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:$src2), sub_ss))>;
555 def : Pat<(v8i32 (X86Movsd VR256:$src1, VR256:$src2)),
556 (SUBREG_TO_REG (i32 0),
557 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
558 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
559 def : Pat<(v8f32 (X86Movsd VR256:$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:$src)))]>,
1865 XD, Requires<[HasSSE2]>;
1868 // Convert with truncation packed single/double fp to doubleword
1869 // SSE2 packed instructions with XS prefix
1870 let neverHasSideEffects = 1 in {
1871 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1872 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1874 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1875 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1876 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1877 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1879 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1880 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1881 } // neverHasSideEffects = 1
1883 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1884 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1886 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1887 XS, VEX, Requires<[HasAVX]>;
1888 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1889 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1890 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1891 (memop addr:$src)))]>,
1892 XS, VEX, Requires<[HasAVX]>;
1894 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1895 "cvttps2dq\t{$src, $dst|$dst, $src}",
1897 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1898 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1899 "cvttps2dq\t{$src, $dst|$dst, $src}",
1901 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1903 let Predicates = [HasAVX] in {
1904 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1905 (Int_VCVTDQ2PSrr VR128:$src)>;
1906 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1907 (VCVTTPS2DQrr VR128:$src)>;
1908 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1909 (VCVTDQ2PSYrr VR256:$src)>;
1910 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1911 (VCVTTPS2DQYrr VR256:$src)>;
1914 let Predicates = [HasSSE2] in {
1915 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1916 (Int_CVTDQ2PSrr VR128:$src)>;
1917 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1918 (CVTTPS2DQrr VR128:$src)>;
1921 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1922 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1924 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1925 let isCodeGenOnly = 1 in
1926 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1927 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1928 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1929 (memop addr:$src)))]>, VEX;
1930 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1931 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1932 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1933 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1934 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1935 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1936 (memop addr:$src)))]>;
1938 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1939 // register, but the same isn't true when using memory operands instead.
1940 // Provide other assembly rr and rm forms to address this explicitly.
1941 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1942 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1945 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1946 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1947 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1948 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1951 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1952 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1953 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1954 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1956 // Convert packed single to packed double
1957 let Predicates = [HasAVX] in {
1958 // SSE2 instructions without OpSize prefix
1959 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1960 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1961 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1962 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1963 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1964 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1965 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1966 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1968 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1969 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1970 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1971 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1973 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1974 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1975 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1976 TB, VEX, Requires<[HasAVX]>;
1977 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1978 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1979 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1980 (load addr:$src)))]>,
1981 TB, VEX, Requires<[HasAVX]>;
1982 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1983 "cvtps2pd\t{$src, $dst|$dst, $src}",
1984 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1985 TB, Requires<[HasSSE2]>;
1986 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1987 "cvtps2pd\t{$src, $dst|$dst, $src}",
1988 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1989 (load addr:$src)))]>,
1990 TB, Requires<[HasSSE2]>;
1992 // Convert packed double to packed single
1993 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1994 // register, but the same isn't true when using memory operands instead.
1995 // Provide other assembly rr and rm forms to address this explicitly.
1996 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1997 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1998 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1999 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
2002 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2003 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
2004 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2005 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
2008 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2009 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
2010 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2011 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
2012 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2013 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
2014 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2015 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
2018 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2019 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2020 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2021 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
2023 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2024 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2025 (memop addr:$src)))]>;
2026 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2027 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2028 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2029 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2030 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2031 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2032 (memop addr:$src)))]>;
2034 // AVX 256-bit register conversion intrinsics
2035 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2036 // whenever possible to avoid declaring two versions of each one.
2037 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2038 (VCVTDQ2PSYrr VR256:$src)>;
2039 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2040 (VCVTDQ2PSYrm addr:$src)>;
2042 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2043 (VCVTPD2PSYrr VR256:$src)>;
2044 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2045 (VCVTPD2PSYrm addr:$src)>;
2047 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2048 (VCVTPS2DQYrr VR256:$src)>;
2049 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2050 (VCVTPS2DQYrm addr:$src)>;
2052 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2053 (VCVTPS2PDYrr VR128:$src)>;
2054 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2055 (VCVTPS2PDYrm addr:$src)>;
2057 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2058 (VCVTTPD2DQYrr VR256:$src)>;
2059 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2060 (VCVTTPD2DQYrm addr:$src)>;
2062 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
2063 (VCVTTPS2DQYrr VR256:$src)>;
2064 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
2065 (VCVTTPS2DQYrm addr:$src)>;
2067 // Match fround and fextend for 128/256-bit conversions
2068 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2069 (VCVTPD2PSYrr VR256:$src)>;
2070 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2071 (VCVTPD2PSYrm addr:$src)>;
2073 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2074 (VCVTPS2PDYrr VR128:$src)>;
2075 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2076 (VCVTPS2PDYrm addr:$src)>;
2078 //===----------------------------------------------------------------------===//
2079 // SSE 1 & 2 - Compare Instructions
2080 //===----------------------------------------------------------------------===//
2082 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2083 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2084 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2085 string asm, string asm_alt> {
2086 def rr : SIi8<0xC2, MRMSrcReg,
2087 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2088 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2089 def rm : SIi8<0xC2, MRMSrcMem,
2090 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2091 [(set RC:$dst, (OpNode (VT RC:$src1),
2092 (ld_frag addr:$src2), imm:$cc))]>;
2094 // Accept explicit immediate argument form instead of comparison code.
2095 let neverHasSideEffects = 1 in {
2096 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2097 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2099 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2100 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2104 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2105 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2106 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2107 XS, VEX_4V, VEX_LIG;
2108 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2109 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2110 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2111 XD, VEX_4V, VEX_LIG;
2113 let Constraints = "$src1 = $dst" in {
2114 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2115 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2116 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2118 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2119 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2120 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2124 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2125 Intrinsic Int, string asm> {
2126 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2127 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2128 [(set VR128:$dst, (Int VR128:$src1,
2129 VR128:$src, imm:$cc))]>;
2130 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2131 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
2132 [(set VR128:$dst, (Int VR128:$src1,
2133 (load addr:$src), imm:$cc))]>;
2136 // Aliases to match intrinsics which expect XMM operand(s).
2137 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2138 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2140 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2141 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2143 let Constraints = "$src1 = $dst" in {
2144 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2145 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2146 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2147 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2151 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2152 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2153 ValueType vt, X86MemOperand x86memop,
2154 PatFrag ld_frag, string OpcodeStr, Domain d> {
2155 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2156 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2157 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2158 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2159 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2160 [(set EFLAGS, (OpNode (vt RC:$src1),
2161 (ld_frag addr:$src2)))], d>;
2164 let Defs = [EFLAGS] in {
2165 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2166 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2167 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2168 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2170 let Pattern = []<dag> in {
2171 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2172 "comiss", SSEPackedSingle>, TB, VEX,
2174 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2175 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2179 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2180 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2181 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2182 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2184 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2185 load, "comiss", SSEPackedSingle>, TB, VEX;
2186 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2187 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2188 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2189 "ucomiss", SSEPackedSingle>, TB;
2190 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2191 "ucomisd", SSEPackedDouble>, TB, OpSize;
2193 let Pattern = []<dag> in {
2194 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2195 "comiss", SSEPackedSingle>, TB;
2196 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2197 "comisd", SSEPackedDouble>, TB, OpSize;
2200 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2201 load, "ucomiss", SSEPackedSingle>, TB;
2202 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2203 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2205 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2206 "comiss", SSEPackedSingle>, TB;
2207 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2208 "comisd", SSEPackedDouble>, TB, OpSize;
2209 } // Defs = [EFLAGS]
2211 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2212 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2213 Intrinsic Int, string asm, string asm_alt,
2215 let isAsmParserOnly = 1 in {
2216 def rri : PIi8<0xC2, MRMSrcReg,
2217 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2218 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2219 def rmi : PIi8<0xC2, MRMSrcMem,
2220 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, SSECC:$cc), asm,
2221 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2224 // Accept explicit immediate argument form instead of comparison code.
2225 def rri_alt : PIi8<0xC2, MRMSrcReg,
2226 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2228 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2229 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, i8imm:$cc),
2233 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2234 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2235 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2236 SSEPackedSingle>, TB, VEX_4V;
2237 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2238 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2239 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2240 SSEPackedDouble>, TB, OpSize, VEX_4V;
2241 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2242 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2243 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2244 SSEPackedSingle>, TB, VEX_4V;
2245 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2246 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2247 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2248 SSEPackedDouble>, TB, OpSize, VEX_4V;
2249 let Constraints = "$src1 = $dst" in {
2250 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2251 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2252 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2253 SSEPackedSingle>, TB;
2254 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2255 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2256 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2257 SSEPackedDouble>, TB, OpSize;
2260 let Predicates = [HasAVX] in {
2261 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2262 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2263 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2264 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2265 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2266 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2267 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2268 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2270 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2271 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2272 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2273 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2274 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2275 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2276 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2277 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2280 let Predicates = [HasSSE1] in {
2281 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2282 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2283 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2284 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2287 let Predicates = [HasSSE2] in {
2288 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2289 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2290 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2291 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2294 //===----------------------------------------------------------------------===//
2295 // SSE 1 & 2 - Shuffle Instructions
2296 //===----------------------------------------------------------------------===//
2298 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2299 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2300 ValueType vt, string asm, PatFrag mem_frag,
2301 Domain d, bit IsConvertibleToThreeAddress = 0> {
2302 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2303 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
2304 [(set RC:$dst, (vt (shufp:$src3
2305 RC:$src1, (mem_frag addr:$src2))))], d>;
2306 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2307 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2308 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2310 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2313 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2314 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2315 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2316 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2317 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2318 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2319 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2320 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2321 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2322 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2323 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2324 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2326 let Constraints = "$src1 = $dst" in {
2327 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2328 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2329 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2331 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2332 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2333 memopv2f64, SSEPackedDouble>, TB, OpSize;
2336 let Predicates = [HasAVX] in {
2337 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2338 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2339 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2340 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2341 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2342 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2343 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2344 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2345 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2346 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2347 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2348 // fall back to this for SSE1)
2349 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2350 (VSHUFPSrri VR128:$src2, VR128:$src1,
2351 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2352 // Special unary SHUFPSrri case.
2353 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2354 (VSHUFPSrri VR128:$src1, VR128:$src1,
2355 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2356 // Special binary v4i32 shuffle cases with SHUFPS.
2357 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2358 (VSHUFPSrri VR128:$src1, VR128:$src2,
2359 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2360 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2361 (bc_v4i32 (memopv2i64 addr:$src2)))),
2362 (VSHUFPSrmi VR128:$src1, addr:$src2,
2363 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2364 // Special unary SHUFPDrri cases.
2365 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2366 (VSHUFPDrri VR128:$src1, VR128:$src1,
2367 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2368 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2369 (VSHUFPDrri VR128:$src1, VR128:$src1,
2370 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2371 // Special binary v2i64 shuffle cases using SHUFPDrri.
2372 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2373 (VSHUFPDrri VR128:$src1, VR128:$src2,
2374 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2376 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2377 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2378 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2379 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2380 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2381 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2382 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2383 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2384 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2385 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2388 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2389 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2390 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2391 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2392 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2394 def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2395 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2396 def : Pat<(v8f32 (X86Shufp VR256:$src1,
2397 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2398 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2400 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2401 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2402 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2403 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2404 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2406 def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2407 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2408 def : Pat<(v4f64 (X86Shufp VR256:$src1,
2409 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2410 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2413 let Predicates = [HasSSE1] in {
2414 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2415 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2416 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2417 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2418 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2419 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2420 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2421 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2422 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2423 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2424 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2425 // fall back to this for SSE1)
2426 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2427 (SHUFPSrri VR128:$src2, VR128:$src1,
2428 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2429 // Special unary SHUFPSrri case.
2430 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2431 (SHUFPSrri VR128:$src1, VR128:$src1,
2432 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2435 let Predicates = [HasSSE2] in {
2436 // Special binary v4i32 shuffle cases with SHUFPS.
2437 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2438 (SHUFPSrri VR128:$src1, VR128:$src2,
2439 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2440 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2441 (bc_v4i32 (memopv2i64 addr:$src2)))),
2442 (SHUFPSrmi VR128:$src1, addr:$src2,
2443 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2444 // Special unary SHUFPDrri cases.
2445 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2446 (SHUFPDrri VR128:$src1, VR128:$src1,
2447 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2448 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2449 (SHUFPDrri VR128:$src1, VR128:$src1,
2450 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2451 // Special binary v2i64 shuffle cases using SHUFPDrri.
2452 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2453 (SHUFPDrri VR128:$src1, VR128:$src2,
2454 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2455 // Generic SHUFPD patterns
2456 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2457 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2458 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2459 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2460 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2461 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2462 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2463 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2464 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2465 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2468 //===----------------------------------------------------------------------===//
2469 // SSE 1 & 2 - Unpack Instructions
2470 //===----------------------------------------------------------------------===//
2472 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2473 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2474 PatFrag mem_frag, RegisterClass RC,
2475 X86MemOperand x86memop, string asm,
2477 def rr : PI<opc, MRMSrcReg,
2478 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2480 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2481 def rm : PI<opc, MRMSrcMem,
2482 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2484 (vt (OpNode RC:$src1,
2485 (mem_frag addr:$src2))))], d>;
2488 let AddedComplexity = 10 in {
2489 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2490 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2491 SSEPackedSingle>, TB, VEX_4V;
2492 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2493 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2494 SSEPackedDouble>, TB, OpSize, VEX_4V;
2495 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2496 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2497 SSEPackedSingle>, TB, VEX_4V;
2498 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2499 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2500 SSEPackedDouble>, TB, OpSize, VEX_4V;
2502 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2503 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2504 SSEPackedSingle>, TB, VEX_4V;
2505 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2506 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2507 SSEPackedDouble>, TB, OpSize, VEX_4V;
2508 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2509 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2510 SSEPackedSingle>, TB, VEX_4V;
2511 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2512 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2513 SSEPackedDouble>, TB, OpSize, VEX_4V;
2515 let Constraints = "$src1 = $dst" in {
2516 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2517 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2518 SSEPackedSingle>, TB;
2519 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2520 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2521 SSEPackedDouble>, TB, OpSize;
2522 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2523 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2524 SSEPackedSingle>, TB;
2525 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2526 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2527 SSEPackedDouble>, TB, OpSize;
2528 } // Constraints = "$src1 = $dst"
2529 } // AddedComplexity
2531 let Predicates = [HasSSE1] in {
2532 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2533 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2534 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2535 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2536 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2537 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2538 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2539 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2542 let Predicates = [HasSSE2] in {
2543 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2544 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2545 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2546 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2547 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2548 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2549 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2550 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2552 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2553 // problem is during lowering, where it's not possible to recognize the load
2554 // fold cause it has two uses through a bitcast. One use disappears at isel
2555 // time and the fold opportunity reappears.
2556 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2557 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2559 let AddedComplexity = 10 in
2560 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2561 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2564 let Predicates = [HasAVX] in {
2565 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2566 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2567 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2568 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2569 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2570 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2571 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2572 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2574 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2575 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2576 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2577 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2578 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2579 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2580 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2581 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2583 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2584 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2585 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2586 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2587 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2588 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2589 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2590 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2592 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2593 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2594 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2595 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2596 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2597 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2598 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2599 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2601 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2602 // problem is during lowering, where it's not possible to recognize the load
2603 // fold cause it has two uses through a bitcast. One use disappears at isel
2604 // time and the fold opportunity reappears.
