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:$src)>;
889 // Use vmovaps/vmovups for AVX integer load/store.
890 let Predicates = [HasAVX] in {
891 // 128-bit load/store
892 def : Pat<(alignedloadv4i32 addr:$src),
893 (VMOVAPSrm addr:$src)>;
894 def : Pat<(loadv4i32 addr:$src),
895 (VMOVUPSrm addr:$src)>;
896 def : Pat<(alignedloadv2i64 addr:$src),
897 (VMOVAPSrm addr:$src)>;
898 def : Pat<(loadv2i64 addr:$src),
899 (VMOVUPSrm addr:$src)>;
901 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
902 (VMOVAPSmr addr:$dst, VR128:$src)>;
903 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
904 (VMOVAPSmr addr:$dst, VR128:$src)>;
905 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
906 (VMOVAPSmr addr:$dst, VR128:$src)>;
907 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
908 (VMOVAPSmr addr:$dst, VR128:$src)>;
909 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
910 (VMOVUPSmr addr:$dst, VR128:$src)>;
911 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
912 (VMOVUPSmr addr:$dst, VR128:$src)>;
913 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
914 (VMOVUPSmr addr:$dst, VR128:$src)>;
915 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
916 (VMOVUPSmr addr:$dst, VR128:$src)>;
918 // 256-bit load/store
919 def : Pat<(alignedloadv4i64 addr:$src),
920 (VMOVAPSYrm addr:$src)>;
921 def : Pat<(loadv4i64 addr:$src),
922 (VMOVUPSYrm addr:$src)>;
923 def : Pat<(alignedloadv8i32 addr:$src),
924 (VMOVAPSYrm addr:$src)>;
925 def : Pat<(loadv8i32 addr:$src),
926 (VMOVUPSYrm addr:$src)>;
927 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
928 (VMOVAPSYmr addr:$dst, VR256:$src)>;
929 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
930 (VMOVAPSYmr addr:$dst, VR256:$src)>;
931 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
932 (VMOVAPSYmr addr:$dst, VR256:$src)>;
933 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
934 (VMOVAPSYmr addr:$dst, VR256:$src)>;
935 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
936 (VMOVUPSYmr addr:$dst, VR256:$src)>;
937 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
938 (VMOVUPSYmr addr:$dst, VR256:$src)>;
939 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
940 (VMOVUPSYmr addr:$dst, VR256:$src)>;
941 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
942 (VMOVUPSYmr addr:$dst, VR256:$src)>;
945 // Use movaps / movups for SSE integer load / store (one byte shorter).
946 // The instructions selected below are then converted to MOVDQA/MOVDQU
947 // during the SSE domain pass.
948 let Predicates = [HasSSE1] in {
949 def : Pat<(alignedloadv4i32 addr:$src),
950 (MOVAPSrm addr:$src)>;
951 def : Pat<(loadv4i32 addr:$src),
952 (MOVUPSrm addr:$src)>;
953 def : Pat<(alignedloadv2i64 addr:$src),
954 (MOVAPSrm addr:$src)>;
955 def : Pat<(loadv2i64 addr:$src),
956 (MOVUPSrm addr:$src)>;
958 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
959 (MOVAPSmr addr:$dst, VR128:$src)>;
960 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
961 (MOVAPSmr addr:$dst, VR128:$src)>;
962 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
963 (MOVAPSmr addr:$dst, VR128:$src)>;
964 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
965 (MOVAPSmr addr:$dst, VR128:$src)>;
966 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
967 (MOVUPSmr addr:$dst, VR128:$src)>;
968 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
969 (MOVUPSmr addr:$dst, VR128:$src)>;
970 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
971 (MOVUPSmr addr:$dst, VR128:$src)>;
972 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
973 (MOVUPSmr addr:$dst, VR128:$src)>;
976 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
977 // bits are disregarded. FIXME: Set encoding to pseudo!
978 let neverHasSideEffects = 1 in {
979 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
980 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
981 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
982 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
983 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
984 "movaps\t{$src, $dst|$dst, $src}", []>;
985 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
986 "movapd\t{$src, $dst|$dst, $src}", []>;
989 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
990 // bits are disregarded. FIXME: Set encoding to pseudo!
991 let canFoldAsLoad = 1, isReMaterializable = 1 in {
992 let isCodeGenOnly = 1 in {
993 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
994 "movaps\t{$src, $dst|$dst, $src}",
995 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
996 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
997 "movapd\t{$src, $dst|$dst, $src}",
998 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
1000 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1001 "movaps\t{$src, $dst|$dst, $src}",
1002 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1003 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1004 "movapd\t{$src, $dst|$dst, $src}",
1005 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1008 //===----------------------------------------------------------------------===//
1009 // SSE 1 & 2 - Move Low packed FP Instructions
1010 //===----------------------------------------------------------------------===//
1012 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
1013 PatFrag mov_frag, string base_opc,
1015 def PSrm : PI<opc, MRMSrcMem,
1016 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1017 !strconcat(base_opc, "s", asm_opr),
1020 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1021 SSEPackedSingle>, TB;
1023 def PDrm : PI<opc, MRMSrcMem,
1024 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1025 !strconcat(base_opc, "d", asm_opr),
1026 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
1027 (scalar_to_vector (loadf64 addr:$src2)))))],
1028 SSEPackedDouble>, TB, OpSize;
1031 let AddedComplexity = 20 in {
1032 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1033 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1035 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1036 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1037 "\t{$src2, $dst|$dst, $src2}">;
1040 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1041 "movlps\t{$src, $dst|$dst, $src}",
1042 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1043 (iPTR 0))), addr:$dst)]>, VEX;
1044 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1045 "movlpd\t{$src, $dst|$dst, $src}",
1046 [(store (f64 (vector_extract (v2f64 VR128:$src),
1047 (iPTR 0))), addr:$dst)]>, VEX;
1048 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1049 "movlps\t{$src, $dst|$dst, $src}",
1050 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1051 (iPTR 0))), addr:$dst)]>;
1052 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1053 "movlpd\t{$src, $dst|$dst, $src}",
1054 [(store (f64 (vector_extract (v2f64 VR128:$src),
1055 (iPTR 0))), addr:$dst)]>;
1057 let Predicates = [HasAVX] in {
1058 let AddedComplexity = 20 in {
1059 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1060 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1061 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1062 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1063 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1064 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1065 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1066 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1067 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1068 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1071 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1072 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1073 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1074 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1075 VR128:$src2)), addr:$src1),
1076 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1078 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1079 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1080 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1081 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1082 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1084 // Shuffle with VMOVLPS
1085 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1086 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1087 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1088 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1089 def : Pat<(X86Movlps VR128:$src1,
1090 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1091 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1093 // Shuffle with VMOVLPD
1094 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1095 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1096 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1097 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1098 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1099 (scalar_to_vector (loadf64 addr:$src2)))),
1100 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1103 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1105 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1106 def : Pat<(store (v4i32 (X86Movlps
1107 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1108 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1109 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1111 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1112 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1114 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1117 let Predicates = [HasSSE1] in {
1118 let AddedComplexity = 20 in {
1119 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1120 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1121 (MOVLPSrm VR128:$src1, addr:$src2)>;
1122 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1123 (MOVLPSrm VR128:$src1, addr:$src2)>;
1126 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1127 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1128 (iPTR 0))), addr:$src1),
1129 (MOVLPSmr addr:$src1, VR128:$src2)>;
1130 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1131 (MOVLPSmr addr:$src1, VR128:$src2)>;
1132 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1133 VR128:$src2)), addr:$src1),
1134 (MOVLPSmr addr:$src1, VR128:$src2)>;
1136 // Shuffle with MOVLPS
1137 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1138 (MOVLPSrm VR128:$src1, addr:$src2)>;
1139 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1140 (MOVLPSrm VR128:$src1, addr:$src2)>;
1141 def : Pat<(X86Movlps VR128:$src1,
1142 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1143 (MOVLPSrm VR128:$src1, addr:$src2)>;
1144 def : Pat<(X86Movlps VR128:$src1,
1145 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1146 (MOVLPSrm VR128:$src1, addr:$src2)>;
1149 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1151 (MOVLPSmr addr:$src1, VR128:$src2)>;
1152 def : Pat<(store (v4i32 (X86Movlps
1153 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1155 (MOVLPSmr addr:$src1, VR128:$src2)>;
1158 let Predicates = [HasSSE2] in {
1159 let AddedComplexity = 20 in {
1160 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1161 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1162 (MOVLPDrm VR128:$src1, addr:$src2)>;
1163 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1164 (MOVLPDrm VR128:$src1, addr:$src2)>;
1167 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1168 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1169 (MOVLPDmr addr:$src1, VR128:$src2)>;
1170 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1171 (MOVLPDmr addr:$src1, VR128:$src2)>;
1173 // Shuffle with MOVLPD
1174 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1175 (MOVLPDrm VR128:$src1, addr:$src2)>;
1176 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1177 (MOVLPDrm VR128:$src1, addr:$src2)>;
1178 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1179 (scalar_to_vector (loadf64 addr:$src2)))),
1180 (MOVLPDrm VR128:$src1, addr:$src2)>;
1183 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1185 (MOVLPDmr addr:$src1, VR128:$src2)>;
1186 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1188 (MOVLPDmr addr:$src1, VR128:$src2)>;
1191 //===----------------------------------------------------------------------===//
1192 // SSE 1 & 2 - Move Hi packed FP Instructions
1193 //===----------------------------------------------------------------------===//
1195 let AddedComplexity = 20 in {
1196 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1197 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1199 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1200 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1201 "\t{$src2, $dst|$dst, $src2}">;
1204 // v2f64 extract element 1 is always custom lowered to unpack high to low
1205 // and extract element 0 so the non-store version isn't too horrible.
1206 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$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:$dst)]>,
1212 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1213 "movhpd\t{$src, $dst|$dst, $src}",
1214 [(store (f64 (vector_extract
1215 (v2f64 (unpckh VR128:$src, (undef))),
1216 (iPTR 0))), addr:$dst)]>,
1218 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1219 "movhps\t{$src, $dst|$dst, $src}",
1220 [(store (f64 (vector_extract
1221 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1222 (undef)), (iPTR 0))), addr:$dst)]>;
1223 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1224 "movhpd\t{$src, $dst|$dst, $src}",
1225 [(store (f64 (vector_extract
1226 (v2f64 (unpckh VR128:$src, (undef))),
1227 (iPTR 0))), addr:$dst)]>;
1229 let Predicates = [HasAVX] in {
1231 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1232 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1233 def : Pat<(X86Movlhps VR128:$src1,
1234 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1235 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1236 def : Pat<(X86Movlhps VR128:$src1,
1237 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1238 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1239 def : Pat<(X86Movlhps VR128:$src1,
1240 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1241 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1243 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1244 // is during lowering, where it's not possible to recognize the load fold
1245 // cause it has two uses through a bitcast. One use disappears at isel time
1246 // and the fold opportunity reappears.
1247 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1248 (scalar_to_vector (loadf64 addr:$src2)))),
1249 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1251 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1252 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1253 (scalar_to_vector (loadf64 addr:$src2)))),
1254 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1257 def : Pat<(store (f64 (vector_extract
1258 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1259 (bc_v2f64 (v4f32 VR128:$src))), (iPTR 0))), addr:$dst),
1260 (VMOVHPSmr addr:$dst, VR128:$src)>;
1261 def : Pat<(store (f64 (vector_extract
1262 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))), addr:$dst),
1263 (VMOVHPDmr addr:$dst, VR128:$src)>;
1266 let Predicates = [HasSSE1] in {
1268 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1269 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1270 def : Pat<(X86Movlhps VR128:$src1,
1271 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1272 (MOVHPSrm VR128:$src1, addr:$src2)>;
1273 def : Pat<(X86Movlhps VR128:$src1,
1274 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1275 (MOVHPSrm VR128:$src1, addr:$src2)>;
1276 def : Pat<(X86Movlhps VR128:$src1,
1277 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1278 (MOVHPSrm VR128:$src1, addr:$src2)>;
1281 def : Pat<(store (f64 (vector_extract
1282 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1283 (bc_v2f64 (v4f32 VR128:$src))), (iPTR 0))), addr:$dst),
1284 (MOVHPSmr addr:$dst, VR128:$src)>;
1287 let Predicates = [HasSSE2] in {
1288 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1289 // is during lowering, where it's not possible to recognize the load fold
1290 // cause it has two uses through a bitcast. One use disappears at isel time
1291 // and the fold opportunity reappears.
1292 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1293 (scalar_to_vector (loadf64 addr:$src2)))),
1294 (MOVHPDrm VR128:$src1, addr:$src2)>;
1296 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1297 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1298 (scalar_to_vector (loadf64 addr:$src2)))),
1299 (MOVHPDrm VR128:$src1, addr:$src2)>;
1302 def : Pat<(store (f64 (vector_extract
1303 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))),addr:$dst),
1304 (MOVHPDmr addr:$dst, VR128:$src)>;
1307 //===----------------------------------------------------------------------===//
1308 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1309 //===----------------------------------------------------------------------===//
1311 let AddedComplexity = 20 in {
1312 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1313 (ins VR128:$src1, VR128:$src2),
1314 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1316 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1318 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1319 (ins VR128:$src1, VR128:$src2),
1320 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1322 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1325 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1326 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1327 (ins VR128:$src1, VR128:$src2),
1328 "movlhps\t{$src2, $dst|$dst, $src2}",
1330 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1331 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1332 (ins VR128:$src1, VR128:$src2),
1333 "movhlps\t{$src2, $dst|$dst, $src2}",
1335 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1338 let Predicates = [HasAVX] in {
1340 let AddedComplexity = 20 in {
1341 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1342 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1343 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1344 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1346 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1347 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1348 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1350 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1351 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1352 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1353 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1354 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1355 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1358 let AddedComplexity = 20 in {
1359 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1360 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1361 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1363 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1364 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1365 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1366 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1367 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1370 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1371 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1372 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1373 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1376 let Predicates = [HasSSE1] in {
1378 let AddedComplexity = 20 in {
1379 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1380 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1381 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1382 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1384 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1385 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1386 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1388 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1389 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1390 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1391 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1392 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1393 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1396 let AddedComplexity = 20 in {
1397 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1398 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1399 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1401 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1402 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1403 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1404 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1405 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1408 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1409 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1410 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1411 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1414 //===----------------------------------------------------------------------===//
1415 // SSE 1 & 2 - Conversion Instructions
1416 //===----------------------------------------------------------------------===//
1418 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1419 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1421 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1422 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1423 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1424 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1427 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1428 X86MemOperand x86memop, string asm> {
1429 let neverHasSideEffects = 1 in {
1430 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, []>;
1432 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, []>;
1433 } // neverHasSideEffects = 1
1436 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1437 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1438 string asm, Domain d> {
1439 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1440 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1441 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1442 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1445 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1446 X86MemOperand x86memop, string asm> {
1447 let neverHasSideEffects = 1 in {
1448 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1449 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1451 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1452 (ins DstRC:$src1, x86memop:$src),
1453 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1454 } // neverHasSideEffects = 1
1457 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1458 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1460 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1461 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1463 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1464 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1466 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1467 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1468 VEX, VEX_W, VEX_LIG;
1470 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1471 // register, but the same isn't true when only using memory operands,
1472 // provide other assembly "l" and "q" forms to address this explicitly
1473 // where appropriate to do so.
1474 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1476 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1477 VEX_4V, VEX_W, VEX_LIG;
1478 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1480 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1482 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1483 VEX_4V, VEX_W, VEX_LIG;
1485 let Predicates = [HasAVX], AddedComplexity = 1 in {
1486 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1487 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1488 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1489 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1490 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1491 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1492 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1493 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1495 def : Pat<(f32 (sint_to_fp GR32:$src)),
1496 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1497 def : Pat<(f32 (sint_to_fp GR64:$src)),
1498 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1499 def : Pat<(f64 (sint_to_fp GR32:$src)),
1500 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1501 def : Pat<(f64 (sint_to_fp GR64:$src)),
1502 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1505 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1506 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1507 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1508 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1509 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1510 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1511 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1512 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1513 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1514 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1515 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1516 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1517 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1518 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1519 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1520 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1522 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1523 // and/or XMM operand(s).
1525 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1526 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1528 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1529 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1530 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1531 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1532 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1533 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1536 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1537 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1538 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1539 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1541 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1542 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1543 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1544 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1545 (ins DstRC:$src1, x86memop:$src2),
1547 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1548 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1549 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1552 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1553 f128mem, load, "cvtsd2si">, XD, VEX;
1554 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1555 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1558 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1559 // Get rid of this hack or rename the intrinsics, there are several
1560 // intructions that only match with the intrinsic form, why create duplicates
1561 // to let them be recognized by the assembler?
