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:$src)))]>;
1415 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1416 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1417 string asm, Domain d> {
1418 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1419 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1420 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1421 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1424 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1425 X86MemOperand x86memop, string asm> {
1426 let neverHasSideEffects = 1 in {
1427 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1428 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1430 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1431 (ins DstRC:$src1, x86memop:$src),
1432 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1433 } // neverHasSideEffects = 1
1436 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1437 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1439 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1440 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1442 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1443 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1445 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1446 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1447 VEX, VEX_W, VEX_LIG;
1449 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1450 // register, but the same isn't true when only using memory operands,
1451 // provide other assembly "l" and "q" forms to address this explicitly
1452 // where appropriate to do so.
1453 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1455 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1456 VEX_4V, VEX_W, VEX_LIG;
1457 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1459 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1461 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1462 VEX_4V, VEX_W, VEX_LIG;
1464 let Predicates = [HasAVX], AddedComplexity = 1 in {
1465 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1466 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1467 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1468 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1469 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1470 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1471 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1472 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1474 def : Pat<(f32 (sint_to_fp GR32:$src)),
1475 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1476 def : Pat<(f32 (sint_to_fp GR64:$src)),
1477 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1478 def : Pat<(f64 (sint_to_fp GR32:$src)),
1479 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1480 def : Pat<(f64 (sint_to_fp GR64:$src)),
1481 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1484 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1485 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1486 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1487 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1488 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1489 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1490 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1491 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1492 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1493 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1494 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1495 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1496 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1497 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1498 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1499 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1501 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1502 // and/or XMM operand(s).
1504 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1505 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1507 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1508 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1509 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1510 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1511 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1512 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1515 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1516 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1517 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1518 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1520 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1521 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1522 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1523 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1524 (ins DstRC:$src1, x86memop:$src2),
1526 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1527 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1528 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1531 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1532 f128mem, load, "cvtsd2si">, XD, VEX, VEX_LIG;
1533 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1534 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1535 XD, VEX, VEX_W, VEX_LIG;
1537 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1538 f128mem, load, "cvtsd2si{l}">, XD;
1539 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1540 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1543 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1544 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1545 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1546 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1548 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1549 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1550 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1551 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1554 let Constraints = "$src1 = $dst" in {
1555 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1556 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1558 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1559 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1560 "cvtsi2ss{q}">, XS, REX_W;
1561 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1562 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1564 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1565 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1566 "cvtsi2sd">, XD, REX_W;
1571 // Aliases for intrinsics
1572 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1573 f32mem, load, "cvttss2si">, XS, VEX;
1574 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1575 int_x86_sse_cvttss2si64, f32mem, load,
1576 "cvttss2si">, XS, VEX, VEX_W;
1577 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1578 f128mem, load, "cvttsd2si">, XD, VEX;
1579 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1580 int_x86_sse2_cvttsd2si64, f128mem, load,
1581 "cvttsd2si">, XD, VEX, VEX_W;
1582 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1583 f32mem, load, "cvttss2si">, XS;
1584 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1585 int_x86_sse_cvttss2si64, f32mem, load,
1586 "cvttss2si{q}">, XS, REX_W;
1587 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1588 f128mem, load, "cvttsd2si">, XD;
1589 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1590 int_x86_sse2_cvttsd2si64, f128mem, load,
1591 "cvttsd2si{q}">, XD, REX_W;
1593 let Pattern = []<dag> in {
1594 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1595 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1597 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1598 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1600 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1601 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1602 SSEPackedSingle>, TB, VEX;
1603 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1604 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1605 SSEPackedSingle>, TB, VEX;
1608 let Pattern = []<dag> in {
1609 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1610 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1611 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1612 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1613 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1614 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1615 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1618 let Predicates = [HasAVX] in {
1619 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1620 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1621 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1622 (VCVTSS2SIrm addr:$src)>;
1623 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1624 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1625 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1626 (VCVTSS2SI64rm addr:$src)>;
1629 let Predicates = [HasSSE1] in {
1630 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1631 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1632 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1633 (CVTSS2SIrm addr:$src)>;
1634 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1635 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1636 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1637 (CVTSS2SI64rm addr:$src)>;
1642 // Convert scalar double to scalar single
1643 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1644 (ins FR64:$src1, FR64:$src2),
1645 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1648 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1649 (ins FR64:$src1, f64mem:$src2),
1650 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1651 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1653 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1656 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1657 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1658 [(set FR32:$dst, (fround FR64:$src))]>;
1659 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1660 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1661 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1662 Requires<[HasSSE2, OptForSize]>;
1664 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1665 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1667 let Constraints = "$src1 = $dst" in
1668 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1669 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1671 // Convert scalar single to scalar double
1672 // SSE2 instructions with XS prefix
1673 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1674 (ins FR32:$src1, FR32:$src2),
1675 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1676 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1678 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1679 (ins FR32:$src1, f32mem:$src2),
1680 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1681 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1683 let Predicates = [HasAVX] in {
1684 def : Pat<(f64 (fextend FR32:$src)),
1685 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1686 def : Pat<(fextend (loadf32 addr:$src)),
1687 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1688 def : Pat<(extloadf32 addr:$src),
1689 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1692 def : Pat<(extloadf32 addr:$src),
1693 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1694 Requires<[HasAVX, OptForSpeed]>;
1696 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1697 "cvtss2sd\t{$src, $dst|$dst, $src}",
1698 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1699 Requires<[HasSSE2]>;
1700 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1701 "cvtss2sd\t{$src, $dst|$dst, $src}",
1702 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1703 Requires<[HasSSE2, OptForSize]>;
1705 // extload f32 -> f64. This matches load+fextend because we have a hack in
1706 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1708 // Since these loads aren't folded into the fextend, we have to match it
1710 def : Pat<(fextend (loadf32 addr:$src)),
1711 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1712 def : Pat<(extloadf32 addr:$src),
1713 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1715 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1716 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1717 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1718 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1719 VR128:$src2))]>, XS, VEX_4V,
1721 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1722 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1723 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1724 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1725 (load addr:$src2)))]>, XS, VEX_4V,
1727 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1728 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1729 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1730 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1731 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1732 VR128:$src2))]>, XS,
1733 Requires<[HasSSE2]>;
1734 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1735 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1736 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1737 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1738 (load addr:$src2)))]>, XS,
1739 Requires<[HasSSE2]>;
1742 // Convert doubleword to packed single/double fp
1743 // SSE2 instructions without OpSize prefix
1744 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1745 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1746 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1747 TB, VEX, Requires<[HasAVX]>;
1748 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1749 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1750 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1751 (bitconvert (memopv2i64 addr:$src))))]>,
1752 TB, VEX, Requires<[HasAVX]>;
1753 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1754 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1755 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1756 TB, Requires<[HasSSE2]>;
1757 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1758 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1759 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1760 (bitconvert (memopv2i64 addr:$src))))]>,
1761 TB, Requires<[HasSSE2]>;
1763 // FIXME: why the non-intrinsic version is described as SSE3?
1764 // SSE2 instructions with XS prefix
1765 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1766 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1767 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1768 XS, VEX, Requires<[HasAVX]>;
1769 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1770 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1771 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1772 (bitconvert (memopv2i64 addr:$src))))]>,
1773 XS, VEX, Requires<[HasAVX]>;
1774 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1775 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1776 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1777 XS, Requires<[HasSSE2]>;
1778 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1779 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1780 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1781 (bitconvert (memopv2i64 addr:$src))))]>,
1782 XS, Requires<[HasSSE2]>;
1785 // Convert packed single/double fp to doubleword
1786 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1787 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1788 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1789 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1790 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1791 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1792 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1793 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1794 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1795 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1796 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1797 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1799 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1800 "cvtps2dq\t{$src, $dst|$dst, $src}",
1801 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1803 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1805 "cvtps2dq\t{$src, $dst|$dst, $src}",
1806 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1807 (memop addr:$src)))]>, VEX;
1808 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1809 "cvtps2dq\t{$src, $dst|$dst, $src}",
1810 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1811 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1812 "cvtps2dq\t{$src, $dst|$dst, $src}",
1813 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1814 (memop addr:$src)))]>;
1816 // SSE2 packed instructions with XD prefix
1817 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1818 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1819 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1820 XD, VEX, Requires<[HasAVX]>;
1821 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1822 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1823 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1824 (memop addr:$src)))]>,
1825 XD, VEX, Requires<[HasAVX]>;
1826 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1827 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1828 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1829 XD, Requires<[HasSSE2]>;
1830 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1831 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1832 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1833 (memop addr:$src)))]>,
1834 XD, Requires<[HasSSE2]>;
1837 // Convert with truncation packed single/double fp to doubleword
1838 // SSE2 packed instructions with XS prefix
1839 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1840 "cvttps2dq\t{$src, $dst|$dst, $src}",
1842 (int_x86_sse2_cvttps2dq VR128:$src))]>, VEX;
1843 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1844 "cvttps2dq\t{$src, $dst|$dst, $src}",
1845 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1846 (memop addr:$src)))]>, VEX;
1847 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1848 "cvttps2dq\t{$src, $dst|$dst, $src}",
1850 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))]>, VEX;
1851 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1852 "cvttps2dq\t{$src, $dst|$dst, $src}",
1853 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1854 (memopv8f32 addr:$src)))]>, VEX;
1856 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1857 "cvttps2dq\t{$src, $dst|$dst, $src}",
1859 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1860 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1861 "cvttps2dq\t{$src, $dst|$dst, $src}",
1863 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1865 let Predicates = [HasAVX] in {
1866 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1867 (Int_VCVTDQ2PSrr VR128:$src)>;
1868 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1869 (Int_VCVTDQ2PSrm addr:$src)>;
1871 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1872 (VCVTTPS2DQrr VR128:$src)>;
1873 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1874 (VCVTTPS2DQrm addr:$src)>;
1876 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1877 (VCVTDQ2PSYrr VR256:$src)>;
1878 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1879 (VCVTDQ2PSYrm addr:$src)>;
1881 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1882 (VCVTTPS2DQYrr VR256:$src)>;
1883 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1884 (VCVTTPS2DQYrm addr:$src)>;
1887 let Predicates = [HasSSE2] in {
1888 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1889 (Int_CVTDQ2PSrr VR128:$src)>;
1890 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1891 (Int_CVTDQ2PSrm addr:$src)>;
1893 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1894 (CVTTPS2DQrr VR128:$src)>;
1895 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1896 (CVTTPS2DQrm addr:$src)>;
1899 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1900 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1902 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1903 let isCodeGenOnly = 1 in
1904 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1905 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1906 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1907 (memop addr:$src)))]>, VEX;
1908 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1909 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1910 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1911 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1912 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1913 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1914 (memop addr:$src)))]>;
1916 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1917 // register, but the same isn't true when using memory operands instead.
1918 // Provide other assembly rr and rm forms to address this explicitly.
1919 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1920 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1923 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1924 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1925 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1926 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1929 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1930 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1931 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1932 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1934 // Convert packed single to packed double
1935 let Predicates = [HasAVX] in {
1936 // SSE2 instructions without OpSize prefix
1937 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1938 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1939 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1940 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1941 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1942 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1943 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1944 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1946 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1947 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1948 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1949 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1951 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1952 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1953 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1954 TB, VEX, Requires<[HasAVX]>;
1955 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1956 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1957 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1958 (load addr:$src)))]>,
1959 TB, VEX, Requires<[HasAVX]>;
1960 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1961 "cvtps2pd\t{$src, $dst|$dst, $src}",
1962 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1963 TB, Requires<[HasSSE2]>;
1964 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1965 "cvtps2pd\t{$src, $dst|$dst, $src}",
1966 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1967 (load addr:$src)))]>,
1968 TB, Requires<[HasSSE2]>;
1970 // Convert packed double to packed single
1971 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1972 // register, but the same isn't true when using memory operands instead.
1973 // Provide other assembly rr and rm forms to address this explicitly.
1974 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1975 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1976 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1977 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1980 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1981 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1982 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1983 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1986 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1987 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1988 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1989 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1990 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1991 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1992 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1993 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1996 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1997 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1998 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1999 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
2001 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2002 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2003 (memop addr:$src)))]>;
2004 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2005 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2006 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2007 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2008 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2009 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2010 (memop addr:$src)))]>;
2012 // AVX 256-bit register conversion intrinsics
2013 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2014 // whenever possible to avoid declaring two versions of each one.
2015 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2016 (VCVTDQ2PSYrr VR256:$src)>;
2017 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2018 (VCVTDQ2PSYrm addr:$src)>;
2020 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2021 (VCVTPD2PSYrr VR256:$src)>;
2022 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2023 (VCVTPD2PSYrm addr:$src)>;
2025 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2026 (VCVTPS2DQYrr VR256:$src)>;
2027 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2028 (VCVTPS2DQYrm addr:$src)>;
2030 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2031 (VCVTPS2PDYrr VR128:$src)>;
2032 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2033 (VCVTPS2PDYrm addr:$src)>;
2035 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2036 (VCVTTPD2DQYrr VR256:$src)>;
2037 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2038 (VCVTTPD2DQYrm addr:$src)>;
2040 // Match fround and fextend for 128/256-bit conversions
2041 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2042 (VCVTPD2PSYrr VR256:$src)>;
2043 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2044 (VCVTPD2PSYrm addr:$src)>;
2046 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2047 (VCVTPS2PDYrr VR128:$src)>;
2048 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2049 (VCVTPS2PDYrm addr:$src)>;
2051 //===----------------------------------------------------------------------===//
2052 // SSE 1 & 2 - Compare Instructions
2053 //===----------------------------------------------------------------------===//
2055 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2056 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2057 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2058 string asm, string asm_alt> {
2059 def rr : SIi8<0xC2, MRMSrcReg,
2060 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2061 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2062 def rm : SIi8<0xC2, MRMSrcMem,
2063 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2064 [(set RC:$dst, (OpNode (VT RC:$src1),
2065 (ld_frag addr:$src2), imm:$cc))]>;
2067 // Accept explicit immediate argument form instead of comparison code.
2068 let neverHasSideEffects = 1 in {
2069 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2070 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2072 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2073 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2077 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2078 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2079 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2080 XS, VEX_4V, VEX_LIG;
2081 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2082 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2083 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2084 XD, VEX_4V, VEX_LIG;
2086 let Constraints = "$src1 = $dst" in {
2087 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2088 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2089 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2091 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2092 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2093 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2097 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2098 Intrinsic Int, string asm> {
2099 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2100 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2101 [(set VR128:$dst, (Int VR128:$src1,
2102 VR128:$src, imm:$cc))]>;
2103 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2104 (ins VR128:$src1, x86memop:$src, SSECC:$cc), asm,
2105 [(set VR128:$dst, (Int VR128:$src1,
2106 (load addr:$src), imm:$cc))]>;
2109 // Aliases to match intrinsics which expect XMM operand(s).
2110 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2111 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2113 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2114 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2116 let Constraints = "$src1 = $dst" in {
2117 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2118 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2119 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2120 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2124 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2125 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2126 ValueType vt, X86MemOperand x86memop,
2127 PatFrag ld_frag, string OpcodeStr, Domain d> {
2128 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2129 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2130 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2131 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2132 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2133 [(set EFLAGS, (OpNode (vt RC:$src1),
2134 (ld_frag addr:$src2)))], d>;
2137 let Defs = [EFLAGS] in {
2138 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2139 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2140 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2141 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2143 let Pattern = []<dag> in {
2144 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2145 "comiss", SSEPackedSingle>, TB, VEX,
2147 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2148 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2152 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2153 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2154 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2155 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2157 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2158 load, "comiss", SSEPackedSingle>, TB, VEX;
2159 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2160 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2161 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2162 "ucomiss", SSEPackedSingle>, TB;
2163 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2164 "ucomisd", SSEPackedDouble>, TB, OpSize;
2166 let Pattern = []<dag> in {
2167 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2168 "comiss", SSEPackedSingle>, TB;
2169 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2170 "comisd", SSEPackedDouble>, TB, OpSize;
2173 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2174 load, "ucomiss", SSEPackedSingle>, TB;
2175 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2176 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2178 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2179 "comiss", SSEPackedSingle>, TB;
2180 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2181 "comisd", SSEPackedDouble>, TB, OpSize;
2182 } // Defs = [EFLAGS]
2184 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2185 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2186 Intrinsic Int, string asm, string asm_alt,
2188 let isAsmParserOnly = 1 in {
2189 def rri : PIi8<0xC2, MRMSrcReg,
2190 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2191 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2192 def rmi : PIi8<0xC2, MRMSrcMem,
2193 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2194 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2197 // Accept explicit immediate argument form instead of comparison code.