2605 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2606 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2607 let AddedComplexity = 10 in
2608 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2609 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2612 //===----------------------------------------------------------------------===//
2613 // SSE 1 & 2 - Extract Floating-Point Sign mask
2614 //===----------------------------------------------------------------------===//
2616 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2617 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2619 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2620 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2621 [(set GR32:$dst, (Int RC:$src))], d>;
2622 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2623 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2626 let Predicates = [HasAVX] in {
2627 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2628 "movmskps", SSEPackedSingle>, TB, VEX;
2629 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2630 "movmskpd", SSEPackedDouble>, TB,
2632 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2633 "movmskps", SSEPackedSingle>, TB, VEX;
2634 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2635 "movmskpd", SSEPackedDouble>, TB,
2638 def : Pat<(i32 (X86fgetsign FR32:$src)),
2639 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2641 def : Pat<(i64 (X86fgetsign FR32:$src)),
2642 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2644 def : Pat<(i32 (X86fgetsign FR64:$src)),
2645 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2647 def : Pat<(i64 (X86fgetsign FR64:$src)),
2648 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2652 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2653 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2654 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2655 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2657 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2658 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2659 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2660 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2664 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2665 SSEPackedSingle>, TB;
2666 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2667 SSEPackedDouble>, TB, OpSize;
2669 def : Pat<(i32 (X86fgetsign FR32:$src)),
2670 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2671 sub_ss))>, Requires<[HasSSE1]>;
2672 def : Pat<(i64 (X86fgetsign FR32:$src)),
2673 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2674 sub_ss))>, Requires<[HasSSE1]>;
2675 def : Pat<(i32 (X86fgetsign FR64:$src)),
2676 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2677 sub_sd))>, Requires<[HasSSE2]>;
2678 def : Pat<(i64 (X86fgetsign FR64:$src)),
2679 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2680 sub_sd))>, Requires<[HasSSE2]>;
2682 //===---------------------------------------------------------------------===//
2683 // SSE2 - Packed Integer Logical Instructions
2684 //===---------------------------------------------------------------------===//
2686 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2688 /// PDI_binop_rm - Simple SSE2 binary operator.
2689 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2690 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2691 X86MemOperand x86memop, bit IsCommutable = 0,
2693 let isCommutable = IsCommutable in
2694 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2695 (ins RC:$src1, RC:$src2),
2697 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2698 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2699 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2700 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2701 (ins RC:$src1, x86memop:$src2),
2703 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2704 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2705 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2706 (bitconvert (memop_frag addr:$src2)))))]>;
2708 } // ExeDomain = SSEPackedInt
2710 // These are ordered here for pattern ordering requirements with the fp versions
2712 let Predicates = [HasAVX] in {
2713 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2714 i128mem, 1, 0>, VEX_4V;
2715 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2716 i128mem, 1, 0>, VEX_4V;
2717 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2718 i128mem, 1, 0>, VEX_4V;
2719 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2720 i128mem, 0, 0>, VEX_4V;
2723 let Constraints = "$src1 = $dst" in {
2724 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2726 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2728 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2730 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2732 } // Constraints = "$src1 = $dst"
2734 let Predicates = [HasAVX2] in {
2735 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2736 i256mem, 1, 0>, VEX_4V;
2737 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2738 i256mem, 1, 0>, VEX_4V;
2739 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2740 i256mem, 1, 0>, VEX_4V;
2741 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2742 i256mem, 0, 0>, VEX_4V;
2745 //===----------------------------------------------------------------------===//
2746 // SSE 1 & 2 - Logical Instructions
2747 //===----------------------------------------------------------------------===//
2749 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2751 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2753 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2754 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2756 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2757 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2759 let Constraints = "$src1 = $dst" in {
2760 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2761 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2763 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2764 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2768 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2769 let mayLoad = 0 in {
2770 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2771 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2772 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2775 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2776 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2778 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2780 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2782 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2783 // are all promoted to v2i64, and the patterns are covered by the int
2784 // version. This is needed in SSE only, because v2i64 isn't supported on
2785 // SSE1, but only on SSE2.
2786 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2787 !strconcat(OpcodeStr, "ps"), f128mem, [],
2788 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2789 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2791 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2792 !strconcat(OpcodeStr, "pd"), f128mem,
2793 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2794 (bc_v2i64 (v2f64 VR128:$src2))))],
2795 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2796 (memopv2i64 addr:$src2)))], 0>,
2798 let Constraints = "$src1 = $dst" in {
2799 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2800 !strconcat(OpcodeStr, "ps"), f128mem,
2801 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2802 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2803 (memopv2i64 addr:$src2)))]>, TB;
2805 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2806 !strconcat(OpcodeStr, "pd"), f128mem,
2807 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2808 (bc_v2i64 (v2f64 VR128:$src2))))],
2809 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2810 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2814 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2816 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2818 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2819 !strconcat(OpcodeStr, "ps"), f256mem,
2820 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2821 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2822 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2824 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2825 !strconcat(OpcodeStr, "pd"), f256mem,
2826 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2827 (bc_v4i64 (v4f64 VR256:$src2))))],
2828 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2829 (memopv4i64 addr:$src2)))], 0>,
2833 // AVX 256-bit packed logical ops forms
2834 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2835 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2836 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2837 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2839 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2840 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2841 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2842 let isCommutable = 0 in
2843 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2845 //===----------------------------------------------------------------------===//
2846 // SSE 1 & 2 - Arithmetic Instructions
2847 //===----------------------------------------------------------------------===//
2849 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2852 /// In addition, we also have a special variant of the scalar form here to
2853 /// represent the associated intrinsic operation. This form is unlike the
2854 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2855 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2857 /// These three forms can each be reg+reg or reg+mem.
2860 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2862 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2864 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2865 OpNode, FR32, f32mem, Is2Addr>, XS;
2866 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2867 OpNode, FR64, f64mem, Is2Addr>, XD;
2870 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2872 let mayLoad = 0 in {
2873 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2874 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2875 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2876 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2880 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2882 let mayLoad = 0 in {
2883 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2884 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2885 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2886 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2890 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2892 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2893 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2894 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2895 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2898 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2900 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2901 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2902 SSEPackedSingle, Is2Addr>, TB;
2904 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2905 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2906 SSEPackedDouble, Is2Addr>, TB, OpSize;
2909 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2910 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2911 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2912 SSEPackedSingle, 0>, TB;
2914 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2915 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2916 SSEPackedDouble, 0>, TB, OpSize;
2919 // Binary Arithmetic instructions
2920 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2921 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2922 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2923 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2924 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2925 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2926 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2927 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2929 let isCommutable = 0 in {
2930 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2931 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2932 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2933 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2934 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2935 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2936 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2937 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2938 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2939 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2940 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2941 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2942 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2943 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2944 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2945 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2946 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2947 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2948 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2949 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2952 let Constraints = "$src1 = $dst" in {
2953 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2954 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2955 basic_sse12_fp_binop_s_int<0x58, "add">;
2956 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2957 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2958 basic_sse12_fp_binop_s_int<0x59, "mul">;
2960 let isCommutable = 0 in {
2961 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2962 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2963 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2964 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2965 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2966 basic_sse12_fp_binop_s_int<0x5E, "div">;
2967 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2968 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2969 basic_sse12_fp_binop_s_int<0x5F, "max">,
2970 basic_sse12_fp_binop_p_int<0x5F, "max">;
2971 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2972 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2973 basic_sse12_fp_binop_s_int<0x5D, "min">,
2974 basic_sse12_fp_binop_p_int<0x5D, "min">;
2979 /// In addition, we also have a special variant of the scalar form here to
2980 /// represent the associated intrinsic operation. This form is unlike the
2981 /// plain scalar form, in that it takes an entire vector (instead of a
2982 /// scalar) and leaves the top elements undefined.
2984 /// And, we have a special variant form for a full-vector intrinsic form.
2986 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2987 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2988 SDNode OpNode, Intrinsic F32Int> {
2989 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2990 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2991 [(set FR32:$dst, (OpNode FR32:$src))]>;
2992 // For scalar unary operations, fold a load into the operation
2993 // only in OptForSize mode. It eliminates an instruction, but it also
2994 // eliminates a whole-register clobber (the load), so it introduces a
2995 // partial register update condition.
2996 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2997 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2998 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2999 Requires<[HasSSE1, OptForSize]>;
3000 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3001 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3002 [(set VR128:$dst, (F32Int VR128:$src))]>;
3003 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3004 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3005 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
3008 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
3009 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3010 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
3011 !strconcat(OpcodeStr,
3012 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3014 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
3015 !strconcat(OpcodeStr,
3016 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3017 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3018 (ins VR128:$src1, ssmem:$src2),
3019 !strconcat(OpcodeStr,
3020 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3023 /// sse1_fp_unop_p - SSE1 unops in packed form.
3024 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3025 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3026 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3027 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
3028 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3029 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3030 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
3033 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3034 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3035 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3036 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3037 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
3038 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3039 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3040 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
3043 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3044 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3045 Intrinsic V4F32Int> {
3046 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3047 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3048 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
3049 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3050 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3051 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
3054 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3055 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3056 Intrinsic V4F32Int> {
3057 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3058 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3059 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
3060 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3061 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3062 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
3065 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3066 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3067 SDNode OpNode, Intrinsic F64Int> {
3068 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3069 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3070 [(set FR64:$dst, (OpNode FR64:$src))]>;
3071 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3072 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3073 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3074 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
3075 Requires<[HasSSE2, OptForSize]>;
3076 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3077 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3078 [(set VR128:$dst, (F64Int VR128:$src))]>;
3079 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3080 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3081 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
3084 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3085 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3086 let neverHasSideEffects = 1 in {
3087 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3088 !strconcat(OpcodeStr,
3089 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3091 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3092 !strconcat(OpcodeStr,
3093 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3095 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3096 (ins VR128:$src1, sdmem:$src2),
3097 !strconcat(OpcodeStr,
3098 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3101 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3102 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3104 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3105 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3106 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3107 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3108 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3109 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3112 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3113 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3114 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3115 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3116 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3117 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3118 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3119 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3122 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3123 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3124 Intrinsic V2F64Int> {
3125 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3126 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3127 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3128 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3129 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3130 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3133 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3134 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3135 Intrinsic V2F64Int> {
3136 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3137 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3138 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3139 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3140 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3141 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3144 let Predicates = [HasAVX] in {
3146 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3147 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3149 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3150 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3151 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3152 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3153 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3154 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3155 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3156 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3159 // Reciprocal approximations. Note that these typically require refinement
3160 // in order to obtain suitable precision.
3161 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3162 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3163 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3164 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3165 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3167 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3168 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3169 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3170 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3171 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3174 let AddedComplexity = 1 in {
3175 def : Pat<(f32 (fsqrt FR32:$src)),
3176 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3177 def : Pat<(f32 (fsqrt (load addr:$src))),
3178 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3179 Requires<[HasAVX, OptForSize]>;
3180 def : Pat<(f64 (fsqrt FR64:$src)),
3181 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3182 def : Pat<(f64 (fsqrt (load addr:$src))),
3183 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3184 Requires<[HasAVX, OptForSize]>;
3186 def : Pat<(f32 (X86frsqrt FR32:$src)),
3187 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3188 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3189 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3190 Requires<[HasAVX, OptForSize]>;
3192 def : Pat<(f32 (X86frcp FR32:$src)),
3193 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3194 def : Pat<(f32 (X86frcp (load addr:$src))),
3195 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3196 Requires<[HasAVX, OptForSize]>;
3199 let Predicates = [HasAVX], AddedComplexity = 1 in {
3200 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3201 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3202 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3203 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3205 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3206 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3208 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3209 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3210 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3211 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3213 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3214 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3216 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3217 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3218 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3219 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3221 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3222 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3224 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3225 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3226 (VRCPSSr (f32 (IMPLICIT_DEF)),
3227 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3229 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3230 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3234 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3235 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3236 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3237 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3238 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3239 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3241 // Reciprocal approximations. Note that these typically require refinement
3242 // in order to obtain suitable precision.
3243 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3244 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3245 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3246 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3247 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3248 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3250 // There is no f64 version of the reciprocal approximation instructions.
3252 //===----------------------------------------------------------------------===//
3253 // SSE 1 & 2 - Non-temporal stores
3254 //===----------------------------------------------------------------------===//
3256 let AddedComplexity = 400 in { // Prefer non-temporal versions
3257 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3258 (ins f128mem:$dst, VR128:$src),
3259 "movntps\t{$src, $dst|$dst, $src}",
3260 [(alignednontemporalstore (v4f32 VR128:$src),
3262 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3263 (ins f128mem:$dst, VR128:$src),
3264 "movntpd\t{$src, $dst|$dst, $src}",
3265 [(alignednontemporalstore (v2f64 VR128:$src),
3268 let ExeDomain = SSEPackedInt in
3269 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3270 (ins f128mem:$dst, VR128:$src),
3271 "movntdq\t{$src, $dst|$dst, $src}",
3272 [(alignednontemporalstore (v2i64 VR128:$src),
3275 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3276 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3278 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3279 (ins f256mem:$dst, VR256:$src),
3280 "movntps\t{$src, $dst|$dst, $src}",
3281 [(alignednontemporalstore (v8f32 VR256:$src),
3283 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3284 (ins f256mem:$dst, VR256:$src),
3285 "movntpd\t{$src, $dst|$dst, $src}",
3286 [(alignednontemporalstore (v4f64 VR256:$src),
3288 let ExeDomain = SSEPackedInt in
3289 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3290 (ins f256mem:$dst, VR256:$src),
3291 "movntdq\t{$src, $dst|$dst, $src}",
3292 [(alignednontemporalstore (v4i64 VR256:$src),
3296 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3297 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3298 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3299 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3300 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3301 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3303 let AddedComplexity = 400 in { // Prefer non-temporal versions
3304 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3305 "movntps\t{$src, $dst|$dst, $src}",
3306 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3307 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3308 "movntpd\t{$src, $dst|$dst, $src}",
3309 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3311 let ExeDomain = SSEPackedInt in
3312 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3313 "movntdq\t{$src, $dst|$dst, $src}",
3314 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)]>;
3316 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3317 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3319 // There is no AVX form for instructions below this point
3320 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3321 "movnti{l}\t{$src, $dst|$dst, $src}",
3322 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3323 TB, Requires<[HasSSE2]>;
3324 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3325 "movnti{q}\t{$src, $dst|$dst, $src}",
3326 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3327 TB, Requires<[HasSSE2]>;
3330 //===----------------------------------------------------------------------===//
3331 // SSE 1 & 2 - Prefetch and memory fence
3332 //===----------------------------------------------------------------------===//
3334 // Prefetch intrinsic.
3335 let Predicates = [HasSSE1] in {
3336 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3337 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3338 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3339 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3340 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3341 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3342 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3343 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3347 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3348 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3349 TB, Requires<[HasSSE2]>;
3351 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3352 // was introduced with SSE2, it's backward compatible.