1562 defm VCVTSD2SI : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
1563 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_LIG;
1564 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
1565 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W,
1568 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1569 f128mem, load, "cvtsd2si{l}">, XD;
1570 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1571 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1574 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1575 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1576 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1577 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1579 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1580 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1581 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1582 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1585 let Constraints = "$src1 = $dst" in {
1586 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1587 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1589 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1590 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1591 "cvtsi2ss{q}">, XS, REX_W;
1592 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1593 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1595 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1596 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1597 "cvtsi2sd">, XD, REX_W;
1602 // Aliases for intrinsics
1603 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1604 f32mem, load, "cvttss2si">, XS, VEX;
1605 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1606 int_x86_sse_cvttss2si64, f32mem, load,
1607 "cvttss2si">, XS, VEX, VEX_W;
1608 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1609 f128mem, load, "cvttsd2si">, XD, VEX;
1610 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1611 int_x86_sse2_cvttsd2si64, f128mem, load,
1612 "cvttsd2si">, XD, VEX, VEX_W;
1613 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1614 f32mem, load, "cvttss2si">, XS;
1615 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1616 int_x86_sse_cvttss2si64, f32mem, load,
1617 "cvttss2si{q}">, XS, REX_W;
1618 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1619 f128mem, load, "cvttsd2si">, XD;
1620 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1621 int_x86_sse2_cvttsd2si64, f128mem, load,
1622 "cvttsd2si{q}">, XD, REX_W;
1624 let Pattern = []<dag> in {
1625 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1626 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1628 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1629 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1631 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1632 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1633 SSEPackedSingle>, TB, VEX;
1634 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1635 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1636 SSEPackedSingle>, TB, VEX;
1639 let Pattern = []<dag> in {
1640 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1641 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1642 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1643 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1644 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1645 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1646 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1649 let Predicates = [HasAVX] in {
1650 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1651 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1652 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1653 (VCVTSS2SIrm addr:$src)>;
1654 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1655 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1656 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1657 (VCVTSS2SI64rm addr:$src)>;
1660 let Predicates = [HasSSE1] in {
1661 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1662 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1663 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1664 (CVTSS2SIrm addr:$src)>;
1665 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1666 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1667 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1668 (CVTSS2SI64rm addr:$src)>;
1673 // Convert scalar double to scalar single
1674 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1675 (ins FR64:$src1, FR64:$src2),
1676 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1679 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1680 (ins FR64:$src1, f64mem:$src2),
1681 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1682 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1684 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1687 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1688 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1689 [(set FR32:$dst, (fround FR64:$src))]>;
1690 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1691 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1692 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1693 Requires<[HasSSE2, OptForSize]>;
1695 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1696 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1698 let Constraints = "$src1 = $dst" in
1699 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1700 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1702 // Convert scalar single to scalar double
1703 // SSE2 instructions with XS prefix
1704 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1705 (ins FR32:$src1, FR32:$src2),
1706 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1707 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1709 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1710 (ins FR32:$src1, f32mem:$src2),
1711 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1712 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1714 let Predicates = [HasAVX] in {
1715 def : Pat<(f64 (fextend FR32:$src)),
1716 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1717 def : Pat<(fextend (loadf32 addr:$src)),
1718 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1719 def : Pat<(extloadf32 addr:$src),
1720 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1723 def : Pat<(extloadf32 addr:$src),
1724 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1725 Requires<[HasAVX, OptForSpeed]>;
1727 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1728 "cvtss2sd\t{$src, $dst|$dst, $src}",
1729 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1730 Requires<[HasSSE2]>;
1731 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1732 "cvtss2sd\t{$src, $dst|$dst, $src}",
1733 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1734 Requires<[HasSSE2, OptForSize]>;
1736 // extload f32 -> f64. This matches load+fextend because we have a hack in
1737 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1739 // Since these loads aren't folded into the fextend, we have to match it
1741 def : Pat<(fextend (loadf32 addr:$src)),
1742 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1743 def : Pat<(extloadf32 addr:$src),
1744 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1746 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1747 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1748 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1749 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1750 VR128:$src2))]>, XS, VEX_4V,
1752 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1753 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1754 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1755 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1756 (load addr:$src2)))]>, XS, VEX_4V,
1758 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1759 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1760 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1761 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1762 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1763 VR128:$src2))]>, XS,
1764 Requires<[HasSSE2]>;
1765 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1766 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1767 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1768 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1769 (load addr:$src2)))]>, XS,
1770 Requires<[HasSSE2]>;
1773 // Convert doubleword to packed single/double fp
1774 // SSE2 instructions without OpSize prefix
1775 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1776 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1777 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1778 TB, VEX, Requires<[HasAVX]>;
1779 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1780 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1781 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1782 (bitconvert (memopv2i64 addr:$src))))]>,
1783 TB, VEX, Requires<[HasAVX]>;
1784 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1785 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1786 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1787 TB, Requires<[HasSSE2]>;
1788 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1789 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1790 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1791 (bitconvert (memopv2i64 addr:$src))))]>,
1792 TB, Requires<[HasSSE2]>;
1794 // FIXME: why the non-intrinsic version is described as SSE3?
1795 // SSE2 instructions with XS prefix
1796 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1797 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1798 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1799 XS, VEX, Requires<[HasAVX]>;
1800 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1801 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1802 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1803 (bitconvert (memopv2i64 addr:$src))))]>,
1804 XS, VEX, Requires<[HasAVX]>;
1805 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1806 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1807 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1808 XS, Requires<[HasSSE2]>;
1809 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1810 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1811 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1812 (bitconvert (memopv2i64 addr:$src))))]>,
1813 XS, Requires<[HasSSE2]>;
1816 // Convert packed single/double fp to doubleword
1817 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1818 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1819 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1820 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1821 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1822 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1823 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1824 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1825 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1826 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1827 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1828 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1830 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1831 "cvtps2dq\t{$src, $dst|$dst, $src}",
1832 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1834 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1836 "cvtps2dq\t{$src, $dst|$dst, $src}",
1837 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1838 (memop addr:$src)))]>, VEX;
1839 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1840 "cvtps2dq\t{$src, $dst|$dst, $src}",
1841 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1842 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1843 "cvtps2dq\t{$src, $dst|$dst, $src}",
1844 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1845 (memop addr:$src)))]>;
1847 // SSE2 packed instructions with XD prefix
1848 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1849 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1850 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1851 XD, VEX, Requires<[HasAVX]>;
1852 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1853 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1854 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1855 (memop addr:$src)))]>,
1856 XD, VEX, Requires<[HasAVX]>;
1857 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1858 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1859 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1860 XD, Requires<[HasSSE2]>;
1861 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1862 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1863 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1864 (memop addr:$src)))]>,
1865 XD, Requires<[HasSSE2]>;
1868 // Convert with truncation packed single/double fp to doubleword
1869 // SSE2 packed instructions with XS prefix
1870 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1871 "cvttps2dq\t{$src, $dst|$dst, $src}",
1873 (int_x86_sse2_cvttps2dq VR128:$src))]>, VEX;
1874 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1875 "cvttps2dq\t{$src, $dst|$dst, $src}",
1876 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1877 (memop addr:$src)))]>, VEX;
1878 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1879 "cvttps2dq\t{$src, $dst|$dst, $src}",
1881 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))]>, VEX;
1882 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1883 "cvttps2dq\t{$src, $dst|$dst, $src}",
1884 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1885 (memopv8f32 addr:$src)))]>, VEX;
1887 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1888 "cvttps2dq\t{$src, $dst|$dst, $src}",
1890 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1891 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1892 "cvttps2dq\t{$src, $dst|$dst, $src}",
1894 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1896 let Predicates = [HasAVX] in {
1897 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1898 (Int_VCVTDQ2PSrr VR128:$src)>;
1899 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1900 (Int_VCVTDQ2PSrm addr:$src)>;
1902 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1903 (VCVTTPS2DQrr VR128:$src)>;
1904 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1905 (VCVTTPS2DQrm addr:$src)>;
1907 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1908 (VCVTDQ2PSYrr VR256:$src)>;
1909 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1910 (VCVTDQ2PSYrm addr:$src)>;
1912 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1913 (VCVTTPS2DQYrr VR256:$src)>;
1914 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1915 (VCVTTPS2DQYrm addr:$src)>;
1918 let Predicates = [HasSSE2] in {
1919 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1920 (Int_CVTDQ2PSrr VR128:$src)>;
1921 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1922 (Int_CVTDQ2PSrm addr:$src)>;
1924 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1925 (CVTTPS2DQrr VR128:$src)>;
1926 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1927 (CVTTPS2DQrm addr:$src)>;
1930 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1931 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1933 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1934 let isCodeGenOnly = 1 in
1935 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1936 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1937 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1938 (memop addr:$src)))]>, VEX;
1939 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1940 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1941 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1942 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1943 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1944 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1945 (memop addr:$src)))]>;
1947 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1948 // register, but the same isn't true when using memory operands instead.
1949 // Provide other assembly rr and rm forms to address this explicitly.
1950 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1951 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1954 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1955 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1956 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1957 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1960 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1961 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1962 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1963 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1965 // Convert packed single to packed double
1966 let Predicates = [HasAVX] in {
1967 // SSE2 instructions without OpSize prefix
1968 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1969 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1970 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1971 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1972 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1973 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1974 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1975 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1977 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1978 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1979 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1980 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1982 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1983 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1984 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1985 TB, VEX, Requires<[HasAVX]>;
1986 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1987 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1988 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1989 (load addr:$src)))]>,
1990 TB, VEX, Requires<[HasAVX]>;
1991 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1992 "cvtps2pd\t{$src, $dst|$dst, $src}",
1993 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1994 TB, Requires<[HasSSE2]>;
1995 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1996 "cvtps2pd\t{$src, $dst|$dst, $src}",
1997 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1998 (load addr:$src)))]>,
1999 TB, Requires<[HasSSE2]>;
2001 // Convert packed double to packed single
2002 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2003 // register, but the same isn't true when using memory operands instead.
2004 // Provide other assembly rr and rm forms to address this explicitly.
2005 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2006 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
2007 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2008 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
2011 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2012 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
2013 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2014 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
2017 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2018 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
2019 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2020 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
2021 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2022 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
2023 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2024 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
2027 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2028 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2029 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2030 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
2032 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2033 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2034 (memop addr:$src)))]>;
2035 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2036 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2037 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2038 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2039 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2040 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2041 (memop addr:$src)))]>;
2043 // AVX 256-bit register conversion intrinsics
2044 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2045 // whenever possible to avoid declaring two versions of each one.
2046 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2047 (VCVTDQ2PSYrr VR256:$src)>;
2048 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2049 (VCVTDQ2PSYrm addr:$src)>;
2051 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2052 (VCVTPD2PSYrr VR256:$src)>;
2053 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2054 (VCVTPD2PSYrm addr:$src)>;
2056 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2057 (VCVTPS2DQYrr VR256:$src)>;
2058 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2059 (VCVTPS2DQYrm addr:$src)>;
2061 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2062 (VCVTPS2PDYrr VR128:$src)>;
2063 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2064 (VCVTPS2PDYrm addr:$src)>;
2066 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2067 (VCVTTPD2DQYrr VR256:$src)>;
2068 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2069 (VCVTTPD2DQYrm addr:$src)>;
2071 // Match fround and fextend for 128/256-bit conversions
2072 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2073 (VCVTPD2PSYrr VR256:$src)>;
2074 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2075 (VCVTPD2PSYrm addr:$src)>;
2077 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2078 (VCVTPS2PDYrr VR128:$src)>;
2079 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2080 (VCVTPS2PDYrm addr:$src)>;
2082 //===----------------------------------------------------------------------===//
2083 // SSE 1 & 2 - Compare Instructions
2084 //===----------------------------------------------------------------------===//
2086 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2087 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2088 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2089 string asm, string asm_alt> {
2090 def rr : SIi8<0xC2, MRMSrcReg,
2091 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2092 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2093 def rm : SIi8<0xC2, MRMSrcMem,
2094 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2095 [(set RC:$dst, (OpNode (VT RC:$src1),
2096 (ld_frag addr:$src2), imm:$cc))]>;
2098 // Accept explicit immediate argument form instead of comparison code.
2099 let neverHasSideEffects = 1 in {
2100 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2101 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2103 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2104 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2108 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2109 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2110 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2111 XS, VEX_4V, VEX_LIG;
2112 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2113 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2114 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2115 XD, VEX_4V, VEX_LIG;
2117 let Constraints = "$src1 = $dst" in {
2118 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2119 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2120 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2122 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2123 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2124 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2128 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2129 Intrinsic Int, string asm> {
2130 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2131 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2132 [(set VR128:$dst, (Int VR128:$src1,
2133 VR128:$src, imm:$cc))]>;
2134 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2135 (ins VR128:$src1, x86memop:$src, SSECC:$cc), asm,
2136 [(set VR128:$dst, (Int VR128:$src1,
2137 (load addr:$src), imm:$cc))]>;
2140 // Aliases to match intrinsics which expect XMM operand(s).
2141 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2142 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2144 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2145 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2147 let Constraints = "$src1 = $dst" in {
2148 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2149 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2150 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2151 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2155 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2156 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2157 ValueType vt, X86MemOperand x86memop,
2158 PatFrag ld_frag, string OpcodeStr, Domain d> {
2159 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2160 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2161 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2162 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2163 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2164 [(set EFLAGS, (OpNode (vt RC:$src1),
2165 (ld_frag addr:$src2)))], d>;
2168 let Defs = [EFLAGS] in {
2169 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2170 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2171 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2172 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2174 let Pattern = []<dag> in {
2175 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2176 "comiss", SSEPackedSingle>, TB, VEX,
2178 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2179 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2183 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2184 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2185 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2186 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2188 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2189 load, "comiss", SSEPackedSingle>, TB, VEX;
2190 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2191 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2192 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2193 "ucomiss", SSEPackedSingle>, TB;
2194 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2195 "ucomisd", SSEPackedDouble>, TB, OpSize;
2197 let Pattern = []<dag> in {
2198 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2199 "comiss", SSEPackedSingle>, TB;
2200 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2201 "comisd", SSEPackedDouble>, TB, OpSize;
2204 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2205 load, "ucomiss", SSEPackedSingle>, TB;
2206 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2207 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2209 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2210 "comiss", SSEPackedSingle>, TB;
2211 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2212 "comisd", SSEPackedDouble>, TB, OpSize;
2213 } // Defs = [EFLAGS]
2215 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2216 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2217 Intrinsic Int, string asm, string asm_alt,
2219 let isAsmParserOnly = 1 in {
2220 def rri : PIi8<0xC2, MRMSrcReg,
2221 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2222 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2223 def rmi : PIi8<0xC2, MRMSrcMem,
2224 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2225 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2228 // Accept explicit immediate argument form instead of comparison code.
2229 def rri_alt : PIi8<0xC2, MRMSrcReg,
2230 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2232 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2233 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2237 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2238 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2239 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2240 SSEPackedSingle>, TB, VEX_4V;
2241 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2242 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2243 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2244 SSEPackedDouble>, TB, OpSize, VEX_4V;
2245 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2246 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2247 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2248 SSEPackedSingle>, TB, VEX_4V;
2249 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2250 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2251 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2252 SSEPackedDouble>, TB, OpSize, VEX_4V;
2253 let Constraints = "$src1 = $dst" in {
2254 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2255 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2256 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2257 SSEPackedSingle>, TB;
2258 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2259 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2260 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2261 SSEPackedDouble>, TB, OpSize;
2264 let Predicates = [HasAVX] in {
2265 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2266 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2267 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2268 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2269 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2270 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2271 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2272 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2274 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2275 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2276 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2277 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2278 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2279 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2280 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2281 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2284 let Predicates = [HasSSE1] in {
2285 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2286 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2287 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2288 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2291 let Predicates = [HasSSE2] in {
2292 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2293 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2294 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2295 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2298 //===----------------------------------------------------------------------===//
2299 // SSE 1 & 2 - Shuffle Instructions
2300 //===----------------------------------------------------------------------===//
2302 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2303 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2304 ValueType vt, string asm, PatFrag mem_frag,
2305 Domain d, bit IsConvertibleToThreeAddress = 0> {
2306 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2307 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2308 [(set RC:$dst, (vt (shufp:$src3
2309 RC:$src1, (mem_frag addr:$src2))))], d>;
2310 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2311 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2312 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2314 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2317 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2318 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2319 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2320 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2321 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2322 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2323 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2324 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2325 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2326 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2327 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2328 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2330 let Constraints = "$src1 = $dst" in {
2331 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2332 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2333 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2335 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2336 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2337 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2341 let Predicates = [HasAVX] in {
2342 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2343 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2344 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2345 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2346 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2347 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2348 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2349 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2350 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2351 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2352 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2353 // fall back to this for SSE1)
2354 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2355 (VSHUFPSrri VR128:$src2, VR128:$src1,
2356 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2357 // Special unary SHUFPSrri case.
2358 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2359 (VSHUFPSrri VR128:$src1, VR128:$src1,
2360 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2361 // Special binary v4i32 shuffle cases with SHUFPS.
2362 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2363 (VSHUFPSrri VR128:$src1, VR128:$src2,
2364 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2365 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2366 (bc_v4i32 (memopv2i64 addr:$src2)))),
2367 (VSHUFPSrmi VR128:$src1, addr:$src2,
2368 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2369 // Special unary SHUFPDrri cases.
2370 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2371 (VSHUFPDrri VR128:$src1, VR128:$src1,
2372 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2373 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2374 (VSHUFPDrri VR128:$src1, VR128:$src1,
2375 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2376 // Special binary v2i64 shuffle cases using SHUFPDrri.
2377 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2378 (VSHUFPDrri VR128:$src1, VR128:$src2,
2379 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2381 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2382 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2383 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2384 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2385 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2386 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2387 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2388 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2389 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2390 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2393 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2394 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2395 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2396 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2397 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2399 def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2400 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2401 def : Pat<(v8f32 (X86Shufp VR256:$src1,
2402 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2403 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2405 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2406 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2407 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2408 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2409 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2411 def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2412 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2413 def : Pat<(v4f64 (X86Shufp VR256:$src1,
2414 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2415 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2418 let Predicates = [HasSSE1] in {
2419 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2420 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2421 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2422 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2423 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2424 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2425 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2426 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2427 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2428 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2429 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2430 // fall back to this for SSE1)
2431 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2432 (SHUFPSrri VR128:$src2, VR128:$src1,
2433 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2434 // Special unary SHUFPSrri case.
2435 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2436 (SHUFPSrri VR128:$src1, VR128:$src1,
2437 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2440 let Predicates = [HasSSE2] in {
2441 // Special binary v4i32 shuffle cases with SHUFPS.
2442 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2443 (SHUFPSrri VR128:$src1, VR128:$src2,
2444 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2445 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2446 (bc_v4i32 (memopv2i64 addr:$src2)))),
2447 (SHUFPSrmi VR128:$src1, addr:$src2,
2448 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2449 // Special unary SHUFPDrri cases.
2450 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2451 (SHUFPDrri VR128:$src1, VR128:$src1,
2452 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2453 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2454 (SHUFPDrri VR128:$src1, VR128:$src1,
2455 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2456 // Special binary v2i64 shuffle cases using SHUFPDrri.
