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:$dst), (ins), "", []>;
266 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
267 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
268 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
269 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
270 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
271 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
274 // The same as done above but for AVX. The 256-bit ISA does not support PI,
275 // and doesn't need it because on sandy bridge the register is set to zero
276 // at the rename stage without using any execution unit, so SET0PSY
277 // and SET0PDY can be used for vector int instructions without penalty
278 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
279 // JIT implementatioan, it does not expand the instructions below like
280 // X86MCInstLower does.
281 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
282 isCodeGenOnly = 1, Predicates = [HasAVX] in {
283 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
284 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
285 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
286 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
290 // AVX has no support for 256-bit integer instructions, but since the 128-bit
291 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
292 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
293 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
294 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
296 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
297 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
298 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
300 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
301 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
302 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
304 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
305 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
306 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
308 // We set canFoldAsLoad because this can be converted to a constant-pool
309 // load of an all-ones value if folding it would be beneficial.
310 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
311 // JIT implementation, it does not expand the instructions below like
312 // X86MCInstLower does.
313 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
314 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
315 let Predicates = [HasAVX] in
316 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
317 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
318 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
319 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
320 let Predicates = [HasAVX2] in
321 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
322 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
326 //===----------------------------------------------------------------------===//
327 // SSE 1 & 2 - Move FP Scalar Instructions
329 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
330 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
331 // is used instead. Register-to-register movss/movsd is not modeled as an
332 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
333 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
334 //===----------------------------------------------------------------------===//
336 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
337 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
338 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
340 // Loading from memory automatically zeroing upper bits.
341 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
342 PatFrag mem_pat, string OpcodeStr> :
343 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
344 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
345 [(set RC:$dst, (mem_pat addr:$src))]>;
348 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
349 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
351 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
352 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
355 // For the disassembler
356 let isCodeGenOnly = 1 in {
357 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
358 (ins VR128:$src1, FR32:$src2),
359 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
361 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
362 (ins VR128:$src1, FR64:$src2),
363 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
367 let canFoldAsLoad = 1, isReMaterializable = 1 in {
368 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
370 let AddedComplexity = 20 in
371 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
375 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
376 "movss\t{$src, $dst|$dst, $src}",
377 [(store FR32:$src, addr:$dst)]>, XS, VEX, VEX_LIG;
378 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
379 "movsd\t{$src, $dst|$dst, $src}",
380 [(store FR64:$src, addr:$dst)]>, XD, VEX, VEX_LIG;
383 let Constraints = "$src1 = $dst" in {
384 def MOVSSrr : sse12_move_rr<FR32, v4f32,
385 "movss\t{$src2, $dst|$dst, $src2}">, XS;
386 def MOVSDrr : sse12_move_rr<FR64, v2f64,
387 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
389 // For the disassembler
390 let isCodeGenOnly = 1 in {
391 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
392 (ins VR128:$src1, FR32:$src2),
393 "movss\t{$src2, $dst|$dst, $src2}", []>, XS;
394 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
395 (ins VR128:$src1, FR64:$src2),
396 "movsd\t{$src2, $dst|$dst, $src2}", []>, XD;
400 let canFoldAsLoad = 1, isReMaterializable = 1 in {
401 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
403 let AddedComplexity = 20 in
404 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
407 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
408 "movss\t{$src, $dst|$dst, $src}",
409 [(store FR32:$src, addr:$dst)]>;
410 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
411 "movsd\t{$src, $dst|$dst, $src}",
412 [(store FR64:$src, addr:$dst)]>;
415 let Predicates = [HasAVX] in {
416 let AddedComplexity = 15 in {
417 // Extract the low 32-bit value from one vector and insert it into another.
418 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
419 (VMOVSSrr (v4f32 VR128:$src1),
420 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
421 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
422 (VMOVSSrr (v4i32 VR128:$src1),
423 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
425 // Extract the low 64-bit value from one vector and insert it into another.
426 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
427 (VMOVSDrr (v2f64 VR128:$src1),
428 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
429 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
430 (VMOVSDrr (v2i64 VR128:$src1),
431 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
433 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
434 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
435 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
436 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
437 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
439 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
440 // MOVS{S,D} to the lower bits.
441 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
442 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
443 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
444 (VMOVSSrr (v4f32 (V_SET0)),
445 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
446 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
447 (VMOVSSrr (v4i32 (V_SET0)),
448 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
449 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
450 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
452 // Move low f32 and clear high bits.
