7379 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7381 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7382 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7383 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7385 let ExeDomain = SSEPackedSingle in {
7386 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7387 int_x86_avx_vbroadcast_ss>;
7388 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7389 int_x86_avx_vbroadcast_ss_256>;
7391 let ExeDomain = SSEPackedDouble in
7392 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7393 int_x86_avx_vbroadcast_sd_256>;
7394 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7395 int_x86_avx_vbroadcastf128_pd_256>;
7397 let ExeDomain = SSEPackedSingle in {
7398 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7399 int_x86_avx2_vbroadcast_ss_ps>;
7400 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7401 int_x86_avx2_vbroadcast_ss_ps_256>;
7403 let ExeDomain = SSEPackedDouble in
7404 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7405 int_x86_avx2_vbroadcast_sd_pd_256>;
7407 let Predicates = [HasAVX2] in
7408 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7409 int_x86_avx2_vbroadcasti128>;
7411 let Predicates = [HasAVX] in
7412 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7413 (VBROADCASTF128 addr:$src)>;
7416 //===----------------------------------------------------------------------===//
7417 // VINSERTF128 - Insert packed floating-point values
7419 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7420 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7421 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7422 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7425 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7426 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7427 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7431 let Predicates = [HasAVX] in {
7432 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7433 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7434 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7435 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7436 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7437 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7440 //===----------------------------------------------------------------------===//
7441 // VEXTRACTF128 - Extract packed floating-point values
7443 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7444 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7445 (ins VR256:$src1, i8imm:$src2),
7446 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7449 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7450 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7451 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7455 let Predicates = [HasAVX] in {
7456 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7457 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7458 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7459 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7460 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7461 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7464 //===----------------------------------------------------------------------===//
7465 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7467 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7468 Intrinsic IntLd, Intrinsic IntLd256,
7469 Intrinsic IntSt, Intrinsic IntSt256> {
7470 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7471 (ins VR128:$src1, f128mem:$src2),
7472 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7473 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7475 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7476 (ins VR256:$src1, f256mem:$src2),
7477 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7478 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7480 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7481 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7482 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7483 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7484 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7485 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7486 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7487 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7490 let ExeDomain = SSEPackedSingle in
7491 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7492 int_x86_avx_maskload_ps,
7493 int_x86_avx_maskload_ps_256,
7494 int_x86_avx_maskstore_ps,
7495 int_x86_avx_maskstore_ps_256>;
7496 let ExeDomain = SSEPackedDouble in
7497 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7498 int_x86_avx_maskload_pd,
7499 int_x86_avx_maskload_pd_256,
7500 int_x86_avx_maskstore_pd,
7501 int_x86_avx_maskstore_pd_256>;
7503 //===----------------------------------------------------------------------===//
7504 // VPERMIL - Permute Single and Double Floating-Point Values
7506 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7507 RegisterClass RC, X86MemOperand x86memop_f,
7508 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7509 Intrinsic IntVar, Intrinsic IntImm> {
7510 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7511 (ins RC:$src1, RC:$src2),
7512 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7513 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7514 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7515 (ins RC:$src1, x86memop_i:$src2),
7516 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7517 [(set RC:$dst, (IntVar RC:$src1,
7518 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7520 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7521 (ins RC:$src1, i8imm:$src2),
7522 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7523 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7524 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7525 (ins x86memop_f:$src1, i8imm:$src2),
7526 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7527 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7530 let ExeDomain = SSEPackedSingle in {
7531 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7532 memopv4f32, memopv2i64,
7533 int_x86_avx_vpermilvar_ps,
7534 int_x86_avx_vpermil_ps>;
7535 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7536 memopv8f32, memopv4i64,
7537 int_x86_avx_vpermilvar_ps_256,
7538 int_x86_avx_vpermil_ps_256>;
7540 let ExeDomain = SSEPackedDouble in {
7541 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7542 memopv2f64, memopv2i64,
7543 int_x86_avx_vpermilvar_pd,
7544 int_x86_avx_vpermil_pd>;
7545 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7546 memopv4f64, memopv4i64,
7547 int_x86_avx_vpermilvar_pd_256,
7548 int_x86_avx_vpermil_pd_256>;
7551 let Predicates = [HasAVX] in {
7552 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7553 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7554 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7555 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7556 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7557 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7558 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7559 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7560 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7561 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7562 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7563 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7564 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7566 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7567 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7568 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7571 //===----------------------------------------------------------------------===//
7572 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7574 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7575 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7576 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7577 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7580 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7581 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7582 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7586 let Predicates = [HasAVX] in {
7587 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7588 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7589 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7590 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7591 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7592 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7594 def : Pat<(int_x86_avx_vperm2f128_ps_256
7595 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7596 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7597 def : Pat<(int_x86_avx_vperm2f128_pd_256
7598 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7599 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7600 def : Pat<(int_x86_avx_vperm2f128_si_256
7601 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7602 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7605 //===----------------------------------------------------------------------===//
7606 // VZERO - Zero YMM registers