3353 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3355 // Load, store, and memory fence
3356 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3357 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3358 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3359 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3360 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3361 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3363 def : Pat<(X86SFence), (SFENCE)>;
3364 def : Pat<(X86LFence), (LFENCE)>;
3365 def : Pat<(X86MFence), (MFENCE)>;
3367 //===----------------------------------------------------------------------===//
3368 // SSE 1 & 2 - Load/Store XCSR register
3369 //===----------------------------------------------------------------------===//
3371 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3372 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3373 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3374 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3376 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3377 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3378 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3379 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3381 //===---------------------------------------------------------------------===//
3382 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3383 //===---------------------------------------------------------------------===//
3385 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3387 let neverHasSideEffects = 1 in {
3388 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3389 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3390 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3391 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3393 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3394 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3395 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3396 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3399 let isCodeGenOnly = 1 in {
3400 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3401 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3402 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3403 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3404 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3405 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3406 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3407 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3410 let canFoldAsLoad = 1, mayLoad = 1 in {
3411 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3412 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3413 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3414 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3415 let Predicates = [HasAVX] in {
3416 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3417 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3418 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3419 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3423 let mayStore = 1 in {
3424 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3425 (ins i128mem:$dst, VR128:$src),
3426 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3427 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3428 (ins i256mem:$dst, VR256:$src),
3429 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3430 let Predicates = [HasAVX] in {
3431 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3432 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3433 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3434 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3438 let neverHasSideEffects = 1 in
3439 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3440 "movdqa\t{$src, $dst|$dst, $src}", []>;
3442 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3443 "movdqu\t{$src, $dst|$dst, $src}",
3444 []>, XS, Requires<[HasSSE2]>;
3447 let isCodeGenOnly = 1 in {
3448 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3449 "movdqa\t{$src, $dst|$dst, $src}", []>;
3451 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3452 "movdqu\t{$src, $dst|$dst, $src}",
3453 []>, XS, Requires<[HasSSE2]>;
3456 let canFoldAsLoad = 1, mayLoad = 1 in {
3457 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3458 "movdqa\t{$src, $dst|$dst, $src}",
3459 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3460 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3461 "movdqu\t{$src, $dst|$dst, $src}",
3462 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3463 XS, Requires<[HasSSE2]>;
3466 let mayStore = 1 in {
3467 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3468 "movdqa\t{$src, $dst|$dst, $src}",
3469 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3470 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3471 "movdqu\t{$src, $dst|$dst, $src}",
3472 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3473 XS, Requires<[HasSSE2]>;
3476 // Intrinsic forms of MOVDQU load and store
3477 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3478 "vmovdqu\t{$src, $dst|$dst, $src}",
3479 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3480 XS, VEX, Requires<[HasAVX]>;
3482 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3483 "movdqu\t{$src, $dst|$dst, $src}",
3484 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3485 XS, Requires<[HasSSE2]>;
3487 } // ExeDomain = SSEPackedInt
3489 let Predicates = [HasAVX] in {
3490 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3491 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3492 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3495 //===---------------------------------------------------------------------===//
3496 // SSE2 - Packed Integer Arithmetic Instructions
3497 //===---------------------------------------------------------------------===//
3499 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3501 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3502 RegisterClass RC, PatFrag memop_frag,
3503 X86MemOperand x86memop, bit IsCommutable = 0,
3505 let isCommutable = IsCommutable in
3506 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3507 (ins RC:$src1, RC:$src2),
3509 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3510 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3511 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3512 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3513 (ins RC:$src1, x86memop:$src2),
3515 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3516 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3517 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3520 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3521 string OpcodeStr, Intrinsic IntId,
3522 Intrinsic IntId2, RegisterClass RC,
3524 // src2 is always 128-bit
3525 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3526 (ins RC:$src1, VR128:$src2),
3528 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3529 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3530 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3531 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3532 (ins RC:$src1, i128mem:$src2),
3534 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3535 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3536 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3537 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3538 (ins RC:$src1, i32i8imm:$src2),
3540 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3541 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3542 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3545 } // ExeDomain = SSEPackedInt
3547 // 128-bit Integer Arithmetic
3549 let Predicates = [HasAVX] in {
3550 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3551 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3552 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3553 i128mem, 1, 0>, VEX_4V;
3554 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3555 i128mem, 1, 0>, VEX_4V;
3556 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3557 i128mem, 1, 0>, VEX_4V;
3558 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3559 i128mem, 1, 0>, VEX_4V;
3560 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3561 i128mem, 0, 0>, VEX_4V;
3562 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3563 i128mem, 0, 0>, VEX_4V;
3564 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3565 i128mem, 0, 0>, VEX_4V;
3566 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3567 i128mem, 0, 0>, VEX_4V;
3570 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3571 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3572 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3573 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3574 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3575 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3576 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3577 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3578 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3579 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3580 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3581 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3582 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3583 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3584 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3585 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3586 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3587 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3588 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3589 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3590 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3591 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3592 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3593 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3594 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3595 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3596 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3597 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3598 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3599 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3600 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3601 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3602 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3603 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3604 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3605 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3606 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3607 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3610 let Predicates = [HasAVX2] in {
3611 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3612 i256mem, 1, 0>, VEX_4V;
3613 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3614 i256mem, 1, 0>, VEX_4V;
3615 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3616 i256mem, 1, 0>, VEX_4V;
3617 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3618 i256mem, 1, 0>, VEX_4V;
3619 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3620 i256mem, 1, 0>, VEX_4V;
3621 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3622 i256mem, 0, 0>, VEX_4V;
3623 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3624 i256mem, 0, 0>, VEX_4V;
3625 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3626 i256mem, 0, 0>, VEX_4V;
3627 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3628 i256mem, 0, 0>, VEX_4V;
3631 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3632 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3633 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3634 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3635 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3636 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3637 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3638 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3639 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3640 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3641 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3642 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3643 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3644 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3645 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3646 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3647 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3648 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3649 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3650 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3651 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3652 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3653 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3654 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3655 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3656 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3657 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3658 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3659 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3660 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3661 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3662 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3663 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3664 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3665 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3666 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3667 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3668 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3671 let Constraints = "$src1 = $dst" in {
3672 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3674 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3676 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3678 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3680 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3682 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3684 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3686 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3688 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3692 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3693 VR128, memopv2i64, i128mem>;
3694 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3695 VR128, memopv2i64, i128mem>;
3696 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3697 VR128, memopv2i64, i128mem>;
3698 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3699 VR128, memopv2i64, i128mem>;
3700 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3701 VR128, memopv2i64, i128mem, 1>;
3702 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3703 VR128, memopv2i64, i128mem, 1>;
3704 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3705 VR128, memopv2i64, i128mem, 1>;
3706 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3707 VR128, memopv2i64, i128mem, 1>;
3708 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3709 VR128, memopv2i64, i128mem, 1>;
3710 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3711 VR128, memopv2i64, i128mem, 1>;
3712 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3713 VR128, memopv2i64, i128mem, 1>;
3714 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3715 VR128, memopv2i64, i128mem, 1>;
3716 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3717 VR128, memopv2i64, i128mem, 1>;
3718 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3719 VR128, memopv2i64, i128mem, 1>;
3720 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3721 VR128, memopv2i64, i128mem, 1>;
3722 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3723 VR128, memopv2i64, i128mem, 1>;
3724 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3725 VR128, memopv2i64, i128mem, 1>;
3726 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3727 VR128, memopv2i64, i128mem, 1>;
3728 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3729 VR128, memopv2i64, i128mem, 1>;
3731 } // Constraints = "$src1 = $dst"
3733 //===---------------------------------------------------------------------===//
3734 // SSE2 - Packed Integer Logical Instructions
3735 //===---------------------------------------------------------------------===//
3737 let Predicates = [HasAVX] in {
3738 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3739 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3741 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3742 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3744 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3745 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3748 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3749 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3751 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3752 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3754 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3755 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3758 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3759 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3761 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3762 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3765 let ExeDomain = SSEPackedInt in {
3766 let neverHasSideEffects = 1 in {
3767 // 128-bit logical shifts.
3768 def VPSLLDQri : PDIi8<0x73, MRM7r,
3769 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3770 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3772 def VPSRLDQri : PDIi8<0x73, MRM3r,
3773 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3774 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3776 // PSRADQri doesn't exist in SSE[1-3].
3781 let Predicates = [HasAVX2] in {
3782 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3783 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3785 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3786 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3788 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3789 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3792 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3793 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3795 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3796 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3798 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3799 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3802 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3803 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3805 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3806 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3809 let ExeDomain = SSEPackedInt in {
3810 let neverHasSideEffects = 1 in {
3811 // 128-bit logical shifts.
3812 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3813 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3814 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3816 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3817 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3818 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3820 // PSRADQYri doesn't exist in SSE[1-3].
3825 let Constraints = "$src1 = $dst" in {
3826 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3827 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3829 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3830 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3832 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3833 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3836 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3837 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3839 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3840 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3842 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3843 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3846 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3847 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3849 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3850 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3853 let ExeDomain = SSEPackedInt in {
3854 let neverHasSideEffects = 1 in {
3855 // 128-bit logical shifts.
3856 def PSLLDQri : PDIi8<0x73, MRM7r,
3857 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3858 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3859 def PSRLDQri : PDIi8<0x73, MRM3r,
3860 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3861 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3862 // PSRADQri doesn't exist in SSE[1-3].
3865 } // Constraints = "$src1 = $dst"
3867 let Predicates = [HasAVX] in {
3868 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3869 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3870 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3871 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3872 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3873 (VPSLLDQri VR128:$src1, imm:$src2)>;
3874 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3875 (VPSRLDQri VR128:$src1, imm:$src2)>;
3876 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3877 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3879 // Shift up / down and insert zero's.
3880 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3881 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3882 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3883 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3886 let Predicates = [HasAVX2] in {
3887 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3888 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3889 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3890 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3891 def : Pat<(int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2),
3892 (VPSLLDQYri VR256:$src1, imm:$src2)>;
3893 def : Pat<(int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2),
3894 (VPSRLDQYri VR256:$src1, imm:$src2)>;
3897 let Predicates = [HasSSE2] in {
3898 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3899 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3900 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3901 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3902 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3903 (PSLLDQri VR128:$src1, imm:$src2)>;
3904 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3905 (PSRLDQri VR128:$src1, imm:$src2)>;
3906 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3907 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3909 // Shift up / down and insert zero's.
3910 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3911 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3912 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3913 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3916 //===---------------------------------------------------------------------===//
3917 // SSE2 - Packed Integer Comparison Instructions
3918 //===---------------------------------------------------------------------===//
3920 let Predicates = [HasAVX] in {
3921 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
3922 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3923 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
3924 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3925 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
3926 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3927 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
3928 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3929 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
3930 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3931 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
3932 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3934 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3935 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3936 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3937 (bc_v16i8 (memopv2i64 addr:$src2)))),
3938 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3939 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3940 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3941 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3942 (bc_v8i16 (memopv2i64 addr:$src2)))),
3943 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3944 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3945 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3946 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3947 (bc_v4i32 (memopv2i64 addr:$src2)))),
3948 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3950 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3951 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3952 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
3953 (bc_v16i8 (memopv2i64 addr:$src2)))),
3954 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3955 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3956 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3957 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
3958 (bc_v8i16 (memopv2i64 addr:$src2)))),
3959 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3960 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3961 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3962 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
3963 (bc_v4i32 (memopv2i64 addr:$src2)))),
3964 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3967 let Predicates = [HasAVX2] in {
3968 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
3969 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3970 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
3971 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3972 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
3973 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3974 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
3975 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3976 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
3977 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3978 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
3979 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3981 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, VR256:$src2)),
3982 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
3983 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1,
3984 (bc_v32i8 (memopv4i64 addr:$src2)))),
3985 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
3986 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, VR256:$src2)),
3987 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
3988 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1,
3989 (bc_v16i16 (memopv4i64 addr:$src2)))),
3990 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
3991 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, VR256:$src2)),
3992 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
3993 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1,
3994 (bc_v8i32 (memopv4i64 addr:$src2)))),
3995 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
3997 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, VR256:$src2)),
3998 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
3999 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1,
4000 (bc_v32i8 (memopv4i64 addr:$src2)))),
4001 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
4002 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, VR256:$src2)),
4003 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
4004 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1,
4005 (bc_v16i16 (memopv4i64 addr:$src2)))),
4006 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
4007 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, VR256:$src2)),
4008 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
4009 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1,
4010 (bc_v8i32 (memopv4i64 addr:$src2)))),
4011 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
4014 let Constraints = "$src1 = $dst" in {
4015 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
4016 VR128, memopv2i64, i128mem, 1>;
4017 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
4018 VR128, memopv2i64, i128mem, 1>;
4019 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
4020 VR128, memopv2i64, i128mem, 1>;
4021 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
4022 VR128, memopv2i64, i128mem>;
4023 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
4024 VR128, memopv2i64, i128mem>;
4025 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
4026 VR128, memopv2i64, i128mem>;
4027 } // Constraints = "$src1 = $dst"
4029 let Predicates = [HasSSE2] in {
4030 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
4031 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
4032 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
4033 (bc_v16i8 (memopv2i64 addr:$src2)))),
4034 (PCMPEQBrm VR128:$src1, addr:$src2)>;
4035 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
4036 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
4037 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
4038 (bc_v8i16 (memopv2i64 addr:$src2)))),
4039 (PCMPEQWrm VR128:$src1, addr:$src2)>;
4040 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
4041 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
4042 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
4043 (bc_v4i32 (memopv2i64 addr:$src2)))),
4044 (PCMPEQDrm VR128:$src1, addr:$src2)>;
4046 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
4047 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
4048 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
4049 (bc_v16i8 (memopv2i64 addr:$src2)))),
4050 (PCMPGTBrm VR128:$src1, addr:$src2)>;
4051 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
4052 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
4053 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
4054 (bc_v8i16 (memopv2i64 addr:$src2)))),
4055 (PCMPGTWrm VR128:$src1, addr:$src2)>;
4056 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
4057 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
4058 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
4059 (bc_v4i32 (memopv2i64 addr:$src2)))),
4060 (PCMPGTDrm VR128:$src1, addr:$src2)>;
4063 //===---------------------------------------------------------------------===//
4064 // SSE2 - Packed Integer Pack Instructions
4065 //===---------------------------------------------------------------------===//
4067 let Predicates = [HasAVX] in {
4068 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4069 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4070 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4071 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4072 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4073 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4076 let Predicates = [HasAVX2] in {
4077 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4078 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4079 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4080 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4081 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4082 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4085 let Constraints = "$src1 = $dst" in {
4086 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4087 VR128, memopv2i64, i128mem>;
4088 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4089 VR128, memopv2i64, i128mem>;
4090 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4091 VR128, memopv2i64, i128mem>;
4092 } // Constraints = "$src1 = $dst"
4094 //===---------------------------------------------------------------------===//
4095 // SSE2 - Packed Integer Shuffle Instructions
4096 //===---------------------------------------------------------------------===//
4098 let ExeDomain = SSEPackedInt in {
4099 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4101 def ri : Ii8<0x70, MRMSrcReg,
4102 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4103 !strconcat(OpcodeStr,
4104 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4105 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4107 def mi : Ii8<0x70, MRMSrcMem,
4108 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4109 !strconcat(OpcodeStr,
4110 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4111 [(set VR128:$dst, (vt (pshuf_frag:$src2
4112 (bc_frag (memopv2i64 addr:$src1)),
4116 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4118 def Yri : Ii8<0x70, MRMSrcReg,
4119 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4120 !strconcat(OpcodeStr,
4121 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4122 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4124 def Ymi : Ii8<0x70, MRMSrcMem,
4125 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4126 !strconcat(OpcodeStr,
4127 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4128 [(set VR256:$dst, (vt (pshuf_frag:$src2
4129 (bc_frag (memopv4i64 addr:$src1)),
4132 } // ExeDomain = SSEPackedInt
4134 let Predicates = [HasAVX] in {
4135 let AddedComplexity = 5 in
4136 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4139 // SSE2 with ImmT == Imm8 and XS prefix.
4140 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4143 // SSE2 with ImmT == Imm8 and XD prefix.