453 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
454 (SUBREG_TO_REG (i32 0),
455 (VMOVSSrr (v4f32 (V_SET0)),
456 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
457 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
458 (SUBREG_TO_REG (i32 0),
459 (VMOVSSrr (v4i32 (V_SET0)),
460 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
463 let AddedComplexity = 20 in {
464 // MOVSSrm zeros the high parts of the register; represent this
465 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
466 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
467 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
468 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
469 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
470 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
471 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
473 // MOVSDrm zeros the high parts of the register; represent this
474 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
475 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
476 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
477 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
478 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
479 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
480 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
481 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
482 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
483 def : Pat<(v2f64 (X86vzload addr:$src)),
484 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
486 // Represent the same patterns above but in the form they appear for
488 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
489 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
490 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
491 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
492 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
493 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
494 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
495 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
496 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
498 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
499 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
500 (SUBREG_TO_REG (i32 0),
501 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
503 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
504 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
505 (SUBREG_TO_REG (i64 0),
506 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
508 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
509 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
510 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
512 // Move low f64 and clear high bits.
513 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
514 (SUBREG_TO_REG (i32 0),
515 (VMOVSDrr (v2f64 (V_SET0)),
516 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
518 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
519 (SUBREG_TO_REG (i32 0),
520 (VMOVSDrr (v2i64 (V_SET0)),
521 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_sd)), sub_xmm)>;
523 // Extract and store.
524 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
527 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
528 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
531 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
533 // Shuffle with VMOVSS
534 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
535 (VMOVSSrr VR128:$src1, FR32:$src2)>;
536 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
537 (VMOVSSrr (v4i32 VR128:$src1),
538 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
539 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
540 (VMOVSSrr (v4f32 VR128:$src1),
541 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
544 def : Pat<(v8i32 (X86Movsd VR256:$src1, VR256:$src2)),
545 (SUBREG_TO_REG (i32 0),
546 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
547 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
548 def : Pat<(v8f32 (X86Movsd VR256:$src1, VR256:$src2)),
549 (SUBREG_TO_REG (i32 0),
550 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
551 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
553 // Shuffle with VMOVSD
554 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
555 (VMOVSDrr VR128:$src1, FR64:$src2)>;
556 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
557 (VMOVSDrr (v2i64 VR128:$src1),
558 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
559 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
560 (VMOVSDrr (v2f64 VR128:$src1),
561 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
562 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
563 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
565 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
566 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
570 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
571 (SUBREG_TO_REG (i32 0),
572 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
573 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
574 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
575 (SUBREG_TO_REG (i32 0),
576 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
577 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
580 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
581 // is during lowering, where it's not possible to recognize the fold cause
582 // it has two uses through a bitcast. One use disappears at isel time and the
583 // fold opportunity reappears.
584 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
585 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
587 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
588 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
590 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
591 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
593 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
594 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
598 let Predicates = [HasSSE1] in {
599 let AddedComplexity = 15 in {
600 // Extract the low 32-bit value from one vector and insert it into another.
601 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
602 (MOVSSrr (v4f32 VR128:$src1),
603 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
604 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
605 (MOVSSrr (v4i32 VR128:$src1),
606 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
608 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
609 // MOVSS to the lower bits.
610 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
611 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
612 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
613 (MOVSSrr (v4f32 (V_SET0)),
614 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
615 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
616 (MOVSSrr (v4i32 (V_SET0)),
617 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
620 let AddedComplexity = 20 in {
621 // MOVSSrm zeros the high parts of the register; represent this
622 // with SUBREG_TO_REG.
623 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
624 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
625 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
626 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
627 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
628 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
631 // Extract and store.
632 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
635 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
637 // Shuffle with MOVSS
638 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
639 (MOVSSrr VR128:$src1, FR32:$src2)>;
640 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
641 (MOVSSrr (v4i32 VR128:$src1),
642 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
643 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
644 (MOVSSrr (v4f32 VR128:$src1),
645 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
648 let Predicates = [HasSSE2] in {
649 let AddedComplexity = 15 in {
650 // Extract the low 64-bit value from one vector and insert it into another.
651 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
652 (MOVSDrr (v2f64 VR128:$src1),
653 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
654 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
655 (MOVSDrr (v2i64 VR128:$src1),
656 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
658 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
659 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
660 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
661 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
662 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
664 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
665 // MOVSD to the lower bits.