2457 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2458 (SHUFPDrri VR128:$src1, VR128:$src2,
2459 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2460 // Generic SHUFPD patterns
2461 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2462 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2463 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2464 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2465 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2466 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2467 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2468 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2469 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2470 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2473 //===----------------------------------------------------------------------===//
2474 // SSE 1 & 2 - Unpack Instructions
2475 //===----------------------------------------------------------------------===//
2477 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2478 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2479 PatFrag mem_frag, RegisterClass RC,
2480 X86MemOperand x86memop, string asm,
2482 def rr : PI<opc, MRMSrcReg,
2483 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2485 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2486 def rm : PI<opc, MRMSrcMem,
2487 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2489 (vt (OpNode RC:$src1,
2490 (mem_frag addr:$src2))))], d>;
2493 let AddedComplexity = 10 in {
2494 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2495 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2496 SSEPackedSingle>, TB, VEX_4V;
2497 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2498 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2499 SSEPackedDouble>, TB, OpSize, VEX_4V;
2500 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2501 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2502 SSEPackedSingle>, TB, VEX_4V;
2503 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2504 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2505 SSEPackedDouble>, TB, OpSize, VEX_4V;
2507 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2508 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2509 SSEPackedSingle>, TB, VEX_4V;
2510 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2511 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2512 SSEPackedDouble>, TB, OpSize, VEX_4V;
2513 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2514 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2515 SSEPackedSingle>, TB, VEX_4V;
2516 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2517 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2518 SSEPackedDouble>, TB, OpSize, VEX_4V;
2520 let Constraints = "$src1 = $dst" in {
2521 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2522 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2523 SSEPackedSingle>, TB;
2524 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2525 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2526 SSEPackedDouble>, TB, OpSize;
2527 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2528 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2529 SSEPackedSingle>, TB;
2530 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2531 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2532 SSEPackedDouble>, TB, OpSize;
2533 } // Constraints = "$src1 = $dst"
2534 } // AddedComplexity
2536 let Predicates = [HasAVX], AddedComplexity = 1 in {
2537 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2538 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2539 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2540 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2541 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2542 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2543 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2544 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2546 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2547 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2548 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2549 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2550 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2551 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2552 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2553 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2555 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2556 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2557 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2558 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2559 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2560 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2561 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2562 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2564 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2565 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2566 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2567 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2568 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2569 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2570 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2571 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2573 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2574 // problem is during lowering, where it's not possible to recognize the load
2575 // fold cause it has two uses through a bitcast. One use disappears at isel
2576 // time and the fold opportunity reappears.
2577 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2578 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2579 let AddedComplexity = 10 in
2580 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2581 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2584 let Predicates = [HasSSE1] in {
2585 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2586 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2587 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2588 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2589 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2590 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2591 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2592 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2595 let Predicates = [HasSSE2] in {
2596 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2597 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2598 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2599 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2600 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2601 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2602 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2603 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2605 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2606 // problem is during lowering, where it's not possible to recognize the load
2607 // fold cause it has two uses through a bitcast. One use disappears at isel
2608 // time and the fold opportunity reappears.
2609 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2610 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2612 let AddedComplexity = 10 in
2613 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2614 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2617 //===----------------------------------------------------------------------===//
2618 // SSE 1 & 2 - Extract Floating-Point Sign mask
2619 //===----------------------------------------------------------------------===//
2621 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2622 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2624 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2625 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2626 [(set GR32:$dst, (Int RC:$src))], d>;
2627 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2628 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2631 let Predicates = [HasAVX] in {
2632 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2633 "movmskps", SSEPackedSingle>, TB, VEX;
2634 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2635 "movmskpd", SSEPackedDouble>, TB,
2637 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2638 "movmskps", SSEPackedSingle>, TB, VEX;
2639 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2640 "movmskpd", SSEPackedDouble>, TB,
2643 def : Pat<(i32 (X86fgetsign FR32:$src)),
2644 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2646 def : Pat<(i64 (X86fgetsign FR32:$src)),
2647 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2649 def : Pat<(i32 (X86fgetsign FR64:$src)),
2650 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2652 def : Pat<(i64 (X86fgetsign FR64:$src)),
2653 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2657 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2658 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2659 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2660 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2662 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2663 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2664 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2665 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2669 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2670 SSEPackedSingle>, TB;
2671 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2672 SSEPackedDouble>, TB, OpSize;
2674 def : Pat<(i32 (X86fgetsign FR32:$src)),
2675 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2676 sub_ss))>, Requires<[HasSSE1]>;
2677 def : Pat<(i64 (X86fgetsign FR32:$src)),
2678 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2679 sub_ss))>, Requires<[HasSSE1]>;
2680 def : Pat<(i32 (X86fgetsign FR64:$src)),
2681 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2682 sub_sd))>, Requires<[HasSSE2]>;
2683 def : Pat<(i64 (X86fgetsign FR64:$src)),
2684 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2685 sub_sd))>, Requires<[HasSSE2]>;
2687 //===---------------------------------------------------------------------===//
2688 // SSE2 - Packed Integer Logical Instructions
2689 //===---------------------------------------------------------------------===//
2691 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2693 /// PDI_binop_rm - Simple SSE2 binary operator.
2694 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2695 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2696 X86MemOperand x86memop, bit IsCommutable = 0,
2698 let isCommutable = IsCommutable in
2699 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2700 (ins RC:$src1, RC:$src2),
2702 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2703 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2704 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2705 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2706 (ins RC:$src1, x86memop:$src2),
2708 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2709 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2710 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2711 (bitconvert (memop_frag addr:$src2)))))]>;
2713 } // ExeDomain = SSEPackedInt
2715 // These are ordered here for pattern ordering requirements with the fp versions
2717 let Predicates = [HasAVX] in {
2718 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2719 i128mem, 1, 0>, VEX_4V;
2720 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2721 i128mem, 1, 0>, VEX_4V;
2722 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2723 i128mem, 1, 0>, VEX_4V;
2724 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2725 i128mem, 0, 0>, VEX_4V;
2728 let Constraints = "$src1 = $dst" in {
2729 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2731 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2733 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2735 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2737 } // Constraints = "$src1 = $dst"
2739 let Predicates = [HasAVX2] in {
2740 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2741 i256mem, 1, 0>, VEX_4V;
2742 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2743 i256mem, 1, 0>, VEX_4V;
2744 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2745 i256mem, 1, 0>, VEX_4V;
2746 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2747 i256mem, 0, 0>, VEX_4V;
2750 //===----------------------------------------------------------------------===//
2751 // SSE 1 & 2 - Logical Instructions
2752 //===----------------------------------------------------------------------===//
2754 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2756 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2758 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2759 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2761 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2762 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2764 let Constraints = "$src1 = $dst" in {
2765 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2766 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2768 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2769 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2773 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2774 let mayLoad = 0 in {
2775 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2776 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2777 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2780 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2781 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2783 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2785 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2787 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2788 // are all promoted to v2i64, and the patterns are covered by the int
2789 // version. This is needed in SSE only, because v2i64 isn't supported on
2790 // SSE1, but only on SSE2.
2791 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2792 !strconcat(OpcodeStr, "ps"), f128mem, [],
2793 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2794 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2796 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2797 !strconcat(OpcodeStr, "pd"), f128mem,
2798 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2799 (bc_v2i64 (v2f64 VR128:$src2))))],
2800 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2801 (memopv2i64 addr:$src2)))], 0>,
2803 let Constraints = "$src1 = $dst" in {
2804 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2805 !strconcat(OpcodeStr, "ps"), f128mem,
2806 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2807 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2808 (memopv2i64 addr:$src2)))]>, TB;
2810 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2811 !strconcat(OpcodeStr, "pd"), f128mem,
2812 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2813 (bc_v2i64 (v2f64 VR128:$src2))))],
2814 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2815 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2819 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2821 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2823 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2824 !strconcat(OpcodeStr, "ps"), f256mem,
2825 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2826 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2827 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2829 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2830 !strconcat(OpcodeStr, "pd"), f256mem,
2831 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2832 (bc_v4i64 (v4f64 VR256:$src2))))],
2833 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2834 (memopv4i64 addr:$src2)))], 0>,
2838 // AVX 256-bit packed logical ops forms
2839 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2840 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2841 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2842 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2844 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2845 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2846 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2847 let isCommutable = 0 in
2848 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2850 //===----------------------------------------------------------------------===//
2851 // SSE 1 & 2 - Arithmetic Instructions
2852 //===----------------------------------------------------------------------===//
2854 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2857 /// In addition, we also have a special variant of the scalar form here to
2858 /// represent the associated intrinsic operation. This form is unlike the
2859 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2860 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2862 /// These three forms can each be reg+reg or reg+mem.
2865 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2867 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2869 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2870 OpNode, FR32, f32mem, Is2Addr>, XS;
2871 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2872 OpNode, FR64, f64mem, Is2Addr>, XD;
2875 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2877 let mayLoad = 0 in {
2878 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2879 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2880 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2881 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2885 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2887 let mayLoad = 0 in {
2888 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2889 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2890 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2891 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2895 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2897 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2898 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2899 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2900 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2903 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2905 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2906 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2907 SSEPackedSingle, Is2Addr>, TB;
2909 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2910 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2911 SSEPackedDouble, Is2Addr>, TB, OpSize;
2914 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2915 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2916 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2917 SSEPackedSingle, 0>, TB;
2919 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2920 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2921 SSEPackedDouble, 0>, TB, OpSize;
2924 // Binary Arithmetic instructions
2925 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2926 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2927 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2928 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2929 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2930 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2931 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2932 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2934 let isCommutable = 0 in {
2935 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2936 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2937 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2938 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2939 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2940 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2941 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2942 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2943 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2944 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2945 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2946 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2947 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2948 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2949 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2950 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2951 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2952 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2953 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2954 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2957 let Constraints = "$src1 = $dst" in {
2958 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2959 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2960 basic_sse12_fp_binop_s_int<0x58, "add">;
2961 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2962 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2963 basic_sse12_fp_binop_s_int<0x59, "mul">;
2965 let isCommutable = 0 in {
2966 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2967 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2968 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2969 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2970 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2971 basic_sse12_fp_binop_s_int<0x5E, "div">;
2972 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2973 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2974 basic_sse12_fp_binop_s_int<0x5F, "max">,
2975 basic_sse12_fp_binop_p_int<0x5F, "max">;
2976 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2977 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2978 basic_sse12_fp_binop_s_int<0x5D, "min">,
2979 basic_sse12_fp_binop_p_int<0x5D, "min">;
2984 /// In addition, we also have a special variant of the scalar form here to
2985 /// represent the associated intrinsic operation. This form is unlike the
2986 /// plain scalar form, in that it takes an entire vector (instead of a
2987 /// scalar) and leaves the top elements undefined.
2989 /// And, we have a special variant form for a full-vector intrinsic form.
2991 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2992 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2993 SDNode OpNode, Intrinsic F32Int> {
2994 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2995 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2996 [(set FR32:$dst, (OpNode FR32:$src))]>;
2997 // For scalar unary operations, fold a load into the operation
2998 // only in OptForSize mode. It eliminates an instruction, but it also
2999 // eliminates a whole-register clobber (the load), so it introduces a
3000 // partial register update condition.
3001 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3002 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3003 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
3004 Requires<[HasSSE1, OptForSize]>;
3005 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3006 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3007 [(set VR128:$dst, (F32Int VR128:$src))]>;
3008 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3009 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3010 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
3013 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
3014 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3015 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
3016 !strconcat(OpcodeStr,
3017 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3019 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
3020 !strconcat(OpcodeStr,
3021 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3022 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3023 (ins VR128:$src1, ssmem:$src2),
3024 !strconcat(OpcodeStr,
3025 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3028 /// sse1_fp_unop_p - SSE1 unops in packed form.
3029 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3030 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3031 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3032 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
3033 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3034 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3035 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
3038 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3039 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3040 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3041 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3042 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
3043 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3044 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3045 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
3048 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3049 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3050 Intrinsic V4F32Int> {
3051 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3052 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3053 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
3054 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3055 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3056 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
3059 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3060 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3061 Intrinsic V4F32Int> {
3062 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3063 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3064 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
3065 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3066 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3067 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
3070 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3071 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3072 SDNode OpNode, Intrinsic F64Int> {
3073 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3074 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3075 [(set FR64:$dst, (OpNode FR64:$src))]>;
3076 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3077 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3078 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3079 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
3080 Requires<[HasSSE2, OptForSize]>;
3081 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3082 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3083 [(set VR128:$dst, (F64Int VR128:$src))]>;
3084 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3085 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3086 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
3089 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3090 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3091 let neverHasSideEffects = 1 in {
3092 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3093 !strconcat(OpcodeStr,
3094 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3096 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3097 !strconcat(OpcodeStr,
3098 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3100 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3101 (ins VR128:$src1, sdmem:$src2),
3102 !strconcat(OpcodeStr,
3103 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3106 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3107 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3109 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3110 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3111 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3112 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3113 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3114 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3117 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3118 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3119 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3120 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3121 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3122 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3123 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3124 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3127 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3128 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3129 Intrinsic V2F64Int> {
3130 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3131 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3132 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3133 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3134 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3135 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3138 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3139 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3140 Intrinsic V2F64Int> {
3141 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3142 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3143 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3144 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3145 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3146 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3149 let Predicates = [HasAVX] in {
3151 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3152 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3154 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3155 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3156 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3157 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3158 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3159 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3160 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3161 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3164 // Reciprocal approximations. Note that these typically require refinement
3165 // in order to obtain suitable precision.
3166 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3167 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3168 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3169 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3170 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3172 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3173 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3174 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3175 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3176 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3179 let AddedComplexity = 1 in {
3180 def : Pat<(f32 (fsqrt FR32:$src)),
3181 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3182 def : Pat<(f32 (fsqrt (load addr:$src))),
3183 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3184 Requires<[HasAVX, OptForSize]>;
3185 def : Pat<(f64 (fsqrt FR64:$src)),
3186 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3187 def : Pat<(f64 (fsqrt (load addr:$src))),
3188 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3189 Requires<[HasAVX, OptForSize]>;
3191 def : Pat<(f32 (X86frsqrt FR32:$src)),
3192 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3193 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3194 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3195 Requires<[HasAVX, OptForSize]>;
3197 def : Pat<(f32 (X86frcp FR32:$src)),
3198 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3199 def : Pat<(f32 (X86frcp (load addr:$src))),
3200 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3201 Requires<[HasAVX, OptForSize]>;
3204 let Predicates = [HasAVX], AddedComplexity = 1 in {
3205 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3206 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3207 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3208 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3210 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3211 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3213 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3214 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3215 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3216 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3218 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3219 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3221 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3222 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3223 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3224 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3226 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3227 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3229 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3230 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3231 (VRCPSSr (f32 (IMPLICIT_DEF)),
3232 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3234 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3235 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3239 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3240 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3241 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3242 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3243 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3244 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3246 // Reciprocal approximations. Note that these typically require refinement
3247 // in order to obtain suitable precision.
3248 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3249 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3250 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3251 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3252 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3253 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3255 // There is no f64 version of the reciprocal approximation instructions.
3257 //===----------------------------------------------------------------------===//
3258 // SSE 1 & 2 - Non-temporal stores
3259 //===----------------------------------------------------------------------===//
3261 let AddedComplexity = 400 in { // Prefer non-temporal versions
3262 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3263 (ins f128mem:$dst, VR128:$src),
3264 "movntps\t{$src, $dst|$dst, $src}",
3265 [(alignednontemporalstore (v4f32 VR128:$src),
3267 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3268 (ins f128mem:$dst, VR128:$src),
3269 "movntpd\t{$src, $dst|$dst, $src}",
3270 [(alignednontemporalstore (v2f64 VR128:$src),
3273 let ExeDomain = SSEPackedInt in
3274 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3275 (ins f128mem:$dst, VR128:$src),
3276 "movntdq\t{$src, $dst|$dst, $src}",
3277 [(alignednontemporalstore (v2i64 VR128:$src),
3280 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3281 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3283 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3284 (ins f256mem:$dst, VR256:$src),
3285 "movntps\t{$src, $dst|$dst, $src}",
3286 [(alignednontemporalstore (v8f32 VR256:$src),
3288 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3289 (ins f256mem:$dst, VR256:$src),
3290 "movntpd\t{$src, $dst|$dst, $src}",
3291 [(alignednontemporalstore (v4f64 VR256:$src),
3293 let ExeDomain = SSEPackedInt in
3294 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3295 (ins f256mem:$dst, VR256:$src),
3296 "movntdq\t{$src, $dst|$dst, $src}",
3297 [(alignednontemporalstore (v4i64 VR256:$src),
3301 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3302 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3303 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3304 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3305 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3306 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3308 let AddedComplexity = 400 in { // Prefer non-temporal versions
3309 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3310 "movntps\t{$src, $dst|$dst, $src}",
3311 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3312 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3313 "movntpd\t{$src, $dst|$dst, $src}",
3314 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3316 let ExeDomain = SSEPackedInt in
3317 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3318 "movntdq\t{$src, $dst|$dst, $src}",
3319 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)]>;
3321 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3322 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3324 // There is no AVX form for instructions below this point
3325 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3326 "movnti{l}\t{$src, $dst|$dst, $src}",
3327 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3328 TB, Requires<[HasSSE2]>;
3329 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3330 "movnti{q}\t{$src, $dst|$dst, $src}",
3331 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3332 TB, Requires<[HasSSE2]>;
3335 //===----------------------------------------------------------------------===//
3336 // SSE 1 & 2 - Prefetch and memory fence
3337 //===----------------------------------------------------------------------===//
3339 // Prefetch intrinsic.
3340 let Predicates = [HasSSE1] in {
3341 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3342 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3343 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3344 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3345 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3346 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3347 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3348 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3352 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3353 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3354 TB, Requires<[HasSSE2]>;
3356 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3357 // was introduced with SSE2, it's backward compatible.