2198 def rri_alt : PIi8<0xC2, MRMSrcReg,
2199 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2201 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2202 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2206 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2207 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2208 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2209 SSEPackedSingle>, TB, VEX_4V;
2210 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2211 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2212 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2213 SSEPackedDouble>, TB, OpSize, VEX_4V;
2214 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2215 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2216 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2217 SSEPackedSingle>, TB, VEX_4V;
2218 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2219 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2220 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2221 SSEPackedDouble>, TB, OpSize, VEX_4V;
2222 let Constraints = "$src1 = $dst" in {
2223 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2224 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2225 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2226 SSEPackedSingle>, TB;
2227 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2228 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2229 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2230 SSEPackedDouble>, TB, OpSize;
2233 let Predicates = [HasAVX] in {
2234 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2235 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2236 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2237 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2238 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2239 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2240 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2241 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2243 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2244 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2245 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2246 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2247 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2248 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2249 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2250 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2253 let Predicates = [HasSSE1] in {
2254 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2255 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2256 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2257 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2260 let Predicates = [HasSSE2] in {
2261 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2262 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2263 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2264 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2267 //===----------------------------------------------------------------------===//
2268 // SSE 1 & 2 - Shuffle Instructions
2269 //===----------------------------------------------------------------------===//
2271 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2272 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2273 ValueType vt, string asm, PatFrag mem_frag,
2274 Domain d, bit IsConvertibleToThreeAddress = 0> {
2275 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2276 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2277 [(set RC:$dst, (vt (shufp:$src3
2278 RC:$src1, (mem_frag addr:$src2))))], d>;
2279 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2280 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2281 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2283 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2286 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2287 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2288 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2289 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2290 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2291 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2292 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2293 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2294 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2295 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2296 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2297 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2299 let Constraints = "$src1 = $dst" in {
2300 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2301 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2302 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2304 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2305 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2306 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2310 let Predicates = [HasAVX] in {
2311 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2312 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2313 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2314 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2315 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2316 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2317 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2318 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2319 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2320 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2321 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2322 // fall back to this for SSE1)
2323 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2324 (VSHUFPSrri VR128:$src2, VR128:$src1,
2325 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2326 // Special unary SHUFPSrri case.
2327 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2328 (VSHUFPSrri VR128:$src1, VR128:$src1,
2329 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2330 // Special binary v4i32 shuffle cases with SHUFPS.
2331 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2332 (VSHUFPSrri VR128:$src1, VR128:$src2,
2333 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2334 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2335 (bc_v4i32 (memopv2i64 addr:$src2)))),
2336 (VSHUFPSrmi VR128:$src1, addr:$src2,
2337 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2338 // Special unary SHUFPDrri cases.
2339 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2340 (VSHUFPDrri VR128:$src1, VR128:$src1,
2341 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2342 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2343 (VSHUFPDrri VR128:$src1, VR128:$src1,
2344 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2345 // Special binary v2i64 shuffle cases using SHUFPDrri.
2346 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2347 (VSHUFPDrri VR128:$src1, VR128:$src2,
2348 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2350 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2351 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2352 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2353 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2354 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2355 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2356 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2357 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2358 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2359 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2362 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2363 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2364 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2365 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2366 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2368 def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2369 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2370 def : Pat<(v8f32 (X86Shufp VR256:$src1,
2371 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2372 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2374 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2375 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2376 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2377 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2378 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2380 def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2381 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2382 def : Pat<(v4f64 (X86Shufp VR256:$src1,
2383 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2384 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2387 let Predicates = [HasSSE1] in {
2388 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2389 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2390 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2391 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2392 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2393 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2394 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2395 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2396 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2397 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2398 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2399 // fall back to this for SSE1)
2400 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2401 (SHUFPSrri VR128:$src2, VR128:$src1,
2402 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2403 // Special unary SHUFPSrri case.
2404 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2405 (SHUFPSrri VR128:$src1, VR128:$src1,
2406 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2409 let Predicates = [HasSSE2] in {
2410 // Special binary v4i32 shuffle cases with SHUFPS.
2411 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2412 (SHUFPSrri VR128:$src1, VR128:$src2,
2413 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2414 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2415 (bc_v4i32 (memopv2i64 addr:$src2)))),
2416 (SHUFPSrmi VR128:$src1, addr:$src2,
2417 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2418 // Special unary SHUFPDrri cases.
2419 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2420 (SHUFPDrri VR128:$src1, VR128:$src1,
2421 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2422 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2423 (SHUFPDrri VR128:$src1, VR128:$src1,
2424 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2425 // Special binary v2i64 shuffle cases using SHUFPDrri.
2426 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2427 (SHUFPDrri VR128:$src1, VR128:$src2,
2428 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2429 // Generic SHUFPD patterns
2430 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2431 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2432 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2433 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2434 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2435 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2436 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2437 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2438 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2439 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2442 //===----------------------------------------------------------------------===//
2443 // SSE 1 & 2 - Unpack Instructions
2444 //===----------------------------------------------------------------------===//
2446 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2447 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2448 PatFrag mem_frag, RegisterClass RC,
2449 X86MemOperand x86memop, string asm,
2451 def rr : PI<opc, MRMSrcReg,
2452 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2454 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2455 def rm : PI<opc, MRMSrcMem,
2456 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2458 (vt (OpNode RC:$src1,
2459 (mem_frag addr:$src2))))], d>;
2462 let AddedComplexity = 10 in {
2463 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2464 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2465 SSEPackedSingle>, TB, VEX_4V;
2466 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2467 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2468 SSEPackedDouble>, TB, OpSize, VEX_4V;
2469 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2470 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2471 SSEPackedSingle>, TB, VEX_4V;
2472 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2473 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2474 SSEPackedDouble>, TB, OpSize, VEX_4V;
2476 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2477 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2478 SSEPackedSingle>, TB, VEX_4V;
2479 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2480 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2481 SSEPackedDouble>, TB, OpSize, VEX_4V;
2482 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2483 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2484 SSEPackedSingle>, TB, VEX_4V;
2485 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2486 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2487 SSEPackedDouble>, TB, OpSize, VEX_4V;
2489 let Constraints = "$src1 = $dst" in {
2490 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2491 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2492 SSEPackedSingle>, TB;
2493 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2494 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2495 SSEPackedDouble>, TB, OpSize;
2496 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2497 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2498 SSEPackedSingle>, TB;
2499 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2500 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2501 SSEPackedDouble>, TB, OpSize;
2502 } // Constraints = "$src1 = $dst"
2503 } // AddedComplexity
2505 let Predicates = [HasAVX], AddedComplexity = 1 in {
2506 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2507 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2508 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2509 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2510 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2511 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2512 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2513 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2515 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2516 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2517 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2518 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2519 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2520 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2521 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2522 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2524 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2525 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2526 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2527 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2528 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2529 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2530 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2531 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2533 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2534 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2535 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2536 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2537 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2538 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2539 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2540 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2542 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2543 // problem is during lowering, where it's not possible to recognize the load
2544 // fold cause it has two uses through a bitcast. One use disappears at isel
2545 // time and the fold opportunity reappears.
2546 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2547 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2548 let AddedComplexity = 10 in
2549 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2550 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2553 let Predicates = [HasSSE1] in {
2554 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2555 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2556 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2557 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2558 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2559 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2560 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2561 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2564 let Predicates = [HasSSE2] in {
2565 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2566 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2567 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2568 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2569 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2570 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2571 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2572 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2574 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2575 // problem is during lowering, where it's not possible to recognize the load
2576 // fold cause it has two uses through a bitcast. One use disappears at isel
2577 // time and the fold opportunity reappears.
2578 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2579 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2581 let AddedComplexity = 10 in
2582 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2583 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2586 //===----------------------------------------------------------------------===//
2587 // SSE 1 & 2 - Extract Floating-Point Sign mask
2588 //===----------------------------------------------------------------------===//
2590 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2591 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2593 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2594 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2595 [(set GR32:$dst, (Int RC:$src))], d>;
2596 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2597 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2600 let Predicates = [HasAVX] in {
2601 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2602 "movmskps", SSEPackedSingle>, TB, VEX;
2603 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2604 "movmskpd", SSEPackedDouble>, TB,
2606 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2607 "movmskps", SSEPackedSingle>, TB, VEX;
2608 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2609 "movmskpd", SSEPackedDouble>, TB,
2612 def : Pat<(i32 (X86fgetsign FR32:$src)),
2613 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2615 def : Pat<(i64 (X86fgetsign FR32:$src)),
2616 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2618 def : Pat<(i32 (X86fgetsign FR64:$src)),
2619 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2621 def : Pat<(i64 (X86fgetsign FR64:$src)),
2622 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2626 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2627 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2628 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2629 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2631 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2632 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2633 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2634 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2638 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2639 SSEPackedSingle>, TB;
2640 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2641 SSEPackedDouble>, TB, OpSize;
2643 def : Pat<(i32 (X86fgetsign FR32:$src)),
2644 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2645 sub_ss))>, Requires<[HasSSE1]>;
2646 def : Pat<(i64 (X86fgetsign FR32:$src)),
2647 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2648 sub_ss))>, Requires<[HasSSE1]>;
2649 def : Pat<(i32 (X86fgetsign FR64:$src)),
2650 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2651 sub_sd))>, Requires<[HasSSE2]>;
2652 def : Pat<(i64 (X86fgetsign FR64:$src)),
2653 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2654 sub_sd))>, Requires<[HasSSE2]>;
2656 //===---------------------------------------------------------------------===//
2657 // SSE2 - Packed Integer Logical Instructions
2658 //===---------------------------------------------------------------------===//
2660 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2662 /// PDI_binop_rm - Simple SSE2 binary operator.
2663 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2664 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2665 X86MemOperand x86memop, bit IsCommutable = 0,
2667 let isCommutable = IsCommutable in
2668 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2669 (ins RC:$src1, RC:$src2),
2671 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2672 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2673 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2674 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2675 (ins RC:$src1, x86memop:$src2),
2677 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2678 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2679 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2680 (bitconvert (memop_frag addr:$src2)))))]>;
2682 } // ExeDomain = SSEPackedInt
2684 // These are ordered here for pattern ordering requirements with the fp versions
2686 let Predicates = [HasAVX] in {
2687 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2688 i128mem, 1, 0>, VEX_4V;
2689 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2690 i128mem, 1, 0>, VEX_4V;
2691 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2692 i128mem, 1, 0>, VEX_4V;
2693 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2694 i128mem, 0, 0>, VEX_4V;
2697 let Constraints = "$src1 = $dst" in {
2698 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2700 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2702 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2704 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2706 } // Constraints = "$src1 = $dst"
2708 let Predicates = [HasAVX2] in {
2709 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2710 i256mem, 1, 0>, VEX_4V;
2711 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2712 i256mem, 1, 0>, VEX_4V;
2713 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2714 i256mem, 1, 0>, VEX_4V;
2715 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2716 i256mem, 0, 0>, VEX_4V;
2719 //===----------------------------------------------------------------------===//
2720 // SSE 1 & 2 - Logical Instructions
2721 //===----------------------------------------------------------------------===//
2723 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2725 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2727 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2728 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2730 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2731 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2733 let Constraints = "$src1 = $dst" in {
2734 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2735 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2737 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2738 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2742 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2743 let mayLoad = 0 in {
2744 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2745 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2746 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2749 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2750 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2752 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2754 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2756 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2757 // are all promoted to v2i64, and the patterns are covered by the int
2758 // version. This is needed in SSE only, because v2i64 isn't supported on
2759 // SSE1, but only on SSE2.
2760 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2761 !strconcat(OpcodeStr, "ps"), f128mem, [],
2762 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2763 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2765 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2766 !strconcat(OpcodeStr, "pd"), f128mem,
2767 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2768 (bc_v2i64 (v2f64 VR128:$src2))))],
2769 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2770 (memopv2i64 addr:$src2)))], 0>,
2772 let Constraints = "$src1 = $dst" in {
2773 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2774 !strconcat(OpcodeStr, "ps"), f128mem,
2775 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2776 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2777 (memopv2i64 addr:$src2)))]>, TB;
2779 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2780 !strconcat(OpcodeStr, "pd"), f128mem,
2781 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2782 (bc_v2i64 (v2f64 VR128:$src2))))],
2783 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2784 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2788 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2790 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2792 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2793 !strconcat(OpcodeStr, "ps"), f256mem,
2794 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2795 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2796 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2798 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2799 !strconcat(OpcodeStr, "pd"), f256mem,
2800 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2801 (bc_v4i64 (v4f64 VR256:$src2))))],
2802 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2803 (memopv4i64 addr:$src2)))], 0>,
2807 // AVX 256-bit packed logical ops forms
2808 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2809 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2810 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2811 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2813 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2814 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2815 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2816 let isCommutable = 0 in
2817 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2819 //===----------------------------------------------------------------------===//
2820 // SSE 1 & 2 - Arithmetic Instructions
2821 //===----------------------------------------------------------------------===//
2823 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2826 /// In addition, we also have a special variant of the scalar form here to
2827 /// represent the associated intrinsic operation. This form is unlike the
2828 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2829 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2831 /// These three forms can each be reg+reg or reg+mem.
2834 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2836 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2838 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2839 OpNode, FR32, f32mem, Is2Addr>, XS;
2840 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2841 OpNode, FR64, f64mem, Is2Addr>, XD;
2844 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2846 let mayLoad = 0 in {
2847 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2848 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2849 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2850 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2854 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2856 let mayLoad = 0 in {
2857 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2858 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2859 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2860 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2864 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2866 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2867 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2868 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2869 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2872 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2874 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2875 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2876 SSEPackedSingle, Is2Addr>, TB;
2878 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2879 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2880 SSEPackedDouble, Is2Addr>, TB, OpSize;
2883 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2884 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2885 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2886 SSEPackedSingle, 0>, TB;
2888 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2889 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2890 SSEPackedDouble, 0>, TB, OpSize;
2893 // Binary Arithmetic instructions
2894 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2895 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2896 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2897 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2898 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2899 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2900 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2901 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2903 let isCommutable = 0 in {
2904 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2905 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2906 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2907 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2908 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2909 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2910 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2911 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2912 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2913 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2914 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2915 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2916 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2917 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2918 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2919 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2920 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2921 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2922 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2923 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2926 let Constraints = "$src1 = $dst" in {
2927 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2928 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2929 basic_sse12_fp_binop_s_int<0x58, "add">;
2930 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2931 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2932 basic_sse12_fp_binop_s_int<0x59, "mul">;
2934 let isCommutable = 0 in {
2935 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2936 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2937 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2938 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2939 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2940 basic_sse12_fp_binop_s_int<0x5E, "div">;
2941 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2942 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2943 basic_sse12_fp_binop_s_int<0x5F, "max">,
2944 basic_sse12_fp_binop_p_int<0x5F, "max">;
2945 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2946 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2947 basic_sse12_fp_binop_s_int<0x5D, "min">,
2948 basic_sse12_fp_binop_p_int<0x5D, "min">;
2953 /// In addition, we also have a special variant of the scalar form here to
2954 /// represent the associated intrinsic operation. This form is unlike the
2955 /// plain scalar form, in that it takes an entire vector (instead of a
2956 /// scalar) and leaves the top elements undefined.
2958 /// And, we have a special variant form for a full-vector intrinsic form.
2960 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2961 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2962 SDNode OpNode, Intrinsic F32Int> {
2963 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2964 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2965 [(set FR32:$dst, (OpNode FR32:$src))]>;
2966 // For scalar unary operations, fold a load into the operation
2967 // only in OptForSize mode. It eliminates an instruction, but it also
2968 // eliminates a whole-register clobber (the load), so it introduces a
2969 // partial register update condition.
2970 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2971 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2972 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2973 Requires<[HasSSE1, OptForSize]>;
2974 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2975 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2976 [(set VR128:$dst, (F32Int VR128:$src))]>;
2977 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2978 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2979 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2982 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2983 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2984 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2985 !strconcat(OpcodeStr,
2986 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2988 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2989 !strconcat(OpcodeStr,
2990 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2991 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2992 (ins VR128:$src1, ssmem:$src2),
2993 !strconcat(OpcodeStr,
2994 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2997 /// sse1_fp_unop_p - SSE1 unops in packed form.