666 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
667 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
670 let AddedComplexity = 20 in {
671 // MOVSDrm zeros the high parts of the register; represent this
672 // with SUBREG_TO_REG.
673 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
674 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
675 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
676 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
677 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
678 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
679 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
680 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
681 def : Pat<(v2f64 (X86vzload addr:$src)),
682 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
685 // Extract and store.
686 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
689 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
691 // Shuffle with MOVSD
692 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
693 (MOVSDrr VR128:$src1, FR64:$src2)>;
694 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
695 (MOVSDrr (v2i64 VR128:$src1),
696 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
697 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
698 (MOVSDrr (v2f64 VR128:$src1),
699 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
700 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
701 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
702 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
703 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
705 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
706 // is during lowering, where it's not possible to recognize the fold cause
707 // it has two uses through a bitcast. One use disappears at isel time and the
708 // fold opportunity reappears.
709 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
710 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
711 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
712 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
713 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
714 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
715 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
716 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
719 //===----------------------------------------------------------------------===//
720 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
721 //===----------------------------------------------------------------------===//
723 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
724 X86MemOperand x86memop, PatFrag ld_frag,
725 string asm, Domain d,
726 bit IsReMaterializable = 1> {
727 let neverHasSideEffects = 1 in
728 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
729 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
730 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
731 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
732 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
733 [(set RC:$dst, (ld_frag addr:$src))], d>;
736 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
737 "movaps", SSEPackedSingle>, TB, VEX;
738 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
739 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
740 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
741 "movups", SSEPackedSingle>, TB, VEX;
742 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
743 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
745 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
746 "movaps", SSEPackedSingle>, TB, VEX;
747 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
748 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
749 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
750 "movups", SSEPackedSingle>, TB, VEX;
751 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
752 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
753 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
754 "movaps", SSEPackedSingle>, TB;
755 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
756 "movapd", SSEPackedDouble>, TB, OpSize;
757 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
758 "movups", SSEPackedSingle>, TB;
759 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
760 "movupd", SSEPackedDouble, 0>, TB, OpSize;
762 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
763 "movaps\t{$src, $dst|$dst, $src}",
764 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
765 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
766 "movapd\t{$src, $dst|$dst, $src}",
767 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
768 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
769 "movups\t{$src, $dst|$dst, $src}",
770 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
771 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
772 "movupd\t{$src, $dst|$dst, $src}",
773 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
774 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
775 "movaps\t{$src, $dst|$dst, $src}",
776 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)]>, VEX;
777 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
778 "movapd\t{$src, $dst|$dst, $src}",
779 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)]>, VEX;
780 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
781 "movups\t{$src, $dst|$dst, $src}",
782 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
783 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
784 "movupd\t{$src, $dst|$dst, $src}",
785 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
788 let isCodeGenOnly = 1 in {
789 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
791 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
792 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
794 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
795 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
797 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
798 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
800 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
801 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
803 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
804 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
806 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
807 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
809 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
810 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
812 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
815 let Predicates = [HasAVX] in {
816 def : Pat<(v8i32 (X86vzmovl
817 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
818 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
819 def : Pat<(v4i64 (X86vzmovl
820 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
821 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
822 def : Pat<(v8f32 (X86vzmovl
823 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
824 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
825 def : Pat<(v4f64 (X86vzmovl
826 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
827 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
831 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
832 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
833 (VMOVUPSYmr addr:$dst, VR256:$src)>;
835 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
836 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
837 (VMOVUPDYmr addr:$dst, VR256:$src)>;
839 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
840 "movaps\t{$src, $dst|$dst, $src}",
841 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
842 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
843 "movapd\t{$src, $dst|$dst, $src}",
844 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
845 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
846 "movups\t{$src, $dst|$dst, $src}",
847 [(store (v4f32 VR128:$src), addr:$dst)]>;
848 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
849 "movupd\t{$src, $dst|$dst, $src}",
850 [(store (v2f64 VR128:$src), addr:$dst)]>;
853 let isCodeGenOnly = 1 in {
854 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
855 "movaps\t{$src, $dst|$dst, $src}", []>;
856 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
857 "movapd\t{$src, $dst|$dst, $src}", []>;
858 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
859 "movups\t{$src, $dst|$dst, $src}", []>;
860 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
861 "movupd\t{$src, $dst|$dst, $src}", []>;
864 let Predicates = [HasAVX] in {
865 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
866 (VMOVUPSmr addr:$dst, VR128:$src)>;
867 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
868 (VMOVUPDmr addr:$dst, VR128:$src)>;
871 let Predicates = [HasSSE1] in
872 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
873 (MOVUPSmr addr:$dst, VR128:$src)>;
874 let Predicates = [HasSSE2] in
875 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
876 (MOVUPDmr addr:$dst, VR128:$src)>;
878 // Use vmovaps/vmovups for AVX integer load/store.