3358 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3360 // Load, store, and memory fence
3361 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3362 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3363 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3364 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3365 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3366 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3368 def : Pat<(X86SFence), (SFENCE)>;
3369 def : Pat<(X86LFence), (LFENCE)>;
3370 def : Pat<(X86MFence), (MFENCE)>;
3372 //===----------------------------------------------------------------------===//
3373 // SSE 1 & 2 - Load/Store XCSR register
3374 //===----------------------------------------------------------------------===//
3376 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3377 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3378 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3379 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3381 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3382 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3383 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3384 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3386 //===---------------------------------------------------------------------===//
3387 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3388 //===---------------------------------------------------------------------===//
3390 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3392 let neverHasSideEffects = 1 in {
3393 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3394 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3395 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3396 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3398 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3399 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3400 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3401 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3404 let isCodeGenOnly = 1 in {
3405 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3406 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3407 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3408 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3409 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3410 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3411 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3412 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3415 let canFoldAsLoad = 1, mayLoad = 1 in {
3416 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3417 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3418 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3419 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3420 let Predicates = [HasAVX] in {
3421 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3422 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3423 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3424 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3428 let mayStore = 1 in {
3429 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3430 (ins i128mem:$dst, VR128:$src),
3431 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3432 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3433 (ins i256mem:$dst, VR256:$src),
3434 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3435 let Predicates = [HasAVX] in {
3436 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3437 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3438 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3439 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3443 let neverHasSideEffects = 1 in
3444 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3445 "movdqa\t{$src, $dst|$dst, $src}", []>;
3447 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3448 "movdqu\t{$src, $dst|$dst, $src}",
3449 []>, XS, Requires<[HasSSE2]>;
3452 let isCodeGenOnly = 1 in {
3453 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3454 "movdqa\t{$src, $dst|$dst, $src}", []>;
3456 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3457 "movdqu\t{$src, $dst|$dst, $src}",
3458 []>, XS, Requires<[HasSSE2]>;
3461 let canFoldAsLoad = 1, mayLoad = 1 in {
3462 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3463 "movdqa\t{$src, $dst|$dst, $src}",
3464 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3465 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3466 "movdqu\t{$src, $dst|$dst, $src}",
3467 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3468 XS, Requires<[HasSSE2]>;
3471 let mayStore = 1 in {
3472 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3473 "movdqa\t{$src, $dst|$dst, $src}",
3474 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3475 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3476 "movdqu\t{$src, $dst|$dst, $src}",
3477 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3478 XS, Requires<[HasSSE2]>;
3481 // Intrinsic forms of MOVDQU load and store
3482 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3483 "vmovdqu\t{$src, $dst|$dst, $src}",
3484 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3485 XS, VEX, Requires<[HasAVX]>;
3487 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3488 "movdqu\t{$src, $dst|$dst, $src}",
3489 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3490 XS, Requires<[HasSSE2]>;
3492 } // ExeDomain = SSEPackedInt
3494 let Predicates = [HasAVX] in {
3495 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3496 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3497 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3500 //===---------------------------------------------------------------------===//
3501 // SSE2 - Packed Integer Arithmetic Instructions
3502 //===---------------------------------------------------------------------===//
3504 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3506 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3507 RegisterClass RC, PatFrag memop_frag,
3508 X86MemOperand x86memop, bit IsCommutable = 0,
3510 let isCommutable = IsCommutable in
3511 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3512 (ins RC:$src1, RC:$src2),
3514 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3515 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3516 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3517 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3518 (ins RC:$src1, x86memop:$src2),
3520 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3521 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3522 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3525 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3526 string OpcodeStr, Intrinsic IntId,
3527 Intrinsic IntId2, RegisterClass RC,
3529 // src2 is always 128-bit
3530 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3531 (ins RC:$src1, VR128:$src2),
3533 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3534 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3535 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3536 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3537 (ins RC:$src1, i128mem:$src2),
3539 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3540 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3541 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3542 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3543 (ins RC:$src1, i32i8imm:$src2),
3545 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3546 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3547 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3550 } // ExeDomain = SSEPackedInt
3552 // 128-bit Integer Arithmetic
3554 let Predicates = [HasAVX] in {
3555 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3556 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3557 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3558 i128mem, 1, 0>, VEX_4V;
3559 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3560 i128mem, 1, 0>, VEX_4V;
3561 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3562 i128mem, 1, 0>, VEX_4V;
3563 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3564 i128mem, 1, 0>, VEX_4V;
3565 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3566 i128mem, 0, 0>, VEX_4V;
3567 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3568 i128mem, 0, 0>, VEX_4V;
3569 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3570 i128mem, 0, 0>, VEX_4V;
3571 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3572 i128mem, 0, 0>, VEX_4V;
3575 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3576 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3577 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3578 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3579 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3580 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3581 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3582 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3583 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3584 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3585 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3586 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3587 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3588 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3589 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3590 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3591 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3592 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3593 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3594 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3595 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3596 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3597 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3598 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3599 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3600 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3601 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3602 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3603 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3604 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3605 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3606 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3607 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3608 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3609 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3610 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3611 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3612 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3615 let Predicates = [HasAVX2] in {
3616 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3617 i256mem, 1, 0>, VEX_4V;
3618 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3619 i256mem, 1, 0>, VEX_4V;
3620 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3621 i256mem, 1, 0>, VEX_4V;
3622 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3623 i256mem, 1, 0>, VEX_4V;
3624 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3625 i256mem, 1, 0>, VEX_4V;
3626 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3627 i256mem, 0, 0>, VEX_4V;
3628 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3629 i256mem, 0, 0>, VEX_4V;
3630 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3631 i256mem, 0, 0>, VEX_4V;
3632 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3633 i256mem, 0, 0>, VEX_4V;
3636 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3637 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3638 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3639 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3640 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3641 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3642 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3643 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3644 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3645 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3646 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3647 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3648 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3649 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3650 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3651 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3652 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3653 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3654 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3655 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3656 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3657 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3658 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3659 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3660 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3661 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3662 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3663 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3664 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3665 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3666 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3667 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3668 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3669 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3670 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3671 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3672 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3673 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3676 let Constraints = "$src1 = $dst" in {
3677 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3679 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3681 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3683 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3685 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3687 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3689 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3691 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3693 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3697 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3698 VR128, memopv2i64, i128mem>;
3699 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3700 VR128, memopv2i64, i128mem>;
3701 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3702 VR128, memopv2i64, i128mem>;
3703 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3704 VR128, memopv2i64, i128mem>;
3705 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3706 VR128, memopv2i64, i128mem, 1>;
3707 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3708 VR128, memopv2i64, i128mem, 1>;
3709 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3710 VR128, memopv2i64, i128mem, 1>;
3711 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3712 VR128, memopv2i64, i128mem, 1>;
3713 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3714 VR128, memopv2i64, i128mem, 1>;
3715 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3716 VR128, memopv2i64, i128mem, 1>;
3717 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3718 VR128, memopv2i64, i128mem, 1>;
3719 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3720 VR128, memopv2i64, i128mem, 1>;
3721 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3722 VR128, memopv2i64, i128mem, 1>;
3723 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3724 VR128, memopv2i64, i128mem, 1>;
3725 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3726 VR128, memopv2i64, i128mem, 1>;
3727 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3728 VR128, memopv2i64, i128mem, 1>;
3729 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3730 VR128, memopv2i64, i128mem, 1>;
3731 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3732 VR128, memopv2i64, i128mem, 1>;
3733 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3734 VR128, memopv2i64, i128mem, 1>;
3736 } // Constraints = "$src1 = $dst"
3738 //===---------------------------------------------------------------------===//
3739 // SSE2 - Packed Integer Logical Instructions
3740 //===---------------------------------------------------------------------===//
3742 let Predicates = [HasAVX] in {
3743 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3744 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3746 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3747 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3749 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3750 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3753 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3754 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3756 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3757 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3759 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3760 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3763 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3764 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3766 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3767 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3770 let ExeDomain = SSEPackedInt in {
3771 // 128-bit logical shifts.
3772 def VPSLLDQri : PDIi8<0x73, MRM7r,
3773 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3774 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3776 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3778 def VPSRLDQri : PDIi8<0x73, MRM3r,
3779 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3780 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3782 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3784 // PSRADQri doesn't exist in SSE[1-3].
3786 } // Predicates = [HasAVX]
3788 let Predicates = [HasAVX2] in {
3789 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3790 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3792 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3793 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3795 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3796 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3799 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3800 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3802 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3803 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3805 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3806 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3809 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3810 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3812 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3813 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3816 let ExeDomain = SSEPackedInt in {
3817 // 256-bit logical shifts.
3818 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3819 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3820 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3822 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3824 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3825 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3826 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3828 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3830 // PSRADQYri doesn't exist in SSE[1-3].
3832 } // Predicates = [HasAVX2]
3834 let Constraints = "$src1 = $dst" in {
3835 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3836 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3838 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3839 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3841 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3842 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3845 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3846 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3848 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3849 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3851 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3852 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3855 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3856 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3858 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3859 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3862 let ExeDomain = SSEPackedInt in {
3863 // 128-bit logical shifts.
3864 def PSLLDQri : PDIi8<0x73, MRM7r,
3865 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3866 "pslldq\t{$src2, $dst|$dst, $src2}",
3868 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3869 def PSRLDQri : PDIi8<0x73, MRM3r,
3870 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3871 "psrldq\t{$src2, $dst|$dst, $src2}",
3873 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3874 // PSRADQri doesn't exist in SSE[1-3].
3876 } // Constraints = "$src1 = $dst"
3878 let Predicates = [HasAVX] in {
3879 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3880 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3881 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3882 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3883 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3884 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3886 // Shift up / down and insert zero's.
3887 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3888 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3889 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3890 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3893 let Predicates = [HasAVX2] in {
3894 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3895 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3896 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3897 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3900 let Predicates = [HasSSE2] in {
3901 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3902 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3903 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3904 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3905 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3906 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3908 // Shift up / down and insert zero's.
3909 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3910 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3911 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3912 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3915 //===---------------------------------------------------------------------===//
3916 // SSE2 - Packed Integer Comparison Instructions
3917 //===---------------------------------------------------------------------===//
3919 let Predicates = [HasAVX] in {
3920 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
3921 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3922 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
3923 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3924 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
3925 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3926 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
3927 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3928 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
3929 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3930 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
3931 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3933 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3934 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3935 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3936 (bc_v16i8 (memopv2i64 addr:$src2)))),
3937 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3938 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3939 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3940 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3941 (bc_v8i16 (memopv2i64 addr:$src2)))),
3942 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3943 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3944 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3945 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3946 (bc_v4i32 (memopv2i64 addr:$src2)))),
3947 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3949 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3950 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3951 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
3952 (bc_v16i8 (memopv2i64 addr:$src2)))),
3953 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3954 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3955 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3956 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
3957 (bc_v8i16 (memopv2i64 addr:$src2)))),
3958 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3959 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3960 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3961 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
3962 (bc_v4i32 (memopv2i64 addr:$src2)))),
3963 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3966 let Predicates = [HasAVX2] in {
3967 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
3968 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3969 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
3970 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3971 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
3972 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3973 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
3974 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3975 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
3976 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3977 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
3978 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3980 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, VR256:$src2)),
3981 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
3982 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1,
3983 (bc_v32i8 (memopv4i64 addr:$src2)))),
3984 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
3985 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, VR256:$src2)),
3986 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
3987 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1,
3988 (bc_v16i16 (memopv4i64 addr:$src2)))),
3989 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
3990 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, VR256:$src2)),
3991 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
3992 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1,
3993 (bc_v8i32 (memopv4i64 addr:$src2)))),
3994 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
3996 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, VR256:$src2)),
3997 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
3998 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1,
3999 (bc_v32i8 (memopv4i64 addr:$src2)))),
4000 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
4001 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, VR256:$src2)),
4002 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
4003 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1,
4004 (bc_v16i16 (memopv4i64 addr:$src2)))),
4005 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
4006 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, VR256:$src2)),
4007 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
4008 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1,
4009 (bc_v8i32 (memopv4i64 addr:$src2)))),
4010 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
4013 let Constraints = "$src1 = $dst" in {
4014 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
4015 VR128, memopv2i64, i128mem, 1>;
4016 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
4017 VR128, memopv2i64, i128mem, 1>;
4018 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
4019 VR128, memopv2i64, i128mem, 1>;
4020 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
4021 VR128, memopv2i64, i128mem>;
4022 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
4023 VR128, memopv2i64, i128mem>;
4024 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
4025 VR128, memopv2i64, i128mem>;
4026 } // Constraints = "$src1 = $dst"
4028 let Predicates = [HasSSE2] in {
4029 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
4030 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
4031 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
4032 (bc_v16i8 (memopv2i64 addr:$src2)))),
4033 (PCMPEQBrm VR128:$src1, addr:$src2)>;
4034 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
4035 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
4036 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
4037 (bc_v8i16 (memopv2i64 addr:$src2)))),
4038 (PCMPEQWrm VR128:$src1, addr:$src2)>;
4039 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
4040 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
4041 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
4042 (bc_v4i32 (memopv2i64 addr:$src2)))),
4043 (PCMPEQDrm VR128:$src1, addr:$src2)>;
4045 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
4046 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
4047 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
4048 (bc_v16i8 (memopv2i64 addr:$src2)))),
4049 (PCMPGTBrm VR128:$src1, addr:$src2)>;
4050 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
4051 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
4052 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
4053 (bc_v8i16 (memopv2i64 addr:$src2)))),
4054 (PCMPGTWrm VR128:$src1, addr:$src2)>;
4055 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
4056 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
4057 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
4058 (bc_v4i32 (memopv2i64 addr:$src2)))),
4059 (PCMPGTDrm VR128:$src1, addr:$src2)>;
4062 //===---------------------------------------------------------------------===//
4063 // SSE2 - Packed Integer Pack Instructions
4064 //===---------------------------------------------------------------------===//
4066 let Predicates = [HasAVX] in {
4067 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4068 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4069 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4070 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4071 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4072 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4075 let Predicates = [HasAVX2] in {
4076 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4077 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4078 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4079 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4080 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4081 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4084 let Constraints = "$src1 = $dst" in {
4085 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4086 VR128, memopv2i64, i128mem>;
4087 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4088 VR128, memopv2i64, i128mem>;
4089 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4090 VR128, memopv2i64, i128mem>;
4091 } // Constraints = "$src1 = $dst"
4093 //===---------------------------------------------------------------------===//
4094 // SSE2 - Packed Integer Shuffle Instructions
4095 //===---------------------------------------------------------------------===//
4097 let ExeDomain = SSEPackedInt in {
4098 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4100 def ri : Ii8<0x70, MRMSrcReg,
4101 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4102 !strconcat(OpcodeStr,
4103 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4104 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4106 def mi : Ii8<0x70, MRMSrcMem,
4107 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4108 !strconcat(OpcodeStr,
4109 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4110 [(set VR128:$dst, (vt (pshuf_frag:$src2
4111 (bc_frag (memopv2i64 addr:$src1)),
4115 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4117 def Yri : Ii8<0x70, MRMSrcReg,
4118 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4119 !strconcat(OpcodeStr,
4120 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4121 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4123 def Ymi : Ii8<0x70, MRMSrcMem,
4124 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4125 !strconcat(OpcodeStr,
4126 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4127 [(set VR256:$dst, (vt (pshuf_frag:$src2
4128 (bc_frag (memopv4i64 addr:$src1)),
4131 } // ExeDomain = SSEPackedInt
4133 let Predicates = [HasAVX] in {
4134 let AddedComplexity = 5 in
4135 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4138 // SSE2 with ImmT == Imm8 and XS prefix.
4139 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4142 // SSE2 with ImmT == Imm8 and XD prefix.
4143 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4146 let AddedComplexity = 5 in
4147 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4148 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4149 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4150 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4151 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4153 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4155 (VPSHUFDmi addr:$src1, imm:$imm)>;
4156 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4158 (VPSHUFDmi addr:$src1, imm:$imm)>;
4159 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4160 (VPSHUFDri VR128:$src1, imm:$imm)>;
4161 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4162 (VPSHUFDri VR128:$src1, imm:$imm)>;
4163 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4164 (VPSHUFHWri VR128:$src, imm:$imm)>;
4165 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4167 (VPSHUFHWmi addr:$src, imm:$imm)>;
4168 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4169 (VPSHUFLWri VR128:$src, imm:$imm)>;
4170 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4172 (VPSHUFLWmi addr:$src, imm:$imm)>;
4175 let Predicates = [HasAVX2] in {
4176 let AddedComplexity = 5 in
4177 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4180 // SSE2 with ImmT == Imm8 and XS prefix.
4181 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4184 // SSE2 with ImmT == Imm8 and XD prefix.
4185 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4189 let Predicates = [HasSSE2] in {
4190 let AddedComplexity = 5 in
4191 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4193 // SSE2 with ImmT == Imm8 and XS prefix.
4194 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4196 // SSE2 with ImmT == Imm8 and XD prefix.