4144 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4147 let AddedComplexity = 5 in
4148 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4149 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4150 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4151 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4152 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4154 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4156 (VPSHUFDmi addr:$src1, imm:$imm)>;
4157 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4159 (VPSHUFDmi addr:$src1, imm:$imm)>;
4160 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4161 (VPSHUFDri VR128:$src1, imm:$imm)>;
4162 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4163 (VPSHUFDri VR128:$src1, imm:$imm)>;
4164 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4165 (VPSHUFHWri VR128:$src, imm:$imm)>;
4166 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4168 (VPSHUFHWmi addr:$src, imm:$imm)>;
4169 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4170 (VPSHUFLWri VR128:$src, imm:$imm)>;
4171 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4173 (VPSHUFLWmi addr:$src, imm:$imm)>;
4176 let Predicates = [HasAVX2] in {
4177 let AddedComplexity = 5 in
4178 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4181 // SSE2 with ImmT == Imm8 and XS prefix.
4182 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4185 // SSE2 with ImmT == Imm8 and XD prefix.
4186 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4190 let Predicates = [HasSSE2] in {
4191 let AddedComplexity = 5 in
4192 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4194 // SSE2 with ImmT == Imm8 and XS prefix.
4195 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4197 // SSE2 with ImmT == Imm8 and XD prefix.
4198 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4200 let AddedComplexity = 5 in
4201 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4202 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4203 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4204 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4205 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4207 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4209 (PSHUFDmi addr:$src1, imm:$imm)>;
4210 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4212 (PSHUFDmi addr:$src1, imm:$imm)>;
4213 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4214 (PSHUFDri VR128:$src1, imm:$imm)>;
4215 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4216 (PSHUFDri VR128:$src1, imm:$imm)>;
4217 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4218 (PSHUFHWri VR128:$src, imm:$imm)>;
4219 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4221 (PSHUFHWmi addr:$src, imm:$imm)>;
4222 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4223 (PSHUFLWri VR128:$src, imm:$imm)>;
4224 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4226 (PSHUFLWmi addr:$src, imm:$imm)>;
4229 //===---------------------------------------------------------------------===//
4230 // SSE2 - Packed Integer Unpack Instructions
4231 //===---------------------------------------------------------------------===//
4233 let ExeDomain = SSEPackedInt in {
4234 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4235 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4236 def rr : PDI<opc, MRMSrcReg,
4237 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4239 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4240 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4241 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4242 def rm : PDI<opc, MRMSrcMem,
4243 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4245 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4246 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4247 [(set VR128:$dst, (OpNode VR128:$src1,
4248 (bc_frag (memopv2i64
4252 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4253 SDNode OpNode, PatFrag bc_frag> {
4254 def Yrr : PDI<opc, MRMSrcReg,
4255 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4256 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4257 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4258 def Yrm : PDI<opc, MRMSrcMem,
4259 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4260 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4261 [(set VR256:$dst, (OpNode VR256:$src1,
4262 (bc_frag (memopv4i64 addr:$src2))))]>;
4265 let Predicates = [HasAVX] in {
4266 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4267 bc_v16i8, 0>, VEX_4V;
4268 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4269 bc_v8i16, 0>, VEX_4V;
4270 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4271 bc_v4i32, 0>, VEX_4V;
4272 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4273 bc_v2i64, 0>, VEX_4V;
4275 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4276 bc_v16i8, 0>, VEX_4V;
4277 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4278 bc_v8i16, 0>, VEX_4V;
4279 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4280 bc_v4i32, 0>, VEX_4V;
4281 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4282 bc_v2i64, 0>, VEX_4V;
4285 let Predicates = [HasAVX2] in {
4286 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4288 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4290 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4292 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4295 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4297 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4299 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4301 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4305 let Constraints = "$src1 = $dst" in {
4306 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4308 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4310 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4312 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4315 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4317 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4319 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4321 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4324 } // ExeDomain = SSEPackedInt
4326 // Patterns for using AVX1 instructions with integer vectors
4327 // Here to give AVX2 priority
4328 let Predicates = [HasAVX] in {
4329 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4330 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4331 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4332 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4333 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4334 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4335 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4336 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4338 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4339 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4340 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4341 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4342 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4343 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4344 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4345 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4348 // Splat v2f64 / v2i64
4349 let AddedComplexity = 10 in {
4350 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4351 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4352 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4353 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4356 //===---------------------------------------------------------------------===//
4357 // SSE2 - Packed Integer Extract and Insert
4358 //===---------------------------------------------------------------------===//
4360 let ExeDomain = SSEPackedInt in {
4361 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4362 def rri : Ii8<0xC4, MRMSrcReg,
4363 (outs VR128:$dst), (ins VR128:$src1,
4364 GR32:$src2, i32i8imm:$src3),
4366 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4367 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4369 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4370 def rmi : Ii8<0xC4, MRMSrcMem,
4371 (outs VR128:$dst), (ins VR128:$src1,
4372 i16mem:$src2, i32i8imm:$src3),
4374 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4375 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4377 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4382 let Predicates = [HasAVX] in
4383 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4384 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4385 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4386 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4387 imm:$src2))]>, TB, OpSize, VEX;
4388 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4389 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4390 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4391 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4395 let Predicates = [HasAVX] in {
4396 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4397 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4398 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4399 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4400 []>, TB, OpSize, VEX_4V;
4403 let Constraints = "$src1 = $dst" in
4404 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4406 } // ExeDomain = SSEPackedInt
4408 //===---------------------------------------------------------------------===//
4409 // SSE2 - Packed Mask Creation
4410 //===---------------------------------------------------------------------===//
4412 let ExeDomain = SSEPackedInt in {
4414 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4415 "pmovmskb\t{$src, $dst|$dst, $src}",
4416 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4417 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4418 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4420 let Predicates = [HasAVX2] in {
4421 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4422 "pmovmskb\t{$src, $dst|$dst, $src}",
4423 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4424 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4425 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4428 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4429 "pmovmskb\t{$src, $dst|$dst, $src}",
4430 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4432 } // ExeDomain = SSEPackedInt
4434 //===---------------------------------------------------------------------===//
4435 // SSE2 - Conditional Store
4436 //===---------------------------------------------------------------------===//
4438 let ExeDomain = SSEPackedInt in {
4441 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4442 (ins VR128:$src, VR128:$mask),
4443 "maskmovdqu\t{$mask, $src|$src, $mask}",
4444 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4446 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4447 (ins VR128:$src, VR128:$mask),
4448 "maskmovdqu\t{$mask, $src|$src, $mask}",
4449 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4452 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4453 "maskmovdqu\t{$mask, $src|$src, $mask}",
4454 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4456 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4457 "maskmovdqu\t{$mask, $src|$src, $mask}",
4458 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4460 } // ExeDomain = SSEPackedInt
4462 //===---------------------------------------------------------------------===//
4463 // SSE2 - Move Doubleword
4464 //===---------------------------------------------------------------------===//
4466 //===---------------------------------------------------------------------===//
4467 // Move Int Doubleword to Packed Double Int
4469 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4470 "movd\t{$src, $dst|$dst, $src}",
4472 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4473 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4474 "movd\t{$src, $dst|$dst, $src}",
4476 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4478 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4479 "mov{d|q}\t{$src, $dst|$dst, $src}",
4481 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4482 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4483 "mov{d|q}\t{$src, $dst|$dst, $src}",
4484 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4486 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4487 "movd\t{$src, $dst|$dst, $src}",
4489 (v4i32 (scalar_to_vector GR32:$src)))]>;
4490 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4491 "movd\t{$src, $dst|$dst, $src}",
4493 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4494 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4495 "mov{d|q}\t{$src, $dst|$dst, $src}",
4497 (v2i64 (scalar_to_vector GR64:$src)))]>;
4498 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4499 "mov{d|q}\t{$src, $dst|$dst, $src}",
4500 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4502 //===---------------------------------------------------------------------===//
4503 // Move Int Doubleword to Single Scalar
4505 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4506 "movd\t{$src, $dst|$dst, $src}",
4507 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4509 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4510 "movd\t{$src, $dst|$dst, $src}",
4511 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4513 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4514 "movd\t{$src, $dst|$dst, $src}",
4515 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4517 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4518 "movd\t{$src, $dst|$dst, $src}",
4519 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4521 //===---------------------------------------------------------------------===//
4522 // Move Packed Doubleword Int to Packed Double Int
4524 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4525 "movd\t{$src, $dst|$dst, $src}",
4526 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4528 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4529 (ins i32mem:$dst, VR128:$src),
4530 "movd\t{$src, $dst|$dst, $src}",
4531 [(store (i32 (vector_extract (v4i32 VR128:$src),
4532 (iPTR 0))), addr:$dst)]>, VEX;
4533 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4534 "movd\t{$src, $dst|$dst, $src}",
4535 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4537 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4538 "movd\t{$src, $dst|$dst, $src}",
4539 [(store (i32 (vector_extract (v4i32 VR128:$src),
4540 (iPTR 0))), addr:$dst)]>;
4542 //===---------------------------------------------------------------------===//
4543 // Move Packed Doubleword Int first element to Doubleword Int
4545 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4546 "mov{d|q}\t{$src, $dst|$dst, $src}",
4547 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4549 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4551 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4552 "mov{d|q}\t{$src, $dst|$dst, $src}",
4553 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4556 //===---------------------------------------------------------------------===//
4557 // Bitcast FR64 <-> GR64
4559 let Predicates = [HasAVX] in
4560 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4561 "vmovq\t{$src, $dst|$dst, $src}",
4562 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4564 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4565 "mov{d|q}\t{$src, $dst|$dst, $src}",
4566 [(set GR64:$dst, (bitconvert FR64:$src))]>, VEX;
4567 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4568 "movq\t{$src, $dst|$dst, $src}",
4569 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>,
4572 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4573 "movq\t{$src, $dst|$dst, $src}",
4574 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4575 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4576 "mov{d|q}\t{$src, $dst|$dst, $src}",
4577 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4578 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4579 "movq\t{$src, $dst|$dst, $src}",
4580 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4582 //===---------------------------------------------------------------------===//
4583 // Move Scalar Single to Double Int
4585 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4586 "movd\t{$src, $dst|$dst, $src}",
4587 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4588 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4589 "movd\t{$src, $dst|$dst, $src}",
4590 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4591 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4592 "movd\t{$src, $dst|$dst, $src}",
4593 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4594 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4595 "movd\t{$src, $dst|$dst, $src}",
4596 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4598 //===---------------------------------------------------------------------===//
4599 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4601 let AddedComplexity = 15 in {
4602 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4603 "movd\t{$src, $dst|$dst, $src}",
4604 [(set VR128:$dst, (v4i32 (X86vzmovl
4605 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4607 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4608 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4609 [(set VR128:$dst, (v2i64 (X86vzmovl
4610 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4613 let AddedComplexity = 15 in {
4614 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4615 "movd\t{$src, $dst|$dst, $src}",
4616 [(set VR128:$dst, (v4i32 (X86vzmovl
4617 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4618 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4619 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4620 [(set VR128:$dst, (v2i64 (X86vzmovl
4621 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4624 let AddedComplexity = 20 in {
4625 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4626 "movd\t{$src, $dst|$dst, $src}",
4628 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4629 (loadi32 addr:$src))))))]>,
4631 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4632 "movd\t{$src, $dst|$dst, $src}",
4634 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4635 (loadi32 addr:$src))))))]>;
4638 let Predicates = [HasAVX] in {
4639 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4640 let AddedComplexity = 20 in {
4641 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4642 (VMOVZDI2PDIrm addr:$src)>;
4643 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4644 (VMOVZDI2PDIrm addr:$src)>;
4645 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4646 (VMOVZDI2PDIrm addr:$src)>;
4648 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4649 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4650 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4651 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4652 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4653 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4654 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4657 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4658 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4659 (MOVZDI2PDIrm addr:$src)>;
4660 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4661 (MOVZDI2PDIrm addr:$src)>;
4662 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4663 (MOVZDI2PDIrm addr:$src)>;
4666 // These are the correct encodings of the instructions so that we know how to
4667 // read correct assembly, even though we continue to emit the wrong ones for
4668 // compatibility with Darwin's buggy assembler.
4669 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4670 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4671 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4672 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4673 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4674 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4675 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4676 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4677 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4678 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4679 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4680 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4682 //===---------------------------------------------------------------------===//
4683 // SSE2 - Move Quadword
4684 //===---------------------------------------------------------------------===//
4686 //===---------------------------------------------------------------------===//
4687 // Move Quadword Int to Packed Quadword Int
4689 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4690 "vmovq\t{$src, $dst|$dst, $src}",
4692 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4693 VEX, Requires<[HasAVX]>;
4694 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4695 "movq\t{$src, $dst|$dst, $src}",
4697 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4698 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4700 //===---------------------------------------------------------------------===//
4701 // Move Packed Quadword Int to Quadword Int
4703 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4704 "movq\t{$src, $dst|$dst, $src}",
4705 [(store (i64 (vector_extract (v2i64 VR128:$src),
4706 (iPTR 0))), addr:$dst)]>, VEX;
4707 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4708 "movq\t{$src, $dst|$dst, $src}",
4709 [(store (i64 (vector_extract (v2i64 VR128:$src),
4710 (iPTR 0))), addr:$dst)]>;
4712 //===---------------------------------------------------------------------===//
4713 // Store / copy lower 64-bits of a XMM register.
4715 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4716 "movq\t{$src, $dst|$dst, $src}",
4717 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4718 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4719 "movq\t{$src, $dst|$dst, $src}",
4720 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4722 let AddedComplexity = 20 in
4723 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4724 "vmovq\t{$src, $dst|$dst, $src}",
4726 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4727 (loadi64 addr:$src))))))]>,
4728 XS, VEX, Requires<[HasAVX]>;
4730 let AddedComplexity = 20 in
4731 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4732 "movq\t{$src, $dst|$dst, $src}",
4734 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4735 (loadi64 addr:$src))))))]>,
4736 XS, Requires<[HasSSE2]>;
4738 let Predicates = [HasAVX], AddedComplexity = 20 in {
4739 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4740 (VMOVZQI2PQIrm addr:$src)>;
4741 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4742 (VMOVZQI2PQIrm addr:$src)>;
4743 def : Pat<(v2i64 (X86vzload addr:$src)),
4744 (VMOVZQI2PQIrm addr:$src)>;
4747 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4748 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4749 (MOVZQI2PQIrm addr:$src)>;
4750 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4751 (MOVZQI2PQIrm addr:$src)>;
4752 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4755 let Predicates = [HasAVX] in {
4756 def : Pat<(v4i64 (X86vzload addr:$src)),
4757 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4760 //===---------------------------------------------------------------------===//
4761 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4762 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4764 let AddedComplexity = 15 in
4765 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4766 "vmovq\t{$src, $dst|$dst, $src}",
4767 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4768 XS, VEX, Requires<[HasAVX]>;
4769 let AddedComplexity = 15 in
4770 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4771 "movq\t{$src, $dst|$dst, $src}",
4772 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4773 XS, Requires<[HasSSE2]>;
4775 let AddedComplexity = 20 in
4776 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4777 "vmovq\t{$src, $dst|$dst, $src}",
4778 [(set VR128:$dst, (v2i64 (X86vzmovl
4779 (loadv2i64 addr:$src))))]>,
4780 XS, VEX, Requires<[HasAVX]>;
4781 let AddedComplexity = 20 in {
4782 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4783 "movq\t{$src, $dst|$dst, $src}",
4784 [(set VR128:$dst, (v2i64 (X86vzmovl
4785 (loadv2i64 addr:$src))))]>,
4786 XS, Requires<[HasSSE2]>;
4789 let AddedComplexity = 20 in {
4790 let Predicates = [HasAVX] in {
4791 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4792 (VMOVZPQILo2PQIrm addr:$src)>;
4793 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4794 (VMOVZPQILo2PQIrr VR128:$src)>;
4796 let Predicates = [HasSSE2] in {
4797 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4798 (MOVZPQILo2PQIrm addr:$src)>;
4799 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4800 (MOVZPQILo2PQIrr VR128:$src)>;
4804 // Instructions to match in the assembler
4805 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4806 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4807 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4808 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4809 // Recognize "movd" with GR64 destination, but encode as a "movq"
4810 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4811 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4813 // Instructions for the disassembler
4814 // xr = XMM register
4817 let Predicates = [HasAVX] in
4818 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4819 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4820 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4821 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4823 //===---------------------------------------------------------------------===//
4824 // SSE3 - Conversion Instructions
4825 //===---------------------------------------------------------------------===//
4827 // Convert Packed Double FP to Packed DW Integers
4828 let Predicates = [HasAVX] in {
4829 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4830 // register, but the same isn't true when using memory operands instead.