879 let Predicates = [HasAVX] in {
880 // 128-bit load/store
881 def : Pat<(alignedloadv4i32 addr:$src),
882 (VMOVAPSrm addr:$src)>;
883 def : Pat<(loadv4i32 addr:$src),
884 (VMOVUPSrm addr:$src)>;
885 def : Pat<(alignedloadv2i64 addr:$src),
886 (VMOVAPSrm addr:$src)>;
887 def : Pat<(loadv2i64 addr:$src),
888 (VMOVUPSrm addr:$src)>;
890 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
891 (VMOVAPSmr addr:$dst, VR128:$src)>;
892 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
893 (VMOVAPSmr addr:$dst, VR128:$src)>;
894 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
895 (VMOVAPSmr addr:$dst, VR128:$src)>;
896 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
897 (VMOVAPSmr addr:$dst, VR128:$src)>;
898 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
899 (VMOVUPSmr addr:$dst, VR128:$src)>;
900 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
901 (VMOVUPSmr addr:$dst, VR128:$src)>;
902 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
903 (VMOVUPSmr addr:$dst, VR128:$src)>;
904 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
905 (VMOVUPSmr addr:$dst, VR128:$src)>;
907 // 256-bit load/store
908 def : Pat<(alignedloadv4i64 addr:$src),
909 (VMOVAPSYrm addr:$src)>;
910 def : Pat<(loadv4i64 addr:$src),
911 (VMOVUPSYrm addr:$src)>;
912 def : Pat<(alignedloadv8i32 addr:$src),
913 (VMOVAPSYrm addr:$src)>;
914 def : Pat<(loadv8i32 addr:$src),
915 (VMOVUPSYrm addr:$src)>;
916 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
917 (VMOVAPSYmr addr:$dst, VR256:$src)>;
918 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
919 (VMOVAPSYmr addr:$dst, VR256:$src)>;
920 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
921 (VMOVAPSYmr addr:$dst, VR256:$src)>;
922 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
923 (VMOVAPSYmr addr:$dst, VR256:$src)>;
924 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
925 (VMOVUPSYmr addr:$dst, VR256:$src)>;
926 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
927 (VMOVUPSYmr addr:$dst, VR256:$src)>;
928 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
929 (VMOVUPSYmr addr:$dst, VR256:$src)>;
930 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
931 (VMOVUPSYmr addr:$dst, VR256:$src)>;
934 // Use movaps / movups for SSE integer load / store (one byte shorter).
935 // The instructions selected below are then converted to MOVDQA/MOVDQU
936 // during the SSE domain pass.
937 let Predicates = [HasSSE1] in {
938 def : Pat<(alignedloadv4i32 addr:$src),
939 (MOVAPSrm addr:$src)>;
940 def : Pat<(loadv4i32 addr:$src),
941 (MOVUPSrm addr:$src)>;
942 def : Pat<(alignedloadv2i64 addr:$src),
943 (MOVAPSrm addr:$src)>;
944 def : Pat<(loadv2i64 addr:$src),
945 (MOVUPSrm addr:$src)>;
947 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
948 (MOVAPSmr addr:$dst, VR128:$src)>;
949 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
950 (MOVAPSmr addr:$dst, VR128:$src)>;
951 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
952 (MOVAPSmr addr:$dst, VR128:$src)>;
953 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
954 (MOVAPSmr addr:$dst, VR128:$src)>;
955 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
956 (MOVUPSmr addr:$dst, VR128:$src)>;
957 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
958 (MOVUPSmr addr:$dst, VR128:$src)>;
959 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
960 (MOVUPSmr addr:$dst, VR128:$src)>;
961 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
962 (MOVUPSmr addr:$dst, VR128:$src)>;
965 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
966 // bits are disregarded. FIXME: Set encoding to pseudo!