4197 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4199 let AddedComplexity = 5 in
4200 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4201 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4202 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4203 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4204 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4206 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4208 (PSHUFDmi addr:$src1, imm:$imm)>;
4209 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4211 (PSHUFDmi addr:$src1, imm:$imm)>;
4212 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4213 (PSHUFDri VR128:$src1, imm:$imm)>;
4214 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4215 (PSHUFDri VR128:$src1, imm:$imm)>;
4216 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4217 (PSHUFHWri VR128:$src, imm:$imm)>;
4218 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4220 (PSHUFHWmi addr:$src, imm:$imm)>;
4221 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4222 (PSHUFLWri VR128:$src, imm:$imm)>;
4223 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4225 (PSHUFLWmi addr:$src, imm:$imm)>;
4228 //===---------------------------------------------------------------------===//
4229 // SSE2 - Packed Integer Unpack Instructions
4230 //===---------------------------------------------------------------------===//
4232 let ExeDomain = SSEPackedInt in {
4233 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4234 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4235 def rr : PDI<opc, MRMSrcReg,
4236 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4238 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4239 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4240 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4241 def rm : PDI<opc, MRMSrcMem,
4242 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4244 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4245 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4246 [(set VR128:$dst, (OpNode VR128:$src1,
4247 (bc_frag (memopv2i64
4251 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4252 SDNode OpNode, PatFrag bc_frag> {
4253 def Yrr : PDI<opc, MRMSrcReg,
4254 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4255 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4256 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4257 def Yrm : PDI<opc, MRMSrcMem,
4258 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4259 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4260 [(set VR256:$dst, (OpNode VR256:$src1,
4261 (bc_frag (memopv4i64 addr:$src2))))]>;
4264 let Predicates = [HasAVX] in {
4265 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4266 bc_v16i8, 0>, VEX_4V;
4267 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4268 bc_v8i16, 0>, VEX_4V;
4269 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4270 bc_v4i32, 0>, VEX_4V;
4271 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4272 bc_v2i64, 0>, VEX_4V;
4274 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4275 bc_v16i8, 0>, VEX_4V;
4276 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4277 bc_v8i16, 0>, VEX_4V;
4278 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4279 bc_v4i32, 0>, VEX_4V;
4280 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4281 bc_v2i64, 0>, VEX_4V;
4284 let Predicates = [HasAVX2] in {
4285 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4287 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4289 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4291 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4294 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4296 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4298 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4300 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4304 let Constraints = "$src1 = $dst" in {
4305 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4307 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4309 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4311 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4314 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4316 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4318 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4320 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4323 } // ExeDomain = SSEPackedInt
4325 // Patterns for using AVX1 instructions with integer vectors
4326 // Here to give AVX2 priority
4327 let Predicates = [HasAVX] in {
4328 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4329 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4330 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4331 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4332 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4333 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4334 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4335 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4337 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4338 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4339 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4340 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4341 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4342 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4343 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4344 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4347 // Splat v2f64 / v2i64
4348 let AddedComplexity = 10 in {
4349 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4350 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4351 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4352 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4355 //===---------------------------------------------------------------------===//
4356 // SSE2 - Packed Integer Extract and Insert
4357 //===---------------------------------------------------------------------===//
4359 let ExeDomain = SSEPackedInt in {
4360 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4361 def rri : Ii8<0xC4, MRMSrcReg,
4362 (outs VR128:$dst), (ins VR128:$src1,
4363 GR32:$src2, i32i8imm:$src3),
4365 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4366 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4368 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4369 def rmi : Ii8<0xC4, MRMSrcMem,
4370 (outs VR128:$dst), (ins VR128:$src1,
4371 i16mem:$src2, i32i8imm:$src3),
4373 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4374 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4376 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4381 let Predicates = [HasAVX] in
4382 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4383 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4384 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4385 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4386 imm:$src2))]>, TB, OpSize, VEX;
4387 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4388 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4389 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4390 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4394 let Predicates = [HasAVX] in {
4395 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4396 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4397 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4398 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4399 []>, TB, OpSize, VEX_4V;
4402 let Constraints = "$src1 = $dst" in
4403 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4405 } // ExeDomain = SSEPackedInt
4407 //===---------------------------------------------------------------------===//
4408 // SSE2 - Packed Mask Creation
4409 //===---------------------------------------------------------------------===//
4411 let ExeDomain = SSEPackedInt in {
4413 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4414 "pmovmskb\t{$src, $dst|$dst, $src}",
4415 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4416 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4417 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4419 let Predicates = [HasAVX2] in {
4420 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4421 "pmovmskb\t{$src, $dst|$dst, $src}",
4422 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4423 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4424 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4427 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4428 "pmovmskb\t{$src, $dst|$dst, $src}",
4429 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4431 } // ExeDomain = SSEPackedInt
4433 //===---------------------------------------------------------------------===//
4434 // SSE2 - Conditional Store
4435 //===---------------------------------------------------------------------===//
4437 let ExeDomain = SSEPackedInt in {
4440 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4441 (ins VR128:$src, VR128:$mask),
4442 "maskmovdqu\t{$mask, $src|$src, $mask}",
4443 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4445 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4446 (ins VR128:$src, VR128:$mask),
4447 "maskmovdqu\t{$mask, $src|$src, $mask}",
4448 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4451 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4452 "maskmovdqu\t{$mask, $src|$src, $mask}",
4453 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4455 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4456 "maskmovdqu\t{$mask, $src|$src, $mask}",
4457 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4459 } // ExeDomain = SSEPackedInt
4461 //===---------------------------------------------------------------------===//
4462 // SSE2 - Move Doubleword
4463 //===---------------------------------------------------------------------===//
4465 //===---------------------------------------------------------------------===//
4466 // Move Int Doubleword to Packed Double Int
4468 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4469 "movd\t{$src, $dst|$dst, $src}",
4471 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4472 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4473 "movd\t{$src, $dst|$dst, $src}",
4475 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4477 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4478 "mov{d|q}\t{$src, $dst|$dst, $src}",
4480 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4481 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4482 "mov{d|q}\t{$src, $dst|$dst, $src}",
4483 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4485 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4486 "movd\t{$src, $dst|$dst, $src}",
4488 (v4i32 (scalar_to_vector GR32:$src)))]>;
4489 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4490 "movd\t{$src, $dst|$dst, $src}",
4492 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4493 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4494 "mov{d|q}\t{$src, $dst|$dst, $src}",
4496 (v2i64 (scalar_to_vector GR64:$src)))]>;
4497 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4498 "mov{d|q}\t{$src, $dst|$dst, $src}",
4499 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4501 //===---------------------------------------------------------------------===//
4502 // Move Int Doubleword to Single Scalar
4504 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4505 "movd\t{$src, $dst|$dst, $src}",
4506 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4508 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4509 "movd\t{$src, $dst|$dst, $src}",
4510 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4512 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4513 "movd\t{$src, $dst|$dst, $src}",
4514 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4516 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4517 "movd\t{$src, $dst|$dst, $src}",
4518 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4520 //===---------------------------------------------------------------------===//
4521 // Move Packed Doubleword Int to Packed Double Int
4523 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4524 "movd\t{$src, $dst|$dst, $src}",
4525 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4527 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4528 (ins i32mem:$dst, VR128:$src),
4529 "movd\t{$src, $dst|$dst, $src}",
4530 [(store (i32 (vector_extract (v4i32 VR128:$src),
4531 (iPTR 0))), addr:$dst)]>, VEX;
4532 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4533 "movd\t{$src, $dst|$dst, $src}",
4534 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4536 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4537 "movd\t{$src, $dst|$dst, $src}",
4538 [(store (i32 (vector_extract (v4i32 VR128:$src),
4539 (iPTR 0))), addr:$dst)]>;
4541 //===---------------------------------------------------------------------===//
4542 // Move Packed Doubleword Int first element to Doubleword Int
4544 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4545 "mov{d|q}\t{$src, $dst|$dst, $src}",
4546 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4548 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4550 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4551 "mov{d|q}\t{$src, $dst|$dst, $src}",
4552 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4555 //===---------------------------------------------------------------------===//
4556 // Bitcast FR64 <-> GR64
4558 let Predicates = [HasAVX] in
4559 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4560 "vmovq\t{$src, $dst|$dst, $src}",
4561 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4563 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4564 "mov{d|q}\t{$src, $dst|$dst, $src}",
4565 [(set GR64:$dst, (bitconvert FR64:$src))]>, VEX;
4566 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4567 "movq\t{$src, $dst|$dst, $src}",
4568 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>,
4571 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4572 "movq\t{$src, $dst|$dst, $src}",
4573 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4574 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4575 "mov{d|q}\t{$src, $dst|$dst, $src}",
4576 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4577 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4578 "movq\t{$src, $dst|$dst, $src}",
4579 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4581 //===---------------------------------------------------------------------===//
4582 // Move Scalar Single to Double Int
4584 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4585 "movd\t{$src, $dst|$dst, $src}",
4586 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4587 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4588 "movd\t{$src, $dst|$dst, $src}",
4589 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4590 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4591 "movd\t{$src, $dst|$dst, $src}",
4592 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4593 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4594 "movd\t{$src, $dst|$dst, $src}",
4595 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4597 //===---------------------------------------------------------------------===//
4598 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4600 let AddedComplexity = 15 in {
4601 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4602 "movd\t{$src, $dst|$dst, $src}",
4603 [(set VR128:$dst, (v4i32 (X86vzmovl
4604 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4606 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4607 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4608 [(set VR128:$dst, (v2i64 (X86vzmovl
4609 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4612 let AddedComplexity = 15 in {
4613 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4614 "movd\t{$src, $dst|$dst, $src}",
4615 [(set VR128:$dst, (v4i32 (X86vzmovl
4616 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4617 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4618 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4619 [(set VR128:$dst, (v2i64 (X86vzmovl
4620 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4623 let AddedComplexity = 20 in {
4624 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4625 "movd\t{$src, $dst|$dst, $src}",
4627 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4628 (loadi32 addr:$src))))))]>,
4630 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4631 "movd\t{$src, $dst|$dst, $src}",
4633 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4634 (loadi32 addr:$src))))))]>;
4637 let Predicates = [HasAVX] in {
4638 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4639 let AddedComplexity = 20 in {
4640 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4641 (VMOVZDI2PDIrm addr:$src)>;
4642 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4643 (VMOVZDI2PDIrm addr:$src)>;
4644 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4645 (VMOVZDI2PDIrm addr:$src)>;
4647 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4648 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4649 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4650 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4651 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4652 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4653 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4656 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4657 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4658 (MOVZDI2PDIrm addr:$src)>;
4659 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4660 (MOVZDI2PDIrm addr:$src)>;
4661 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4662 (MOVZDI2PDIrm addr:$src)>;
4665 // These are the correct encodings of the instructions so that we know how to
4666 // read correct assembly, even though we continue to emit the wrong ones for
4667 // compatibility with Darwin's buggy assembler.
4668 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4669 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4670 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4671 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4672 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4673 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4674 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4675 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4676 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4677 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4678 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4679 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4681 //===---------------------------------------------------------------------===//
4682 // SSE2 - Move Quadword
4683 //===---------------------------------------------------------------------===//
4685 //===---------------------------------------------------------------------===//
4686 // Move Quadword Int to Packed Quadword Int
4688 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4689 "vmovq\t{$src, $dst|$dst, $src}",
4691 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4692 VEX, Requires<[HasAVX]>;
4693 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4694 "movq\t{$src, $dst|$dst, $src}",
4696 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4697 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4699 //===---------------------------------------------------------------------===//
4700 // Move Packed Quadword Int to Quadword Int
4702 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4703 "movq\t{$src, $dst|$dst, $src}",
4704 [(store (i64 (vector_extract (v2i64 VR128:$src),
4705 (iPTR 0))), addr:$dst)]>, VEX;
4706 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4707 "movq\t{$src, $dst|$dst, $src}",
4708 [(store (i64 (vector_extract (v2i64 VR128:$src),
4709 (iPTR 0))), addr:$dst)]>;
4711 //===---------------------------------------------------------------------===//
4712 // Store / copy lower 64-bits of a XMM register.
4714 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4715 "movq\t{$src, $dst|$dst, $src}",
4716 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4717 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4718 "movq\t{$src, $dst|$dst, $src}",
4719 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4721 let AddedComplexity = 20 in
4722 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4723 "vmovq\t{$src, $dst|$dst, $src}",
4725 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4726 (loadi64 addr:$src))))))]>,
4727 XS, VEX, Requires<[HasAVX]>;
4729 let AddedComplexity = 20 in
4730 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4731 "movq\t{$src, $dst|$dst, $src}",
4733 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4734 (loadi64 addr:$src))))))]>,
4735 XS, Requires<[HasSSE2]>;
4737 let Predicates = [HasAVX], AddedComplexity = 20 in {
4738 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4739 (VMOVZQI2PQIrm addr:$src)>;
4740 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4741 (VMOVZQI2PQIrm addr:$src)>;
4742 def : Pat<(v2i64 (X86vzload addr:$src)),
4743 (VMOVZQI2PQIrm addr:$src)>;
4746 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4747 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4748 (MOVZQI2PQIrm addr:$src)>;
4749 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4750 (MOVZQI2PQIrm addr:$src)>;
4751 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4754 let Predicates = [HasAVX] in {
4755 def : Pat<(v4i64 (X86vzload addr:$src)),
4756 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4759 //===---------------------------------------------------------------------===//
4760 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4761 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4763 let AddedComplexity = 15 in
4764 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4765 "vmovq\t{$src, $dst|$dst, $src}",
4766 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4767 XS, VEX, Requires<[HasAVX]>;
4768 let AddedComplexity = 15 in
4769 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4770 "movq\t{$src, $dst|$dst, $src}",
4771 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4772 XS, Requires<[HasSSE2]>;
4774 let AddedComplexity = 20 in
4775 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4776 "vmovq\t{$src, $dst|$dst, $src}",
4777 [(set VR128:$dst, (v2i64 (X86vzmovl
4778 (loadv2i64 addr:$src))))]>,
4779 XS, VEX, Requires<[HasAVX]>;
4780 let AddedComplexity = 20 in {
4781 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4782 "movq\t{$src, $dst|$dst, $src}",
4783 [(set VR128:$dst, (v2i64 (X86vzmovl
4784 (loadv2i64 addr:$src))))]>,
4785 XS, Requires<[HasSSE2]>;
4788 let AddedComplexity = 20 in {
4789 let Predicates = [HasAVX] in {
4790 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4791 (VMOVZPQILo2PQIrm addr:$src)>;
4792 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4793 (VMOVZPQILo2PQIrr VR128:$src)>;
4795 let Predicates = [HasSSE2] in {
4796 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4797 (MOVZPQILo2PQIrm addr:$src)>;
4798 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4799 (MOVZPQILo2PQIrr VR128:$src)>;
4803 // Instructions to match in the assembler
4804 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4805 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4806 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4807 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4808 // Recognize "movd" with GR64 destination, but encode as a "movq"
4809 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4810 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4812 // Instructions for the disassembler
4813 // xr = XMM register
4816 let Predicates = [HasAVX] in
4817 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4818 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4819 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4820 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4822 //===---------------------------------------------------------------------===//
4823 // SSE3 - Conversion Instructions
4824 //===---------------------------------------------------------------------===//
4826 // Convert Packed Double FP to Packed DW Integers
4827 let Predicates = [HasAVX] in {
4828 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4829 // register, but the same isn't true when using memory operands instead.
4830 // Provide other assembly rr and rm forms to address this explicitly.
4831 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4832 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4833 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4834 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4837 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4838 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4839 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4840 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4843 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4844 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4845 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4846 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4849 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4850 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4851 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4852 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4854 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4855 (VCVTPD2DQYrr VR256:$src)>;
4856 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4857 (VCVTPD2DQYrm addr:$src)>;
4859 // Convert Packed DW Integers to Packed Double FP
4860 let Predicates = [HasAVX] in {
4861 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4862 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4863 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4864 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4865 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4866 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4867 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4868 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4871 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4872 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4873 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4874 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4876 // AVX 256-bit register conversion intrinsics
4877 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4878 (VCVTDQ2PDYrr VR128:$src)>;
4879 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4880 (VCVTDQ2PDYrm addr:$src)>;
4882 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4883 (VCVTPD2DQYrr VR256:$src)>;
4884 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4885 (VCVTPD2DQYrm addr:$src)>;
4887 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4888 (VCVTDQ2PDYrr VR128:$src)>;
4889 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4890 (VCVTDQ2PDYrm addr:$src)>;
4892 //===---------------------------------------------------------------------===//
4893 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4894 //===---------------------------------------------------------------------===//
4895 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4896 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4897 X86MemOperand x86memop> {
4898 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4899 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4900 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4901 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4902 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4903 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4906 let Predicates = [HasAVX] in {
4907 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4908 v4f32, VR128, memopv4f32, f128mem>, VEX;
4909 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4910 v4f32, VR128, memopv4f32, f128mem>, VEX;
4911 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4912 v8f32, VR256, memopv8f32, f256mem>, VEX;
4913 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4914 v8f32, VR256, memopv8f32, f256mem>, VEX;
4916 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4917 memopv4f32, f128mem>;
4918 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4919 memopv4f32, f128mem>;
4921 let Predicates = [HasAVX] in {
4922 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4923 (VMOVSHDUPrr VR128:$src)>;
4924 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4925 (VMOVSHDUPrm addr:$src)>;
4926 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4927 (VMOVSLDUPrr VR128:$src)>;
4928 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4929 (VMOVSLDUPrm addr:$src)>;
4930 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4931 (VMOVSHDUPYrr VR256:$src)>;
4932 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4933 (VMOVSHDUPYrm addr:$src)>;
4934 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4935 (VMOVSLDUPYrr VR256:$src)>;
4936 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4937 (VMOVSLDUPYrm addr:$src)>;
4940 let Predicates = [HasSSE3] in {
4941 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4942 (MOVSHDUPrr VR128:$src)>;
4943 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4944 (MOVSHDUPrm addr:$src)>;
4945 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4946 (MOVSLDUPrr VR128:$src)>;
4947 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4948 (MOVSLDUPrm addr:$src)>;
4951 //===---------------------------------------------------------------------===//
4952 // SSE3 - Replicate Double FP - MOVDDUP
4953 //===---------------------------------------------------------------------===//
4955 multiclass sse3_replicate_dfp<string OpcodeStr> {
4956 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4957 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4958 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4959 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4960 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4962 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4966 // FIXME: Merge with above classe when there're patterns for the ymm version
4967 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4968 let Predicates = [HasAVX] in {
4969 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4970 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4972 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4973 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4978 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4979 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4980 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4982 let Predicates = [HasAVX] in {
4983 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4985 (VMOVDDUPrm addr:$src)>;
4986 let AddedComplexity = 5 in {
4987 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4988 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4989 (VMOVDDUPrm addr:$src)>;
4990 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4991 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4992 (VMOVDDUPrm addr:$src)>;
4994 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4995 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4996 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4997 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4998 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4999 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5000 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5001 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5002 def : Pat<(X86Movddup (bc_v2f64
5003 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5004 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5007 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
5008 (VMOVDDUPYrm addr:$src)>;
5009 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
5010 (VMOVDDUPYrm addr:$src)>;
5011 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
5012 (VMOVDDUPYrm addr:$src)>;
5013 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
5014 (VMOVDDUPYrm addr:$src)>;
5015 def : Pat<(X86Movddup (v4f64 VR256:$src)),
5016 (VMOVDDUPYrr VR256:$src)>;
5017 def : Pat<(X86Movddup (v4i64 VR256:$src)),
5018 (VMOVDDUPYrr VR256:$src)>;
5021 let Predicates = [HasSSE3] in {
5022 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
5024 (MOVDDUPrm addr:$src)>;
5025 let AddedComplexity = 5 in {
5026 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
5027 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
5028 (MOVDDUPrm addr:$src)>;
5029 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
5030 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
5031 (MOVDDUPrm addr:$src)>;
5033 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5034 (MOVDDUPrm addr:$src)>;
5035 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5036 (MOVDDUPrm addr:$src)>;
5037 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5038 (MOVDDUPrm addr:$src)>;
5039 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5040 (MOVDDUPrm addr:$src)>;
5041 def : Pat<(X86Movddup (bc_v2f64
5042 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5043 (MOVDDUPrm addr:$src)>;
5046 //===---------------------------------------------------------------------===//
5047 // SSE3 - Move Unaligned Integer
5048 //===---------------------------------------------------------------------===//
5050 let Predicates = [HasAVX] in {
5051 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5052 "vlddqu\t{$src, $dst|$dst, $src}",
5053 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5054 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5055 "vlddqu\t{$src, $dst|$dst, $src}",
5056 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5058 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5059 "lddqu\t{$src, $dst|$dst, $src}",
5060 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
5062 //===---------------------------------------------------------------------===//
5063 // SSE3 - Arithmetic
5064 //===---------------------------------------------------------------------===//
5066 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5067 X86MemOperand x86memop, bit Is2Addr = 1> {
5068 def rr : I<0xD0, MRMSrcReg,
5069 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5071 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5072 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5073 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
5074 def rm : I<0xD0, MRMSrcMem,
5075 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5077 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5078 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5079 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
5082 let Predicates = [HasAVX] in {
5083 let ExeDomain = SSEPackedSingle in {
5084 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5085 f128mem, 0>, TB, XD, VEX_4V;
5086 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5087 f256mem, 0>, TB, XD, VEX_4V;
5089 let ExeDomain = SSEPackedDouble in {
5090 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5091 f128mem, 0>, TB, OpSize, VEX_4V;
5092 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5093 f256mem, 0>, TB, OpSize, VEX_4V;
5096 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5097 let ExeDomain = SSEPackedSingle in
5098 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5100 let ExeDomain = SSEPackedDouble in
5101 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5102 f128mem>, TB, OpSize;
5105 //===---------------------------------------------------------------------===//
5106 // SSE3 Instructions
5107 //===---------------------------------------------------------------------===//
5110 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5111 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5112 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5114 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5115 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5116 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5118 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5120 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5121 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5122 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5124 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5125 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5126 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5128 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5129 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5130 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5132 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5134 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5135 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5136 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5139 let Predicates = [HasAVX] in {
5140 let ExeDomain = SSEPackedSingle in {
5141 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5142 X86fhadd, 0>, VEX_4V;
5143 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5144 X86fhsub, 0>, VEX_4V;
5145 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5146 X86fhadd, 0>, VEX_4V;
5147 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5148 X86fhsub, 0>, VEX_4V;
5150 let ExeDomain = SSEPackedDouble in {
5151 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5152 X86fhadd, 0>, VEX_4V;
5153 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5154 X86fhsub, 0>, VEX_4V;
5155 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5156 X86fhadd, 0>, VEX_4V;
5157 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5158 X86fhsub, 0>, VEX_4V;
5162 let Constraints = "$src1 = $dst" in {
5163 let ExeDomain = SSEPackedSingle in {
5164 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5165 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5167 let ExeDomain = SSEPackedDouble in {
5168 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5169 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5173 //===---------------------------------------------------------------------===//
5174 // SSSE3 - Packed Absolute Instructions
5175 //===---------------------------------------------------------------------===//
5178 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5179 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5180 Intrinsic IntId128> {
5181 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5183 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5184 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5187 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5189 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5192 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5195 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5196 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5197 Intrinsic IntId256> {
5198 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5200 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5201 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5204 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5206 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5209 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5212 let Predicates = [HasAVX] in {
5213 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5214 int_x86_ssse3_pabs_b_128>, VEX;
5215 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5216 int_x86_ssse3_pabs_w_128>, VEX;
5217 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5218 int_x86_ssse3_pabs_d_128>, VEX;
5221 let Predicates = [HasAVX2] in {
5222 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5223 int_x86_avx2_pabs_b>, VEX;
5224 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5225 int_x86_avx2_pabs_w>, VEX;
5226 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5227 int_x86_avx2_pabs_d>, VEX;
5230 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5231 int_x86_ssse3_pabs_b_128>;
5232 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5233 int_x86_ssse3_pabs_w_128>;
5234 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5235 int_x86_ssse3_pabs_d_128>;
5237 //===---------------------------------------------------------------------===//
5238 // SSSE3 - Packed Binary Operator Instructions
5239 //===---------------------------------------------------------------------===//
5241 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5242 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5243 Intrinsic IntId128, bit Is2Addr = 1> {
5244 let isCommutable = 1 in
5245 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5246 (ins VR128:$src1, VR128:$src2),
5248 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5249 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5250 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5252 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5253 (ins VR128:$src1, i128mem:$src2),
5255 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5256 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5258 (IntId128 VR128:$src1,
5259 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5262 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5263 Intrinsic IntId256> {
5264 let isCommutable = 1 in
5265 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5266 (ins VR256:$src1, VR256:$src2),
5267 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5268 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5270 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5271 (ins VR256:$src1, i256mem:$src2),
5272 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5274 (IntId256 VR256:$src1,
5275 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5278 let ImmT = NoImm, Predicates = [HasAVX] in {
5279 let isCommutable = 0 in {
5280 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw",
5281 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5282 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd",
5283 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5284 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5285 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5286 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw",
5287 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5288 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd",
5289 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5290 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5291 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5292 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5293 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5294 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb",
5295 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5296 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb",
5297 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5298 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw",
5299 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5300 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd",
5301 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5303 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5304 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5307 let ImmT = NoImm, Predicates = [HasAVX2] in {
5308 let isCommutable = 0 in {
5309 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw",
5310 int_x86_avx2_phadd_w>, VEX_4V;
5311 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd",
5312 int_x86_avx2_phadd_d>, VEX_4V;
5313 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5314 int_x86_avx2_phadd_sw>, VEX_4V;
5315 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw",
5316 int_x86_avx2_phsub_w>, VEX_4V;
5317 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd",
5318 int_x86_avx2_phsub_d>, VEX_4V;
5319 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5320 int_x86_avx2_phsub_sw>, VEX_4V;
5321 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5322 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5323 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb",
5324 int_x86_avx2_pshuf_b>, VEX_4V;
5325 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb",
5326 int_x86_avx2_psign_b>, VEX_4V;
5327 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw",
5328 int_x86_avx2_psign_w>, VEX_4V;
5329 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd",
5330 int_x86_avx2_psign_d>, VEX_4V;
5332 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5333 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5336 // None of these have i8 immediate fields.