4831 // Provide other assembly rr and rm forms to address this explicitly.
4832 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4833 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4834 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4835 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4838 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4839 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4840 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4841 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4844 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4845 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4846 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4847 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4850 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4851 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4852 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4853 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4855 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4856 (VCVTPD2DQYrr VR256:$src)>;
4857 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4858 (VCVTPD2DQYrm addr:$src)>;
4860 // Convert Packed DW Integers to Packed Double FP
4861 let Predicates = [HasAVX] in {
4862 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4863 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4864 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4865 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4866 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4867 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4868 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4869 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4872 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4873 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4874 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4875 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4877 // AVX 256-bit register conversion intrinsics
4878 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4879 (VCVTDQ2PDYrr VR128:$src)>;
4880 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4881 (VCVTDQ2PDYrm addr:$src)>;
4883 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4884 (VCVTPD2DQYrr VR256:$src)>;
4885 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4886 (VCVTPD2DQYrm addr:$src)>;
4888 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4889 (VCVTDQ2PDYrr VR128:$src)>;
4890 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4891 (VCVTDQ2PDYrm addr:$src)>;
4893 //===---------------------------------------------------------------------===//
4894 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4895 //===---------------------------------------------------------------------===//
4896 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4897 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4898 X86MemOperand x86memop> {
4899 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4900 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4901 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4902 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4903 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4904 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4907 let Predicates = [HasAVX] in {
4908 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4909 v4f32, VR128, memopv4f32, f128mem>, VEX;
4910 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4911 v4f32, VR128, memopv4f32, f128mem>, VEX;
4912 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4913 v8f32, VR256, memopv8f32, f256mem>, VEX;
4914 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4915 v8f32, VR256, memopv8f32, f256mem>, VEX;
4917 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4918 memopv4f32, f128mem>;
4919 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4920 memopv4f32, f128mem>;
4922 let Predicates = [HasAVX] in {
4923 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4924 (VMOVSHDUPrr VR128:$src)>;
4925 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4926 (VMOVSHDUPrm addr:$src)>;
4927 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4928 (VMOVSLDUPrr VR128:$src)>;
4929 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4930 (VMOVSLDUPrm addr:$src)>;
4931 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4932 (VMOVSHDUPYrr VR256:$src)>;
4933 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4934 (VMOVSHDUPYrm addr:$src)>;
4935 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4936 (VMOVSLDUPYrr VR256:$src)>;
4937 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4938 (VMOVSLDUPYrm addr:$src)>;
4941 let Predicates = [HasSSE3] in {
4942 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4943 (MOVSHDUPrr VR128:$src)>;
4944 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4945 (MOVSHDUPrm addr:$src)>;
4946 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4947 (MOVSLDUPrr VR128:$src)>;
4948 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4949 (MOVSLDUPrm addr:$src)>;
4952 //===---------------------------------------------------------------------===//
4953 // SSE3 - Replicate Double FP - MOVDDUP
4954 //===---------------------------------------------------------------------===//
4956 multiclass sse3_replicate_dfp<string OpcodeStr> {
4957 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4958 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4959 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4960 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4961 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4963 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4967 // FIXME: Merge with above classe when there're patterns for the ymm version
4968 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4969 let Predicates = [HasAVX] in {
4970 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4971 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4973 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4974 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4979 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4980 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4981 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4983 let Predicates = [HasAVX] in {
4984 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4986 (VMOVDDUPrm addr:$src)>;
4987 let AddedComplexity = 5 in {
4988 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4989 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4990 (VMOVDDUPrm addr:$src)>;
4991 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4992 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4993 (VMOVDDUPrm addr:$src)>;
4995 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4996 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4997 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4998 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4999 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5000 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5001 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5002 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5003 def : Pat<(X86Movddup (bc_v2f64
5004 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5005 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5008 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
5009 (VMOVDDUPYrm addr:$src)>;
5010 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
5011 (VMOVDDUPYrm addr:$src)>;
5012 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
5013 (VMOVDDUPYrm addr:$src)>;
5014 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
5015 (VMOVDDUPYrm addr:$src)>;
5016 def : Pat<(X86Movddup (v4f64 VR256:$src)),
5017 (VMOVDDUPYrr VR256:$src)>;
5018 def : Pat<(X86Movddup (v4i64 VR256:$src)),
5019 (VMOVDDUPYrr VR256:$src)>;
5022 let Predicates = [HasSSE3] in {
5023 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
5025 (MOVDDUPrm addr:$src)>;
5026 let AddedComplexity = 5 in {
5027 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
5028 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
5029 (MOVDDUPrm addr:$src)>;
5030 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
5031 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
5032 (MOVDDUPrm addr:$src)>;
5034 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5035 (MOVDDUPrm addr:$src)>;
5036 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5037 (MOVDDUPrm addr:$src)>;
5038 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5039 (MOVDDUPrm addr:$src)>;
5040 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5041 (MOVDDUPrm addr:$src)>;
5042 def : Pat<(X86Movddup (bc_v2f64
5043 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5044 (MOVDDUPrm addr:$src)>;
5047 //===---------------------------------------------------------------------===//
5048 // SSE3 - Move Unaligned Integer
5049 //===---------------------------------------------------------------------===//
5051 let Predicates = [HasAVX] in {
5052 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5053 "vlddqu\t{$src, $dst|$dst, $src}",
5054 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5055 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5056 "vlddqu\t{$src, $dst|$dst, $src}",
5057 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5059 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5060 "lddqu\t{$src, $dst|$dst, $src}",
5061 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
5063 //===---------------------------------------------------------------------===//
5064 // SSE3 - Arithmetic
5065 //===---------------------------------------------------------------------===//
5067 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5068 X86MemOperand x86memop, bit Is2Addr = 1> {
5069 def rr : I<0xD0, MRMSrcReg,
5070 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5072 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5073 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5074 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
5075 def rm : I<0xD0, MRMSrcMem,
5076 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5078 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5079 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5080 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
5083 let Predicates = [HasAVX] in {
5084 let ExeDomain = SSEPackedSingle in {
5085 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5086 f128mem, 0>, TB, XD, VEX_4V;
5087 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5088 f256mem, 0>, TB, XD, VEX_4V;
5090 let ExeDomain = SSEPackedDouble in {
5091 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5092 f128mem, 0>, TB, OpSize, VEX_4V;
5093 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5094 f256mem, 0>, TB, OpSize, VEX_4V;
5097 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5098 let ExeDomain = SSEPackedSingle in
5099 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5101 let ExeDomain = SSEPackedDouble in
5102 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5103 f128mem>, TB, OpSize;
5106 //===---------------------------------------------------------------------===//
5107 // SSE3 Instructions
5108 //===---------------------------------------------------------------------===//
5111 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5112 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5113 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5115 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5116 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5117 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5119 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5121 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5122 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5123 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5125 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5126 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5127 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5129 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5130 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5131 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5133 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5135 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5136 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5137 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5140 let Predicates = [HasAVX] in {
5141 let ExeDomain = SSEPackedSingle in {
5142 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5143 X86fhadd, 0>, VEX_4V;
5144 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5145 X86fhsub, 0>, VEX_4V;
5146 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5147 X86fhadd, 0>, VEX_4V;
5148 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5149 X86fhsub, 0>, VEX_4V;
5151 let ExeDomain = SSEPackedDouble in {
5152 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5153 X86fhadd, 0>, VEX_4V;
5154 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5155 X86fhsub, 0>, VEX_4V;
5156 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5157 X86fhadd, 0>, VEX_4V;
5158 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5159 X86fhsub, 0>, VEX_4V;
5163 let Constraints = "$src1 = $dst" in {
5164 let ExeDomain = SSEPackedSingle in {
5165 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5166 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5168 let ExeDomain = SSEPackedDouble in {
5169 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5170 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5174 //===---------------------------------------------------------------------===//
5175 // SSSE3 - Packed Absolute Instructions
5176 //===---------------------------------------------------------------------===//
5179 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5180 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5181 Intrinsic IntId128> {
5182 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5184 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5185 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5188 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5190 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5193 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5196 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5197 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5198 Intrinsic IntId256> {
5199 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5201 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5202 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5205 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5207 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5210 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5213 let Predicates = [HasAVX] in {
5214 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5215 int_x86_ssse3_pabs_b_128>, VEX;
5216 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5217 int_x86_ssse3_pabs_w_128>, VEX;
5218 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5219 int_x86_ssse3_pabs_d_128>, VEX;
5222 let Predicates = [HasAVX2] in {
5223 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5224 int_x86_avx2_pabs_b>, VEX;
5225 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5226 int_x86_avx2_pabs_w>, VEX;
5227 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5228 int_x86_avx2_pabs_d>, VEX;
5231 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5232 int_x86_ssse3_pabs_b_128>;
5233 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5234 int_x86_ssse3_pabs_w_128>;
5235 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5236 int_x86_ssse3_pabs_d_128>;
5238 //===---------------------------------------------------------------------===//
5239 // SSSE3 - Packed Binary Operator Instructions
5240 //===---------------------------------------------------------------------===//
5242 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5243 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5244 Intrinsic IntId128, bit Is2Addr = 1> {
5245 let isCommutable = 1 in
5246 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5247 (ins VR128:$src1, VR128:$src2),
5249 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5250 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5251 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5253 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5254 (ins VR128:$src1, i128mem:$src2),
5256 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5257 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5259 (IntId128 VR128:$src1,
5260 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5263 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5264 Intrinsic IntId256> {
5265 let isCommutable = 1 in
5266 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5267 (ins VR256:$src1, VR256:$src2),
5268 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5269 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5271 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5272 (ins VR256:$src1, i256mem:$src2),
5273 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5275 (IntId256 VR256:$src1,
5276 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5279 let ImmT = NoImm, Predicates = [HasAVX] in {
5280 let isCommutable = 0 in {
5281 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw",
5282 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5283 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd",
5284 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5285 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5286 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5287 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw",
5288 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5289 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd",
5290 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5291 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5292 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5293 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5294 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5295 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb",
5296 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5297 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb",
5298 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5299 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw",
5300 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5301 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd",
5302 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5304 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5305 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5308 let ImmT = NoImm, Predicates = [HasAVX2] in {
5309 let isCommutable = 0 in {
5310 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw",
5311 int_x86_avx2_phadd_w>, VEX_4V;
5312 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd",
5313 int_x86_avx2_phadd_d>, VEX_4V;
5314 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5315 int_x86_avx2_phadd_sw>, VEX_4V;
5316 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw",
5317 int_x86_avx2_phsub_w>, VEX_4V;
5318 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd",
5319 int_x86_avx2_phsub_d>, VEX_4V;
5320 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5321 int_x86_avx2_phsub_sw>, VEX_4V;
5322 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5323 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5324 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb",
5325 int_x86_avx2_pshuf_b>, VEX_4V;
5326 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb",
5327 int_x86_avx2_psign_b>, VEX_4V;
5328 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw",
5329 int_x86_avx2_psign_w>, VEX_4V;
5330 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd",
5331 int_x86_avx2_psign_d>, VEX_4V;
5333 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5334 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5337 // None of these have i8 immediate fields.
5338 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5339 let isCommutable = 0 in {
5340 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw",
5341 int_x86_ssse3_phadd_w_128>;
5342 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd",
5343 int_x86_ssse3_phadd_d_128>;
5344 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5345 int_x86_ssse3_phadd_sw_128>;
5346 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw",
5347 int_x86_ssse3_phsub_w_128>;
5348 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd",
5349 int_x86_ssse3_phsub_d_128>;
5350 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5351 int_x86_ssse3_phsub_sw_128>;
5352 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5353 int_x86_ssse3_pmadd_ub_sw_128>;
5354 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb",
5355 int_x86_ssse3_pshuf_b_128>;
5356 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb",
5357 int_x86_ssse3_psign_b_128>;
5358 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw",
5359 int_x86_ssse3_psign_w_128>;
5360 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd",
5361 int_x86_ssse3_psign_d_128>;
5363 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5364 int_x86_ssse3_pmul_hr_sw_128>;
5367 let Predicates = [HasAVX] in {
5368 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5369 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5370 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5371 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5373 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5374 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5375 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5376 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5377 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5378 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5380 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5381 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5382 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5383 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5384 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5385 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5386 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5387 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5390 let Predicates = [HasAVX2] in {
5391 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5392 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5393 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5394 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5395 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5396 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5398 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5399 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5400 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5401 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5402 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5403 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5404 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5405 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5408 let Predicates = [HasSSSE3] in {
5409 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5410 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5411 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5412 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5414 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5415 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5416 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5417 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5418 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5419 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5421 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5422 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5423 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5424 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5425 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5426 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5427 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5428 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5431 //===---------------------------------------------------------------------===//
5432 // SSSE3 - Packed Align Instruction Patterns
5433 //===---------------------------------------------------------------------===//
5435 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5436 let neverHasSideEffects = 1 in {
5437 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5438 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5440 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5442 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5445 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5446 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5448 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5450 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5455 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5456 let neverHasSideEffects = 1 in {
5457 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5458 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5460 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5463 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5464 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5466 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5471 let Predicates = [HasAVX] in
5472 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5473 let Predicates = [HasAVX2] in
5474 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5475 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5476 defm PALIGN : ssse3_palign<"palignr">;
5478 let Predicates = [HasAVX] in {
5479 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5480 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5481 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5482 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5483 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5484 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5485 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5486 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5489 let Predicates = [HasSSSE3] in {
5490 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5491 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5492 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5493 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5494 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5495 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5496 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5497 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5500 //===---------------------------------------------------------------------===//
5501 // SSSE3 - Thread synchronization
5502 //===---------------------------------------------------------------------===//
5504 let usesCustomInserter = 1 in {
5505 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5506 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5507 Requires<[HasSSE3]>;
5508 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5509 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5510 Requires<[HasSSE3]>;
5513 let Uses = [EAX, ECX, EDX] in
5514 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5515 Requires<[HasSSE3]>;
5516 let Uses = [ECX, EAX] in
5517 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5518 Requires<[HasSSE3]>;
5520 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5521 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5523 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5524 Requires<[In32BitMode]>;
5525 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5526 Requires<[In64BitMode]>;
5528 //===----------------------------------------------------------------------===//
5529 // SSE4.1 - Packed Move with Sign/Zero Extend
5530 //===----------------------------------------------------------------------===//
5532 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5533 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5534 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5535 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5537 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5538 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5540 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5544 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5546 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5547 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5548 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5550 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5551 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5552 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5555 let Predicates = [HasAVX] in {
5556 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5558 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5560 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5562 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5564 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5566 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5570 let Predicates = [HasAVX2] in {
5571 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5572 int_x86_avx2_pmovsxbw>, VEX;
5573 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5574 int_x86_avx2_pmovsxwd>, VEX;
5575 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5576 int_x86_avx2_pmovsxdq>, VEX;
5577 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5578 int_x86_avx2_pmovzxbw>, VEX;
5579 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5580 int_x86_avx2_pmovzxwd>, VEX;
5581 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5582 int_x86_avx2_pmovzxdq>, VEX;
5585 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5586 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5587 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5588 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5589 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5590 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5592 let Predicates = [HasAVX] in {
5593 // Common patterns involving scalar load.