2998 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2999 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3000 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3001 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
3002 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3003 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3004 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
3007 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3008 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3009 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3010 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3011 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
3012 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3013 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3014 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
3017 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3018 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3019 Intrinsic V4F32Int> {
3020 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3021 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3022 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
3023 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3024 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3025 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
3028 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3029 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3030 Intrinsic V4F32Int> {
3031 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3032 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3033 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
3034 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3035 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3036 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
3039 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3040 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3041 SDNode OpNode, Intrinsic F64Int> {
3042 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3043 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3044 [(set FR64:$dst, (OpNode FR64:$src))]>;
3045 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3046 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3047 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3048 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
3049 Requires<[HasSSE2, OptForSize]>;
3050 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3051 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3052 [(set VR128:$dst, (F64Int VR128:$src))]>;
3053 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3054 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3055 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
3058 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3059 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3060 let neverHasSideEffects = 1 in {
3061 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3062 !strconcat(OpcodeStr,
3063 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3065 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3066 !strconcat(OpcodeStr,
3067 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3069 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3070 (ins VR128:$src1, sdmem:$src2),
3071 !strconcat(OpcodeStr,
3072 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3075 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3076 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3078 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3079 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3080 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3081 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3082 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3083 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3086 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3087 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3088 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3089 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3090 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3091 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3092 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3093 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3096 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3097 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3098 Intrinsic V2F64Int> {
3099 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3100 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3101 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3102 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3103 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3104 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3107 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3108 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3109 Intrinsic V2F64Int> {
3110 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3111 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3112 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3113 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3114 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3115 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3118 let Predicates = [HasAVX] in {
3120 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3121 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3123 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3124 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3125 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3126 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3127 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3128 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3129 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3130 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3133 // Reciprocal approximations. Note that these typically require refinement
3134 // in order to obtain suitable precision.
3135 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3136 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3137 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3138 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3139 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3141 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3142 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3143 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3144 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3145 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3148 let AddedComplexity = 1 in {
3149 def : Pat<(f32 (fsqrt FR32:$src)),
3150 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3151 def : Pat<(f32 (fsqrt (load addr:$src))),
3152 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3153 Requires<[HasAVX, OptForSize]>;
3154 def : Pat<(f64 (fsqrt FR64:$src)),
3155 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3156 def : Pat<(f64 (fsqrt (load addr:$src))),
3157 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3158 Requires<[HasAVX, OptForSize]>;
3160 def : Pat<(f32 (X86frsqrt FR32:$src)),
3161 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3162 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3163 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3164 Requires<[HasAVX, OptForSize]>;
3166 def : Pat<(f32 (X86frcp FR32:$src)),
3167 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3168 def : Pat<(f32 (X86frcp (load addr:$src))),
3169 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3170 Requires<[HasAVX, OptForSize]>;
3173 let Predicates = [HasAVX], AddedComplexity = 1 in {
3174 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3175 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3176 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3177 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3179 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3180 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3182 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3183 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3184 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3185 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3187 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3188 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3190 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3191 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3192 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3193 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3195 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3196 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3198 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3199 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3200 (VRCPSSr (f32 (IMPLICIT_DEF)),
3201 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3203 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3204 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3208 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3209 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3210 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3211 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3212 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3213 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3215 // Reciprocal approximations. Note that these typically require refinement
3216 // in order to obtain suitable precision.
3217 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3218 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3219 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3220 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3221 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3222 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3224 // There is no f64 version of the reciprocal approximation instructions.
3226 //===----------------------------------------------------------------------===//
3227 // SSE 1 & 2 - Non-temporal stores
3228 //===----------------------------------------------------------------------===//
3230 let AddedComplexity = 400 in { // Prefer non-temporal versions
3231 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3232 (ins f128mem:$dst, VR128:$src),
3233 "movntps\t{$src, $dst|$dst, $src}",
3234 [(alignednontemporalstore (v4f32 VR128:$src),
3236 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3237 (ins f128mem:$dst, VR128:$src),
3238 "movntpd\t{$src, $dst|$dst, $src}",
3239 [(alignednontemporalstore (v2f64 VR128:$src),
3242 let ExeDomain = SSEPackedInt in
3243 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3244 (ins f128mem:$dst, VR128:$src),
3245 "movntdq\t{$src, $dst|$dst, $src}",
3246 [(alignednontemporalstore (v2i64 VR128:$src),
3249 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3250 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3252 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3253 (ins f256mem:$dst, VR256:$src),
3254 "movntps\t{$src, $dst|$dst, $src}",
3255 [(alignednontemporalstore (v8f32 VR256:$src),
3257 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3258 (ins f256mem:$dst, VR256:$src),
3259 "movntpd\t{$src, $dst|$dst, $src}",
3260 [(alignednontemporalstore (v4f64 VR256:$src),
3262 let ExeDomain = SSEPackedInt in
3263 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3264 (ins f256mem:$dst, VR256:$src),
3265 "movntdq\t{$src, $dst|$dst, $src}",
3266 [(alignednontemporalstore (v4i64 VR256:$src),
3270 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3271 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3272 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3273 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3274 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3275 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3277 let AddedComplexity = 400 in { // Prefer non-temporal versions
3278 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3279 "movntps\t{$src, $dst|$dst, $src}",
3280 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3281 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3282 "movntpd\t{$src, $dst|$dst, $src}",
3283 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3285 let ExeDomain = SSEPackedInt in
3286 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3287 "movntdq\t{$src, $dst|$dst, $src}",
3288 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)]>;
3290 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3291 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3293 // There is no AVX form for instructions below this point
3294 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3295 "movnti{l}\t{$src, $dst|$dst, $src}",
3296 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3297 TB, Requires<[HasSSE2]>;
3298 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3299 "movnti{q}\t{$src, $dst|$dst, $src}",
3300 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3301 TB, Requires<[HasSSE2]>;
3304 //===----------------------------------------------------------------------===//
3305 // SSE 1 & 2 - Prefetch and memory fence
3306 //===----------------------------------------------------------------------===//
3308 // Prefetch intrinsic.
3309 let Predicates = [HasSSE1] in {
3310 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3311 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3312 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3313 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3314 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3315 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3316 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3317 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3321 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3322 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3323 TB, Requires<[HasSSE2]>;
3325 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3326 // was introduced with SSE2, it's backward compatible.
3327 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3329 // Load, store, and memory fence
3330 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3331 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3332 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3333 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3334 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3335 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3337 def : Pat<(X86SFence), (SFENCE)>;
3338 def : Pat<(X86LFence), (LFENCE)>;
3339 def : Pat<(X86MFence), (MFENCE)>;
3341 //===----------------------------------------------------------------------===//
3342 // SSE 1 & 2 - Load/Store XCSR register
3343 //===----------------------------------------------------------------------===//
3345 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3346 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3347 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3348 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3350 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3351 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3352 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3353 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3355 //===---------------------------------------------------------------------===//
3356 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3357 //===---------------------------------------------------------------------===//
3359 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3361 let neverHasSideEffects = 1 in {
3362 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3363 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3364 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3365 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3367 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3368 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3369 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3370 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3373 let isCodeGenOnly = 1 in {
3374 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3375 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3376 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3377 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3378 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3379 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3380 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3381 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3384 let canFoldAsLoad = 1, mayLoad = 1 in {
3385 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3386 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3387 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3388 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3389 let Predicates = [HasAVX] in {
3390 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3391 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3392 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3393 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3397 let mayStore = 1 in {
3398 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3399 (ins i128mem:$dst, VR128:$src),
3400 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3401 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3402 (ins i256mem:$dst, VR256:$src),
3403 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3404 let Predicates = [HasAVX] in {
3405 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3406 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3407 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3408 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3412 let neverHasSideEffects = 1 in
3413 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3414 "movdqa\t{$src, $dst|$dst, $src}", []>;
3416 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3417 "movdqu\t{$src, $dst|$dst, $src}",
3418 []>, XS, Requires<[HasSSE2]>;
3421 let isCodeGenOnly = 1 in {
3422 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3423 "movdqa\t{$src, $dst|$dst, $src}", []>;
3425 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3426 "movdqu\t{$src, $dst|$dst, $src}",
3427 []>, XS, Requires<[HasSSE2]>;
3430 let canFoldAsLoad = 1, mayLoad = 1 in {
3431 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3432 "movdqa\t{$src, $dst|$dst, $src}",
3433 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3434 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3435 "movdqu\t{$src, $dst|$dst, $src}",
3436 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3437 XS, Requires<[HasSSE2]>;
3440 let mayStore = 1 in {
3441 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3442 "movdqa\t{$src, $dst|$dst, $src}",
3443 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3444 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3445 "movdqu\t{$src, $dst|$dst, $src}",
3446 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3447 XS, Requires<[HasSSE2]>;
3450 // Intrinsic forms of MOVDQU load and store
3451 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3452 "vmovdqu\t{$src, $dst|$dst, $src}",
3453 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3454 XS, VEX, Requires<[HasAVX]>;
3456 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3457 "movdqu\t{$src, $dst|$dst, $src}",
3458 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3459 XS, Requires<[HasSSE2]>;
3461 } // ExeDomain = SSEPackedInt
3463 let Predicates = [HasAVX] in {
3464 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3465 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3466 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3469 //===---------------------------------------------------------------------===//
3470 // SSE2 - Packed Integer Arithmetic Instructions
3471 //===---------------------------------------------------------------------===//
3473 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3475 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3476 RegisterClass RC, PatFrag memop_frag,
3477 X86MemOperand x86memop, bit IsCommutable = 0,
3479 let isCommutable = IsCommutable in
3480 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3481 (ins RC:$src1, RC:$src2),
3483 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3484 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3485 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3486 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3487 (ins RC:$src1, x86memop:$src2),
3489 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3490 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3491 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3494 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3495 string OpcodeStr, SDNode OpNode,
3496 SDNode OpNode2, RegisterClass RC,
3497 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3499 // src2 is always 128-bit
3500 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3501 (ins RC:$src1, VR128:$src2),
3503 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3504 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3505 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))]>;
3506 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3507 (ins RC:$src1, i128mem:$src2),
3509 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3510 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3511 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3512 (bc_frag (memopv2i64 addr:$src2)))))]>;
3513 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3514 (ins RC:$src1, i32i8imm:$src2),
3516 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3517 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3518 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))]>;
3521 } // ExeDomain = SSEPackedInt
3523 // 128-bit Integer Arithmetic
3525 let Predicates = [HasAVX] in {
3526 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3527 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3528 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3529 i128mem, 1, 0>, VEX_4V;
3530 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3531 i128mem, 1, 0>, VEX_4V;
3532 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3533 i128mem, 1, 0>, VEX_4V;
3534 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3535 i128mem, 1, 0>, VEX_4V;
3536 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3537 i128mem, 0, 0>, VEX_4V;
3538 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3539 i128mem, 0, 0>, VEX_4V;
3540 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3541 i128mem, 0, 0>, VEX_4V;
3542 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3543 i128mem, 0, 0>, VEX_4V;
3546 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3547 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3548 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3549 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3550 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3551 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3552 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3553 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3554 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3555 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3556 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3557 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3558 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3559 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3560 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3561 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3562 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3563 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3564 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3565 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3566 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3567 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3568 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3569 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3570 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3571 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3572 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3573 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3574 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3575 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3576 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3577 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3578 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3579 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3580 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3581 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3582 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3583 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3586 let Predicates = [HasAVX2] in {
3587 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3588 i256mem, 1, 0>, VEX_4V;
3589 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3590 i256mem, 1, 0>, VEX_4V;
3591 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3592 i256mem, 1, 0>, VEX_4V;
3593 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3594 i256mem, 1, 0>, VEX_4V;
3595 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3596 i256mem, 1, 0>, VEX_4V;
3597 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3598 i256mem, 0, 0>, VEX_4V;
3599 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3600 i256mem, 0, 0>, VEX_4V;
3601 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3602 i256mem, 0, 0>, VEX_4V;
3603 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3604 i256mem, 0, 0>, VEX_4V;
3607 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3608 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3609 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3610 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3611 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3612 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3613 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3614 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3615 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3616 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3617 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3618 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3619 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3620 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3621 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3622 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3623 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3624 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3625 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3626 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3627 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3628 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3629 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3630 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3631 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3632 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3633 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3634 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3635 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3636 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3637 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3638 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3639 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3640 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3641 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3642 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3643 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3644 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3647 let Constraints = "$src1 = $dst" in {
3648 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3650 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3652 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3654 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3656 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3658 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3660 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3662 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3664 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3668 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3669 VR128, memopv2i64, i128mem>;
3670 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3671 VR128, memopv2i64, i128mem>;
3672 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3673 VR128, memopv2i64, i128mem>;
3674 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3675 VR128, memopv2i64, i128mem>;
3676 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3677 VR128, memopv2i64, i128mem, 1>;
3678 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3679 VR128, memopv2i64, i128mem, 1>;
3680 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3681 VR128, memopv2i64, i128mem, 1>;
3682 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3683 VR128, memopv2i64, i128mem, 1>;
3684 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3685 VR128, memopv2i64, i128mem, 1>;
3686 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3687 VR128, memopv2i64, i128mem, 1>;
3688 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3689 VR128, memopv2i64, i128mem, 1>;
3690 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3691 VR128, memopv2i64, i128mem, 1>;
3692 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3693 VR128, memopv2i64, i128mem, 1>;
3694 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3695 VR128, memopv2i64, i128mem, 1>;
3696 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3697 VR128, memopv2i64, i128mem, 1>;
3698 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3699 VR128, memopv2i64, i128mem, 1>;
3700 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3701 VR128, memopv2i64, i128mem, 1>;
3702 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3703 VR128, memopv2i64, i128mem, 1>;
3704 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3705 VR128, memopv2i64, i128mem, 1>;
3707 } // Constraints = "$src1 = $dst"
3709 //===---------------------------------------------------------------------===//
3710 // SSE2 - Packed Integer Logical Instructions
3711 //===---------------------------------------------------------------------===//
3713 let Predicates = [HasAVX] in {
3714 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3715 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3716 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3717 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3718 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3719 VR128, v2i64, v2i64, bc_v2i64, 0>, VEX_4V;
3721 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3722 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3723 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3724 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3725 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3726 VR128, v2i64, v2i64, bc_v2i64, 0>, VEX_4V;
3728 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3729 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3730 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3731 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3733 let ExeDomain = SSEPackedInt in {
3734 // 128-bit logical shifts.
3735 def VPSLLDQri : PDIi8<0x73, MRM7r,
3736 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3737 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3739 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3741 def VPSRLDQri : PDIi8<0x73, MRM3r,
3742 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3743 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3745 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3747 // PSRADQri doesn't exist in SSE[1-3].
3749 } // Predicates = [HasAVX]
3751 let Predicates = [HasAVX2] in {
3752 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3753 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3754 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3755 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3756 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3757 VR256, v4i64, v2i64, bc_v2i64, 0>, VEX_4V;
3759 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3760 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3761 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3762 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3763 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3764 VR256, v4i64, v2i64, bc_v2i64, 0>, VEX_4V;
3766 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3767 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3768 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3769 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3771 let ExeDomain = SSEPackedInt in {
3772 // 256-bit logical shifts.
3773 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3774 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3775 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3777 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3779 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3780 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3781 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3783 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3785 // PSRADQYri doesn't exist in SSE[1-3].
3787 } // Predicates = [HasAVX2]
3789 let Constraints = "$src1 = $dst" in {
3790 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3791 VR128, v8i16, v8i16, bc_v8i16>;
3792 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3793 VR128, v4i32, v4i32, bc_v4i32>;
3794 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3795 VR128, v2i64, v2i64, bc_v2i64>;
3797 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3798 VR128, v8i16, v8i16, bc_v8i16>;
3799 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3800 VR128, v4i32, v4i32, bc_v4i32>;
3801 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3802 VR128, v2i64, v2i64, bc_v2i64>;
3804 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3805 VR128, v8i16, v8i16, bc_v8i16>;
3806 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3807 VR128, v4i32, v4i32, bc_v4i32>;
3809 let ExeDomain = SSEPackedInt in {
3810 // 128-bit logical shifts.
3811 def PSLLDQri : PDIi8<0x73, MRM7r,
3812 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3813 "pslldq\t{$src2, $dst|$dst, $src2}",
3815 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3816 def PSRLDQri : PDIi8<0x73, MRM3r,
3817 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3818 "psrldq\t{$src2, $dst|$dst, $src2}",
3820 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3821 // PSRADQri doesn't exist in SSE[1-3].