967 let neverHasSideEffects = 1 in {
968 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
969 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
970 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
971 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
972 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
973 "movaps\t{$src, $dst|$dst, $src}", []>;
974 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
975 "movapd\t{$src, $dst|$dst, $src}", []>;
978 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
979 // bits are disregarded. FIXME: Set encoding to pseudo!
980 let canFoldAsLoad = 1, isReMaterializable = 1 in {
981 let isCodeGenOnly = 1 in {
982 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
983 "movaps\t{$src, $dst|$dst, $src}",
984 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
985 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
986 "movapd\t{$src, $dst|$dst, $src}",
987 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
989 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
990 "movaps\t{$src, $dst|$dst, $src}",
991 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
992 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
993 "movapd\t{$src, $dst|$dst, $src}",
994 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
997 //===----------------------------------------------------------------------===//
998 // SSE 1 & 2 - Move Low packed FP Instructions
999 //===----------------------------------------------------------------------===//
1001 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
1002 PatFrag mov_frag, string base_opc,
1004 def PSrm : PI<opc, MRMSrcMem,
1005 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1006 !strconcat(base_opc, "s", asm_opr),
1009 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1010 SSEPackedSingle>, TB;
1012 def PDrm : PI<opc, MRMSrcMem,
1013 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1014 !strconcat(base_opc, "d", asm_opr),
1015 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
1016 (scalar_to_vector (loadf64 addr:$src2)))))],
1017 SSEPackedDouble>, TB, OpSize;
1020 let AddedComplexity = 20 in {
1021 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1022 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1024 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1025 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1026 "\t{$src2, $dst|$dst, $src2}">;
1029 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1030 "movlps\t{$src, $dst|$dst, $src}",
1031 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1032 (iPTR 0))), addr:$dst)]>, VEX;
1033 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1034 "movlpd\t{$src, $dst|$dst, $src}",
1035 [(store (f64 (vector_extract (v2f64 VR128:$src),
1036 (iPTR 0))), addr:$dst)]>, VEX;
1037 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1038 "movlps\t{$src, $dst|$dst, $src}",
1039 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1040 (iPTR 0))), addr:$dst)]>;
1041 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1042 "movlpd\t{$src, $dst|$dst, $src}",
1043 [(store (f64 (vector_extract (v2f64 VR128:$src),
1044 (iPTR 0))), addr:$dst)]>;
1046 let Predicates = [HasAVX] in {
1047 let AddedComplexity = 20 in {
1048 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1049 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1050 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1051 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1052 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1053 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1054 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1055 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1056 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1057 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1060 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1061 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1062 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1063 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1064 VR128:$src2)), addr:$src1),
1065 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1067 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1068 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1069 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1070 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1071 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1073 // Shuffle with VMOVLPS
1074 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1075 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1076 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1077 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1078 def : Pat<(X86Movlps VR128:$src1,
1079 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1080 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1082 // Shuffle with VMOVLPD
1083 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1084 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1085 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1086 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1087 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1088 (scalar_to_vector (loadf64 addr:$src2)))),
1089 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1092 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1094 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1095 def : Pat<(store (v4i32 (X86Movlps
1096 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1097 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1098 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1100 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1101 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1103 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1106 let Predicates = [HasSSE1] in {
1107 let AddedComplexity = 20 in {
1108 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1109 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1110 (MOVLPSrm VR128:$src1, addr:$src2)>;
1111 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1112 (MOVLPSrm VR128:$src1, addr:$src2)>;
1115 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1116 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1117 (iPTR 0))), addr:$src1),
1118 (MOVLPSmr addr:$src1, VR128:$src2)>;