7608 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7609 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7610 // Zero All YMM registers
7611 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7612 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7614 // Zero Upper bits of YMM registers
7615 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7616 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7619 //===----------------------------------------------------------------------===//
7620 // Half precision conversion instructions
7621 //===----------------------------------------------------------------------===//
7622 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7623 let Predicates = [HasAVX, HasF16C] in {
7624 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7625 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7626 [(set RC:$dst, (Int VR128:$src))]>,
7628 let neverHasSideEffects = 1, mayLoad = 1 in
7629 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7630 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7634 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7635 let Predicates = [HasAVX, HasF16C] in {
7636 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7637 (ins RC:$src1, i32i8imm:$src2),
7638 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7639 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7641 let neverHasSideEffects = 1, mayLoad = 1 in
7642 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7643 (ins RC:$src1, i32i8imm:$src2),
7644 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7649 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7650 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7651 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7652 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7654 //===----------------------------------------------------------------------===//
7655 // AVX2 Instructions
7656 //===----------------------------------------------------------------------===//
7658 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7659 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7660 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7661 X86MemOperand x86memop> {
7662 let isCommutable = 1 in
7663 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7664 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7665 !strconcat(OpcodeStr,
7666 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7667 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7669 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7670 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7671 !strconcat(OpcodeStr,
7672 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7675 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7679 let isCommutable = 0 in {
7680 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7681 VR128, memopv2i64, i128mem>;
7682 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7683 VR256, memopv4i64, i256mem>;
7686 //===----------------------------------------------------------------------===//
7687 // VPBROADCAST - Load from memory and broadcast to all elements of the
7688 // destination operand
7690 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7691 X86MemOperand x86memop, PatFrag ld_frag,
7692 Intrinsic Int128, Intrinsic Int256> {
7693 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7694 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7695 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7696 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7697 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7699 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7700 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7701 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7702 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7703 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7704 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7706 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7709 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7710 int_x86_avx2_pbroadcastb_128,
7711 int_x86_avx2_pbroadcastb_256>;
7712 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7713 int_x86_avx2_pbroadcastw_128,
7714 int_x86_avx2_pbroadcastw_256>;
7715 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7716 int_x86_avx2_pbroadcastd_128,
7717 int_x86_avx2_pbroadcastd_256>;
7718 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7719 int_x86_avx2_pbroadcastq_128,
7720 int_x86_avx2_pbroadcastq_256>;
7722 let Predicates = [HasAVX2] in {
7723 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7724 (VPBROADCASTBrm addr:$src)>;
7725 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7726 (VPBROADCASTBYrm addr:$src)>;
7727 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7728 (VPBROADCASTWrm addr:$src)>;
7729 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7730 (VPBROADCASTWYrm addr:$src)>;
7731 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7732 (VPBROADCASTDrm addr:$src)>;
7733 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7734 (VPBROADCASTDYrm addr:$src)>;
7735 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7736 (VPBROADCASTQrm addr:$src)>;
7737 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7738 (VPBROADCASTQYrm addr:$src)>;
7741 // AVX1 broadcast patterns
7742 let Predicates = [HasAVX] in {
7743 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7744 (VBROADCASTSSYrm addr:$src)>;
7745 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7746 (VBROADCASTSDrm addr:$src)>;
7747 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7748 (VBROADCASTSSYrm addr:$src)>;
7749 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7750 (VBROADCASTSDrm addr:$src)>;
7752 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7753 (VBROADCASTSSrm addr:$src)>;
7754 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7755 (VBROADCASTSSrm addr:$src)>;
7758 //===----------------------------------------------------------------------===//
7759 // VPERM - Permute instructions
7762 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7764 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7765 (ins VR256:$src1, VR256:$src2),
7766 !strconcat(OpcodeStr,
7767 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7768 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7769 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7770 (ins VR256:$src1, i256mem:$src2),
7771 !strconcat(OpcodeStr,
7772 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7773 [(set VR256:$dst, (Int VR256:$src1,
7774 (bitconvert (mem_frag addr:$src2))))]>,
7778 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7779 let ExeDomain = SSEPackedSingle in
7780 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7782 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7784 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7785 (ins VR256:$src1, i8imm:$src2),
7786 !strconcat(OpcodeStr,
7787 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7788 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7789 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7790 (ins i256mem:$src1, i8imm:$src2),
7791 !strconcat(OpcodeStr,
7792 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7793 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7797 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7799 let ExeDomain = SSEPackedDouble in
7800 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7803 //===----------------------------------------------------------------------===//
7804 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7806 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7807 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7808 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7810 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7812 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7813 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7814 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7816 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7820 let Predicates = [HasAVX2] in {
7821 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7822 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7823 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7824 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7825 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7826 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7827 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7828 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7830 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7832 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7833 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7834 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7835 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7836 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7838 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7839 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7841 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7845 let Predicates = [HasAVX] in {