3823 } // Constraints = "$src1 = $dst"
3825 let Predicates = [HasAVX] in {
3826 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3827 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3828 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3829 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3830 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3831 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3833 // Shift up / down and insert zero's.
3834 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3835 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3836 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3837 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3840 let Predicates = [HasAVX2] in {
3841 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3842 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3843 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3844 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3847 let Predicates = [HasSSE2] in {
3848 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3849 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3850 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3851 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3852 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3853 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3855 // Shift up / down and insert zero's.
3856 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3857 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3858 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3859 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3862 //===---------------------------------------------------------------------===//
3863 // SSE2 - Packed Integer Comparison Instructions
3864 //===---------------------------------------------------------------------===//
3866 let Predicates = [HasAVX] in {
3867 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
3868 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3869 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
3870 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3871 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
3872 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3873 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
3874 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3875 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
3876 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3877 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
3878 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3881 let Predicates = [HasAVX2] in {
3882 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
3883 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3884 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
3885 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3886 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
3887 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3888 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
3889 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3890 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
3891 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3892 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
3893 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3896 let Constraints = "$src1 = $dst" in {
3897 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
3898 VR128, memopv2i64, i128mem, 1>;
3899 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
3900 VR128, memopv2i64, i128mem, 1>;
3901 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
3902 VR128, memopv2i64, i128mem, 1>;
3903 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
3904 VR128, memopv2i64, i128mem>;
3905 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
3906 VR128, memopv2i64, i128mem>;
3907 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
3908 VR128, memopv2i64, i128mem>;
3909 } // Constraints = "$src1 = $dst"
3911 //===---------------------------------------------------------------------===//
3912 // SSE2 - Packed Integer Pack Instructions
3913 //===---------------------------------------------------------------------===//
3915 let Predicates = [HasAVX] in {
3916 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3917 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3918 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3919 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3920 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3921 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3924 let Predicates = [HasAVX2] in {
3925 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
3926 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3927 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
3928 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3929 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
3930 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3933 let Constraints = "$src1 = $dst" in {
3934 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
3935 VR128, memopv2i64, i128mem>;
3936 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
3937 VR128, memopv2i64, i128mem>;
3938 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
3939 VR128, memopv2i64, i128mem>;
3940 } // Constraints = "$src1 = $dst"
3942 //===---------------------------------------------------------------------===//
3943 // SSE2 - Packed Integer Shuffle Instructions
3944 //===---------------------------------------------------------------------===//
3946 let ExeDomain = SSEPackedInt in {
3947 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
3949 def ri : Ii8<0x70, MRMSrcReg,
3950 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3951 !strconcat(OpcodeStr,
3952 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3953 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
3955 def mi : Ii8<0x70, MRMSrcMem,
3956 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3957 !strconcat(OpcodeStr,
3958 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3959 [(set VR128:$dst, (vt (pshuf_frag:$src2
3960 (bc_frag (memopv2i64 addr:$src1)),
3964 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
3966 def Yri : Ii8<0x70, MRMSrcReg,
3967 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
3968 !strconcat(OpcodeStr,
3969 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3970 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
3972 def Ymi : Ii8<0x70, MRMSrcMem,
3973 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
3974 !strconcat(OpcodeStr,
3975 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3976 [(set VR256:$dst, (vt (pshuf_frag:$src2
3977 (bc_frag (memopv4i64 addr:$src1)),
3980 } // ExeDomain = SSEPackedInt
3982 let Predicates = [HasAVX] in {
3983 let AddedComplexity = 5 in
3984 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
3987 // SSE2 with ImmT == Imm8 and XS prefix.
3988 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
3991 // SSE2 with ImmT == Imm8 and XD prefix.
3992 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
3995 let AddedComplexity = 5 in
3996 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3997 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3998 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
3999 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4000 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4002 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4004 (VPSHUFDmi addr:$src1, imm:$imm)>;
4005 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4007 (VPSHUFDmi addr:$src1, imm:$imm)>;
4008 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4009 (VPSHUFDri VR128:$src1, imm:$imm)>;
4010 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4011 (VPSHUFDri VR128:$src1, imm:$imm)>;
4012 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4013 (VPSHUFHWri VR128:$src, imm:$imm)>;
4014 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4016 (VPSHUFHWmi addr:$src, imm:$imm)>;
4017 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4018 (VPSHUFLWri VR128:$src, imm:$imm)>;
4019 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4021 (VPSHUFLWmi addr:$src, imm:$imm)>;
4024 let Predicates = [HasAVX2] in {
4025 let AddedComplexity = 5 in
4026 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4029 // SSE2 with ImmT == Imm8 and XS prefix.
4030 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4033 // SSE2 with ImmT == Imm8 and XD prefix.
4034 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4038 let Predicates = [HasSSE2] in {
4039 let AddedComplexity = 5 in
4040 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4042 // SSE2 with ImmT == Imm8 and XS prefix.
4043 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4045 // SSE2 with ImmT == Imm8 and XD prefix.
4046 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4048 let AddedComplexity = 5 in
4049 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4050 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4051 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4052 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4053 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4055 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4057 (PSHUFDmi addr:$src1, imm:$imm)>;
4058 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4060 (PSHUFDmi addr:$src1, imm:$imm)>;
4061 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4062 (PSHUFDri VR128:$src1, imm:$imm)>;
4063 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4064 (PSHUFDri VR128:$src1, imm:$imm)>;
4065 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4066 (PSHUFHWri VR128:$src, imm:$imm)>;
4067 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4069 (PSHUFHWmi addr:$src, imm:$imm)>;
4070 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4071 (PSHUFLWri VR128:$src, imm:$imm)>;
4072 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4074 (PSHUFLWmi addr:$src, imm:$imm)>;
4077 //===---------------------------------------------------------------------===//
4078 // SSE2 - Packed Integer Unpack Instructions
4079 //===---------------------------------------------------------------------===//
4081 let ExeDomain = SSEPackedInt in {
4082 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4083 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4084 def rr : PDI<opc, MRMSrcReg,
4085 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4087 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4088 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4089 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4090 def rm : PDI<opc, MRMSrcMem,
4091 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4093 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4094 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4095 [(set VR128:$dst, (OpNode VR128:$src1,
4096 (bc_frag (memopv2i64
4100 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4101 SDNode OpNode, PatFrag bc_frag> {
4102 def Yrr : PDI<opc, MRMSrcReg,
4103 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4104 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4105 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4106 def Yrm : PDI<opc, MRMSrcMem,
4107 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4108 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4109 [(set VR256:$dst, (OpNode VR256:$src1,
4110 (bc_frag (memopv4i64 addr:$src2))))]>;
4113 let Predicates = [HasAVX] in {
4114 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4115 bc_v16i8, 0>, VEX_4V;
4116 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4117 bc_v8i16, 0>, VEX_4V;
4118 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4119 bc_v4i32, 0>, VEX_4V;
4120 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4121 bc_v2i64, 0>, VEX_4V;
4123 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4124 bc_v16i8, 0>, VEX_4V;
4125 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4126 bc_v8i16, 0>, VEX_4V;
4127 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4128 bc_v4i32, 0>, VEX_4V;
4129 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4130 bc_v2i64, 0>, VEX_4V;
4133 let Predicates = [HasAVX2] in {
4134 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4136 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4138 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4140 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4143 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4145 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4147 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4149 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4153 let Constraints = "$src1 = $dst" in {
4154 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4156 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4158 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4160 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4163 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4165 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4167 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4169 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4172 } // ExeDomain = SSEPackedInt
4174 // Patterns for using AVX1 instructions with integer vectors
4175 // Here to give AVX2 priority
4176 let Predicates = [HasAVX] in {
4177 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4178 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4179 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4180 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4181 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4182 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4183 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4184 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4186 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4187 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4188 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4189 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4190 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4191 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4192 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4193 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4196 // Splat v2f64 / v2i64
4197 let AddedComplexity = 10 in {
4198 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4199 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4200 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4201 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4204 //===---------------------------------------------------------------------===//
4205 // SSE2 - Packed Integer Extract and Insert
4206 //===---------------------------------------------------------------------===//
4208 let ExeDomain = SSEPackedInt in {
4209 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4210 def rri : Ii8<0xC4, MRMSrcReg,
4211 (outs VR128:$dst), (ins VR128:$src1,
4212 GR32:$src2, i32i8imm:$src3),
4214 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4215 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4217 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4218 def rmi : Ii8<0xC4, MRMSrcMem,
4219 (outs VR128:$dst), (ins VR128:$src1,
4220 i16mem:$src2, i32i8imm:$src3),
4222 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4223 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4225 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4230 let Predicates = [HasAVX] in
4231 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4232 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4233 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4234 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4235 imm:$src2))]>, TB, OpSize, VEX;
4236 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4237 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4238 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4239 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4243 let Predicates = [HasAVX] in {
4244 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4245 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4246 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4247 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4248 []>, TB, OpSize, VEX_4V;
4251 let Constraints = "$src1 = $dst" in
4252 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4254 } // ExeDomain = SSEPackedInt
4256 //===---------------------------------------------------------------------===//
4257 // SSE2 - Packed Mask Creation
4258 //===---------------------------------------------------------------------===//
4260 let ExeDomain = SSEPackedInt in {
4262 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4263 "pmovmskb\t{$src, $dst|$dst, $src}",
4264 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4265 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4266 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4268 let Predicates = [HasAVX2] in {
4269 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4270 "pmovmskb\t{$src, $dst|$dst, $src}",
4271 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4272 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4273 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4276 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4277 "pmovmskb\t{$src, $dst|$dst, $src}",
4278 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4280 } // ExeDomain = SSEPackedInt
4282 //===---------------------------------------------------------------------===//
4283 // SSE2 - Conditional Store
4284 //===---------------------------------------------------------------------===//
4286 let ExeDomain = SSEPackedInt in {
4289 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4290 (ins VR128:$src, VR128:$mask),
4291 "maskmovdqu\t{$mask, $src|$src, $mask}",
4292 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4294 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4295 (ins VR128:$src, VR128:$mask),
4296 "maskmovdqu\t{$mask, $src|$src, $mask}",
4297 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4300 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4301 "maskmovdqu\t{$mask, $src|$src, $mask}",
4302 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4304 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4305 "maskmovdqu\t{$mask, $src|$src, $mask}",
4306 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4308 } // ExeDomain = SSEPackedInt
4310 //===---------------------------------------------------------------------===//
4311 // SSE2 - Move Doubleword
4312 //===---------------------------------------------------------------------===//
4314 //===---------------------------------------------------------------------===//
4315 // Move Int Doubleword to Packed Double Int
4317 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4318 "movd\t{$src, $dst|$dst, $src}",
4320 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4321 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4322 "movd\t{$src, $dst|$dst, $src}",
4324 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4326 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4327 "mov{d|q}\t{$src, $dst|$dst, $src}",
4329 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4330 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4331 "mov{d|q}\t{$src, $dst|$dst, $src}",
4332 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4334 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4335 "movd\t{$src, $dst|$dst, $src}",
4337 (v4i32 (scalar_to_vector GR32:$src)))]>;
4338 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4339 "movd\t{$src, $dst|$dst, $src}",
4341 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4342 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4343 "mov{d|q}\t{$src, $dst|$dst, $src}",
4345 (v2i64 (scalar_to_vector GR64:$src)))]>;
4346 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4347 "mov{d|q}\t{$src, $dst|$dst, $src}",
4348 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4350 //===---------------------------------------------------------------------===//
4351 // Move Int Doubleword to Single Scalar
4353 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4354 "movd\t{$src, $dst|$dst, $src}",
4355 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4357 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4358 "movd\t{$src, $dst|$dst, $src}",
4359 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4361 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4362 "movd\t{$src, $dst|$dst, $src}",
4363 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4365 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4366 "movd\t{$src, $dst|$dst, $src}",
4367 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4369 //===---------------------------------------------------------------------===//
4370 // Move Packed Doubleword Int to Packed Double Int
4372 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4373 "movd\t{$src, $dst|$dst, $src}",
4374 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4376 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4377 (ins i32mem:$dst, VR128:$src),
4378 "movd\t{$src, $dst|$dst, $src}",
4379 [(store (i32 (vector_extract (v4i32 VR128:$src),
4380 (iPTR 0))), addr:$dst)]>, VEX;
4381 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4382 "movd\t{$src, $dst|$dst, $src}",
4383 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4385 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4386 "movd\t{$src, $dst|$dst, $src}",
4387 [(store (i32 (vector_extract (v4i32 VR128:$src),
4388 (iPTR 0))), addr:$dst)]>;
4390 //===---------------------------------------------------------------------===//
4391 // Move Packed Doubleword Int first element to Doubleword Int
4393 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4394 "mov{d|q}\t{$src, $dst|$dst, $src}",
4395 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4397 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4399 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4400 "mov{d|q}\t{$src, $dst|$dst, $src}",
4401 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4404 //===---------------------------------------------------------------------===//
4405 // Bitcast FR64 <-> GR64
4407 let Predicates = [HasAVX] in
4408 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4409 "vmovq\t{$src, $dst|$dst, $src}",
4410 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4412 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4413 "mov{d|q}\t{$src, $dst|$dst, $src}",
4414 [(set GR64:$dst, (bitconvert FR64:$src))]>, VEX;
4415 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4416 "movq\t{$src, $dst|$dst, $src}",
4417 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>,
4420 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4421 "movq\t{$src, $dst|$dst, $src}",
4422 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4423 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4424 "mov{d|q}\t{$src, $dst|$dst, $src}",
4425 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4426 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4427 "movq\t{$src, $dst|$dst, $src}",
4428 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4430 //===---------------------------------------------------------------------===//
4431 // Move Scalar Single to Double Int
4433 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4434 "movd\t{$src, $dst|$dst, $src}",
4435 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4436 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4437 "movd\t{$src, $dst|$dst, $src}",
4438 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4439 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4440 "movd\t{$src, $dst|$dst, $src}",
4441 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4442 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4443 "movd\t{$src, $dst|$dst, $src}",
4444 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4446 //===---------------------------------------------------------------------===//
4447 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4449 let AddedComplexity = 15 in {
4450 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4451 "movd\t{$src, $dst|$dst, $src}",
4452 [(set VR128:$dst, (v4i32 (X86vzmovl
4453 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4455 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4456 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4457 [(set VR128:$dst, (v2i64 (X86vzmovl
4458 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4461 let AddedComplexity = 15 in {
4462 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4463 "movd\t{$src, $dst|$dst, $src}",
4464 [(set VR128:$dst, (v4i32 (X86vzmovl
4465 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4466 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4467 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4468 [(set VR128:$dst, (v2i64 (X86vzmovl
4469 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4472 let AddedComplexity = 20 in {
4473 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4474 "movd\t{$src, $dst|$dst, $src}",
4476 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4477 (loadi32 addr:$src))))))]>,
4479 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4480 "movd\t{$src, $dst|$dst, $src}",
4482 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4483 (loadi32 addr:$src))))))]>;
4486 let Predicates = [HasAVX] in {
4487 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4488 let AddedComplexity = 20 in {
4489 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4490 (VMOVZDI2PDIrm addr:$src)>;
4491 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4492 (VMOVZDI2PDIrm addr:$src)>;
4494 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4495 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4496 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4497 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4498 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4499 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4500 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4503 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4504 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4505 (MOVZDI2PDIrm addr:$src)>;
4506 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4507 (MOVZDI2PDIrm addr:$src)>;
4510 // These are the correct encodings of the instructions so that we know how to
4511 // read correct assembly, even though we continue to emit the wrong ones for
4512 // compatibility with Darwin's buggy assembler.
4513 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4514 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4515 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4516 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4517 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4518 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4519 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4520 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4521 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4522 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4523 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4524 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4526 //===---------------------------------------------------------------------===//
4527 // SSE2 - Move Quadword
4528 //===---------------------------------------------------------------------===//
4530 //===---------------------------------------------------------------------===//
4531 // Move Quadword Int to Packed Quadword Int
4533 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4534 "vmovq\t{$src, $dst|$dst, $src}",
4536 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4537 VEX, Requires<[HasAVX]>;
4538 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4539 "movq\t{$src, $dst|$dst, $src}",
4541 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4542 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4544 //===---------------------------------------------------------------------===//
4545 // Move Packed Quadword Int to Quadword Int
4547 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4548 "movq\t{$src, $dst|$dst, $src}",
4549 [(store (i64 (vector_extract (v2i64 VR128:$src),
4550 (iPTR 0))), addr:$dst)]>, VEX;
4551 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4552 "movq\t{$src, $dst|$dst, $src}",
4553 [(store (i64 (vector_extract (v2i64 VR128:$src),
4554 (iPTR 0))), addr:$dst)]>;
4556 //===---------------------------------------------------------------------===//
4557 // Store / copy lower 64-bits of a XMM register.