1119 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1120 (MOVLPSmr addr:$src1, VR128:$src2)>;
1121 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1122 VR128:$src2)), addr:$src1),
1123 (MOVLPSmr addr:$src1, VR128:$src2)>;
1125 // Shuffle with MOVLPS
1126 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1127 (MOVLPSrm VR128:$src1, addr:$src2)>;
1128 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1129 (MOVLPSrm VR128:$src1, addr:$src2)>;
1130 def : Pat<(X86Movlps VR128:$src1,
1131 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1132 (MOVLPSrm VR128:$src1, addr:$src2)>;
1133 def : Pat<(X86Movlps VR128:$src1,
1134 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1135 (MOVLPSrm VR128:$src1, addr:$src2)>;
1138 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1140 (MOVLPSmr addr:$src1, VR128:$src2)>;
1141 def : Pat<(store (v4i32 (X86Movlps
1142 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1144 (MOVLPSmr addr:$src1, VR128:$src2)>;
1147 let Predicates = [HasSSE2] in {
1148 let AddedComplexity = 20 in {
1149 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1150 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1151 (MOVLPDrm VR128:$src1, addr:$src2)>;
1152 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1153 (MOVLPDrm VR128:$src1, addr:$src2)>;
1156 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1157 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1158 (MOVLPDmr addr:$src1, VR128:$src2)>;
1159 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1160 (MOVLPDmr addr:$src1, VR128:$src2)>;
1162 // Shuffle with MOVLPD
1163 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1164 (MOVLPDrm VR128:$src1, addr:$src2)>;
1165 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1166 (MOVLPDrm VR128:$src1, addr:$src2)>;
1167 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1168 (scalar_to_vector (loadf64 addr:$src2)))),
1169 (MOVLPDrm VR128:$src1, addr:$src2)>;
1172 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1174 (MOVLPDmr addr:$src1, VR128:$src2)>;
1175 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1177 (MOVLPDmr addr:$src1, VR128:$src2)>;
1180 //===----------------------------------------------------------------------===//
1181 // SSE 1 & 2 - Move Hi packed FP Instructions
1182 //===----------------------------------------------------------------------===//
1184 let AddedComplexity = 20 in {
1185 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1186 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1188 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1189 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1190 "\t{$src2, $dst|$dst, $src2}">;
1193 // v2f64 extract element 1 is always custom lowered to unpack high to low
1194 // and extract element 0 so the non-store version isn't too horrible.
1195 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1196 "movhps\t{$src, $dst|$dst, $src}",
1197 [(store (f64 (vector_extract
1198 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1199 (undef)), (iPTR 0))), addr:$dst)]>,
1201 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1202 "movhpd\t{$src, $dst|$dst, $src}",
1203 [(store (f64 (vector_extract
1204 (v2f64 (unpckh VR128:$src, (undef))),
1205 (iPTR 0))), addr:$dst)]>,
1207 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1208 "movhps\t{$src, $dst|$dst, $src}",
1209 [(store (f64 (vector_extract
1210 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1211 (undef)), (iPTR 0))), addr:$dst)]>;
1212 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$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:$dst)]>;
1218 let Predicates = [HasAVX] in {
1220 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1221 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1222 def : Pat<(X86Movlhps VR128:$src1,
1223 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1224 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1225 def : Pat<(X86Movlhps VR128:$src1,
1226 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1227 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1228 def : Pat<(X86Movlhps VR128:$src1,
1229 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1230 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1232 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1233 // is during lowering, where it's not possible to recognize the load fold
1234 // cause it has two uses through a bitcast. One use disappears at isel time
1235 // and the fold opportunity reappears.
1236 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1237 (scalar_to_vector (loadf64 addr:$src2)))),
1238 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1240 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1241 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1242 (scalar_to_vector (loadf64 addr:$src2)))),
1243 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1246 def : Pat<(store (f64 (vector_extract
1247 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1248 (bc_v2f64 (v4f32 VR128:$src))), (iPTR 0))), addr:$dst),
1249 (VMOVHPSmr addr:$dst, VR128:$src)>;
1250 def : Pat<(store (f64 (vector_extract
1251 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))), addr:$dst),
1252 (VMOVHPDmr addr:$dst, VR128:$src)>;
1255 let Predicates = [HasSSE1] in {
1257 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1258 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1259 def : Pat<(X86Movlhps VR128:$src1,
1260 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1261 (MOVHPSrm VR128:$src1, addr:$src2)>;
1262 def : Pat<(X86Movlhps VR128:$src1,
1263 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1264 (MOVHPSrm VR128:$src1, addr:$src2)>;
1265 def : Pat<(X86Movlhps VR128:$src1,
1266 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1267 (MOVHPSrm VR128:$src1, addr:$src2)>;
1270 def : Pat<(store (f64 (vector_extract
1271 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1272 (bc_v2f64 (v4f32 VR128:$src))), (iPTR 0))), addr:$dst),
1273 (MOVHPSmr addr:$dst, VR128:$src)>;
1276 let Predicates = [HasSSE2] in {
1277 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1278 // is during lowering, where it's not possible to recognize the load fold
1279 // cause it has two uses through a bitcast. One use disappears at isel time
1280 // and the fold opportunity reappears.