7846 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7847 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7848 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7849 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7850 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7851 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7852 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7853 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7854 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7855 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7856 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7857 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7859 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7860 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7861 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7862 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7863 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7864 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7865 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7866 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7867 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7868 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7869 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7870 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7871 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7872 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7873 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7874 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7875 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7876 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7880 //===----------------------------------------------------------------------===//
7881 // VINSERTI128 - Insert packed integer values
7883 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7884 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7885 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7887 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7889 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7890 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7891 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7893 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7894 imm:$src3))]>, VEX_4V;
7896 let Predicates = [HasAVX2] in {
7897 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7899 (VINSERTI128rr VR256:$src1, VR128:$src2,
7900 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7901 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7903 (VINSERTI128rr VR256:$src1, VR128:$src2,
7904 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7905 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7907 (VINSERTI128rr VR256:$src1, VR128:$src2,
7908 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7909 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7911 (VINSERTI128rr VR256:$src1, VR128:$src2,
7912 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7916 let Predicates = [HasAVX] in {
7917 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7919 (VINSERTF128rr VR256:$src1, VR128:$src2,
7920 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7921 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7923 (VINSERTF128rr VR256:$src1, VR128:$src2,
7924 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7925 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7927 (VINSERTF128rr VR256:$src1, VR128:$src2,
7928 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7929 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7931 (VINSERTF128rr VR256:$src1, VR128:$src2,
7932 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7933 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7935 (VINSERTF128rr VR256:$src1, VR128:$src2,
7936 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7937 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7939 (VINSERTF128rr VR256:$src1, VR128:$src2,
7940 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7943 //===----------------------------------------------------------------------===//
7944 // VEXTRACTI128 - Extract packed integer values
7946 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7947 (ins VR256:$src1, i8imm:$src2),
7948 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7950 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7952 let neverHasSideEffects = 1, mayStore = 1 in
7953 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7954 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7955 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7957 let Predicates = [HasAVX2] in {
7958 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7959 (v2i64 (VEXTRACTI128rr
7960 (v4i64 VR256:$src1),
7961 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7962 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7963 (v4i32 (VEXTRACTI128rr
7964 (v8i32 VR256:$src1),
7965 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7966 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7967 (v8i16 (VEXTRACTI128rr
7968 (v16i16 VR256:$src1),
7969 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7970 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7971 (v16i8 (VEXTRACTI128rr
7972 (v32i8 VR256:$src1),
7973 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7977 let Predicates = [HasAVX] in {
7978 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7979 (v4f32 (VEXTRACTF128rr
7980 (v8f32 VR256:$src1),
7981 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7982 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7983 (v2f64 (VEXTRACTF128rr
7984 (v4f64 VR256:$src1),
7985 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7986 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7987 (v2i64 (VEXTRACTF128rr
7988 (v4i64 VR256:$src1),
7989 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7990 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7991 (v4i32 (VEXTRACTF128rr
7992 (v8i32 VR256:$src1),
7993 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7994 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7995 (v8i16 (VEXTRACTF128rr
7996 (v16i16 VR256:$src1),
7997 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7998 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7999 (v16i8 (VEXTRACTF128rr
8000 (v32i8 VR256:$src1),
8001 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
8004 //===----------------------------------------------------------------------===//
8005 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
8007 multiclass avx2_pmovmask<string OpcodeStr,
8008 Intrinsic IntLd128, Intrinsic IntLd256,
8009 Intrinsic IntSt128, Intrinsic IntSt256> {
8010 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
8011 (ins VR128:$src1, i128mem:$src2),
8012 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8013 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
8014 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
8015 (ins VR256:$src1, i256mem:$src2),
8016 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8017 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
8018 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8019 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8020 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8021 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8022 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8023 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8024 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8025 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
8028 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8029 int_x86_avx2_maskload_d,
8030 int_x86_avx2_maskload_d_256,
8031 int_x86_avx2_maskstore_d,
8032 int_x86_avx2_maskstore_d_256>;
8033 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8034 int_x86_avx2_maskload_q,
8035 int_x86_avx2_maskload_q_256,
8036 int_x86_avx2_maskstore_q,
8037 int_x86_avx2_maskstore_q_256>, VEX_W;
8040 //===----------------------------------------------------------------------===//
8041 // Variable Bit Shifts
8043 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8044 ValueType vt128, ValueType vt256> {
8045 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8046 (ins VR128:$src1, VR128:$src2),
8047 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8049 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8051 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8052 (ins VR128:$src1, i128mem:$src2),
8053 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8055 (vt128 (OpNode VR128:$src1,
8056 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8058 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8059 (ins VR256:$src1, VR256:$src2),
8060 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8062 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8064 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8065 (ins VR256:$src1, i256mem:$src2),
8066 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8068 (vt256 (OpNode VR256:$src1,
8069 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8073 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8074 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8075 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8076 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8077 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;