5594 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5595 (VPMOVSXBWrm addr:$src)>;
5596 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5597 (VPMOVSXBWrm addr:$src)>;
5599 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5600 (VPMOVSXWDrm addr:$src)>;
5601 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5602 (VPMOVSXWDrm addr:$src)>;
5604 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5605 (VPMOVSXDQrm addr:$src)>;
5606 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5607 (VPMOVSXDQrm addr:$src)>;
5609 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5610 (VPMOVZXBWrm addr:$src)>;
5611 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5612 (VPMOVZXBWrm addr:$src)>;
5614 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5615 (VPMOVZXWDrm addr:$src)>;
5616 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5617 (VPMOVZXWDrm addr:$src)>;
5619 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5620 (VPMOVZXDQrm addr:$src)>;
5621 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5622 (VPMOVZXDQrm addr:$src)>;
5625 let Predicates = [HasSSE41] in {
5626 // Common patterns involving scalar load.
5627 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5628 (PMOVSXBWrm addr:$src)>;
5629 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5630 (PMOVSXBWrm addr:$src)>;
5632 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5633 (PMOVSXWDrm addr:$src)>;
5634 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5635 (PMOVSXWDrm addr:$src)>;
5637 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5638 (PMOVSXDQrm addr:$src)>;
5639 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5640 (PMOVSXDQrm addr:$src)>;
5642 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5643 (PMOVZXBWrm addr:$src)>;
5644 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5645 (PMOVZXBWrm addr:$src)>;
5647 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5648 (PMOVZXWDrm addr:$src)>;
5649 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5650 (PMOVZXWDrm addr:$src)>;
5652 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5653 (PMOVZXDQrm addr:$src)>;
5654 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5655 (PMOVZXDQrm addr:$src)>;
5659 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5660 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5661 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5662 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5664 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5665 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5667 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5671 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5673 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5674 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5675 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5677 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5678 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5680 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5684 let Predicates = [HasAVX] in {
5685 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5687 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5689 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5691 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5695 let Predicates = [HasAVX2] in {
5696 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5697 int_x86_avx2_pmovsxbd>, VEX;
5698 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5699 int_x86_avx2_pmovsxwq>, VEX;
5700 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5701 int_x86_avx2_pmovzxbd>, VEX;
5702 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5703 int_x86_avx2_pmovzxwq>, VEX;
5706 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5707 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5708 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5709 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5711 let Predicates = [HasAVX] in {
5712 // Common patterns involving scalar load
5713 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5714 (VPMOVSXBDrm addr:$src)>;
5715 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5716 (VPMOVSXWQrm addr:$src)>;
5718 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5719 (VPMOVZXBDrm addr:$src)>;
5720 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5721 (VPMOVZXWQrm addr:$src)>;
5724 let Predicates = [HasSSE41] in {
5725 // Common patterns involving scalar load
5726 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5727 (PMOVSXBDrm addr:$src)>;
5728 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5729 (PMOVSXWQrm addr:$src)>;
5731 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5732 (PMOVZXBDrm addr:$src)>;
5733 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5734 (PMOVZXWQrm addr:$src)>;
5737 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5738 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5739 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5740 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5742 // Expecting a i16 load any extended to i32 value.
5743 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5744 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5745 [(set VR128:$dst, (IntId (bitconvert
5746 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5750 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5752 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5753 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5754 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5756 // Expecting a i16 load any extended to i32 value.
5757 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5758 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5759 [(set VR256:$dst, (IntId (bitconvert
5760 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5764 let Predicates = [HasAVX] in {
5765 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5767 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5770 let Predicates = [HasAVX2] in {
5771 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5772 int_x86_avx2_pmovsxbq>, VEX;
5773 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5774 int_x86_avx2_pmovzxbq>, VEX;
5776 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5777 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5779 let Predicates = [HasAVX] in {
5780 // Common patterns involving scalar load
5781 def : Pat<(int_x86_sse41_pmovsxbq
5782 (bitconvert (v4i32 (X86vzmovl
5783 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5784 (VPMOVSXBQrm addr:$src)>;
5786 def : Pat<(int_x86_sse41_pmovzxbq
5787 (bitconvert (v4i32 (X86vzmovl
5788 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5789 (VPMOVZXBQrm addr:$src)>;
5792 let Predicates = [HasSSE41] in {
5793 // Common patterns involving scalar load
5794 def : Pat<(int_x86_sse41_pmovsxbq
5795 (bitconvert (v4i32 (X86vzmovl
5796 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5797 (PMOVSXBQrm addr:$src)>;
5799 def : Pat<(int_x86_sse41_pmovzxbq
5800 (bitconvert (v4i32 (X86vzmovl
5801 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5802 (PMOVZXBQrm addr:$src)>;
5805 //===----------------------------------------------------------------------===//
5806 // SSE4.1 - Extract Instructions
5807 //===----------------------------------------------------------------------===//
5809 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5810 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5811 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5812 (ins VR128:$src1, i32i8imm:$src2),
5813 !strconcat(OpcodeStr,
5814 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5815 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5817 let neverHasSideEffects = 1, mayStore = 1 in
5818 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5819 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5820 !strconcat(OpcodeStr,
5821 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5824 // There's an AssertZext in the way of writing the store pattern
5825 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5828 let Predicates = [HasAVX] in {
5829 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5830 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5831 (ins VR128:$src1, i32i8imm:$src2),
5832 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5835 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5838 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5839 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5840 let neverHasSideEffects = 1, mayStore = 1 in
5841 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5842 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5843 !strconcat(OpcodeStr,
5844 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5847 // There's an AssertZext in the way of writing the store pattern
5848 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5851 let Predicates = [HasAVX] in
5852 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5854 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5857 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5858 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5859 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5860 (ins VR128:$src1, i32i8imm:$src2),
5861 !strconcat(OpcodeStr,
5862 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5864 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5865 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5866 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5867 !strconcat(OpcodeStr,
5868 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5869 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5870 addr:$dst)]>, OpSize;
5873 let Predicates = [HasAVX] in
5874 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5876 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5878 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5879 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5880 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5881 (ins VR128:$src1, i32i8imm:$src2),
5882 !strconcat(OpcodeStr,
5883 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5885 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5886 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5887 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5888 !strconcat(OpcodeStr,
5889 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5890 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5891 addr:$dst)]>, OpSize, REX_W;
5894 let Predicates = [HasAVX] in
5895 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5897 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5899 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5901 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5902 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5903 (ins VR128:$src1, i32i8imm:$src2),
5904 !strconcat(OpcodeStr,
5905 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5907 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5909 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5910 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5911 !strconcat(OpcodeStr,
5912 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5913 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5914 addr:$dst)]>, OpSize;
5917 let ExeDomain = SSEPackedSingle in {
5918 let Predicates = [HasAVX] in {
5919 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5920 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5921 (ins VR128:$src1, i32i8imm:$src2),
5922 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5925 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5928 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5929 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5932 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5934 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5937 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5938 Requires<[HasSSE41]>;
5940 //===----------------------------------------------------------------------===//
5941 // SSE4.1 - Insert Instructions
5942 //===----------------------------------------------------------------------===//
5944 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5945 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5946 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5948 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5950 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5952 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5953 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5954 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5956 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5958 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5960 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5961 imm:$src3))]>, OpSize;
5964 let Predicates = [HasAVX] in
5965 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5966 let Constraints = "$src1 = $dst" in
5967 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5969 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5970 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5971 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5973 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5975 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5977 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5979 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5980 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5982 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5984 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5986 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5987 imm:$src3)))]>, OpSize;
5990 let Predicates = [HasAVX] in
5991 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5992 let Constraints = "$src1 = $dst" in
5993 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5995 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5996 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5997 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5999 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6001 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6003 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6005 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6006 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6008 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6010 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6012 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6013 imm:$src3)))]>, OpSize;
6016 let Predicates = [HasAVX] in
6017 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6018 let Constraints = "$src1 = $dst" in
6019 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6021 // insertps has a few different modes, there's the first two here below which
6022 // are optimized inserts that won't zero arbitrary elements in the destination
6023 // vector. The next one matches the intrinsic and could zero arbitrary elements
6024 // in the target vector.
6025 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6026 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6027 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6029 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6031 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6033 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6035 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6036 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6038 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6040 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6042 (X86insrtps VR128:$src1,
6043 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6044 imm:$src3))]>, OpSize;
6047 let ExeDomain = SSEPackedSingle in {
6048 let Predicates = [HasAVX] in
6049 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6050 let Constraints = "$src1 = $dst" in
6051 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6054 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6055 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6057 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6058 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6059 Requires<[HasSSE41]>;
6061 //===----------------------------------------------------------------------===//
6062 // SSE4.1 - Round Instructions
6063 //===----------------------------------------------------------------------===//
6065 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6066 X86MemOperand x86memop, RegisterClass RC,
6067 PatFrag mem_frag32, PatFrag mem_frag64,
6068 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6069 let ExeDomain = SSEPackedSingle in {
6070 // Intrinsic operation, reg.
6071 // Vector intrinsic operation, reg
6072 def PSr : SS4AIi8<opcps, MRMSrcReg,
6073 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6074 !strconcat(OpcodeStr,
6075 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6076 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6079 // Vector intrinsic operation, mem
6080 def PSm : SS4AIi8<opcps, MRMSrcMem,
6081 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6082 !strconcat(OpcodeStr,
6083 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6085 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6087 } // ExeDomain = SSEPackedSingle
6089 let ExeDomain = SSEPackedDouble in {
6090 // Vector intrinsic operation, reg
6091 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6092 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6093 !strconcat(OpcodeStr,
6094 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6095 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6098 // Vector intrinsic operation, mem
6099 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6100 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6101 !strconcat(OpcodeStr,
6102 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6104 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6106 } // ExeDomain = SSEPackedDouble
6109 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6112 Intrinsic F64Int, bit Is2Addr = 1> {
6113 let ExeDomain = GenericDomain in {
6115 def SSr : SS4AIi8<opcss, MRMSrcReg,
6116 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6118 !strconcat(OpcodeStr,
6119 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6120 !strconcat(OpcodeStr,
6121 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6124 // Intrinsic operation, reg.
6125 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6126 (outs VR128:$dst), (ins VR128:$src1, VR128:$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, $src3}")),
6132 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6135 // Intrinsic operation, mem.
6136 def SSm : SS4AIi8<opcss, MRMSrcMem,
6137 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6139 !strconcat(OpcodeStr,
6140 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6141 !strconcat(OpcodeStr,
6142 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6144 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6148 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6149 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6151 !strconcat(OpcodeStr,
6152 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6153 !strconcat(OpcodeStr,
6154 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6157 // Intrinsic operation, reg.
6158 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6159 (outs VR128:$dst), (ins VR128:$src1, VR128:$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, $src3}")),
6165 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6168 // Intrinsic operation, mem.
6169 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6170 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6172 !strconcat(OpcodeStr,
6173 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6174 !strconcat(OpcodeStr,
6175 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6177 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6179 } // ExeDomain = GenericDomain
6182 // FP round - roundss, roundps, roundsd, roundpd
6183 let Predicates = [HasAVX] in {
6185 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6186 memopv4f32, memopv2f64,
6187 int_x86_sse41_round_ps,
6188 int_x86_sse41_round_pd>, VEX;
6189 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6190 memopv8f32, memopv4f64,
6191 int_x86_avx_round_ps_256,
6192 int_x86_avx_round_pd_256>, VEX;
6193 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6194 int_x86_sse41_round_ss,
6195 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6197 def : Pat<(ffloor FR32:$src),
6198 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6199 def : Pat<(f64 (ffloor FR64:$src)),
6200 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6201 def : Pat<(f32 (fnearbyint FR32:$src)),
6202 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6203 def : Pat<(f64 (fnearbyint FR64:$src)),
6204 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6205 def : Pat<(f32 (fceil FR32:$src)),
6206 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6207 def : Pat<(f64 (fceil FR64:$src)),
6208 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6209 def : Pat<(f32 (frint FR32:$src)),
6210 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6211 def : Pat<(f64 (frint FR64:$src)),
6212 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6213 def : Pat<(f32 (ftrunc FR32:$src)),
6214 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6215 def : Pat<(f64 (ftrunc FR64:$src)),
6216 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6219 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6220 memopv4f32, memopv2f64,
6221 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6222 let Constraints = "$src1 = $dst" in
6223 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6224 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6226 def : Pat<(ffloor FR32:$src),
6227 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6228 def : Pat<(f64 (ffloor FR64:$src)),
6229 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6230 def : Pat<(f32 (fnearbyint FR32:$src)),
6231 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6232 def : Pat<(f64 (fnearbyint FR64:$src)),
6233 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6234 def : Pat<(f32 (fceil FR32:$src)),
6235 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6236 def : Pat<(f64 (fceil FR64:$src)),
6237 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6238 def : Pat<(f32 (frint FR32:$src)),
6239 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6240 def : Pat<(f64 (frint FR64:$src)),
6241 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6242 def : Pat<(f32 (ftrunc FR32:$src)),
6243 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6244 def : Pat<(f64 (ftrunc FR64:$src)),
6245 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6247 //===----------------------------------------------------------------------===//
6248 // SSE4.1 - Packed Bit Test
6249 //===----------------------------------------------------------------------===//
6251 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6252 // the intel intrinsic that corresponds to this.
6253 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6254 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6255 "vptest\t{$src2, $src1|$src1, $src2}",
6256 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6258 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6259 "vptest\t{$src2, $src1|$src1, $src2}",
6260 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6263 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6264 "vptest\t{$src2, $src1|$src1, $src2}",
6265 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6267 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6268 "vptest\t{$src2, $src1|$src1, $src2}",
6269 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6273 let Defs = [EFLAGS] in {
6274 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6275 "ptest\t{$src2, $src1|$src1, $src2}",
6276 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6278 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6279 "ptest\t{$src2, $src1|$src1, $src2}",
6280 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6284 // The bit test instructions below are AVX only
6285 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6286 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6287 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6288 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6289 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6290 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6291 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6292 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6296 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6297 let ExeDomain = SSEPackedSingle in {
6298 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6299 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6301 let ExeDomain = SSEPackedDouble in {
6302 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6303 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6307 //===----------------------------------------------------------------------===//
6308 // SSE4.1 - Misc Instructions
6309 //===----------------------------------------------------------------------===//
6311 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6312 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6313 "popcnt{w}\t{$src, $dst|$dst, $src}",
6314 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6316 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6317 "popcnt{w}\t{$src, $dst|$dst, $src}",
6318 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6319 (implicit EFLAGS)]>, OpSize, XS;
6321 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6322 "popcnt{l}\t{$src, $dst|$dst, $src}",
6323 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6325 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6326 "popcnt{l}\t{$src, $dst|$dst, $src}",
6327 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6328 (implicit EFLAGS)]>, XS;
6330 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6331 "popcnt{q}\t{$src, $dst|$dst, $src}",
6332 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6334 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6335 "popcnt{q}\t{$src, $dst|$dst, $src}",
6336 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6337 (implicit EFLAGS)]>, XS;
6342 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6343 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6344 Intrinsic IntId128> {
6345 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6347 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6348 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6349 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6351 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6354 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6357 let Predicates = [HasAVX] in
6358 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6359 int_x86_sse41_phminposuw>, VEX;
6360 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6361 int_x86_sse41_phminposuw>;
6363 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6364 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6365 Intrinsic IntId128, bit Is2Addr = 1> {
6366 let isCommutable = 1 in
6367 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6368 (ins VR128:$src1, VR128:$src2),
6370 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6371 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6372 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6373 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6374 (ins VR128:$src1, i128mem:$src2),
6376 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6377 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6379 (IntId128 VR128:$src1,
6380 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6383 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6384 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6385 Intrinsic IntId256> {
6386 let isCommutable = 1 in
6387 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6388 (ins VR256:$src1, VR256:$src2),
6389 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6390 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6391 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6392 (ins VR256:$src1, i256mem:$src2),
6393 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6395 (IntId256 VR256:$src1,
6396 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6399 let Predicates = [HasAVX] in {
6400 let isCommutable = 0 in
6401 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6403 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6405 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6407 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6409 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6411 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6413 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6415 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6417 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6419 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6421 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6424 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6425 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6426 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6427 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6430 let Predicates = [HasAVX2] in {
6431 let isCommutable = 0 in
6432 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6433 int_x86_avx2_packusdw>, VEX_4V;
6434 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6435 int_x86_avx2_pcmpeq_q>, VEX_4V;
6436 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6437 int_x86_avx2_pmins_b>, VEX_4V;
6438 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6439 int_x86_avx2_pmins_d>, VEX_4V;
6440 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6441 int_x86_avx2_pminu_d>, VEX_4V;
6442 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6443 int_x86_avx2_pminu_w>, VEX_4V;
6444 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6445 int_x86_avx2_pmaxs_b>, VEX_4V;
6446 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6447 int_x86_avx2_pmaxs_d>, VEX_4V;
6448 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6449 int_x86_avx2_pmaxu_d>, VEX_4V;
6450 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6451 int_x86_avx2_pmaxu_w>, VEX_4V;
6452 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6453 int_x86_avx2_pmul_dq>, VEX_4V;
6455 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, VR256:$src2)),
6456 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6457 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, (memop addr:$src2))),
6458 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6461 let Constraints = "$src1 = $dst" in {
6462 let isCommutable = 0 in
6463 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6464 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6465 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6466 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6467 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6468 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6469 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6470 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6471 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6472 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6473 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6476 let Predicates = [HasSSE41] in {
6477 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6478 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6479 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6480 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6483 /// SS48I_binop_rm - Simple SSE41 binary operator.