4559 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4560 "movq\t{$src, $dst|$dst, $src}",
4561 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4562 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4563 "movq\t{$src, $dst|$dst, $src}",
4564 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4566 let AddedComplexity = 20 in
4567 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4568 "vmovq\t{$src, $dst|$dst, $src}",
4570 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4571 (loadi64 addr:$src))))))]>,
4572 XS, VEX, Requires<[HasAVX]>;
4574 let AddedComplexity = 20 in
4575 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4576 "movq\t{$src, $dst|$dst, $src}",
4578 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4579 (loadi64 addr:$src))))))]>,
4580 XS, Requires<[HasSSE2]>;
4582 let Predicates = [HasAVX], AddedComplexity = 20 in {
4583 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4584 (VMOVZQI2PQIrm addr:$src)>;
4585 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4586 (VMOVZQI2PQIrm addr:$src)>;
4587 def : Pat<(v2i64 (X86vzload addr:$src)),
4588 (VMOVZQI2PQIrm addr:$src)>;
4591 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4592 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4593 (MOVZQI2PQIrm addr:$src)>;
4594 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4595 (MOVZQI2PQIrm addr:$src)>;
4596 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4599 let Predicates = [HasAVX] in {
4600 def : Pat<(v4i64 (X86vzload addr:$src)),
4601 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4604 //===---------------------------------------------------------------------===//
4605 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4606 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4608 let AddedComplexity = 15 in
4609 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4610 "vmovq\t{$src, $dst|$dst, $src}",
4611 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4612 XS, VEX, Requires<[HasAVX]>;
4613 let AddedComplexity = 15 in
4614 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4615 "movq\t{$src, $dst|$dst, $src}",
4616 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4617 XS, Requires<[HasSSE2]>;
4619 let AddedComplexity = 20 in
4620 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4621 "vmovq\t{$src, $dst|$dst, $src}",
4622 [(set VR128:$dst, (v2i64 (X86vzmovl
4623 (loadv2i64 addr:$src))))]>,
4624 XS, VEX, Requires<[HasAVX]>;
4625 let AddedComplexity = 20 in {
4626 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4627 "movq\t{$src, $dst|$dst, $src}",
4628 [(set VR128:$dst, (v2i64 (X86vzmovl
4629 (loadv2i64 addr:$src))))]>,
4630 XS, Requires<[HasSSE2]>;
4633 let AddedComplexity = 20 in {
4634 let Predicates = [HasAVX] in {
4635 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4636 (VMOVZPQILo2PQIrm addr:$src)>;
4637 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4638 (VMOVZPQILo2PQIrr VR128:$src)>;
4640 let Predicates = [HasSSE2] in {
4641 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4642 (MOVZPQILo2PQIrm addr:$src)>;
4643 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4644 (MOVZPQILo2PQIrr VR128:$src)>;
4648 // Instructions to match in the assembler
4649 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4650 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4651 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4652 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4653 // Recognize "movd" with GR64 destination, but encode as a "movq"
4654 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4655 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4657 // Instructions for the disassembler
4658 // xr = XMM register
4661 let Predicates = [HasAVX] in
4662 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4663 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4664 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4665 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4667 //===---------------------------------------------------------------------===//
4668 // SSE3 - Conversion Instructions
4669 //===---------------------------------------------------------------------===//
4671 // Convert Packed Double FP to Packed DW Integers
4672 let Predicates = [HasAVX] in {
4673 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4674 // register, but the same isn't true when using memory operands instead.
4675 // Provide other assembly rr and rm forms to address this explicitly.
4676 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4677 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4678 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4679 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4682 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4683 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4684 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4685 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4688 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4689 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4690 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4691 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4694 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4695 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4696 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4697 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4699 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4700 (VCVTPD2DQYrr VR256:$src)>;
4701 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4702 (VCVTPD2DQYrm addr:$src)>;
4704 // Convert Packed DW Integers to Packed Double FP
4705 let Predicates = [HasAVX] in {
4706 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4707 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4708 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4709 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4710 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4711 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4712 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4713 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4716 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4717 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4718 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4719 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4721 // AVX 256-bit register conversion intrinsics
4722 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4723 (VCVTDQ2PDYrr VR128:$src)>;
4724 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4725 (VCVTDQ2PDYrm addr:$src)>;
4727 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4728 (VCVTPD2DQYrr VR256:$src)>;
4729 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4730 (VCVTPD2DQYrm addr:$src)>;
4732 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4733 (VCVTDQ2PDYrr VR128:$src)>;
4734 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4735 (VCVTDQ2PDYrm addr:$src)>;
4737 //===---------------------------------------------------------------------===//
4738 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4739 //===---------------------------------------------------------------------===//
4740 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4741 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4742 X86MemOperand x86memop> {
4743 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4744 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4745 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4746 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4747 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4748 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4751 let Predicates = [HasAVX] in {
4752 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4753 v4f32, VR128, memopv4f32, f128mem>, VEX;
4754 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4755 v4f32, VR128, memopv4f32, f128mem>, VEX;
4756 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4757 v8f32, VR256, memopv8f32, f256mem>, VEX;
4758 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4759 v8f32, VR256, memopv8f32, f256mem>, VEX;
4761 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4762 memopv4f32, f128mem>;
4763 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4764 memopv4f32, f128mem>;
4766 let Predicates = [HasAVX] in {
4767 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4768 (VMOVSHDUPrr VR128:$src)>;
4769 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4770 (VMOVSHDUPrm addr:$src)>;
4771 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4772 (VMOVSLDUPrr VR128:$src)>;
4773 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4774 (VMOVSLDUPrm addr:$src)>;
4775 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4776 (VMOVSHDUPYrr VR256:$src)>;
4777 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4778 (VMOVSHDUPYrm addr:$src)>;
4779 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4780 (VMOVSLDUPYrr VR256:$src)>;
4781 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4782 (VMOVSLDUPYrm addr:$src)>;
4785 let Predicates = [HasSSE3] in {
4786 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4787 (MOVSHDUPrr VR128:$src)>;
4788 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4789 (MOVSHDUPrm addr:$src)>;
4790 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4791 (MOVSLDUPrr VR128:$src)>;
4792 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4793 (MOVSLDUPrm addr:$src)>;
4796 //===---------------------------------------------------------------------===//
4797 // SSE3 - Replicate Double FP - MOVDDUP
4798 //===---------------------------------------------------------------------===//
4800 multiclass sse3_replicate_dfp<string OpcodeStr> {
4801 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4802 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4803 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4804 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4805 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4807 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4811 // FIXME: Merge with above classe when there're patterns for the ymm version
4812 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4813 let Predicates = [HasAVX] in {
4814 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4815 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4817 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4818 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4823 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4824 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4825 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4827 let Predicates = [HasAVX] in {
4828 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4830 (VMOVDDUPrm addr:$src)>;
4831 let AddedComplexity = 5 in {
4832 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4833 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4834 (VMOVDDUPrm addr:$src)>;
4835 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4836 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4837 (VMOVDDUPrm addr:$src)>;
4839 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4840 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4841 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4842 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4843 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4844 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4845 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4846 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4847 def : Pat<(X86Movddup (bc_v2f64
4848 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4849 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4852 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4853 (VMOVDDUPYrm addr:$src)>;
4854 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4855 (VMOVDDUPYrm addr:$src)>;
4856 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
4857 (VMOVDDUPYrm addr:$src)>;
4858 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4859 (VMOVDDUPYrm addr:$src)>;
4860 def : Pat<(X86Movddup (v4f64 VR256:$src)),
4861 (VMOVDDUPYrr VR256:$src)>;
4862 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4863 (VMOVDDUPYrr VR256:$src)>;
4866 let Predicates = [HasSSE3] in {
4867 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4869 (MOVDDUPrm addr:$src)>;
4870 let AddedComplexity = 5 in {
4871 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4872 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4873 (MOVDDUPrm addr:$src)>;
4874 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4875 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4876 (MOVDDUPrm addr:$src)>;
4878 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4879 (MOVDDUPrm addr:$src)>;
4880 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4881 (MOVDDUPrm addr:$src)>;
4882 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4883 (MOVDDUPrm addr:$src)>;
4884 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4885 (MOVDDUPrm addr:$src)>;
4886 def : Pat<(X86Movddup (bc_v2f64
4887 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4888 (MOVDDUPrm addr:$src)>;
4891 //===---------------------------------------------------------------------===//
4892 // SSE3 - Move Unaligned Integer
4893 //===---------------------------------------------------------------------===//
4895 let Predicates = [HasAVX] in {
4896 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4897 "vlddqu\t{$src, $dst|$dst, $src}",
4898 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4899 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4900 "vlddqu\t{$src, $dst|$dst, $src}",
4901 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4903 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4904 "lddqu\t{$src, $dst|$dst, $src}",
4905 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
4907 //===---------------------------------------------------------------------===//
4908 // SSE3 - Arithmetic
4909 //===---------------------------------------------------------------------===//
4911 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4912 X86MemOperand x86memop, bit Is2Addr = 1> {
4913 def rr : I<0xD0, MRMSrcReg,
4914 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4916 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4917 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4918 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
4919 def rm : I<0xD0, MRMSrcMem,
4920 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4922 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4923 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4924 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
4927 let Predicates = [HasAVX] in {
4928 let ExeDomain = SSEPackedSingle in {
4929 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4930 f128mem, 0>, TB, XD, VEX_4V;
4931 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4932 f256mem, 0>, TB, XD, VEX_4V;
4934 let ExeDomain = SSEPackedDouble in {
4935 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4936 f128mem, 0>, TB, OpSize, VEX_4V;
4937 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4938 f256mem, 0>, TB, OpSize, VEX_4V;
4941 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
4942 let ExeDomain = SSEPackedSingle in
4943 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4945 let ExeDomain = SSEPackedDouble in
4946 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4947 f128mem>, TB, OpSize;
4950 //===---------------------------------------------------------------------===//
4951 // SSE3 Instructions
4952 //===---------------------------------------------------------------------===//
4955 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4956 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4957 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4959 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4960 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4961 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
4963 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4965 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4966 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4967 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
4969 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4970 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4971 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4973 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4974 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4975 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
4977 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4979 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4980 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4981 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
4984 let Predicates = [HasAVX] in {
4985 let ExeDomain = SSEPackedSingle in {
4986 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4987 X86fhadd, 0>, VEX_4V;
4988 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4989 X86fhsub, 0>, VEX_4V;
4990 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4991 X86fhadd, 0>, VEX_4V;
4992 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4993 X86fhsub, 0>, VEX_4V;
4995 let ExeDomain = SSEPackedDouble in {
4996 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4997 X86fhadd, 0>, VEX_4V;
4998 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4999 X86fhsub, 0>, VEX_4V;
5000 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5001 X86fhadd, 0>, VEX_4V;
5002 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5003 X86fhsub, 0>, VEX_4V;
5007 let Constraints = "$src1 = $dst" in {
5008 let ExeDomain = SSEPackedSingle in {
5009 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5010 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5012 let ExeDomain = SSEPackedDouble in {
5013 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5014 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5018 //===---------------------------------------------------------------------===//
5019 // SSSE3 - Packed Absolute Instructions
5020 //===---------------------------------------------------------------------===//
5023 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5024 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5025 Intrinsic IntId128> {
5026 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5028 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5029 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5032 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5034 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5037 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5040 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5041 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5042 Intrinsic IntId256> {
5043 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5045 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5046 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5049 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5051 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5054 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5057 let Predicates = [HasAVX] in {
5058 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5059 int_x86_ssse3_pabs_b_128>, VEX;
5060 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5061 int_x86_ssse3_pabs_w_128>, VEX;
5062 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5063 int_x86_ssse3_pabs_d_128>, VEX;
5066 let Predicates = [HasAVX2] in {
5067 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5068 int_x86_avx2_pabs_b>, VEX;
5069 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5070 int_x86_avx2_pabs_w>, VEX;
5071 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5072 int_x86_avx2_pabs_d>, VEX;
5075 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5076 int_x86_ssse3_pabs_b_128>;
5077 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5078 int_x86_ssse3_pabs_w_128>;
5079 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5080 int_x86_ssse3_pabs_d_128>;
5082 //===---------------------------------------------------------------------===//
5083 // SSSE3 - Packed Binary Operator Instructions
5084 //===---------------------------------------------------------------------===//
5086 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5087 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5088 Intrinsic IntId128, bit Is2Addr = 1> {
5089 let isCommutable = 1 in
5090 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5091 (ins VR128:$src1, VR128:$src2),
5093 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5094 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5095 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5097 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5098 (ins VR128:$src1, i128mem:$src2),
5100 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5101 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5103 (IntId128 VR128:$src1,
5104 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5107 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5108 Intrinsic IntId256> {
5109 let isCommutable = 1 in
5110 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5111 (ins VR256:$src1, VR256:$src2),
5112 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5113 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5115 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5116 (ins VR256:$src1, i256mem:$src2),
5117 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5119 (IntId256 VR256:$src1,
5120 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5123 let ImmT = NoImm, Predicates = [HasAVX] in {
5124 let isCommutable = 0 in {
5125 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw",
5126 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5127 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd",
5128 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5129 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5130 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5131 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw",
5132 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5133 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd",
5134 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5135 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5136 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5137 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5138 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5139 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb",
5140 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5141 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb",
5142 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5143 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw",
5144 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5145 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd",
5146 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5148 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5149 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5152 let ImmT = NoImm, Predicates = [HasAVX2] in {
5153 let isCommutable = 0 in {
5154 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw",
5155 int_x86_avx2_phadd_w>, VEX_4V;
5156 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd",
5157 int_x86_avx2_phadd_d>, VEX_4V;
5158 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5159 int_x86_avx2_phadd_sw>, VEX_4V;
5160 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw",
5161 int_x86_avx2_phsub_w>, VEX_4V;
5162 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd",
5163 int_x86_avx2_phsub_d>, VEX_4V;
5164 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5165 int_x86_avx2_phsub_sw>, VEX_4V;
5166 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5167 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5168 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb",
5169 int_x86_avx2_pshuf_b>, VEX_4V;
5170 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb",
5171 int_x86_avx2_psign_b>, VEX_4V;
5172 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw",
5173 int_x86_avx2_psign_w>, VEX_4V;
5174 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd",
5175 int_x86_avx2_psign_d>, VEX_4V;
5177 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5178 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5181 // None of these have i8 immediate fields.