1281 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1282 (scalar_to_vector (loadf64 addr:$src2)))),
1283 (MOVHPDrm VR128:$src1, addr:$src2)>;
1285 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1286 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1287 (scalar_to_vector (loadf64 addr:$src2)))),
1288 (MOVHPDrm VR128:$src1, addr:$src2)>;
1291 def : Pat<(store (f64 (vector_extract
1292 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))),addr:$dst),
1293 (MOVHPDmr addr:$dst, VR128:$src)>;
1296 //===----------------------------------------------------------------------===//
1297 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1298 //===----------------------------------------------------------------------===//
1300 let AddedComplexity = 20 in {
1301 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1302 (ins VR128:$src1, VR128:$src2),
1303 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1305 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1307 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1308 (ins VR128:$src1, VR128:$src2),
1309 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1311 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1314 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1315 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1316 (ins VR128:$src1, VR128:$src2),
1317 "movlhps\t{$src2, $dst|$dst, $src2}",
1319 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1320 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1321 (ins VR128:$src1, VR128:$src2),
1322 "movhlps\t{$src2, $dst|$dst, $src2}",
1324 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1327 let Predicates = [HasAVX] in {
1329 let AddedComplexity = 20 in {
1330 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1331 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1332 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1333 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1335 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1336 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1337 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1339 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1340 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1341 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1342 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1343 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1344 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1347 let AddedComplexity = 20 in {
1348 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1349 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1350 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1352 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1353 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1354 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1355 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1356 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1359 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1360 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1361 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1362 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1365 let Predicates = [HasSSE1] in {
1367 let AddedComplexity = 20 in {
1368 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1369 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1370 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1371 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1373 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1374 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1375 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1377 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1378 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1379 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1380 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1381 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1382 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1385 let AddedComplexity = 20 in {
1386 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1387 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1388 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1390 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1391 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1392 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1393 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1394 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1397 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1398 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1399 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1400 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1403 //===----------------------------------------------------------------------===//
1404 // SSE 1 & 2 - Conversion Instructions
1405 //===----------------------------------------------------------------------===//
1407 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1408 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1410 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1411 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1412 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1413 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1416 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1417 X86MemOperand x86memop, string asm> {
1418 let neverHasSideEffects = 1 in {
1419 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, []>;
1421 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, []>;
1422 } // neverHasSideEffects = 1
1425 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1426 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1427 string asm, Domain d> {
1428 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1429 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1430 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1431 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1434 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1435 X86MemOperand x86memop, string asm> {
1436 let neverHasSideEffects = 1 in {
1437 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1438 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1440 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1441 (ins DstRC:$src1, x86memop:$src),
1442 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1443 } // neverHasSideEffects = 1
1446 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1447 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1449 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1450 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1452 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1453 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1455 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1456 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1457 VEX, VEX_W, VEX_LIG;
1459 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1460 // register, but the same isn't true when only using memory operands,
1461 // provide other assembly "l" and "q" forms to address this explicitly
1462 // where appropriate to do so.
1463 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1465 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1466 VEX_4V, VEX_W, VEX_LIG;
1467 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1469 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1471 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1472 VEX_4V, VEX_W, VEX_LIG;
1474 let Predicates = [HasAVX], AddedComplexity = 1 in {
1475 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1476 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1477 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1478 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1479 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1480 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1481 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1482 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1484 def : Pat<(f32 (sint_to_fp GR32:$src)),
1485 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1486 def : Pat<(f32 (sint_to_fp GR64:$src)),
1487 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1488 def : Pat<(f64 (sint_to_fp GR32:$src)),
1489 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1490 def : Pat<(f64 (sint_to_fp GR64:$src)),
1491 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1494 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1495 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1496 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1497 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1498 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1499 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1500 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1501 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1502 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1503 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1504 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1505 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1506 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1507 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1508 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1509 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1511 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1512 // and/or XMM operand(s).
1514 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1515 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1517 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1518 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1519 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1520 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1521 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1522 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1525 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1526 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1527 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1528 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1530 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1531 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1532 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1533 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1534 (ins DstRC:$src1, x86memop:$src2),
1536 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1537 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1538 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1541 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1542 f128mem, load, "cvtsd2si">, XD, VEX;
1543 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1544 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1547 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1548 // Get rid of this hack or rename the intrinsics, there are several
1549 // intructions that only match with the intrinsic form, why create duplicates
1550 // to let them be recognized by the assembler?