6484 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6485 ValueType OpVT, bit Is2Addr = 1> {
6486 let isCommutable = 1 in
6487 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6488 (ins VR128:$src1, VR128:$src2),
6490 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6491 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6492 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6494 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6495 (ins VR128:$src1, i128mem:$src2),
6497 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6498 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6499 [(set VR128:$dst, (OpNode VR128:$src1,
6500 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6504 /// SS48I_binop_rm - Simple SSE41 binary operator.
6505 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6507 let isCommutable = 1 in
6508 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6509 (ins VR256:$src1, VR256:$src2),
6510 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6511 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6513 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6514 (ins VR256:$src1, i256mem:$src2),
6515 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6516 [(set VR256:$dst, (OpNode VR256:$src1,
6517 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6521 let Predicates = [HasAVX] in
6522 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6523 let Predicates = [HasAVX2] in
6524 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6525 let Constraints = "$src1 = $dst" in
6526 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6528 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6529 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6530 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6531 X86MemOperand x86memop, bit Is2Addr = 1> {
6532 let isCommutable = 1 in
6533 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6534 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6536 !strconcat(OpcodeStr,
6537 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6538 !strconcat(OpcodeStr,
6539 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6540 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6542 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6543 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6545 !strconcat(OpcodeStr,
6546 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6547 !strconcat(OpcodeStr,
6548 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6551 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6555 let Predicates = [HasAVX] in {
6556 let isCommutable = 0 in {
6557 let ExeDomain = SSEPackedSingle in {
6558 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6559 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6560 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6561 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6563 let ExeDomain = SSEPackedDouble in {
6564 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6565 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6566 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6567 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6569 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6570 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6571 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6572 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6574 let ExeDomain = SSEPackedSingle in
6575 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6576 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6577 let ExeDomain = SSEPackedDouble in
6578 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6579 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6580 let ExeDomain = SSEPackedSingle in
6581 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6582 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6585 let Predicates = [HasAVX2] in {
6586 let isCommutable = 0 in {
6587 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6588 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6589 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6590 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6594 let Constraints = "$src1 = $dst" in {
6595 let isCommutable = 0 in {
6596 let ExeDomain = SSEPackedSingle in
6597 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6598 VR128, memopv4f32, i128mem>;
6599 let ExeDomain = SSEPackedDouble in
6600 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6601 VR128, memopv2f64, i128mem>;
6602 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6603 VR128, memopv2i64, i128mem>;
6604 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6605 VR128, memopv2i64, i128mem>;
6607 let ExeDomain = SSEPackedSingle in
6608 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6609 VR128, memopv4f32, i128mem>;
6610 let ExeDomain = SSEPackedDouble in
6611 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6612 VR128, memopv2f64, i128mem>;
6615 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6616 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6617 RegisterClass RC, X86MemOperand x86memop,
6618 PatFrag mem_frag, Intrinsic IntId> {
6619 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6620 (ins RC:$src1, RC:$src2, RC:$src3),
6621 !strconcat(OpcodeStr,
6622 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6623 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6624 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6626 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6627 (ins RC:$src1, x86memop:$src2, RC:$src3),
6628 !strconcat(OpcodeStr,
6629 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6631 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6633 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6636 let Predicates = [HasAVX] in {
6637 let ExeDomain = SSEPackedDouble in {
6638 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6639 memopv2f64, int_x86_sse41_blendvpd>;
6640 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6641 memopv4f64, int_x86_avx_blendv_pd_256>;
6642 } // ExeDomain = SSEPackedDouble
6643 let ExeDomain = SSEPackedSingle in {
6644 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6645 memopv4f32, int_x86_sse41_blendvps>;
6646 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6647 memopv8f32, int_x86_avx_blendv_ps_256>;
6648 } // ExeDomain = SSEPackedSingle
6649 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6650 memopv2i64, int_x86_sse41_pblendvb>;
6653 let Predicates = [HasAVX2] in {
6654 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6655 memopv4i64, int_x86_avx2_pblendvb>;
6658 let Predicates = [HasAVX] in {
6659 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6660 (v16i8 VR128:$src2))),
6661 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6662 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6663 (v4i32 VR128:$src2))),
6664 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6665 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6666 (v4f32 VR128:$src2))),
6667 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6668 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6669 (v2i64 VR128:$src2))),
6670 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6671 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6672 (v2f64 VR128:$src2))),
6673 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6674 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6675 (v8i32 VR256:$src2))),
6676 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6677 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6678 (v8f32 VR256:$src2))),
6679 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6680 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6681 (v4i64 VR256:$src2))),
6682 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6683 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6684 (v4f64 VR256:$src2))),
6685 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6688 let Predicates = [HasAVX2] in {
6689 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6690 (v32i8 VR256:$src2))),
6691 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6694 /// SS41I_ternary_int - SSE 4.1 ternary operator
6695 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6696 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6698 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6699 (ins VR128:$src1, VR128:$src2),
6700 !strconcat(OpcodeStr,
6701 "\t{$src2, $dst|$dst, $src2}"),
6702 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6705 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6706 (ins VR128:$src1, i128mem:$src2),
6707 !strconcat(OpcodeStr,
6708 "\t{$src2, $dst|$dst, $src2}"),
6711 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6715 let ExeDomain = SSEPackedDouble in
6716 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6717 int_x86_sse41_blendvpd>;
6718 let ExeDomain = SSEPackedSingle in
6719 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6720 int_x86_sse41_blendvps>;
6721 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6722 int_x86_sse41_pblendvb>;
6724 let Predicates = [HasSSE41] in {
6725 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6726 (v16i8 VR128:$src2))),
6727 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6728 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6729 (v4i32 VR128:$src2))),
6730 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6731 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6732 (v4f32 VR128:$src2))),
6733 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6734 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6735 (v2i64 VR128:$src2))),
6736 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6737 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6738 (v2f64 VR128:$src2))),
6739 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6742 let Predicates = [HasAVX] in
6743 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6744 "vmovntdqa\t{$src, $dst|$dst, $src}",
6745 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6747 let Predicates = [HasAVX2] in
6748 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6749 "vmovntdqa\t{$src, $dst|$dst, $src}",
6750 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6752 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6753 "movntdqa\t{$src, $dst|$dst, $src}",
6754 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6757 //===----------------------------------------------------------------------===//
6758 // SSE4.2 - Compare Instructions
6759 //===----------------------------------------------------------------------===//
6761 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6762 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6763 Intrinsic IntId128, bit Is2Addr = 1> {
6764 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6765 (ins VR128:$src1, VR128:$src2),
6767 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6768 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6769 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6771 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6772 (ins VR128:$src1, i128mem:$src2),
6774 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6775 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6777 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6780 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6781 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6782 Intrinsic IntId256> {
6783 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6784 (ins VR256:$src1, VR256:$src2),
6785 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6786 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6788 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6789 (ins VR256:$src1, i256mem:$src2),
6790 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6792 (IntId256 VR256:$src1, (memopv4i64 addr:$src2)))]>, OpSize;
6795 let Predicates = [HasAVX] in {
6796 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6799 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6800 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6801 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6802 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6805 let Predicates = [HasAVX2] in {
6806 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6809 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, VR256:$src2)),
6810 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6811 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, (memop addr:$src2))),
6812 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6815 let Constraints = "$src1 = $dst" in
6816 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6818 let Predicates = [HasSSE42] in {
6819 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6820 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6821 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6822 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6825 //===----------------------------------------------------------------------===//
6826 // SSE4.2 - String/text Processing Instructions
6827 //===----------------------------------------------------------------------===//
6829 // Packed Compare Implicit Length Strings, Return Mask
6830 multiclass pseudo_pcmpistrm<string asm> {
6831 def REG : PseudoI<(outs VR128:$dst),
6832 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6833 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6835 def MEM : PseudoI<(outs VR128:$dst),
6836 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6837 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6838 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6841 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6842 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6843 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6846 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6847 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6848 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6849 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6851 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6852 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6853 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6856 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6857 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6858 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6859 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6861 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6862 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6863 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6866 // Packed Compare Explicit Length Strings, Return Mask
6867 multiclass pseudo_pcmpestrm<string asm> {
6868 def REG : PseudoI<(outs VR128:$dst),
6869 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6870 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6871 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6872 def MEM : PseudoI<(outs VR128:$dst),
6873 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6874 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6875 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6878 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6879 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6880 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6883 let Predicates = [HasAVX],
6884 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6885 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6886 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6887 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6889 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6890 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6891 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6894 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6895 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6896 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6897 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6899 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6900 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6901 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6904 // Packed Compare Implicit Length Strings, Return Index
6905 let Defs = [ECX, EFLAGS] in {
6906 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6907 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6908 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6909 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6910 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6911 (implicit EFLAGS)]>, OpSize;
6912 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6913 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6914 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6915 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6916 (implicit EFLAGS)]>, OpSize;
6920 let Predicates = [HasAVX] in {
6921 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6923 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6925 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6927 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6929 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6931 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6935 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6936 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6937 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6938 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6939 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6940 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6942 // Packed Compare Explicit Length Strings, Return Index
6943 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6944 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6945 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6946 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6947 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6948 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6949 (implicit EFLAGS)]>, OpSize;
6950 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6951 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6952 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6954 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6955 (implicit EFLAGS)]>, OpSize;
6959 let Predicates = [HasAVX] in {
6960 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6962 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6964 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6966 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6968 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6970 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6974 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6975 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6976 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6977 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6978 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6979 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6981 //===----------------------------------------------------------------------===//
6982 // SSE4.2 - CRC Instructions
6983 //===----------------------------------------------------------------------===//
6985 // No CRC instructions have AVX equivalents
6987 // crc intrinsic instruction
6988 // This set of instructions are only rm, the only difference is the size
6990 let Constraints = "$src1 = $dst" in {
6991 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6992 (ins GR32:$src1, i8mem:$src2),
6993 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6995 (int_x86_sse42_crc32_32_8 GR32:$src1,
6996 (load addr:$src2)))]>;
6997 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6998 (ins GR32:$src1, GR8:$src2),
6999 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7001 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
7002 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7003 (ins GR32:$src1, i16mem:$src2),
7004 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7006 (int_x86_sse42_crc32_32_16 GR32:$src1,
7007 (load addr:$src2)))]>,
7009 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7010 (ins GR32:$src1, GR16:$src2),
7011 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7013 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7015 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7016 (ins GR32:$src1, i32mem:$src2),
7017 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7019 (int_x86_sse42_crc32_32_32 GR32:$src1,
7020 (load addr:$src2)))]>;
7021 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7022 (ins GR32:$src1, GR32:$src2),
7023 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7025 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7026 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7027 (ins GR64:$src1, i8mem:$src2),
7028 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7030 (int_x86_sse42_crc32_64_8 GR64:$src1,
7031 (load addr:$src2)))]>,
7033 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7034 (ins GR64:$src1, GR8:$src2),
7035 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7037 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7039 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7040 (ins GR64:$src1, i64mem:$src2),
7041 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7043 (int_x86_sse42_crc32_64_64 GR64:$src1,
7044 (load addr:$src2)))]>,
7046 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7047 (ins GR64:$src1, GR64:$src2),
7048 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7050 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7054 //===----------------------------------------------------------------------===//
7055 // AES-NI Instructions
7056 //===----------------------------------------------------------------------===//
7058 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7059 Intrinsic IntId128, bit Is2Addr = 1> {
7060 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7061 (ins VR128:$src1, VR128:$src2),
7063 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7064 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7065 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7067 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7068 (ins VR128:$src1, i128mem:$src2),
7070 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7071 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7073 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7076 // Perform One Round of an AES Encryption/Decryption Flow
7077 let Predicates = [HasAVX, HasAES] in {
7078 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7079 int_x86_aesni_aesenc, 0>, VEX_4V;
7080 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7081 int_x86_aesni_aesenclast, 0>, VEX_4V;
7082 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7083 int_x86_aesni_aesdec, 0>, VEX_4V;
7084 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7085 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7088 let Constraints = "$src1 = $dst" in {
7089 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7090 int_x86_aesni_aesenc>;
7091 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7092 int_x86_aesni_aesenclast>;
7093 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7094 int_x86_aesni_aesdec>;
7095 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7096 int_x86_aesni_aesdeclast>;
7099 // Perform the AES InvMixColumn Transformation
7100 let Predicates = [HasAVX, HasAES] in {
7101 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7103 "vaesimc\t{$src1, $dst|$dst, $src1}",
7105 (int_x86_aesni_aesimc VR128:$src1))]>,
7107 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7108 (ins i128mem:$src1),
7109 "vaesimc\t{$src1, $dst|$dst, $src1}",
7110 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7113 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7115 "aesimc\t{$src1, $dst|$dst, $src1}",
7117 (int_x86_aesni_aesimc VR128:$src1))]>,
7119 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7120 (ins i128mem:$src1),
7121 "aesimc\t{$src1, $dst|$dst, $src1}",
7122 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7125 // AES Round Key Generation Assist
7126 let Predicates = [HasAVX, HasAES] in {
7127 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7128 (ins VR128:$src1, i8imm:$src2),
7129 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7131 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7133 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7134 (ins i128mem:$src1, i8imm:$src2),
7135 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7137 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7140 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7141 (ins VR128:$src1, i8imm:$src2),
7142 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7144 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7146 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7147 (ins i128mem:$src1, i8imm:$src2),
7148 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7150 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7153 //===----------------------------------------------------------------------===//
7154 // CLMUL Instructions
7155 //===----------------------------------------------------------------------===//
7157 // Carry-less Multiplication instructions
7158 let neverHasSideEffects = 1 in {
7159 // AVX carry-less Multiplication instructions
7160 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7161 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7162 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7166 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7167 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7168 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7171 let Constraints = "$src1 = $dst" in {
7172 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7173 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7174 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7178 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7179 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7180 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7182 } // Constraints = "$src1 = $dst"
7183 } // neverHasSideEffects = 1
7186 multiclass pclmul_alias<string asm, int immop> {
7187 def : InstAlias<!strconcat("pclmul", asm,
7188 "dq {$src, $dst|$dst, $src}"),
7189 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7191 def : InstAlias<!strconcat("pclmul", asm,
7192 "dq {$src, $dst|$dst, $src}"),
7193 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7195 def : InstAlias<!strconcat("vpclmul", asm,
7196 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7197 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7199 def : InstAlias<!strconcat("vpclmul", asm,
7200 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7201 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7203 defm : pclmul_alias<"hqhq", 0x11>;
7204 defm : pclmul_alias<"hqlq", 0x01>;
7205 defm : pclmul_alias<"lqhq", 0x10>;
7206 defm : pclmul_alias<"lqlq", 0x00>;
7208 //===----------------------------------------------------------------------===//
7210 //===----------------------------------------------------------------------===//
7212 //===----------------------------------------------------------------------===//
7213 // VBROADCAST - Load from memory and broadcast to all elements of the
7214 // destination operand
7216 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7217 X86MemOperand x86memop, Intrinsic Int> :
7218 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7219 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7220 [(set RC:$dst, (Int addr:$src))]>, VEX;
7222 // AVX2 adds register forms