5182 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5183 let isCommutable = 0 in {
5184 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw",
5185 int_x86_ssse3_phadd_w_128>;
5186 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd",
5187 int_x86_ssse3_phadd_d_128>;
5188 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5189 int_x86_ssse3_phadd_sw_128>;
5190 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw",
5191 int_x86_ssse3_phsub_w_128>;
5192 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd",
5193 int_x86_ssse3_phsub_d_128>;
5194 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5195 int_x86_ssse3_phsub_sw_128>;
5196 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5197 int_x86_ssse3_pmadd_ub_sw_128>;
5198 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb",
5199 int_x86_ssse3_pshuf_b_128>;
5200 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb",
5201 int_x86_ssse3_psign_b_128>;
5202 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw",
5203 int_x86_ssse3_psign_w_128>;
5204 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd",
5205 int_x86_ssse3_psign_d_128>;
5207 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5208 int_x86_ssse3_pmul_hr_sw_128>;
5211 let Predicates = [HasAVX] in {
5212 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5213 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5214 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5215 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5217 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5218 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5219 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5220 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5221 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5222 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5224 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5225 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5226 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5227 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5228 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5229 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5230 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5231 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5234 let Predicates = [HasAVX2] in {
5235 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5236 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5237 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5238 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5239 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5240 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5242 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5243 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5244 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5245 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5246 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5247 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5248 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5249 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5252 let Predicates = [HasSSSE3] in {
5253 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5254 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5255 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5256 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5258 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5259 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5260 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5261 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5262 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5263 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5265 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5266 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5267 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5268 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5269 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5270 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5271 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5272 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5275 //===---------------------------------------------------------------------===//
5276 // SSSE3 - Packed Align Instruction Patterns
5277 //===---------------------------------------------------------------------===//
5279 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5280 let neverHasSideEffects = 1 in {
5281 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5282 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5284 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5286 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5289 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5290 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5292 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5294 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5299 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5300 let neverHasSideEffects = 1 in {
5301 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5302 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5304 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5307 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5308 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5310 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5315 let Predicates = [HasAVX] in
5316 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5317 let Predicates = [HasAVX2] in
5318 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5319 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5320 defm PALIGN : ssse3_palign<"palignr">;
5322 let Predicates = [HasAVX2] in {
5323 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5324 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5325 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5326 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5327 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5328 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5329 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5330 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5333 let Predicates = [HasAVX] in {
5334 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5335 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5336 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5337 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5338 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5339 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5340 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5341 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5344 let Predicates = [HasSSSE3] in {
5345 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5346 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5347 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5348 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5349 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5350 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5351 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5352 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5355 //===---------------------------------------------------------------------===//
5356 // SSSE3 - Thread synchronization
5357 //===---------------------------------------------------------------------===//
5359 let usesCustomInserter = 1 in {
5360 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5361 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5362 Requires<[HasSSE3]>;
5363 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5364 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5365 Requires<[HasSSE3]>;
5368 let Uses = [EAX, ECX, EDX] in
5369 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5370 Requires<[HasSSE3]>;
5371 let Uses = [ECX, EAX] in
5372 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5373 Requires<[HasSSE3]>;
5375 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5376 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5378 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5379 Requires<[In32BitMode]>;
5380 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5381 Requires<[In64BitMode]>;
5383 //===----------------------------------------------------------------------===//
5384 // SSE4.1 - Packed Move with Sign/Zero Extend
5385 //===----------------------------------------------------------------------===//
5387 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5388 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5389 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5390 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5392 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5393 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5395 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5399 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5401 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5402 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5403 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5405 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5406 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5407 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5410 let Predicates = [HasAVX] in {
5411 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5413 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5415 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5417 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5419 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5421 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5425 let Predicates = [HasAVX2] in {
5426 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5427 int_x86_avx2_pmovsxbw>, VEX;
5428 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5429 int_x86_avx2_pmovsxwd>, VEX;
5430 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5431 int_x86_avx2_pmovsxdq>, VEX;
5432 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5433 int_x86_avx2_pmovzxbw>, VEX;
5434 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5435 int_x86_avx2_pmovzxwd>, VEX;
5436 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5437 int_x86_avx2_pmovzxdq>, VEX;
5440 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5441 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5442 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5443 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5444 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5445 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5447 let Predicates = [HasAVX] in {
5448 // Common patterns involving scalar load.
5449 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5450 (VPMOVSXBWrm addr:$src)>;
5451 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5452 (VPMOVSXBWrm addr:$src)>;
5454 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5455 (VPMOVSXWDrm addr:$src)>;
5456 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5457 (VPMOVSXWDrm addr:$src)>;
5459 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5460 (VPMOVSXDQrm addr:$src)>;
5461 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5462 (VPMOVSXDQrm addr:$src)>;
5464 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5465 (VPMOVZXBWrm addr:$src)>;
5466 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5467 (VPMOVZXBWrm addr:$src)>;
5469 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5470 (VPMOVZXWDrm addr:$src)>;
5471 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5472 (VPMOVZXWDrm addr:$src)>;
5474 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5475 (VPMOVZXDQrm addr:$src)>;
5476 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5477 (VPMOVZXDQrm addr:$src)>;
5480 let Predicates = [HasSSE41] in {
5481 // Common patterns involving scalar load.
5482 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5483 (PMOVSXBWrm addr:$src)>;
5484 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5485 (PMOVSXBWrm addr:$src)>;
5487 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5488 (PMOVSXWDrm addr:$src)>;
5489 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5490 (PMOVSXWDrm addr:$src)>;
5492 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5493 (PMOVSXDQrm addr:$src)>;
5494 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5495 (PMOVSXDQrm addr:$src)>;
5497 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5498 (PMOVZXBWrm addr:$src)>;
5499 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5500 (PMOVZXBWrm addr:$src)>;
5502 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5503 (PMOVZXWDrm addr:$src)>;
5504 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5505 (PMOVZXWDrm addr:$src)>;
5507 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5508 (PMOVZXDQrm addr:$src)>;
5509 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5510 (PMOVZXDQrm addr:$src)>;
5514 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5515 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5516 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5517 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5519 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5520 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5522 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5526 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5528 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5529 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5530 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5532 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5533 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5535 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5539 let Predicates = [HasAVX] in {
5540 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5542 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5544 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5546 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5550 let Predicates = [HasAVX2] in {
5551 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5552 int_x86_avx2_pmovsxbd>, VEX;
5553 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5554 int_x86_avx2_pmovsxwq>, VEX;
5555 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5556 int_x86_avx2_pmovzxbd>, VEX;
5557 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5558 int_x86_avx2_pmovzxwq>, VEX;
5561 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5562 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5563 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5564 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5566 let Predicates = [HasAVX] in {
5567 // Common patterns involving scalar load
5568 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5569 (VPMOVSXBDrm addr:$src)>;
5570 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5571 (VPMOVSXWQrm addr:$src)>;
5573 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5574 (VPMOVZXBDrm addr:$src)>;
5575 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5576 (VPMOVZXWQrm addr:$src)>;
5579 let Predicates = [HasSSE41] in {
5580 // Common patterns involving scalar load
5581 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5582 (PMOVSXBDrm addr:$src)>;
5583 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5584 (PMOVSXWQrm addr:$src)>;
5586 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5587 (PMOVZXBDrm addr:$src)>;
5588 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5589 (PMOVZXWQrm addr:$src)>;
5592 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5593 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5594 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5595 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5597 // Expecting a i16 load any extended to i32 value.
5598 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5599 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5600 [(set VR128:$dst, (IntId (bitconvert
5601 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5605 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5607 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5608 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5609 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5611 // Expecting a i16 load any extended to i32 value.
5612 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5613 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5614 [(set VR256:$dst, (IntId (bitconvert
5615 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5619 let Predicates = [HasAVX] in {
5620 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5622 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5625 let Predicates = [HasAVX2] in {
5626 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5627 int_x86_avx2_pmovsxbq>, VEX;
5628 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5629 int_x86_avx2_pmovzxbq>, VEX;
5631 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5632 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5634 let Predicates = [HasAVX] in {
5635 // Common patterns involving scalar load
5636 def : Pat<(int_x86_sse41_pmovsxbq
5637 (bitconvert (v4i32 (X86vzmovl
5638 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5639 (VPMOVSXBQrm addr:$src)>;
5641 def : Pat<(int_x86_sse41_pmovzxbq
5642 (bitconvert (v4i32 (X86vzmovl
5643 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5644 (VPMOVZXBQrm addr:$src)>;
5647 let Predicates = [HasSSE41] in {
5648 // Common patterns involving scalar load
5649 def : Pat<(int_x86_sse41_pmovsxbq
5650 (bitconvert (v4i32 (X86vzmovl
5651 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5652 (PMOVSXBQrm addr:$src)>;
5654 def : Pat<(int_x86_sse41_pmovzxbq
5655 (bitconvert (v4i32 (X86vzmovl
5656 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5657 (PMOVZXBQrm addr:$src)>;
5660 //===----------------------------------------------------------------------===//
5661 // SSE4.1 - Extract Instructions
5662 //===----------------------------------------------------------------------===//
5664 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5665 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5666 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5667 (ins VR128:$src1, i32i8imm:$src2),
5668 !strconcat(OpcodeStr,
5669 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5670 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5672 let neverHasSideEffects = 1, mayStore = 1 in
5673 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5674 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5675 !strconcat(OpcodeStr,
5676 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5679 // There's an AssertZext in the way of writing the store pattern
5680 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5683 let Predicates = [HasAVX] in {
5684 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5685 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5686 (ins VR128:$src1, i32i8imm:$src2),
5687 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5690 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5693 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5694 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5695 let neverHasSideEffects = 1, mayStore = 1 in
5696 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5697 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5698 !strconcat(OpcodeStr,
5699 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5702 // There's an AssertZext in the way of writing the store pattern
5703 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5706 let Predicates = [HasAVX] in
5707 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5709 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5712 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5713 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5714 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5715 (ins VR128:$src1, i32i8imm:$src2),
5716 !strconcat(OpcodeStr,
5717 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5719 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5720 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5721 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5722 !strconcat(OpcodeStr,
5723 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5724 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5725 addr:$dst)]>, OpSize;
5728 let Predicates = [HasAVX] in
5729 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5731 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5733 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5734 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5735 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5736 (ins VR128:$src1, i32i8imm:$src2),
5737 !strconcat(OpcodeStr,
5738 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5740 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5741 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5742 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5743 !strconcat(OpcodeStr,
5744 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5745 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5746 addr:$dst)]>, OpSize, REX_W;
5749 let Predicates = [HasAVX] in
5750 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5752 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5754 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5756 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5757 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5758 (ins VR128:$src1, i32i8imm:$src2),
5759 !strconcat(OpcodeStr,
5760 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5762 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5764 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5765 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5766 !strconcat(OpcodeStr,
5767 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5768 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5769 addr:$dst)]>, OpSize;
5772 let ExeDomain = SSEPackedSingle in {
5773 let Predicates = [HasAVX] in {
5774 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5775 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5776 (ins VR128:$src1, i32i8imm:$src2),
5777 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5780 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5783 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5784 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5787 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5789 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5792 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5793 Requires<[HasSSE41]>;
5795 //===----------------------------------------------------------------------===//
5796 // SSE4.1 - Insert Instructions
5797 //===----------------------------------------------------------------------===//
5799 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5800 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5801 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5803 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5805 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5807 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5808 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5809 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5811 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5813 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5815 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5816 imm:$src3))]>, OpSize;
5819 let Predicates = [HasAVX] in
5820 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5821 let Constraints = "$src1 = $dst" in
5822 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5824 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5825 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5826 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5828 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5830 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5832 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5834 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5835 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5837 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5839 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5841 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5842 imm:$src3)))]>, OpSize;
5845 let Predicates = [HasAVX] in
5846 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5847 let Constraints = "$src1 = $dst" in
5848 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5850 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5851 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5852 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5854 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5856 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5858 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5860 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5861 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5863 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5865 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5867 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5868 imm:$src3)))]>, OpSize;
5871 let Predicates = [HasAVX] in
5872 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5873 let Constraints = "$src1 = $dst" in
5874 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5876 // insertps has a few different modes, there's the first two here below which
5877 // are optimized inserts that won't zero arbitrary elements in the destination
5878 // vector. The next one matches the intrinsic and could zero arbitrary elements
5879 // in the target vector.
5880 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5881 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5882 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5884 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5886 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5888 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5890 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5891 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5893 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5895 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5897 (X86insrtps VR128:$src1,
5898 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5899 imm:$src3))]>, OpSize;
5902 let ExeDomain = SSEPackedSingle in {
5903 let Predicates = [HasAVX] in
5904 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5905 let Constraints = "$src1 = $dst" in
5906 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5909 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5910 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5912 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5913 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5914 Requires<[HasSSE41]>;
5916 //===----------------------------------------------------------------------===//
5917 // SSE4.1 - Round Instructions
5918 //===----------------------------------------------------------------------===//
5920 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5921 X86MemOperand x86memop, RegisterClass RC,
5922 PatFrag mem_frag32, PatFrag mem_frag64,
5923 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5924 let ExeDomain = SSEPackedSingle in {
5925 // Intrinsic operation, reg.
5926 // Vector intrinsic operation, reg
5927 def PSr : SS4AIi8<opcps, MRMSrcReg,
5928 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5929 !strconcat(OpcodeStr,
5930 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5931 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5934 // Vector intrinsic operation, mem
5935 def PSm : SS4AIi8<opcps, MRMSrcMem,
5936 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5937 !strconcat(OpcodeStr,
5938 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5940 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
5942 } // ExeDomain = SSEPackedSingle
5944 let ExeDomain = SSEPackedDouble in {
5945 // Vector intrinsic operation, reg
5946 def PDr : SS4AIi8<opcpd, MRMSrcReg,
5947 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5948 !strconcat(OpcodeStr,
5949 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5950 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
5953 // Vector intrinsic operation, mem
5954 def PDm : SS4AIi8<opcpd, MRMSrcMem,
5955 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5956 !strconcat(OpcodeStr,
5957 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5959 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
5961 } // ExeDomain = SSEPackedDouble
5964 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
5967 Intrinsic F64Int, bit Is2Addr = 1> {
5968 let ExeDomain = GenericDomain in {
5970 def SSr : SS4AIi8<opcss, MRMSrcReg,
5971 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
5973 !strconcat(OpcodeStr,
5974 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5975 !strconcat(OpcodeStr,
5976 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5979 // Intrinsic operation, reg.
5980 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
5981 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5983 !strconcat(OpcodeStr,
5984 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5985 !strconcat(OpcodeStr,
5986 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5987 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5990 // Intrinsic operation, mem.
5991 def SSm : SS4AIi8<opcss, MRMSrcMem,
5992 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5994 !strconcat(OpcodeStr,
5995 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5996 !strconcat(OpcodeStr,
5997 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5999 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6003 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6004 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6006 !strconcat(OpcodeStr,
6007 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6008 !strconcat(OpcodeStr,
6009 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6012 // Intrinsic operation, reg.
6013 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6014 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6016 !strconcat(OpcodeStr,
6017 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6018 !strconcat(OpcodeStr,
6019 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6020 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6023 // Intrinsic operation, mem.
6024 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6025 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6027 !strconcat(OpcodeStr,
6028 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6029 !strconcat(OpcodeStr,
6030 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6032 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6034 } // ExeDomain = GenericDomain
6037 // FP round - roundss, roundps, roundsd, roundpd
6038 let Predicates = [HasAVX] in {
6040 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6041 memopv4f32, memopv2f64,
6042 int_x86_sse41_round_ps,
6043 int_x86_sse41_round_pd>, VEX;
6044 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6045 memopv8f32, memopv4f64,
6046 int_x86_avx_round_ps_256,
6047 int_x86_avx_round_pd_256>, VEX;
6048 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6049 int_x86_sse41_round_ss,
6050 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6052 def : Pat<(ffloor FR32:$src),
6053 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6054 def : Pat<(f64 (ffloor FR64:$src)),
6055 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6056 def : Pat<(f32 (fnearbyint FR32:$src)),
6057 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6058 def : Pat<(f64 (fnearbyint FR64:$src)),
6059 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6060 def : Pat<(f32 (fceil FR32:$src)),
6061 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6062 def : Pat<(f64 (fceil FR64:$src)),
6063 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6064 def : Pat<(f32 (frint FR32:$src)),
6065 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6066 def : Pat<(f64 (frint FR64:$src)),
6067 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6068 def : Pat<(f32 (ftrunc FR32:$src)),
6069 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6070 def : Pat<(f64 (ftrunc FR64:$src)),
6071 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6074 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6075 memopv4f32, memopv2f64,
6076 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6077 let Constraints = "$src1 = $dst" in
6078 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6079 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6081 def : Pat<(ffloor FR32:$src),
6082 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6083 def : Pat<(f64 (ffloor FR64:$src)),
6084 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6085 def : Pat<(f32 (fnearbyint FR32:$src)),
6086 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6087 def : Pat<(f64 (fnearbyint FR64:$src)),
6088 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6089 def : Pat<(f32 (fceil FR32:$src)),
6090 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6091 def : Pat<(f64 (fceil FR64:$src)),
6092 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6093 def : Pat<(f32 (frint FR32:$src)),
6094 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6095 def : Pat<(f64 (frint FR64:$src)),
6096 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6097 def : Pat<(f32 (ftrunc FR32:$src)),
6098 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6099 def : Pat<(f64 (ftrunc FR64:$src)),
6100 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6102 //===----------------------------------------------------------------------===//
6103 // SSE4.1 - Packed Bit Test
6104 //===----------------------------------------------------------------------===//
6106 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6107 // the intel intrinsic that corresponds to this.