5337 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5338 let isCommutable = 0 in {
5339 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw",
5340 int_x86_ssse3_phadd_w_128>;
5341 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd",
5342 int_x86_ssse3_phadd_d_128>;
5343 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5344 int_x86_ssse3_phadd_sw_128>;
5345 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw",
5346 int_x86_ssse3_phsub_w_128>;
5347 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd",
5348 int_x86_ssse3_phsub_d_128>;
5349 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5350 int_x86_ssse3_phsub_sw_128>;
5351 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5352 int_x86_ssse3_pmadd_ub_sw_128>;
5353 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb",
5354 int_x86_ssse3_pshuf_b_128>;
5355 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb",
5356 int_x86_ssse3_psign_b_128>;
5357 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw",
5358 int_x86_ssse3_psign_w_128>;
5359 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd",
5360 int_x86_ssse3_psign_d_128>;
5362 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5363 int_x86_ssse3_pmul_hr_sw_128>;
5366 let Predicates = [HasAVX] in {
5367 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5368 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5369 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5370 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5372 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5373 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5374 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5375 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5376 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5377 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5379 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5380 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5381 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5382 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5383 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5384 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5385 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5386 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5389 let Predicates = [HasAVX2] in {
5390 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5391 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5392 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5393 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5394 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5395 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5397 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5398 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5399 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5400 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5401 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5402 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5403 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5404 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5407 let Predicates = [HasSSSE3] in {
5408 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5409 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5410 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5411 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5413 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5414 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5415 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5416 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5417 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5418 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5420 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5421 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5422 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5423 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5424 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5425 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5426 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5427 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5430 //===---------------------------------------------------------------------===//
5431 // SSSE3 - Packed Align Instruction Patterns
5432 //===---------------------------------------------------------------------===//
5434 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5435 let neverHasSideEffects = 1 in {
5436 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5437 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5439 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5441 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5444 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5445 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5447 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5449 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5454 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5455 let neverHasSideEffects = 1 in {
5456 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5457 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5459 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5462 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5463 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5465 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5470 let Predicates = [HasAVX] in
5471 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5472 let Predicates = [HasAVX2] in
5473 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5474 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5475 defm PALIGN : ssse3_palign<"palignr">;
5477 let Predicates = [HasAVX2] in {
5478 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5479 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5480 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5481 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5482 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5483 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5484 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5485 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5488 let Predicates = [HasAVX] in {
5489 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5490 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5491 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5492 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5493 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5494 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5495 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5496 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5499 let Predicates = [HasSSSE3] in {
5500 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5501 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5502 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5503 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5504 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5505 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5506 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5507 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5510 //===---------------------------------------------------------------------===//
5511 // SSSE3 - Thread synchronization
5512 //===---------------------------------------------------------------------===//
5514 let usesCustomInserter = 1 in {
5515 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5516 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5517 Requires<[HasSSE3]>;
5518 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5519 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5520 Requires<[HasSSE3]>;
5523 let Uses = [EAX, ECX, EDX] in
5524 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5525 Requires<[HasSSE3]>;
5526 let Uses = [ECX, EAX] in
5527 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5528 Requires<[HasSSE3]>;
5530 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5531 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5533 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5534 Requires<[In32BitMode]>;
5535 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5536 Requires<[In64BitMode]>;
5538 //===----------------------------------------------------------------------===//
5539 // SSE4.1 - Packed Move with Sign/Zero Extend
5540 //===----------------------------------------------------------------------===//
5542 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5543 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5544 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5545 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5547 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5548 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5550 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5554 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5556 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5557 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5558 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5560 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5561 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5562 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5565 let Predicates = [HasAVX] in {
5566 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5568 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5570 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5572 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5574 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5576 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5580 let Predicates = [HasAVX2] in {
5581 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5582 int_x86_avx2_pmovsxbw>, VEX;
5583 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5584 int_x86_avx2_pmovsxwd>, VEX;
5585 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5586 int_x86_avx2_pmovsxdq>, VEX;
5587 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5588 int_x86_avx2_pmovzxbw>, VEX;
5589 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5590 int_x86_avx2_pmovzxwd>, VEX;
5591 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5592 int_x86_avx2_pmovzxdq>, VEX;
5595 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5596 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5597 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5598 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5599 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5600 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5602 let Predicates = [HasAVX] in {
5603 // Common patterns involving scalar load.
5604 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5605 (VPMOVSXBWrm addr:$src)>;
5606 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5607 (VPMOVSXBWrm addr:$src)>;
5609 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5610 (VPMOVSXWDrm addr:$src)>;
5611 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5612 (VPMOVSXWDrm addr:$src)>;
5614 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5615 (VPMOVSXDQrm addr:$src)>;
5616 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5617 (VPMOVSXDQrm addr:$src)>;
5619 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5620 (VPMOVZXBWrm addr:$src)>;
5621 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5622 (VPMOVZXBWrm addr:$src)>;
5624 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5625 (VPMOVZXWDrm addr:$src)>;
5626 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5627 (VPMOVZXWDrm addr:$src)>;
5629 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5630 (VPMOVZXDQrm addr:$src)>;
5631 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5632 (VPMOVZXDQrm addr:$src)>;
5635 let Predicates = [HasSSE41] in {
5636 // Common patterns involving scalar load.
5637 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5638 (PMOVSXBWrm addr:$src)>;
5639 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5640 (PMOVSXBWrm addr:$src)>;
5642 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5643 (PMOVSXWDrm addr:$src)>;
5644 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5645 (PMOVSXWDrm addr:$src)>;
5647 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5648 (PMOVSXDQrm addr:$src)>;
5649 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5650 (PMOVSXDQrm addr:$src)>;
5652 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5653 (PMOVZXBWrm addr:$src)>;
5654 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5655 (PMOVZXBWrm addr:$src)>;
5657 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5658 (PMOVZXWDrm addr:$src)>;
5659 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5660 (PMOVZXWDrm addr:$src)>;
5662 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5663 (PMOVZXDQrm addr:$src)>;
5664 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5665 (PMOVZXDQrm addr:$src)>;
5669 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5670 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5671 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5672 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5674 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5675 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5677 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5681 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5683 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5684 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5685 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5687 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5688 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5690 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5694 let Predicates = [HasAVX] in {
5695 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5697 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5699 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5701 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5705 let Predicates = [HasAVX2] in {
5706 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5707 int_x86_avx2_pmovsxbd>, VEX;
5708 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5709 int_x86_avx2_pmovsxwq>, VEX;
5710 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5711 int_x86_avx2_pmovzxbd>, VEX;
5712 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5713 int_x86_avx2_pmovzxwq>, VEX;
5716 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5717 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5718 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5719 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5721 let Predicates = [HasAVX] in {
5722 // Common patterns involving scalar load
5723 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5724 (VPMOVSXBDrm addr:$src)>;
5725 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5726 (VPMOVSXWQrm addr:$src)>;
5728 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5729 (VPMOVZXBDrm addr:$src)>;
5730 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5731 (VPMOVZXWQrm addr:$src)>;
5734 let Predicates = [HasSSE41] in {
5735 // Common patterns involving scalar load
5736 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5737 (PMOVSXBDrm addr:$src)>;
5738 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5739 (PMOVSXWQrm addr:$src)>;
5741 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5742 (PMOVZXBDrm addr:$src)>;
5743 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5744 (PMOVZXWQrm addr:$src)>;
5747 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5748 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5749 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5750 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5752 // Expecting a i16 load any extended to i32 value.
5753 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5754 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5755 [(set VR128:$dst, (IntId (bitconvert
5756 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5760 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5762 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5763 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5764 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5766 // Expecting a i16 load any extended to i32 value.
5767 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5768 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5769 [(set VR256:$dst, (IntId (bitconvert
5770 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5774 let Predicates = [HasAVX] in {
5775 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5777 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5780 let Predicates = [HasAVX2] in {
5781 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5782 int_x86_avx2_pmovsxbq>, VEX;
5783 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5784 int_x86_avx2_pmovzxbq>, VEX;
5786 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5787 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5789 let Predicates = [HasAVX] in {
5790 // Common patterns involving scalar load
5791 def : Pat<(int_x86_sse41_pmovsxbq
5792 (bitconvert (v4i32 (X86vzmovl
5793 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5794 (VPMOVSXBQrm addr:$src)>;
5796 def : Pat<(int_x86_sse41_pmovzxbq
5797 (bitconvert (v4i32 (X86vzmovl
5798 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5799 (VPMOVZXBQrm addr:$src)>;
5802 let Predicates = [HasSSE41] in {
5803 // Common patterns involving scalar load
5804 def : Pat<(int_x86_sse41_pmovsxbq
5805 (bitconvert (v4i32 (X86vzmovl
5806 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5807 (PMOVSXBQrm addr:$src)>;
5809 def : Pat<(int_x86_sse41_pmovzxbq
5810 (bitconvert (v4i32 (X86vzmovl
5811 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5812 (PMOVZXBQrm addr:$src)>;
5815 //===----------------------------------------------------------------------===//
5816 // SSE4.1 - Extract Instructions
5817 //===----------------------------------------------------------------------===//
5819 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5820 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5821 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5822 (ins VR128:$src1, i32i8imm:$src2),
5823 !strconcat(OpcodeStr,
5824 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5825 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5827 let neverHasSideEffects = 1, mayStore = 1 in
5828 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5829 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5830 !strconcat(OpcodeStr,
5831 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5834 // There's an AssertZext in the way of writing the store pattern
5835 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5838 let Predicates = [HasAVX] in {
5839 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5840 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5841 (ins VR128:$src1, i32i8imm:$src2),
5842 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5845 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5848 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5849 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5850 let neverHasSideEffects = 1, mayStore = 1 in
5851 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5852 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5853 !strconcat(OpcodeStr,
5854 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5857 // There's an AssertZext in the way of writing the store pattern
5858 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5861 let Predicates = [HasAVX] in
5862 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5864 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5867 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5868 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5869 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5870 (ins VR128:$src1, i32i8imm:$src2),
5871 !strconcat(OpcodeStr,
5872 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5874 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5875 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5876 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5877 !strconcat(OpcodeStr,
5878 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5879 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5880 addr:$dst)]>, OpSize;
5883 let Predicates = [HasAVX] in
5884 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5886 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5888 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5889 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5890 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5891 (ins VR128:$src1, i32i8imm:$src2),
5892 !strconcat(OpcodeStr,
5893 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5895 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5896 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5897 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5898 !strconcat(OpcodeStr,
5899 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5900 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5901 addr:$dst)]>, OpSize, REX_W;
5904 let Predicates = [HasAVX] in
5905 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5907 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5909 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5911 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5912 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5913 (ins VR128:$src1, i32i8imm:$src2),
5914 !strconcat(OpcodeStr,
5915 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5917 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5919 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5920 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5921 !strconcat(OpcodeStr,
5922 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5923 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5924 addr:$dst)]>, OpSize;
5927 let ExeDomain = SSEPackedSingle in {
5928 let Predicates = [HasAVX] in {
5929 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5930 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5931 (ins VR128:$src1, i32i8imm:$src2),
5932 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5935 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5938 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5939 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5942 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5944 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5947 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5948 Requires<[HasSSE41]>;
5950 //===----------------------------------------------------------------------===//
5951 // SSE4.1 - Insert Instructions
5952 //===----------------------------------------------------------------------===//
5954 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5955 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5956 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5958 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5960 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5962 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5963 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5964 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5966 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5968 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5970 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5971 imm:$src3))]>, OpSize;
5974 let Predicates = [HasAVX] in
5975 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5976 let Constraints = "$src1 = $dst" in
5977 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5979 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5980 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5981 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5983 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5985 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5987 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5989 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5990 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5992 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5994 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5996 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5997 imm:$src3)))]>, OpSize;
6000 let Predicates = [HasAVX] in
6001 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6002 let Constraints = "$src1 = $dst" in
6003 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6005 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6006 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6007 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6009 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6011 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6013 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6015 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6016 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6018 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6020 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6022 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6023 imm:$src3)))]>, OpSize;
6026 let Predicates = [HasAVX] in
6027 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6028 let Constraints = "$src1 = $dst" in
6029 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6031 // insertps has a few different modes, there's the first two here below which
6032 // are optimized inserts that won't zero arbitrary elements in the destination
6033 // vector. The next one matches the intrinsic and could zero arbitrary elements
6034 // in the target vector.
6035 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6036 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6037 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6039 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6041 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6043 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6045 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6046 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6048 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6050 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6052 (X86insrtps VR128:$src1,
6053 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6054 imm:$src3))]>, OpSize;
6057 let ExeDomain = SSEPackedSingle in {
6058 let Predicates = [HasAVX] in
6059 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6060 let Constraints = "$src1 = $dst" in
6061 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6064 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6065 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6067 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6068 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6069 Requires<[HasSSE41]>;
6071 //===----------------------------------------------------------------------===//
6072 // SSE4.1 - Round Instructions
6073 //===----------------------------------------------------------------------===//
6075 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6076 X86MemOperand x86memop, RegisterClass RC,
6077 PatFrag mem_frag32, PatFrag mem_frag64,
6078 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6079 let ExeDomain = SSEPackedSingle in {
6080 // Intrinsic operation, reg.
6081 // Vector intrinsic operation, reg
6082 def PSr : SS4AIi8<opcps, MRMSrcReg,
6083 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6084 !strconcat(OpcodeStr,
6085 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6086 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6089 // Vector intrinsic operation, mem
6090 def PSm : SS4AIi8<opcps, MRMSrcMem,
6091 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6092 !strconcat(OpcodeStr,
6093 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6095 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6097 } // ExeDomain = SSEPackedSingle
6099 let ExeDomain = SSEPackedDouble in {
6100 // Vector intrinsic operation, reg
6101 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6102 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6103 !strconcat(OpcodeStr,
6104 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6105 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6108 // Vector intrinsic operation, mem
6109 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6110 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6111 !strconcat(OpcodeStr,
6112 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6114 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6116 } // ExeDomain = SSEPackedDouble
6119 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6122 Intrinsic F64Int, bit Is2Addr = 1> {
6123 let ExeDomain = GenericDomain in {
6125 def SSr : SS4AIi8<opcss, MRMSrcReg,
6126 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6128 !strconcat(OpcodeStr,
6129 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6130 !strconcat(OpcodeStr,
6131 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6134 // Intrinsic operation, reg.