7223 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7225 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7226 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7227 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7229 let ExeDomain = SSEPackedSingle in {
7230 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7231 int_x86_avx_vbroadcast_ss>;
7232 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7233 int_x86_avx_vbroadcast_ss_256>;
7235 let ExeDomain = SSEPackedDouble in
7236 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7237 int_x86_avx_vbroadcast_sd_256>;
7238 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7239 int_x86_avx_vbroadcastf128_pd_256>;
7241 let ExeDomain = SSEPackedSingle in {
7242 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7243 int_x86_avx2_vbroadcast_ss_ps>;
7244 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7245 int_x86_avx2_vbroadcast_ss_ps_256>;
7247 let ExeDomain = SSEPackedDouble in
7248 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7249 int_x86_avx2_vbroadcast_sd_pd_256>;
7251 let Predicates = [HasAVX2] in
7252 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7253 int_x86_avx2_vbroadcasti128>;
7255 let Predicates = [HasAVX] in
7256 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7257 (VBROADCASTF128 addr:$src)>;
7260 //===----------------------------------------------------------------------===//
7261 // VINSERTF128 - Insert packed floating-point values
7263 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7264 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7265 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7266 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7269 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7270 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7271 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7275 let Predicates = [HasAVX] in {
7276 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7277 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7278 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7279 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7280 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7281 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7284 //===----------------------------------------------------------------------===//
7285 // VEXTRACTF128 - Extract packed floating-point values
7287 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7288 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7289 (ins VR256:$src1, i8imm:$src2),
7290 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7293 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7294 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7295 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7299 let Predicates = [HasAVX] in {
7300 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7301 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7302 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7303 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7304 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7305 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7308 //===----------------------------------------------------------------------===//
7309 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7311 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7312 Intrinsic IntLd, Intrinsic IntLd256,
7313 Intrinsic IntSt, Intrinsic IntSt256> {
7314 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7315 (ins VR128:$src1, f128mem:$src2),
7316 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7317 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7319 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7320 (ins VR256:$src1, f256mem:$src2),
7321 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7322 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7324 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7325 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7326 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7327 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7328 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7329 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7330 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7331 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7334 let ExeDomain = SSEPackedSingle in
7335 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7336 int_x86_avx_maskload_ps,
7337 int_x86_avx_maskload_ps_256,
7338 int_x86_avx_maskstore_ps,
7339 int_x86_avx_maskstore_ps_256>;
7340 let ExeDomain = SSEPackedDouble in
7341 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7342 int_x86_avx_maskload_pd,
7343 int_x86_avx_maskload_pd_256,
7344 int_x86_avx_maskstore_pd,
7345 int_x86_avx_maskstore_pd_256>;
7347 //===----------------------------------------------------------------------===//
7348 // VPERMIL - Permute Single and Double Floating-Point Values
7350 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7351 RegisterClass RC, X86MemOperand x86memop_f,
7352 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7353 Intrinsic IntVar, Intrinsic IntImm> {
7354 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7355 (ins RC:$src1, RC:$src2),
7356 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7357 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7358 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7359 (ins RC:$src1, x86memop_i:$src2),
7360 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7361 [(set RC:$dst, (IntVar RC:$src1,
7362 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7364 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7365 (ins RC:$src1, i8imm:$src2),
7366 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7367 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7368 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7369 (ins x86memop_f:$src1, i8imm:$src2),
7370 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7371 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7374 let ExeDomain = SSEPackedSingle in {
7375 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7376 memopv4f32, memopv2i64,
7377 int_x86_avx_vpermilvar_ps,
7378 int_x86_avx_vpermil_ps>;
7379 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7380 memopv8f32, memopv4i64,
7381 int_x86_avx_vpermilvar_ps_256,
7382 int_x86_avx_vpermil_ps_256>;
7384 let ExeDomain = SSEPackedDouble in {
7385 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7386 memopv2f64, memopv2i64,
7387 int_x86_avx_vpermilvar_pd,
7388 int_x86_avx_vpermil_pd>;
7389 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7390 memopv4f64, memopv4i64,
7391 int_x86_avx_vpermilvar_pd_256,
7392 int_x86_avx_vpermil_pd_256>;
7395 let Predicates = [HasAVX] in {
7396 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7397 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7398 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7399 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7400 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7401 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7402 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7403 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7404 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7405 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7406 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7407 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7408 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7410 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7411 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7412 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7415 //===----------------------------------------------------------------------===//
7416 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7418 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7419 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7420 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7421 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7424 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7425 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7426 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7430 let Predicates = [HasAVX] in {
7431 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7432 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7433 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7434 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7435 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7436 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7438 def : Pat<(int_x86_avx_vperm2f128_ps_256
7439 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7440 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7441 def : Pat<(int_x86_avx_vperm2f128_pd_256
7442 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7443 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7444 def : Pat<(int_x86_avx_vperm2f128_si_256
7445 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7446 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7449 //===----------------------------------------------------------------------===//
7450 // VZERO - Zero YMM registers
7452 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7453 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7454 // Zero All YMM registers
7455 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7456 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7458 // Zero Upper bits of YMM registers
7459 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7460 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7463 //===----------------------------------------------------------------------===//
7464 // Half precision conversion instructions
7465 //===----------------------------------------------------------------------===//
7466 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7467 let Predicates = [HasAVX, HasF16C] in {
7468 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7469 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7470 [(set RC:$dst, (Int VR128:$src))]>,
7472 let neverHasSideEffects = 1, mayLoad = 1 in
7473 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7474 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7478 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7479 let Predicates = [HasAVX, HasF16C] in {
7480 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7481 (ins RC:$src1, i32i8imm:$src2),
7482 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7483 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7485 let neverHasSideEffects = 1, mayLoad = 1 in
7486 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7487 (ins RC:$src1, i32i8imm:$src2),
7488 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7493 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7494 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7495 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7496 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7498 //===----------------------------------------------------------------------===//
7499 // AVX2 Instructions
7500 //===----------------------------------------------------------------------===//
7502 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7503 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7504 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7505 X86MemOperand x86memop> {
7506 let isCommutable = 1 in
7507 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7508 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7509 !strconcat(OpcodeStr,
7510 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7511 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7513 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7514 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7515 !strconcat(OpcodeStr,
7516 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7519 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7523 let isCommutable = 0 in {
7524 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7525 VR128, memopv2i64, i128mem>;
7526 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7527 VR256, memopv4i64, i256mem>;
7530 //===----------------------------------------------------------------------===//
7531 // VPBROADCAST - Load from memory and broadcast to all elements of the
7532 // destination operand
7534 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7535 X86MemOperand x86memop, PatFrag ld_frag,
7536 Intrinsic Int128, Intrinsic Int256> {
7537 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7538 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7539 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7540 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7541 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7543 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7544 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7545 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7546 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7547 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7548 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7550 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7553 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7554 int_x86_avx2_pbroadcastb_128,
7555 int_x86_avx2_pbroadcastb_256>;
7556 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7557 int_x86_avx2_pbroadcastw_128,
7558 int_x86_avx2_pbroadcastw_256>;
7559 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7560 int_x86_avx2_pbroadcastd_128,
7561 int_x86_avx2_pbroadcastd_256>;
7562 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7563 int_x86_avx2_pbroadcastq_128,
7564 int_x86_avx2_pbroadcastq_256>;
7566 let Predicates = [HasAVX2] in {
7567 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7568 (VPBROADCASTBrm addr:$src)>;
7569 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7570 (VPBROADCASTBYrm addr:$src)>;
7571 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7572 (VPBROADCASTWrm addr:$src)>;
7573 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7574 (VPBROADCASTWYrm addr:$src)>;
7575 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7576 (VPBROADCASTDrm addr:$src)>;
7577 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7578 (VPBROADCASTDYrm addr:$src)>;
7579 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7580 (VPBROADCASTQrm addr:$src)>;
7581 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7582 (VPBROADCASTQYrm addr:$src)>;
7585 // AVX1 broadcast patterns
7586 let Predicates = [HasAVX] in {
7587 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7588 (VBROADCASTSSYrm addr:$src)>;
7589 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7590 (VBROADCASTSDrm addr:$src)>;
7591 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7592 (VBROADCASTSSYrm addr:$src)>;
7593 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7594 (VBROADCASTSDrm addr:$src)>;
7596 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7597 (VBROADCASTSSrm addr:$src)>;
7598 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7599 (VBROADCASTSSrm addr:$src)>;
7602 //===----------------------------------------------------------------------===//
7603 // VPERM - Permute instructions
7606 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7608 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7609 (ins VR256:$src1, VR256:$src2),
7610 !strconcat(OpcodeStr,
7611 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7612 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7613 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7614 (ins VR256:$src1, i256mem:$src2),
7615 !strconcat(OpcodeStr,
7616 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7617 [(set VR256:$dst, (Int VR256:$src1,
7618 (bitconvert (mem_frag addr:$src2))))]>,
7622 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7623 let ExeDomain = SSEPackedSingle in
7624 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7626 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7628 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7629 (ins VR256:$src1, i8imm:$src2),
7630 !strconcat(OpcodeStr,
7631 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7632 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7633 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7634 (ins i256mem:$src1, i8imm:$src2),
7635 !strconcat(OpcodeStr,
7636 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7637 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7641 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7643 let ExeDomain = SSEPackedDouble in
7644 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7647 //===----------------------------------------------------------------------===//
7648 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7650 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7651 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7652 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7654 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7656 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7657 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7658 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7660 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7664 let Predicates = [HasAVX2] in {
7665 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7666 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7667 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7668 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7669 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7670 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7671 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7672 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7674 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7676 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7677 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7678 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7679 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7680 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7682 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7683 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7685 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7689 let Predicates = [HasAVX] in {
7690 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7691 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7692 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7693 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7694 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7695 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7696 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7697 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7698 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7699 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7700 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7701 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7703 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7704 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7705 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7706 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7707 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7708 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7709 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7710 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7711 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7712 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7713 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7714 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7715 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7716 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7717 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7718 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7719 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7720 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7724 //===----------------------------------------------------------------------===//
7725 // VINSERTI128 - Insert packed integer values
7727 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7728 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7729 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7731 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7733 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7734 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7735 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7737 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7738 imm:$src3))]>, VEX_4V;
7740 let Predicates = [HasAVX2] in {
7741 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7743 (VINSERTI128rr VR256:$src1, VR128:$src2,
7744 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7745 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7747 (VINSERTI128rr VR256:$src1, VR128:$src2,
7748 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7749 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7751 (VINSERTI128rr VR256:$src1, VR128:$src2,
7752 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7753 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7755 (VINSERTI128rr VR256:$src1, VR128:$src2,
7756 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7760 let Predicates = [HasAVX] in {
7761 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7763 (VINSERTF128rr VR256:$src1, VR128:$src2,
7764 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7765 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7767 (VINSERTF128rr VR256:$src1, VR128:$src2,
7768 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7769 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7771 (VINSERTF128rr VR256:$src1, VR128:$src2,
7772 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7773 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7775 (VINSERTF128rr VR256:$src1, VR128:$src2,
7776 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7777 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7779 (VINSERTF128rr VR256:$src1, VR128:$src2,
7780 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7781 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7783 (VINSERTF128rr VR256:$src1, VR128:$src2,
7784 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7787 //===----------------------------------------------------------------------===//
7788 // VEXTRACTI128 - Extract packed integer values
7790 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7791 (ins VR256:$src1, i8imm:$src2),
7792 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7794 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7796 let neverHasSideEffects = 1, mayStore = 1 in
7797 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7798 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7799 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7801 let Predicates = [HasAVX2] in {
7802 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7803 (v2i64 (VEXTRACTI128rr
7804 (v4i64 VR256:$src1),
7805 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7806 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7807 (v4i32 (VEXTRACTI128rr
7808 (v8i32 VR256:$src1),
7809 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7810 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7811 (v8i16 (VEXTRACTI128rr
7812 (v16i16 VR256:$src1),
7813 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7814 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7815 (v16i8 (VEXTRACTI128rr
7816 (v32i8 VR256:$src1),
7817 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7821 let Predicates = [HasAVX] in {
7822 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7823 (v4f32 (VEXTRACTF128rr
7824 (v8f32 VR256:$src1),
7825 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7826 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7827 (v2f64 (VEXTRACTF128rr
7828 (v4f64 VR256:$src1),
7829 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7830 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7831 (v2i64 (VEXTRACTF128rr
7832 (v4i64 VR256:$src1),
7833 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7834 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7835 (v4i32 (VEXTRACTF128rr
7836 (v8i32 VR256:$src1),
7837 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7838 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7839 (v8i16 (VEXTRACTF128rr
7840 (v16i16 VR256:$src1),
7841 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7842 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7843 (v16i8 (VEXTRACTF128rr
7844 (v32i8 VR256:$src1),
7845 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7848 //===----------------------------------------------------------------------===//
7849 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7851 multiclass avx2_pmovmask<string OpcodeStr,
7852 Intrinsic IntLd128, Intrinsic IntLd256,
7853 Intrinsic IntSt128, Intrinsic IntSt256> {
7854 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7855 (ins VR128:$src1, i128mem:$src2),
7856 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7857 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7858 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7859 (ins VR256:$src1, i256mem:$src2),
7860 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7861 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7862 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7863 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7864 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7865 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7866 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7867 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7868 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7869 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7872 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7873 int_x86_avx2_maskload_d,
7874 int_x86_avx2_maskload_d_256,
7875 int_x86_avx2_maskstore_d,
7876 int_x86_avx2_maskstore_d_256>;
7877 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7878 int_x86_avx2_maskload_q,
7879 int_x86_avx2_maskload_q_256,
7880 int_x86_avx2_maskstore_q,
7881 int_x86_avx2_maskstore_q_256>, VEX_W;
7884 //===----------------------------------------------------------------------===//
7885 // Variable Bit Shifts
7887 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7888 ValueType vt128, ValueType vt256> {
7889 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7890 (ins VR128:$src1, VR128:$src2),
7891 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7893 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7895 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7896 (ins VR128:$src1, i128mem:$src2),
7897 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7899 (vt128 (OpNode VR128:$src1,
7900 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7902 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7903 (ins VR256:$src1, VR256:$src2),
7904 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7906 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7908 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7909 (ins VR256:$src1, i256mem:$src2),
7910 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7912 (vt256 (OpNode VR256:$src1,
7913 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7917 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7918 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7919 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7920 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7921 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;