6108 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6109 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6110 "vptest\t{$src2, $src1|$src1, $src2}",
6111 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6113 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6114 "vptest\t{$src2, $src1|$src1, $src2}",
6115 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6118 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6119 "vptest\t{$src2, $src1|$src1, $src2}",
6120 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6122 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6123 "vptest\t{$src2, $src1|$src1, $src2}",
6124 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6128 let Defs = [EFLAGS] in {
6129 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6130 "ptest\t{$src2, $src1|$src1, $src2}",
6131 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6133 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6134 "ptest\t{$src2, $src1|$src1, $src2}",
6135 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6139 // The bit test instructions below are AVX only
6140 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6141 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6142 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6143 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6144 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6145 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6146 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6147 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6151 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6152 let ExeDomain = SSEPackedSingle in {
6153 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6154 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6156 let ExeDomain = SSEPackedDouble in {
6157 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6158 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6162 //===----------------------------------------------------------------------===//
6163 // SSE4.1 - Misc Instructions
6164 //===----------------------------------------------------------------------===//
6166 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6167 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6168 "popcnt{w}\t{$src, $dst|$dst, $src}",
6169 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6171 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6172 "popcnt{w}\t{$src, $dst|$dst, $src}",
6173 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6174 (implicit EFLAGS)]>, OpSize, XS;
6176 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6177 "popcnt{l}\t{$src, $dst|$dst, $src}",
6178 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6180 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6181 "popcnt{l}\t{$src, $dst|$dst, $src}",
6182 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6183 (implicit EFLAGS)]>, XS;
6185 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6186 "popcnt{q}\t{$src, $dst|$dst, $src}",
6187 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6189 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6190 "popcnt{q}\t{$src, $dst|$dst, $src}",
6191 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6192 (implicit EFLAGS)]>, XS;
6197 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6198 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6199 Intrinsic IntId128> {
6200 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6202 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6203 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6204 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6206 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6209 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6212 let Predicates = [HasAVX] in
6213 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6214 int_x86_sse41_phminposuw>, VEX;
6215 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6216 int_x86_sse41_phminposuw>;
6218 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6219 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6220 Intrinsic IntId128, bit Is2Addr = 1> {
6221 let isCommutable = 1 in
6222 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6223 (ins VR128:$src1, VR128:$src2),
6225 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6226 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6227 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6228 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6229 (ins VR128:$src1, i128mem:$src2),
6231 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6232 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6234 (IntId128 VR128:$src1,
6235 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6238 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6239 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6240 Intrinsic IntId256> {
6241 let isCommutable = 1 in
6242 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6243 (ins VR256:$src1, VR256:$src2),
6244 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6245 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6246 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6247 (ins VR256:$src1, i256mem:$src2),
6248 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6250 (IntId256 VR256:$src1,
6251 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6254 let Predicates = [HasAVX] in {
6255 let isCommutable = 0 in
6256 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6258 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6260 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6262 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6264 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6266 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6268 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6270 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6272 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6274 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6278 let Predicates = [HasAVX2] in {
6279 let isCommutable = 0 in
6280 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6281 int_x86_avx2_packusdw>, VEX_4V;
6282 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6283 int_x86_avx2_pmins_b>, VEX_4V;
6284 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6285 int_x86_avx2_pmins_d>, VEX_4V;
6286 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6287 int_x86_avx2_pminu_d>, VEX_4V;
6288 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6289 int_x86_avx2_pminu_w>, VEX_4V;
6290 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6291 int_x86_avx2_pmaxs_b>, VEX_4V;
6292 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6293 int_x86_avx2_pmaxs_d>, VEX_4V;
6294 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6295 int_x86_avx2_pmaxu_d>, VEX_4V;
6296 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6297 int_x86_avx2_pmaxu_w>, VEX_4V;
6298 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6299 int_x86_avx2_pmul_dq>, VEX_4V;
6302 let Constraints = "$src1 = $dst" in {
6303 let isCommutable = 0 in
6304 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6305 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6306 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6307 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6308 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6309 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6310 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6311 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6312 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6313 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6316 /// SS48I_binop_rm - Simple SSE41 binary operator.
6317 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6318 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6319 X86MemOperand x86memop, bit Is2Addr = 1> {
6320 let isCommutable = 1 in
6321 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6322 (ins RC:$src1, RC:$src2),
6324 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6325 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6326 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6327 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6328 (ins RC:$src1, x86memop:$src2),
6330 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6331 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6333 (OpVT (OpNode RC:$src1,
6334 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6337 let Predicates = [HasAVX] in {
6338 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6339 memopv2i64, i128mem, 0>, VEX_4V;
6340 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6341 memopv2i64, i128mem, 0>, VEX_4V;
6343 let Predicates = [HasAVX2] in {
6344 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6345 memopv4i64, i256mem, 0>, VEX_4V;
6346 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6347 memopv4i64, i256mem, 0>, VEX_4V;
6350 let Constraints = "$src1 = $dst" in {
6351 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6352 memopv2i64, i128mem>;
6353 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6354 memopv2i64, i128mem>;
6357 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6358 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6359 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6360 X86MemOperand x86memop, bit Is2Addr = 1> {
6361 let isCommutable = 1 in
6362 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6363 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6365 !strconcat(OpcodeStr,
6366 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6367 !strconcat(OpcodeStr,
6368 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6369 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6371 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6372 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6374 !strconcat(OpcodeStr,
6375 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6376 !strconcat(OpcodeStr,
6377 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6380 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6384 let Predicates = [HasAVX] in {
6385 let isCommutable = 0 in {
6386 let ExeDomain = SSEPackedSingle in {
6387 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6388 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6389 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6390 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6392 let ExeDomain = SSEPackedDouble in {
6393 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6394 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6395 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6396 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6398 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6399 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6400 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6401 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6403 let ExeDomain = SSEPackedSingle in
6404 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6405 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6406 let ExeDomain = SSEPackedDouble in
6407 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6408 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6409 let ExeDomain = SSEPackedSingle in
6410 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6411 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6414 let Predicates = [HasAVX2] in {
6415 let isCommutable = 0 in {
6416 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6417 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6418 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6419 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6423 let Constraints = "$src1 = $dst" in {
6424 let isCommutable = 0 in {
6425 let ExeDomain = SSEPackedSingle in
6426 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6427 VR128, memopv4f32, i128mem>;
6428 let ExeDomain = SSEPackedDouble in
6429 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6430 VR128, memopv2f64, i128mem>;
6431 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6432 VR128, memopv2i64, i128mem>;
6433 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6434 VR128, memopv2i64, i128mem>;
6436 let ExeDomain = SSEPackedSingle in
6437 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6438 VR128, memopv4f32, i128mem>;
6439 let ExeDomain = SSEPackedDouble in
6440 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6441 VR128, memopv2f64, i128mem>;
6444 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6445 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6446 RegisterClass RC, X86MemOperand x86memop,
6447 PatFrag mem_frag, Intrinsic IntId> {
6448 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6449 (ins RC:$src1, RC:$src2, RC:$src3),
6450 !strconcat(OpcodeStr,
6451 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6452 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6453 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6455 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6456 (ins RC:$src1, x86memop:$src2, RC:$src3),
6457 !strconcat(OpcodeStr,
6458 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6460 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6462 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6465 let Predicates = [HasAVX] in {
6466 let ExeDomain = SSEPackedDouble in {
6467 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6468 memopv2f64, int_x86_sse41_blendvpd>;
6469 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6470 memopv4f64, int_x86_avx_blendv_pd_256>;
6471 } // ExeDomain = SSEPackedDouble
6472 let ExeDomain = SSEPackedSingle in {
6473 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6474 memopv4f32, int_x86_sse41_blendvps>;
6475 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6476 memopv8f32, int_x86_avx_blendv_ps_256>;
6477 } // ExeDomain = SSEPackedSingle
6478 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6479 memopv2i64, int_x86_sse41_pblendvb>;
6482 let Predicates = [HasAVX2] in {
6483 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6484 memopv4i64, int_x86_avx2_pblendvb>;
6487 let Predicates = [HasAVX] in {
6488 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6489 (v16i8 VR128:$src2))),
6490 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6491 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6492 (v4i32 VR128:$src2))),
6493 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6494 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6495 (v4f32 VR128:$src2))),
6496 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6497 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6498 (v2i64 VR128:$src2))),
6499 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6500 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6501 (v2f64 VR128:$src2))),
6502 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6503 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6504 (v8i32 VR256:$src2))),
6505 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6506 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6507 (v8f32 VR256:$src2))),
6508 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6509 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6510 (v4i64 VR256:$src2))),
6511 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6512 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6513 (v4f64 VR256:$src2))),
6514 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6517 let Predicates = [HasAVX2] in {
6518 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6519 (v32i8 VR256:$src2))),
6520 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6523 /// SS41I_ternary_int - SSE 4.1 ternary operator
6524 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6525 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6527 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6528 (ins VR128:$src1, VR128:$src2),
6529 !strconcat(OpcodeStr,
6530 "\t{$src2, $dst|$dst, $src2}"),
6531 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6534 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6535 (ins VR128:$src1, i128mem:$src2),
6536 !strconcat(OpcodeStr,
6537 "\t{$src2, $dst|$dst, $src2}"),
6540 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6544 let ExeDomain = SSEPackedDouble in
6545 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6546 int_x86_sse41_blendvpd>;
6547 let ExeDomain = SSEPackedSingle in
6548 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6549 int_x86_sse41_blendvps>;
6550 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6551 int_x86_sse41_pblendvb>;
6553 let Predicates = [HasSSE41] in {
6554 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6555 (v16i8 VR128:$src2))),
6556 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6557 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6558 (v4i32 VR128:$src2))),
6559 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6560 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6561 (v4f32 VR128:$src2))),
6562 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6563 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6564 (v2i64 VR128:$src2))),
6565 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6566 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6567 (v2f64 VR128:$src2))),
6568 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6571 let Predicates = [HasAVX] in
6572 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6573 "vmovntdqa\t{$src, $dst|$dst, $src}",
6574 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6576 let Predicates = [HasAVX2] in
6577 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6578 "vmovntdqa\t{$src, $dst|$dst, $src}",
6579 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6581 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6582 "movntdqa\t{$src, $dst|$dst, $src}",
6583 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6586 //===----------------------------------------------------------------------===//
6587 // SSE4.2 - Compare Instructions
6588 //===----------------------------------------------------------------------===//
6590 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6591 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6592 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6593 X86MemOperand x86memop, bit Is2Addr = 1> {
6594 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6595 (ins RC:$src1, RC:$src2),
6597 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6598 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6599 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6601 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6602 (ins RC:$src1, x86memop:$src2),
6604 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6605 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6607 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6610 let Predicates = [HasAVX] in
6611 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6612 memopv2i64, i128mem, 0>, VEX_4V;
6614 let Predicates = [HasAVX2] in
6615 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6616 memopv4i64, i256mem, 0>, VEX_4V;
6618 let Constraints = "$src1 = $dst" in
6619 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6620 memopv2i64, i128mem>;
6622 //===----------------------------------------------------------------------===//
6623 // SSE4.2 - String/text Processing Instructions
6624 //===----------------------------------------------------------------------===//
6626 // Packed Compare Implicit Length Strings, Return Mask
6627 multiclass pseudo_pcmpistrm<string asm> {
6628 def REG : PseudoI<(outs VR128:$dst),
6629 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6630 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6632 def MEM : PseudoI<(outs VR128:$dst),
6633 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6634 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6635 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6638 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6639 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6640 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6643 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6644 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6645 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6646 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6648 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6649 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6650 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6653 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6654 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6655 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6656 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6658 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6659 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6660 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6663 // Packed Compare Explicit Length Strings, Return Mask
6664 multiclass pseudo_pcmpestrm<string asm> {
6665 def REG : PseudoI<(outs VR128:$dst),
6666 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6667 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6668 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6669 def MEM : PseudoI<(outs VR128:$dst),
6670 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6671 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6672 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6675 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6676 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6677 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6680 let Predicates = [HasAVX],
6681 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6682 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6683 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6684 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6686 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6687 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6688 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6691 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6692 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6693 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6694 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6696 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6697 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6698 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6701 // Packed Compare Implicit Length Strings, Return Index
6702 let Defs = [ECX, EFLAGS] in {
6703 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6704 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6705 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6706 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6707 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6708 (implicit EFLAGS)]>, OpSize;
6709 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6710 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6711 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6712 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6713 (implicit EFLAGS)]>, OpSize;
6717 let Predicates = [HasAVX] in {
6718 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6720 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6722 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6724 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6726 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6728 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6732 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6733 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6734 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6735 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6736 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6737 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6739 // Packed Compare Explicit Length Strings, Return Index
6740 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6741 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6742 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6743 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6744 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6745 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6746 (implicit EFLAGS)]>, OpSize;
6747 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6748 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6749 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6751 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6752 (implicit EFLAGS)]>, OpSize;
6756 let Predicates = [HasAVX] in {
6757 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6759 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6761 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6763 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6765 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6767 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6771 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6772 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6773 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6774 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6775 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6776 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6778 //===----------------------------------------------------------------------===//
6779 // SSE4.2 - CRC Instructions
6780 //===----------------------------------------------------------------------===//
6782 // No CRC instructions have AVX equivalents
6784 // crc intrinsic instruction
6785 // This set of instructions are only rm, the only difference is the size
6787 let Constraints = "$src1 = $dst" in {
6788 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6789 (ins GR32:$src1, i8mem:$src2),
6790 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6792 (int_x86_sse42_crc32_32_8 GR32:$src1,
6793 (load addr:$src2)))]>;
6794 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6795 (ins GR32:$src1, GR8:$src2),
6796 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6798 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6799 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6800 (ins GR32:$src1, i16mem:$src2),
6801 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6803 (int_x86_sse42_crc32_32_16 GR32:$src1,
6804 (load addr:$src2)))]>,
6806 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6807 (ins GR32:$src1, GR16:$src2),
6808 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6810 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6812 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6813 (ins GR32:$src1, i32mem:$src2),
6814 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6816 (int_x86_sse42_crc32_32_32 GR32:$src1,
6817 (load addr:$src2)))]>;
6818 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6819 (ins GR32:$src1, GR32:$src2),
6820 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6822 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6823 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6824 (ins GR64:$src1, i8mem:$src2),
6825 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6827 (int_x86_sse42_crc32_64_8 GR64:$src1,
6828 (load addr:$src2)))]>,
6830 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6831 (ins GR64:$src1, GR8:$src2),
6832 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6834 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6836 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6837 (ins GR64:$src1, i64mem:$src2),
6838 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6840 (int_x86_sse42_crc32_64_64 GR64:$src1,
6841 (load addr:$src2)))]>,
6843 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6844 (ins GR64:$src1, GR64:$src2),
6845 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6847 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6851 //===----------------------------------------------------------------------===//
6852 // AES-NI Instructions
6853 //===----------------------------------------------------------------------===//
6855 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6856 Intrinsic IntId128, bit Is2Addr = 1> {
6857 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6858 (ins VR128:$src1, VR128:$src2),
6860 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6861 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6862 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6864 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6865 (ins VR128:$src1, i128mem:$src2),
6867 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6868 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6870 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6873 // Perform One Round of an AES Encryption/Decryption Flow
6874 let Predicates = [HasAVX, HasAES] in {
6875 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
6876 int_x86_aesni_aesenc, 0>, VEX_4V;
6877 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
6878 int_x86_aesni_aesenclast, 0>, VEX_4V;
6879 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
6880 int_x86_aesni_aesdec, 0>, VEX_4V;
6881 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
6882 int_x86_aesni_aesdeclast, 0>, VEX_4V;
6885 let Constraints = "$src1 = $dst" in {
6886 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
6887 int_x86_aesni_aesenc>;
6888 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
6889 int_x86_aesni_aesenclast>;
6890 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
6891 int_x86_aesni_aesdec>;
6892 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
6893 int_x86_aesni_aesdeclast>;
6896 // Perform the AES InvMixColumn Transformation
6897 let Predicates = [HasAVX, HasAES] in {
6898 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6900 "vaesimc\t{$src1, $dst|$dst, $src1}",
6902 (int_x86_aesni_aesimc VR128:$src1))]>,
6904 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6905 (ins i128mem:$src1),
6906 "vaesimc\t{$src1, $dst|$dst, $src1}",
6907 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
6910 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6912 "aesimc\t{$src1, $dst|$dst, $src1}",
6914 (int_x86_aesni_aesimc VR128:$src1))]>,
6916 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6917 (ins i128mem:$src1),
6918 "aesimc\t{$src1, $dst|$dst, $src1}",
6919 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
6922 // AES Round Key Generation Assist
6923 let Predicates = [HasAVX, HasAES] in {
6924 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6925 (ins VR128:$src1, i8imm:$src2),
6926 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6928 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6930 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6931 (ins i128mem:$src1, i8imm:$src2),
6932 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6934 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