6135 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6136 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6138 !strconcat(OpcodeStr,
6139 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6140 !strconcat(OpcodeStr,
6141 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6142 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6145 // Intrinsic operation, mem.
6146 def SSm : SS4AIi8<opcss, MRMSrcMem,
6147 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6149 !strconcat(OpcodeStr,
6150 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6151 !strconcat(OpcodeStr,
6152 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6154 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6158 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6159 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6161 !strconcat(OpcodeStr,
6162 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6163 !strconcat(OpcodeStr,
6164 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6167 // Intrinsic operation, reg.
6168 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6169 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6171 !strconcat(OpcodeStr,
6172 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6173 !strconcat(OpcodeStr,
6174 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6175 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6178 // Intrinsic operation, mem.
6179 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6180 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6182 !strconcat(OpcodeStr,
6183 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6184 !strconcat(OpcodeStr,
6185 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6187 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6189 } // ExeDomain = GenericDomain
6192 // FP round - roundss, roundps, roundsd, roundpd
6193 let Predicates = [HasAVX] in {
6195 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6196 memopv4f32, memopv2f64,
6197 int_x86_sse41_round_ps,
6198 int_x86_sse41_round_pd>, VEX;
6199 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6200 memopv8f32, memopv4f64,
6201 int_x86_avx_round_ps_256,
6202 int_x86_avx_round_pd_256>, VEX;
6203 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6204 int_x86_sse41_round_ss,
6205 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6207 def : Pat<(ffloor FR32:$src),
6208 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6209 def : Pat<(f64 (ffloor FR64:$src)),
6210 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6211 def : Pat<(f32 (fnearbyint FR32:$src)),
6212 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6213 def : Pat<(f64 (fnearbyint FR64:$src)),
6214 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6215 def : Pat<(f32 (fceil FR32:$src)),
6216 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6217 def : Pat<(f64 (fceil FR64:$src)),
6218 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6219 def : Pat<(f32 (frint FR32:$src)),
6220 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6221 def : Pat<(f64 (frint FR64:$src)),
6222 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6223 def : Pat<(f32 (ftrunc FR32:$src)),
6224 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6225 def : Pat<(f64 (ftrunc FR64:$src)),
6226 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6229 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6230 memopv4f32, memopv2f64,
6231 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6232 let Constraints = "$src1 = $dst" in
6233 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6234 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6236 def : Pat<(ffloor FR32:$src),
6237 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6238 def : Pat<(f64 (ffloor FR64:$src)),
6239 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6240 def : Pat<(f32 (fnearbyint FR32:$src)),
6241 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6242 def : Pat<(f64 (fnearbyint FR64:$src)),
6243 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6244 def : Pat<(f32 (fceil FR32:$src)),
6245 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6246 def : Pat<(f64 (fceil FR64:$src)),
6247 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6248 def : Pat<(f32 (frint FR32:$src)),
6249 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6250 def : Pat<(f64 (frint FR64:$src)),
6251 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6252 def : Pat<(f32 (ftrunc FR32:$src)),
6253 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6254 def : Pat<(f64 (ftrunc FR64:$src)),
6255 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6257 //===----------------------------------------------------------------------===//
6258 // SSE4.1 - Packed Bit Test
6259 //===----------------------------------------------------------------------===//
6261 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6262 // the intel intrinsic that corresponds to this.
6263 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6264 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6265 "vptest\t{$src2, $src1|$src1, $src2}",
6266 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6268 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6269 "vptest\t{$src2, $src1|$src1, $src2}",
6270 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6273 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6274 "vptest\t{$src2, $src1|$src1, $src2}",
6275 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6277 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6278 "vptest\t{$src2, $src1|$src1, $src2}",
6279 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6283 let Defs = [EFLAGS] in {
6284 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6285 "ptest\t{$src2, $src1|$src1, $src2}",
6286 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6288 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6289 "ptest\t{$src2, $src1|$src1, $src2}",
6290 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6294 // The bit test instructions below are AVX only
6295 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6296 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6297 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6298 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6299 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6300 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6301 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6302 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6306 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6307 let ExeDomain = SSEPackedSingle in {
6308 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6309 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6311 let ExeDomain = SSEPackedDouble in {
6312 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6313 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6317 //===----------------------------------------------------------------------===//
6318 // SSE4.1 - Misc Instructions
6319 //===----------------------------------------------------------------------===//
6321 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6322 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6323 "popcnt{w}\t{$src, $dst|$dst, $src}",
6324 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6326 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6327 "popcnt{w}\t{$src, $dst|$dst, $src}",
6328 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6329 (implicit EFLAGS)]>, OpSize, XS;
6331 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6332 "popcnt{l}\t{$src, $dst|$dst, $src}",
6333 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6335 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6336 "popcnt{l}\t{$src, $dst|$dst, $src}",
6337 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6338 (implicit EFLAGS)]>, XS;
6340 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6341 "popcnt{q}\t{$src, $dst|$dst, $src}",
6342 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6344 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6345 "popcnt{q}\t{$src, $dst|$dst, $src}",
6346 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6347 (implicit EFLAGS)]>, XS;
6352 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6353 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6354 Intrinsic IntId128> {
6355 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6357 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6358 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6359 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6361 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6364 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6367 let Predicates = [HasAVX] in
6368 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6369 int_x86_sse41_phminposuw>, VEX;
6370 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6371 int_x86_sse41_phminposuw>;
6373 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6374 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6375 Intrinsic IntId128, bit Is2Addr = 1> {
6376 let isCommutable = 1 in
6377 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6378 (ins VR128:$src1, VR128:$src2),
6380 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6381 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6382 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6383 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6384 (ins VR128:$src1, i128mem:$src2),
6386 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6387 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6389 (IntId128 VR128:$src1,
6390 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6393 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6394 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6395 Intrinsic IntId256> {
6396 let isCommutable = 1 in
6397 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6398 (ins VR256:$src1, VR256:$src2),
6399 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6400 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6401 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6402 (ins VR256:$src1, i256mem:$src2),
6403 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6405 (IntId256 VR256:$src1,
6406 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6409 let Predicates = [HasAVX] in {
6410 let isCommutable = 0 in
6411 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6413 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6415 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6417 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6419 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6421 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6423 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6425 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6427 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6429 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6431 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6434 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6435 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6436 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6437 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6440 let Predicates = [HasAVX2] in {
6441 let isCommutable = 0 in
6442 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6443 int_x86_avx2_packusdw>, VEX_4V;
6444 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6445 int_x86_avx2_pcmpeq_q>, VEX_4V;
6446 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6447 int_x86_avx2_pmins_b>, VEX_4V;
6448 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6449 int_x86_avx2_pmins_d>, VEX_4V;
6450 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6451 int_x86_avx2_pminu_d>, VEX_4V;
6452 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6453 int_x86_avx2_pminu_w>, VEX_4V;
6454 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6455 int_x86_avx2_pmaxs_b>, VEX_4V;
6456 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6457 int_x86_avx2_pmaxs_d>, VEX_4V;
6458 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6459 int_x86_avx2_pmaxu_d>, VEX_4V;
6460 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6461 int_x86_avx2_pmaxu_w>, VEX_4V;
6462 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6463 int_x86_avx2_pmul_dq>, VEX_4V;
6465 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, VR256:$src2)),
6466 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6467 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, (memop addr:$src2))),
6468 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6471 let Constraints = "$src1 = $dst" in {
6472 let isCommutable = 0 in
6473 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6474 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6475 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6476 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6477 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6478 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6479 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6480 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6481 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6482 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6483 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6486 let Predicates = [HasSSE41] in {
6487 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6488 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6489 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6490 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6493 /// SS48I_binop_rm - Simple SSE41 binary operator.
6494 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6495 ValueType OpVT, bit Is2Addr = 1> {
6496 let isCommutable = 1 in
6497 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6498 (ins VR128:$src1, VR128:$src2),
6500 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6501 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6502 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6504 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6505 (ins VR128:$src1, i128mem:$src2),
6507 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6508 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6509 [(set VR128:$dst, (OpNode VR128:$src1,
6510 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6514 /// SS48I_binop_rm - Simple SSE41 binary operator.
6515 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6517 let isCommutable = 1 in
6518 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6519 (ins VR256:$src1, VR256:$src2),
6520 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6521 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6523 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6524 (ins VR256:$src1, i256mem:$src2),
6525 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6526 [(set VR256:$dst, (OpNode VR256:$src1,
6527 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6531 let Predicates = [HasAVX] in
6532 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6533 let Predicates = [HasAVX2] in
6534 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6535 let Constraints = "$src1 = $dst" in
6536 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6538 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6539 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6540 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6541 X86MemOperand x86memop, bit Is2Addr = 1> {
6542 let isCommutable = 1 in
6543 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6544 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6546 !strconcat(OpcodeStr,
6547 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6548 !strconcat(OpcodeStr,
6549 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6550 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6552 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6553 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6555 !strconcat(OpcodeStr,
6556 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6557 !strconcat(OpcodeStr,
6558 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6561 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6565 let Predicates = [HasAVX] in {
6566 let isCommutable = 0 in {
6567 let ExeDomain = SSEPackedSingle in {
6568 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6569 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6570 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6571 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6573 let ExeDomain = SSEPackedDouble in {
6574 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6575 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6576 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6577 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6579 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6580 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6581 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6582 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6584 let ExeDomain = SSEPackedSingle in
6585 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6586 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6587 let ExeDomain = SSEPackedDouble in
6588 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6589 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6590 let ExeDomain = SSEPackedSingle in
6591 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6592 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6595 let Predicates = [HasAVX2] in {
6596 let isCommutable = 0 in {
6597 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6598 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6599 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6600 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6604 let Constraints = "$src1 = $dst" in {
6605 let isCommutable = 0 in {
6606 let ExeDomain = SSEPackedSingle in
6607 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6608 VR128, memopv4f32, i128mem>;
6609 let ExeDomain = SSEPackedDouble in
6610 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6611 VR128, memopv2f64, i128mem>;
6612 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6613 VR128, memopv2i64, i128mem>;
6614 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6615 VR128, memopv2i64, i128mem>;
6617 let ExeDomain = SSEPackedSingle in
6618 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6619 VR128, memopv4f32, i128mem>;
6620 let ExeDomain = SSEPackedDouble in
6621 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6622 VR128, memopv2f64, i128mem>;
6625 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6626 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6627 RegisterClass RC, X86MemOperand x86memop,
6628 PatFrag mem_frag, Intrinsic IntId> {
6629 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6630 (ins RC:$src1, RC:$src2, RC:$src3),
6631 !strconcat(OpcodeStr,
6632 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6633 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6634 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6636 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6637 (ins RC:$src1, x86memop:$src2, RC:$src3),
6638 !strconcat(OpcodeStr,
6639 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6641 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6643 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6646 let Predicates = [HasAVX] in {
6647 let ExeDomain = SSEPackedDouble in {
6648 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6649 memopv2f64, int_x86_sse41_blendvpd>;
6650 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6651 memopv4f64, int_x86_avx_blendv_pd_256>;
6652 } // ExeDomain = SSEPackedDouble
6653 let ExeDomain = SSEPackedSingle in {
6654 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6655 memopv4f32, int_x86_sse41_blendvps>;
6656 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6657 memopv8f32, int_x86_avx_blendv_ps_256>;
6658 } // ExeDomain = SSEPackedSingle
6659 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6660 memopv2i64, int_x86_sse41_pblendvb>;
6663 let Predicates = [HasAVX2] in {
6664 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6665 memopv4i64, int_x86_avx2_pblendvb>;
6668 let Predicates = [HasAVX] in {
6669 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6670 (v16i8 VR128:$src2))),
6671 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6672 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6673 (v4i32 VR128:$src2))),
6674 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6675 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6676 (v4f32 VR128:$src2))),
6677 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6678 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6679 (v2i64 VR128:$src2))),
6680 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6681 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6682 (v2f64 VR128:$src2))),
6683 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6684 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6685 (v8i32 VR256:$src2))),
6686 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6687 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6688 (v8f32 VR256:$src2))),
6689 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6690 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6691 (v4i64 VR256:$src2))),
6692 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6693 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6694 (v4f64 VR256:$src2))),
6695 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6698 let Predicates = [HasAVX2] in {
6699 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6700 (v32i8 VR256:$src2))),
6701 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6704 /// SS41I_ternary_int - SSE 4.1 ternary operator
6705 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6706 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6708 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6709 (ins VR128:$src1, VR128:$src2),
6710 !strconcat(OpcodeStr,
6711 "\t{$src2, $dst|$dst, $src2}"),
6712 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6715 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6716 (ins VR128:$src1, i128mem:$src2),
6717 !strconcat(OpcodeStr,
6718 "\t{$src2, $dst|$dst, $src2}"),
6721 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6725 let ExeDomain = SSEPackedDouble in
6726 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6727 int_x86_sse41_blendvpd>;
6728 let ExeDomain = SSEPackedSingle in
6729 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6730 int_x86_sse41_blendvps>;
6731 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6732 int_x86_sse41_pblendvb>;
6734 let Predicates = [HasSSE41] in {
6735 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6736 (v16i8 VR128:$src2))),
6737 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6738 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6739 (v4i32 VR128:$src2))),
6740 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6741 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6742 (v4f32 VR128:$src2))),
6743 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6744 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6745 (v2i64 VR128:$src2))),
6746 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6747 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6748 (v2f64 VR128:$src2))),
6749 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6752 let Predicates = [HasAVX] in
6753 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6754 "vmovntdqa\t{$src, $dst|$dst, $src}",
6755 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6757 let Predicates = [HasAVX2] in
6758 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6759 "vmovntdqa\t{$src, $dst|$dst, $src}",
6760 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6762 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6763 "movntdqa\t{$src, $dst|$dst, $src}",
6764 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6767 //===----------------------------------------------------------------------===//
6768 // SSE4.2 - Compare Instructions
6769 //===----------------------------------------------------------------------===//
6771 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6772 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6773 Intrinsic IntId128, bit Is2Addr = 1> {
6774 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6775 (ins VR128:$src1, VR128:$src2),
6777 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6778 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6779 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6781 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6782 (ins VR128:$src1, i128mem:$src2),
6784 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6785 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6787 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6790 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6791 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6792 Intrinsic IntId256> {
6793 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6794 (ins VR256:$src1, VR256:$src2),
6795 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6796 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6798 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6799 (ins VR256:$src1, i256mem:$src2),
6800 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6802 (IntId256 VR256:$src1, (memopv4i64 addr:$src2)))]>, OpSize;
6805 let Predicates = [HasAVX] in {
6806 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6809 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6810 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6811 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6812 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6815 let Predicates = [HasAVX2] in {
6816 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6819 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, VR256:$src2)),
6820 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6821 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, (memop addr:$src2))),
6822 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6825 let Constraints = "$src1 = $dst" in
6826 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6828 let Predicates = [HasSSE42] in {
6829 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6830 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6831 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6832 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6835 //===----------------------------------------------------------------------===//
6836 // SSE4.2 - String/text Processing Instructions
6837 //===----------------------------------------------------------------------===//
6839 // Packed Compare Implicit Length Strings, Return Mask
6840 multiclass pseudo_pcmpistrm<string asm> {
6841 def REG : PseudoI<(outs VR128:$dst),
6842 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6843 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6845 def MEM : PseudoI<(outs VR128:$dst),
6846 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6847 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6848 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6851 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6852 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6853 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6856 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6857 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6858 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6859 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6861 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6862 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6863 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6866 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6867 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6868 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6869 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6871 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6872 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6873 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6876 // Packed Compare Explicit Length Strings, Return Mask
6877 multiclass pseudo_pcmpestrm<string asm> {
6878 def REG : PseudoI<(outs VR128:$dst),
6879 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6880 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6881 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6882 def MEM : PseudoI<(outs VR128:$dst),
6883 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6884 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6885 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6888 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6889 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6890 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6893 let Predicates = [HasAVX],
6894 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6895 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6896 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6897 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6899 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6900 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6901 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6904 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6905 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6906 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6907 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6909 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6910 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6911 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6914 // Packed Compare Implicit Length Strings, Return Index
6915 let Defs = [ECX, EFLAGS] in {
6916 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6917 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6918 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6919 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6920 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6921 (implicit EFLAGS)]>, OpSize;
6922 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6923 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6924 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6925 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6926 (implicit EFLAGS)]>, OpSize;
6930 let Predicates = [HasAVX] in {
6931 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6933 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6935 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6937 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6939 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6941 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6945 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6946 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6947 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6948 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6949 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6950 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6952 // Packed Compare Explicit Length Strings, Return Index
6953 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6954 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6955 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6956 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6957 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6958 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6959 (implicit EFLAGS)]>, OpSize;
6960 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6961 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6962 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6964 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6965 (implicit EFLAGS)]>, OpSize;
6969 let Predicates = [HasAVX] in {
6970 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6972 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6974 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6976 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6978 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6980 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6984 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6985 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6986 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6987 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6988 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6989 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6991 //===----------------------------------------------------------------------===//
6992 // SSE4.2 - CRC Instructions
6993 //===----------------------------------------------------------------------===//
6995 // No CRC instructions have AVX equivalents
6997 // crc intrinsic instruction
6998 // This set of instructions are only rm, the only difference is the size
7000 let Constraints = "$src1 = $dst" in {
7001 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
7002 (ins GR32:$src1, i8mem:$src2),
7003 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7005 (int_x86_sse42_crc32_32_8 GR32:$src1,
7006 (load addr:$src2)))]>;
7007 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
7008 (ins GR32:$src1, GR8:$src2),
7009 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7011 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
7012 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7013 (ins GR32:$src1, i16mem:$src2),
7014 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7016 (int_x86_sse42_crc32_32_16 GR32:$src1,
7017 (load addr:$src2)))]>,
7019 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7020 (ins GR32:$src1, GR16:$src2),
7021 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7023 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7025 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7026 (ins GR32:$src1, i32mem:$src2),
7027 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7029 (int_x86_sse42_crc32_32_32 GR32:$src1,
7030 (load addr:$src2)))]>;
7031 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7032 (ins GR32:$src1, GR32:$src2),
7033 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7035 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7036 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7037 (ins GR64:$src1, i8mem:$src2),
7038 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7040 (int_x86_sse42_crc32_64_8 GR64:$src1,