6937 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6938 (ins VR128:$src1, i8imm:$src2),
6939 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6941 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6943 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6944 (ins i128mem:$src1, i8imm:$src2),
6945 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6947 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
6950 //===----------------------------------------------------------------------===//
6951 // CLMUL Instructions
6952 //===----------------------------------------------------------------------===//
6954 // Carry-less Multiplication instructions
6955 let neverHasSideEffects = 1 in {
6956 // AVX carry-less Multiplication instructions
6957 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6958 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6959 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6963 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6964 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6965 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6968 let Constraints = "$src1 = $dst" in {
6969 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6970 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6971 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6975 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6976 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6977 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6979 } // Constraints = "$src1 = $dst"
6980 } // neverHasSideEffects = 1
6983 multiclass pclmul_alias<string asm, int immop> {
6984 def : InstAlias<!strconcat("pclmul", asm,
6985 "dq {$src, $dst|$dst, $src}"),
6986 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
6988 def : InstAlias<!strconcat("pclmul", asm,
6989 "dq {$src, $dst|$dst, $src}"),
6990 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
6992 def : InstAlias<!strconcat("vpclmul", asm,
6993 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6994 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
6996 def : InstAlias<!strconcat("vpclmul", asm,
6997 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6998 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7000 defm : pclmul_alias<"hqhq", 0x11>;
7001 defm : pclmul_alias<"hqlq", 0x01>;
7002 defm : pclmul_alias<"lqhq", 0x10>;
7003 defm : pclmul_alias<"lqlq", 0x00>;
7005 //===----------------------------------------------------------------------===//
7007 //===----------------------------------------------------------------------===//
7009 //===----------------------------------------------------------------------===//
7010 // VBROADCAST - Load from memory and broadcast to all elements of the
7011 // destination operand
7013 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7014 X86MemOperand x86memop, Intrinsic Int> :
7015 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7016 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7017 [(set RC:$dst, (Int addr:$src))]>, VEX;
7019 // AVX2 adds register forms
7020 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7022 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7023 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7024 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7026 let ExeDomain = SSEPackedSingle in {
7027 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7028 int_x86_avx_vbroadcast_ss>;
7029 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7030 int_x86_avx_vbroadcast_ss_256>;
7032 let ExeDomain = SSEPackedDouble in
7033 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7034 int_x86_avx_vbroadcast_sd_256>;
7035 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7036 int_x86_avx_vbroadcastf128_pd_256>;
7038 let ExeDomain = SSEPackedSingle in {
7039 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7040 int_x86_avx2_vbroadcast_ss_ps>;
7041 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7042 int_x86_avx2_vbroadcast_ss_ps_256>;
7044 let ExeDomain = SSEPackedDouble in
7045 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7046 int_x86_avx2_vbroadcast_sd_pd_256>;
7048 let Predicates = [HasAVX2] in
7049 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7050 int_x86_avx2_vbroadcasti128>;
7052 let Predicates = [HasAVX] in
7053 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7054 (VBROADCASTF128 addr:$src)>;
7057 //===----------------------------------------------------------------------===//
7058 // VINSERTF128 - Insert packed floating-point values
7060 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7061 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7062 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7063 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7066 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7067 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7068 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7072 let Predicates = [HasAVX] in {
7073 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7074 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7075 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7076 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7077 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7078 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7081 //===----------------------------------------------------------------------===//
7082 // VEXTRACTF128 - Extract packed floating-point values
7084 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7085 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7086 (ins VR256:$src1, i8imm:$src2),
7087 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7090 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7091 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7092 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7096 let Predicates = [HasAVX] in {
7097 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7098 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7099 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7100 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7101 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7102 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7105 //===----------------------------------------------------------------------===//
7106 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7108 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7109 Intrinsic IntLd, Intrinsic IntLd256,
7110 Intrinsic IntSt, Intrinsic IntSt256> {
7111 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7112 (ins VR128:$src1, f128mem:$src2),
7113 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7114 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7116 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7117 (ins VR256:$src1, f256mem:$src2),
7118 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7119 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7121 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7122 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7123 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7124 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7125 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7126 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7127 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7128 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7131 let ExeDomain = SSEPackedSingle in
7132 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7133 int_x86_avx_maskload_ps,
7134 int_x86_avx_maskload_ps_256,
7135 int_x86_avx_maskstore_ps,
7136 int_x86_avx_maskstore_ps_256>;
7137 let ExeDomain = SSEPackedDouble in
7138 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7139 int_x86_avx_maskload_pd,
7140 int_x86_avx_maskload_pd_256,
7141 int_x86_avx_maskstore_pd,
7142 int_x86_avx_maskstore_pd_256>;
7144 //===----------------------------------------------------------------------===//
7145 // VPERMIL - Permute Single and Double Floating-Point Values
7147 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7148 RegisterClass RC, X86MemOperand x86memop_f,
7149 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7150 Intrinsic IntVar, Intrinsic IntImm> {
7151 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7152 (ins RC:$src1, RC:$src2),
7153 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7154 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7155 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7156 (ins RC:$src1, x86memop_i:$src2),
7157 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7158 [(set RC:$dst, (IntVar RC:$src1,
7159 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7161 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7162 (ins RC:$src1, i8imm:$src2),
7163 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7164 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7165 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7166 (ins x86memop_f:$src1, i8imm:$src2),
7167 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7168 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7171 let ExeDomain = SSEPackedSingle in {
7172 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7173 memopv4f32, memopv2i64,
7174 int_x86_avx_vpermilvar_ps,
7175 int_x86_avx_vpermil_ps>;
7176 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7177 memopv8f32, memopv4i64,
7178 int_x86_avx_vpermilvar_ps_256,
7179 int_x86_avx_vpermil_ps_256>;
7181 let ExeDomain = SSEPackedDouble in {
7182 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7183 memopv2f64, memopv2i64,
7184 int_x86_avx_vpermilvar_pd,
7185 int_x86_avx_vpermil_pd>;
7186 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7187 memopv4f64, memopv4i64,
7188 int_x86_avx_vpermilvar_pd_256,
7189 int_x86_avx_vpermil_pd_256>;
7192 let Predicates = [HasAVX] in {
7193 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7194 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7195 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7196 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7197 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7198 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7199 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7200 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7201 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7202 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7203 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7204 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7205 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7207 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7208 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7209 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7212 //===----------------------------------------------------------------------===//
7213 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7215 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7216 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7217 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7218 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7221 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7222 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7223 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7227 let Predicates = [HasAVX] in {
7228 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7229 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7230 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7231 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7232 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7233 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7235 def : Pat<(int_x86_avx_vperm2f128_ps_256
7236 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7237 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7238 def : Pat<(int_x86_avx_vperm2f128_pd_256
7239 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7240 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7241 def : Pat<(int_x86_avx_vperm2f128_si_256
7242 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7243 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7246 //===----------------------------------------------------------------------===//
7247 // VZERO - Zero YMM registers
7249 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7250 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7251 // Zero All YMM registers
7252 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7253 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7255 // Zero Upper bits of YMM registers
7256 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7257 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7260 //===----------------------------------------------------------------------===//
7261 // Half precision conversion instructions
7262 //===----------------------------------------------------------------------===//
7263 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7264 let Predicates = [HasAVX, HasF16C] in {
7265 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7266 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7267 [(set RC:$dst, (Int VR128:$src))]>,
7269 let neverHasSideEffects = 1, mayLoad = 1 in
7270 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7271 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7275 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7276 let Predicates = [HasAVX, HasF16C] in {
7277 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7278 (ins RC:$src1, i32i8imm:$src2),
7279 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7280 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7282 let neverHasSideEffects = 1, mayLoad = 1 in
7283 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7284 (ins RC:$src1, i32i8imm:$src2),
7285 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7290 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7291 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7292 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7293 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7295 //===----------------------------------------------------------------------===//
7296 // AVX2 Instructions
7297 //===----------------------------------------------------------------------===//
7299 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7300 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7301 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7302 X86MemOperand x86memop> {
7303 let isCommutable = 1 in
7304 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7305 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7306 !strconcat(OpcodeStr,
7307 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7308 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7310 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7311 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7312 !strconcat(OpcodeStr,
7313 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7316 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7320 let isCommutable = 0 in {
7321 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7322 VR128, memopv2i64, i128mem>;
7323 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7324 VR256, memopv4i64, i256mem>;
7327 //===----------------------------------------------------------------------===//
7328 // VPBROADCAST - Load from memory and broadcast to all elements of the
7329 // destination operand
7331 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7332 X86MemOperand x86memop, PatFrag ld_frag,
7333 Intrinsic Int128, Intrinsic Int256> {
7334 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7335 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7336 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7337 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7338 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7340 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7341 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7342 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7343 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7344 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7345 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7347 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7350 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7351 int_x86_avx2_pbroadcastb_128,
7352 int_x86_avx2_pbroadcastb_256>;
7353 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7354 int_x86_avx2_pbroadcastw_128,
7355 int_x86_avx2_pbroadcastw_256>;
7356 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7357 int_x86_avx2_pbroadcastd_128,
7358 int_x86_avx2_pbroadcastd_256>;
7359 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7360 int_x86_avx2_pbroadcastq_128,
7361 int_x86_avx2_pbroadcastq_256>;
7363 let Predicates = [HasAVX2] in {
7364 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7365 (VPBROADCASTBrm addr:$src)>;
7366 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7367 (VPBROADCASTBYrm addr:$src)>;
7368 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7369 (VPBROADCASTWrm addr:$src)>;
7370 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7371 (VPBROADCASTWYrm addr:$src)>;
7372 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7373 (VPBROADCASTDrm addr:$src)>;
7374 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7375 (VPBROADCASTDYrm addr:$src)>;
7376 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7377 (VPBROADCASTQrm addr:$src)>;
7378 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7379 (VPBROADCASTQYrm addr:$src)>;
7382 // AVX1 broadcast patterns
7383 let Predicates = [HasAVX] in {
7384 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7385 (VBROADCASTSSYrm addr:$src)>;
7386 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7387 (VBROADCASTSDrm addr:$src)>;
7388 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7389 (VBROADCASTSSYrm addr:$src)>;
7390 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7391 (VBROADCASTSDrm addr:$src)>;
7393 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7394 (VBROADCASTSSrm addr:$src)>;
7395 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7396 (VBROADCASTSSrm addr:$src)>;
7399 //===----------------------------------------------------------------------===//
7400 // VPERM - Permute instructions
7403 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7405 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7406 (ins VR256:$src1, VR256:$src2),
7407 !strconcat(OpcodeStr,
7408 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7409 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7410 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7411 (ins VR256:$src1, i256mem:$src2),
7412 !strconcat(OpcodeStr,
7413 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7414 [(set VR256:$dst, (Int VR256:$src1,
7415 (bitconvert (mem_frag addr:$src2))))]>,
7419 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7420 let ExeDomain = SSEPackedSingle in
7421 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7423 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7425 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7426 (ins VR256:$src1, i8imm:$src2),
7427 !strconcat(OpcodeStr,
7428 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7429 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7430 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7431 (ins i256mem:$src1, i8imm:$src2),
7432 !strconcat(OpcodeStr,
7433 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7434 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7438 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7440 let ExeDomain = SSEPackedDouble in
7441 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7444 //===----------------------------------------------------------------------===//
7445 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7447 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7448 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7449 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7451 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7453 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7454 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7455 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7457 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7461 let Predicates = [HasAVX2] in {
7462 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7463 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7464 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7465 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7466 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7467 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7468 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7469 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7471 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7473 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7474 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7475 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7476 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7477 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7479 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7480 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7482 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7486 let Predicates = [HasAVX] in {
7487 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7488 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7489 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7490 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7491 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7492 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7493 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7494 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7495 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7496 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7497 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7498 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7500 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7501 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7502 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7503 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7504 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7505 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7506 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7507 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7508 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7509 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7510 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7511 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7512 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7513 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7514 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7515 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7516 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7517 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7521 //===----------------------------------------------------------------------===//
7522 // VINSERTI128 - Insert packed integer values
7524 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7525 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7526 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7528 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7530 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7531 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7532 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7534 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7535 imm:$src3))]>, VEX_4V;
7537 let Predicates = [HasAVX2] in {
7538 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7540 (VINSERTI128rr VR256:$src1, VR128:$src2,
7541 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7542 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7544 (VINSERTI128rr VR256:$src1, VR128:$src2,
7545 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7546 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7548 (VINSERTI128rr VR256:$src1, VR128:$src2,
7549 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7550 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7552 (VINSERTI128rr VR256:$src1, VR128:$src2,
7553 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7557 let Predicates = [HasAVX] in {
7558 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7560 (VINSERTF128rr VR256:$src1, VR128:$src2,
7561 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7562 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7564 (VINSERTF128rr VR256:$src1, VR128:$src2,
7565 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7566 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7568 (VINSERTF128rr VR256:$src1, VR128:$src2,
7569 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7570 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7572 (VINSERTF128rr VR256:$src1, VR128:$src2,
7573 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7574 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7576 (VINSERTF128rr VR256:$src1, VR128:$src2,
7577 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7578 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7580 (VINSERTF128rr VR256:$src1, VR128:$src2,
7581 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7584 //===----------------------------------------------------------------------===//
7585 // VEXTRACTI128 - Extract packed integer values
7587 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7588 (ins VR256:$src1, i8imm:$src2),
7589 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7591 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7593 let neverHasSideEffects = 1, mayStore = 1 in
7594 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7595 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7596 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7598 let Predicates = [HasAVX2] in {
7599 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7600 (v2i64 (VEXTRACTI128rr
7601 (v4i64 VR256:$src1),
7602 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7603 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7604 (v4i32 (VEXTRACTI128rr
7605 (v8i32 VR256:$src1),
7606 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7607 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7608 (v8i16 (VEXTRACTI128rr
7609 (v16i16 VR256:$src1),
7610 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7611 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7612 (v16i8 (VEXTRACTI128rr
7613 (v32i8 VR256:$src1),
7614 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7618 let Predicates = [HasAVX] in {
7619 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7620 (v4f32 (VEXTRACTF128rr
7621 (v8f32 VR256:$src1),
7622 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7623 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7624 (v2f64 (VEXTRACTF128rr
7625 (v4f64 VR256:$src1),
7626 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7627 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7628 (v2i64 (VEXTRACTF128rr
7629 (v4i64 VR256:$src1),
7630 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7631 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7632 (v4i32 (VEXTRACTF128rr
7633 (v8i32 VR256:$src1),
7634 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7635 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7636 (v8i16 (VEXTRACTF128rr
7637 (v16i16 VR256:$src1),
7638 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7639 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7640 (v16i8 (VEXTRACTF128rr
7641 (v32i8 VR256:$src1),
7642 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7645 //===----------------------------------------------------------------------===//
7646 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7648 multiclass avx2_pmovmask<string OpcodeStr,
7649 Intrinsic IntLd128, Intrinsic IntLd256,
7650 Intrinsic IntSt128, Intrinsic IntSt256> {
7651 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7652 (ins VR128:$src1, i128mem:$src2),
7653 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7654 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7655 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7656 (ins VR256:$src1, i256mem:$src2),
7657 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7658 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7659 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7660 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7661 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7662 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7663 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7664 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7665 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7666 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7669 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7670 int_x86_avx2_maskload_d,
7671 int_x86_avx2_maskload_d_256,
7672 int_x86_avx2_maskstore_d,
7673 int_x86_avx2_maskstore_d_256>;
7674 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7675 int_x86_avx2_maskload_q,
7676 int_x86_avx2_maskload_q_256,
7677 int_x86_avx2_maskstore_q,
7678 int_x86_avx2_maskstore_q_256>, VEX_W;
7681 //===----------------------------------------------------------------------===//
7682 // Variable Bit Shifts
7684 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7685 ValueType vt128, ValueType vt256> {
7686 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7687 (ins VR128:$src1, VR128:$src2),
7688 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7690 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7692 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7693 (ins VR128:$src1, i128mem:$src2),
7694 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7696 (vt128 (OpNode VR128:$src1,
7697 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7699 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7700 (ins VR256:$src1, VR256:$src2),
7701 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7703 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7705 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7706 (ins VR256:$src1, i256mem:$src2),
7707 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7709 (vt256 (OpNode VR256:$src1,
7710 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7714 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7715 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7716 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7717 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7718 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;