7041 (load addr:$src2)))]>,
7043 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7044 (ins GR64:$src1, GR8:$src2),
7045 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7047 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7049 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7050 (ins GR64:$src1, i64mem:$src2),
7051 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7053 (int_x86_sse42_crc32_64_64 GR64:$src1,
7054 (load addr:$src2)))]>,
7056 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7057 (ins GR64:$src1, GR64:$src2),
7058 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7060 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7064 //===----------------------------------------------------------------------===//
7065 // AES-NI Instructions
7066 //===----------------------------------------------------------------------===//
7068 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7069 Intrinsic IntId128, bit Is2Addr = 1> {
7070 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7071 (ins VR128:$src1, VR128:$src2),
7073 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7074 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7075 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7077 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7078 (ins VR128:$src1, i128mem:$src2),
7080 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7081 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7083 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7086 // Perform One Round of an AES Encryption/Decryption Flow
7087 let Predicates = [HasAVX, HasAES] in {
7088 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7089 int_x86_aesni_aesenc, 0>, VEX_4V;
7090 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7091 int_x86_aesni_aesenclast, 0>, VEX_4V;
7092 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7093 int_x86_aesni_aesdec, 0>, VEX_4V;
7094 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7095 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7098 let Constraints = "$src1 = $dst" in {
7099 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7100 int_x86_aesni_aesenc>;
7101 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7102 int_x86_aesni_aesenclast>;
7103 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7104 int_x86_aesni_aesdec>;
7105 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7106 int_x86_aesni_aesdeclast>;
7109 // Perform the AES InvMixColumn Transformation
7110 let Predicates = [HasAVX, HasAES] in {
7111 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7113 "vaesimc\t{$src1, $dst|$dst, $src1}",
7115 (int_x86_aesni_aesimc VR128:$src1))]>,
7117 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7118 (ins i128mem:$src1),
7119 "vaesimc\t{$src1, $dst|$dst, $src1}",
7120 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7123 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7125 "aesimc\t{$src1, $dst|$dst, $src1}",
7127 (int_x86_aesni_aesimc VR128:$src1))]>,
7129 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7130 (ins i128mem:$src1),
7131 "aesimc\t{$src1, $dst|$dst, $src1}",
7132 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7135 // AES Round Key Generation Assist
7136 let Predicates = [HasAVX, HasAES] in {
7137 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7138 (ins VR128:$src1, i8imm:$src2),
7139 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7141 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7143 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7144 (ins i128mem:$src1, i8imm:$src2),
7145 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7147 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7150 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7151 (ins VR128:$src1, i8imm:$src2),
7152 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7154 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7156 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7157 (ins i128mem:$src1, i8imm:$src2),
7158 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7160 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7163 //===----------------------------------------------------------------------===//
7164 // CLMUL Instructions
7165 //===----------------------------------------------------------------------===//
7167 // Carry-less Multiplication instructions
7168 let neverHasSideEffects = 1 in {
7169 // AVX carry-less Multiplication instructions
7170 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7171 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7172 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7176 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7177 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7178 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7181 let Constraints = "$src1 = $dst" in {
7182 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7183 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7184 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7188 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7189 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7190 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7192 } // Constraints = "$src1 = $dst"
7193 } // neverHasSideEffects = 1
7196 multiclass pclmul_alias<string asm, int immop> {
7197 def : InstAlias<!strconcat("pclmul", asm,
7198 "dq {$src, $dst|$dst, $src}"),
7199 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7201 def : InstAlias<!strconcat("pclmul", asm,
7202 "dq {$src, $dst|$dst, $src}"),
7203 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7205 def : InstAlias<!strconcat("vpclmul", asm,
7206 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7207 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7209 def : InstAlias<!strconcat("vpclmul", asm,
7210 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7211 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7213 defm : pclmul_alias<"hqhq", 0x11>;
7214 defm : pclmul_alias<"hqlq", 0x01>;
7215 defm : pclmul_alias<"lqhq", 0x10>;
7216 defm : pclmul_alias<"lqlq", 0x00>;
7218 //===----------------------------------------------------------------------===//
7220 //===----------------------------------------------------------------------===//
7222 //===----------------------------------------------------------------------===//
7223 // VBROADCAST - Load from memory and broadcast to all elements of the
7224 // destination operand
7226 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7227 X86MemOperand x86memop, Intrinsic Int> :
7228 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7229 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7230 [(set RC:$dst, (Int addr:$src))]>, VEX;
7232 // AVX2 adds register forms
7233 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7235 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7236 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7237 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7239 let ExeDomain = SSEPackedSingle in {
7240 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7241 int_x86_avx_vbroadcast_ss>;
7242 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7243 int_x86_avx_vbroadcast_ss_256>;
7245 let ExeDomain = SSEPackedDouble in
7246 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7247 int_x86_avx_vbroadcast_sd_256>;
7248 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7249 int_x86_avx_vbroadcastf128_pd_256>;
7251 let ExeDomain = SSEPackedSingle in {
7252 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7253 int_x86_avx2_vbroadcast_ss_ps>;
7254 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7255 int_x86_avx2_vbroadcast_ss_ps_256>;
7257 let ExeDomain = SSEPackedDouble in
7258 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7259 int_x86_avx2_vbroadcast_sd_pd_256>;
7261 let Predicates = [HasAVX2] in
7262 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7263 int_x86_avx2_vbroadcasti128>;
7265 let Predicates = [HasAVX] in
7266 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7267 (VBROADCASTF128 addr:$src)>;
7270 //===----------------------------------------------------------------------===//
7271 // VINSERTF128 - Insert packed floating-point values
7273 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7274 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7275 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7276 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7279 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7280 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7281 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7285 let Predicates = [HasAVX] in {
7286 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7287 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7288 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7289 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7290 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7291 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7294 //===----------------------------------------------------------------------===//
7295 // VEXTRACTF128 - Extract packed floating-point values
7297 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7298 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7299 (ins VR256:$src1, i8imm:$src2),
7300 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7303 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7304 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7305 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7309 let Predicates = [HasAVX] in {
7310 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7311 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7312 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7313 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7314 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7315 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7318 //===----------------------------------------------------------------------===//
7319 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7321 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7322 Intrinsic IntLd, Intrinsic IntLd256,
7323 Intrinsic IntSt, Intrinsic IntSt256> {
7324 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7325 (ins VR128:$src1, f128mem:$src2),
7326 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7327 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7329 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7330 (ins VR256:$src1, f256mem:$src2),
7331 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7332 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7334 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7335 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7336 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7337 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7338 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7339 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7340 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7341 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7344 let ExeDomain = SSEPackedSingle in
7345 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7346 int_x86_avx_maskload_ps,
7347 int_x86_avx_maskload_ps_256,
7348 int_x86_avx_maskstore_ps,
7349 int_x86_avx_maskstore_ps_256>;
7350 let ExeDomain = SSEPackedDouble in
7351 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7352 int_x86_avx_maskload_pd,
7353 int_x86_avx_maskload_pd_256,
7354 int_x86_avx_maskstore_pd,
7355 int_x86_avx_maskstore_pd_256>;
7357 //===----------------------------------------------------------------------===//
7358 // VPERMIL - Permute Single and Double Floating-Point Values
7360 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7361 RegisterClass RC, X86MemOperand x86memop_f,
7362 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7363 Intrinsic IntVar, Intrinsic IntImm> {
7364 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7365 (ins RC:$src1, RC:$src2),
7366 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7367 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7368 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7369 (ins RC:$src1, x86memop_i:$src2),
7370 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7371 [(set RC:$dst, (IntVar RC:$src1,
7372 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7374 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7375 (ins RC:$src1, i8imm:$src2),
7376 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7377 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7378 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7379 (ins x86memop_f:$src1, i8imm:$src2),
7380 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7381 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7384 let ExeDomain = SSEPackedSingle in {
7385 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7386 memopv4f32, memopv2i64,
7387 int_x86_avx_vpermilvar_ps,
7388 int_x86_avx_vpermil_ps>;
7389 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7390 memopv8f32, memopv4i64,
7391 int_x86_avx_vpermilvar_ps_256,
7392 int_x86_avx_vpermil_ps_256>;
7394 let ExeDomain = SSEPackedDouble in {
7395 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7396 memopv2f64, memopv2i64,
7397 int_x86_avx_vpermilvar_pd,
7398 int_x86_avx_vpermil_pd>;
7399 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7400 memopv4f64, memopv4i64,
7401 int_x86_avx_vpermilvar_pd_256,
7402 int_x86_avx_vpermil_pd_256>;
7405 let Predicates = [HasAVX] in {
7406 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7407 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7408 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7409 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7410 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7411 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7412 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7413 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7414 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7415 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7416 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7417 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7418 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7420 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7421 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7422 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7425 //===----------------------------------------------------------------------===//
7426 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7428 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7429 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7430 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7431 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7434 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7435 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7436 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7440 let Predicates = [HasAVX] in {
7441 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7442 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7443 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7444 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7445 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7446 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7448 def : Pat<(int_x86_avx_vperm2f128_ps_256
7449 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7450 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7451 def : Pat<(int_x86_avx_vperm2f128_pd_256
7452 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7453 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7454 def : Pat<(int_x86_avx_vperm2f128_si_256
7455 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7456 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7459 //===----------------------------------------------------------------------===//
7460 // VZERO - Zero YMM registers
7462 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7463 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7464 // Zero All YMM registers
7465 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7466 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7468 // Zero Upper bits of YMM registers
7469 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7470 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7473 //===----------------------------------------------------------------------===//
7474 // Half precision conversion instructions
7475 //===----------------------------------------------------------------------===//
7476 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7477 let Predicates = [HasAVX, HasF16C] in {
7478 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7479 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7480 [(set RC:$dst, (Int VR128:$src))]>,
7482 let neverHasSideEffects = 1, mayLoad = 1 in
7483 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7484 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7488 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7489 let Predicates = [HasAVX, HasF16C] in {
7490 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7491 (ins RC:$src1, i32i8imm:$src2),
7492 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7493 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7495 let neverHasSideEffects = 1, mayLoad = 1 in
7496 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7497 (ins RC:$src1, i32i8imm:$src2),
7498 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7503 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7504 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7505 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7506 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7508 //===----------------------------------------------------------------------===//
7509 // AVX2 Instructions
7510 //===----------------------------------------------------------------------===//
7512 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7513 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7514 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7515 X86MemOperand x86memop> {
7516 let isCommutable = 1 in
7517 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7518 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7519 !strconcat(OpcodeStr,
7520 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7521 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7523 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7524 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7525 !strconcat(OpcodeStr,
7526 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7529 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7533 let isCommutable = 0 in {
7534 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7535 VR128, memopv2i64, i128mem>;
7536 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7537 VR256, memopv4i64, i256mem>;
7540 //===----------------------------------------------------------------------===//
7541 // VPBROADCAST - Load from memory and broadcast to all elements of the
7542 // destination operand
7544 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7545 X86MemOperand x86memop, PatFrag ld_frag,
7546 Intrinsic Int128, Intrinsic Int256> {
7547 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7548 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7549 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7550 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7551 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7553 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7554 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7555 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7556 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7557 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7558 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7560 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7563 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7564 int_x86_avx2_pbroadcastb_128,
7565 int_x86_avx2_pbroadcastb_256>;
7566 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7567 int_x86_avx2_pbroadcastw_128,
7568 int_x86_avx2_pbroadcastw_256>;
7569 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7570 int_x86_avx2_pbroadcastd_128,
7571 int_x86_avx2_pbroadcastd_256>;
7572 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7573 int_x86_avx2_pbroadcastq_128,
7574 int_x86_avx2_pbroadcastq_256>;
7576 let Predicates = [HasAVX2] in {
7577 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7578 (VPBROADCASTBrm addr:$src)>;
7579 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7580 (VPBROADCASTBYrm addr:$src)>;
7581 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7582 (VPBROADCASTWrm addr:$src)>;
7583 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7584 (VPBROADCASTWYrm addr:$src)>;
7585 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7586 (VPBROADCASTDrm addr:$src)>;
7587 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7588 (VPBROADCASTDYrm addr:$src)>;
7589 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7590 (VPBROADCASTQrm addr:$src)>;
7591 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7592 (VPBROADCASTQYrm addr:$src)>;
7595 // AVX1 broadcast patterns
7596 let Predicates = [HasAVX] in {
7597 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7598 (VBROADCASTSSYrm addr:$src)>;
7599 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7600 (VBROADCASTSDrm addr:$src)>;
7601 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7602 (VBROADCASTSSYrm addr:$src)>;
7603 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7604 (VBROADCASTSDrm addr:$src)>;
7606 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7607 (VBROADCASTSSrm addr:$src)>;
7608 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7609 (VBROADCASTSSrm addr:$src)>;
7612 //===----------------------------------------------------------------------===//
7613 // VPERM - Permute instructions
7616 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7618 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7619 (ins VR256:$src1, VR256:$src2),
7620 !strconcat(OpcodeStr,
7621 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7622 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7623 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7624 (ins VR256:$src1, i256mem:$src2),
7625 !strconcat(OpcodeStr,
7626 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7627 [(set VR256:$dst, (Int VR256:$src1,
7628 (bitconvert (mem_frag addr:$src2))))]>,
7632 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7633 let ExeDomain = SSEPackedSingle in
7634 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7636 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7638 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7639 (ins VR256:$src1, i8imm:$src2),
7640 !strconcat(OpcodeStr,
7641 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7642 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7643 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7644 (ins i256mem:$src1, i8imm:$src2),
7645 !strconcat(OpcodeStr,
7646 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7647 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7651 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7653 let ExeDomain = SSEPackedDouble in
7654 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7657 //===----------------------------------------------------------------------===//
7658 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7660 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7661 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7662 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7664 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7666 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7667 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7668 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7670 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7674 let Predicates = [HasAVX2] in {
7675 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7676 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7677 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7678 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7679 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7680 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7681 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7682 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7684 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7686 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7687 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7688 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7689 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7690 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7692 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7693 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7695 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7699 let Predicates = [HasAVX] in {
7700 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7701 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7702 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7703 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7704 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7705 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7706 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7707 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7708 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7709 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7710 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7711 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7713 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7714 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7715 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7716 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7717 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7718 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7719 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7720 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7721 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7722 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7723 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7724 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7725 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7726 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7727 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7728 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7729 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7730 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7734 //===----------------------------------------------------------------------===//
7735 // VINSERTI128 - Insert packed integer values
7737 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7738 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7739 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7741 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7743 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7744 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7745 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7747 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7748 imm:$src3))]>, VEX_4V;
7750 let Predicates = [HasAVX2] in {
7751 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7753 (VINSERTI128rr VR256:$src1, VR128:$src2,
7754 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7755 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7757 (VINSERTI128rr VR256:$src1, VR128:$src2,
7758 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7759 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7761 (VINSERTI128rr VR256:$src1, VR128:$src2,
7762 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7763 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7765 (VINSERTI128rr VR256:$src1, VR128:$src2,
7766 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7770 let Predicates = [HasAVX] in {
7771 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7773 (VINSERTF128rr VR256:$src1, VR128:$src2,
7774 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7775 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7777 (VINSERTF128rr VR256:$src1, VR128:$src2,
7778 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7779 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7781 (VINSERTF128rr VR256:$src1, VR128:$src2,
7782 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7783 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7785 (VINSERTF128rr VR256:$src1, VR128:$src2,
7786 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7787 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7789 (VINSERTF128rr VR256:$src1, VR128:$src2,
7790 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7791 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7793 (VINSERTF128rr VR256:$src1, VR128:$src2,
7794 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7797 //===----------------------------------------------------------------------===//
7798 // VEXTRACTI128 - Extract packed integer values
7800 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7801 (ins VR256:$src1, i8imm:$src2),
7802 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7804 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7806 let neverHasSideEffects = 1, mayStore = 1 in
7807 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7808 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7809 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7811 let Predicates = [HasAVX2] in {
7812 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7813 (v2i64 (VEXTRACTI128rr
7814 (v4i64 VR256:$src1),
7815 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7816 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7817 (v4i32 (VEXTRACTI128rr
7818 (v8i32 VR256:$src1),
7819 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7820 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7821 (v8i16 (VEXTRACTI128rr
7822 (v16i16 VR256:$src1),
7823 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7824 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7825 (v16i8 (VEXTRACTI128rr
7826 (v32i8 VR256:$src1),
7827 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7831 let Predicates = [HasAVX] in {
7832 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7833 (v4f32 (VEXTRACTF128rr
7834 (v8f32 VR256:$src1),
7835 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7836 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7837 (v2f64 (VEXTRACTF128rr
7838 (v4f64 VR256:$src1),
7839 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7840 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7841 (v2i64 (VEXTRACTF128rr
7842 (v4i64 VR256:$src1),
7843 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7844 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7845 (v4i32 (VEXTRACTF128rr
7846 (v8i32 VR256:$src1),
7847 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7848 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7849 (v8i16 (VEXTRACTF128rr
7850 (v16i16 VR256:$src1),
7851 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7852 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7853 (v16i8 (VEXTRACTF128rr
7854 (v32i8 VR256:$src1),
7855 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7858 //===----------------------------------------------------------------------===//
7859 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7861 multiclass avx2_pmovmask<string OpcodeStr,
7862 Intrinsic IntLd128, Intrinsic IntLd256,
7863 Intrinsic IntSt128, Intrinsic IntSt256> {
7864 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7865 (ins VR128:$src1, i128mem:$src2),
7866 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7867 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7868 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7869 (ins VR256:$src1, i256mem:$src2),
7870 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7871 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7872 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7873 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7874 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7875 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7876 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7877 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7878 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7879 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7882 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7883 int_x86_avx2_maskload_d,
7884 int_x86_avx2_maskload_d_256,
7885 int_x86_avx2_maskstore_d,
7886 int_x86_avx2_maskstore_d_256>;
7887 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7888 int_x86_avx2_maskload_q,
7889 int_x86_avx2_maskload_q_256,
7890 int_x86_avx2_maskstore_q,
7891 int_x86_avx2_maskstore_q_256>, VEX_W;
7894 //===----------------------------------------------------------------------===//
7895 // Variable Bit Shifts
7897 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7898 ValueType vt128, ValueType vt256> {
7899 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7900 (ins VR128:$src1, VR128:$src2),
7901 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7903 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7905 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7906 (ins VR128:$src1, i128mem:$src2),
7907 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7909 (vt128 (OpNode VR128:$src1,
7910 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7912 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7913 (ins VR256:$src1, VR256:$src2),
7914 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7916 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7918 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7919 (ins VR256:$src1, i256mem:$src2),
7920 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7922 (vt256 (OpNode VR256:$src1,
7923 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7927 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7928 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7929 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7930 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7931 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;