1551 defm VCVTSD2SI : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
1552 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_LIG;
1553 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
1554 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W,
1557 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1558 f128mem, load, "cvtsd2si{l}">, XD;
1559 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1560 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1563 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1564 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1565 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1566 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1568 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1569 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1570 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1571 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1574 let Constraints = "$src1 = $dst" in {
1575 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1576 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1578 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1579 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1580 "cvtsi2ss{q}">, XS, REX_W;
1581 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1582 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1584 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1585 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1586 "cvtsi2sd">, XD, REX_W;
1591 // Aliases for intrinsics
1592 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1593 f32mem, load, "cvttss2si">, XS, VEX;
1594 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1595 int_x86_sse_cvttss2si64, f32mem, load,
1596 "cvttss2si">, XS, VEX, VEX_W;
1597 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1598 f128mem, load, "cvttsd2si">, XD, VEX;
1599 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1600 int_x86_sse2_cvttsd2si64, f128mem, load,
1601 "cvttsd2si">, XD, VEX, VEX_W;
1602 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1603 f32mem, load, "cvttss2si">, XS;
1604 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1605 int_x86_sse_cvttss2si64, f32mem, load,
1606 "cvttss2si{q}">, XS, REX_W;
1607 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1608 f128mem, load, "cvttsd2si">, XD;
1609 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1610 int_x86_sse2_cvttsd2si64, f128mem, load,
1611 "cvttsd2si{q}">, XD, REX_W;
1613 let Pattern = []<dag> in {
1614 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1615 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1617 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1618 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1620 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1621 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1622 SSEPackedSingle>, TB, VEX;
1623 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1624 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1625 SSEPackedSingle>, TB, VEX;
1628 let Pattern = []<dag> in {
1629 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1630 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1631 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1632 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1633 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1634 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1635 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1638 let Predicates = [HasAVX] in {
1639 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1640 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1641 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1642 (VCVTSS2SIrm addr:$src)>;
1643 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1644 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1645 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1646 (VCVTSS2SI64rm addr:$src)>;
1649 let Predicates = [HasSSE1] in {
1650 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1651 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1652 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1653 (CVTSS2SIrm addr:$src)>;
1654 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1655 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1656 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1657 (CVTSS2SI64rm addr:$src)>;
1662 // Convert scalar double to scalar single
1663 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1664 (ins FR64:$src1, FR64:$src2),
1665 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1668 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1669 (ins FR64:$src1, f64mem:$src2),
1670 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1671 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1673 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1676 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1677 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1678 [(set FR32:$dst, (fround FR64:$src))]>;
1679 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1680 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1681 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1682 Requires<[HasSSE2, OptForSize]>;
1684 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1685 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1687 let Constraints = "$src1 = $dst" in
1688 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1689 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1691 // Convert scalar single to scalar double
1692 // SSE2 instructions with XS prefix
1693 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1694 (ins FR32:$src1, FR32:$src2),
1695 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1696 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1698 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1699 (ins FR32:$src1, f32mem:$src2),
1700 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1701 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1703 let Predicates = [HasAVX] in {
1704 def : Pat<(f64 (fextend FR32:$src)),
1705 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1706 def : Pat<(fextend (loadf32 addr:$src)),
1707 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1708 def : Pat<(extloadf32 addr:$src),
1709 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1712 def : Pat<(extloadf32 addr:$src),
1713 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1714 Requires<[HasAVX, OptForSpeed]>;
1716 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1717 "cvtss2sd\t{$src, $dst|$dst, $src}",
1718 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1719 Requires<[HasSSE2]>;
1720 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1721 "cvtss2sd\t{$src, $dst|$dst, $src}",
1722 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1723 Requires<[HasSSE2, OptForSize]>;
1725 // extload f32 -> f64. This matches load+fextend because we have a hack in
1726 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1728 // Since these loads aren't folded into the fextend, we have to match it
1730 def : Pat<(fextend (loadf32 addr:$src)),
1731 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1732 def : Pat<(extloadf32 addr:$src),
1733 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1735 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1736 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1737 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1738 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1739 VR128:$src2))]>, XS, VEX_4V,
1741 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1742 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1743 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1744 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1745 (load addr:$src2)))]>, XS, VEX_4V,
1747 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1748 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1749 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1750 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1751 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1752 VR128:$src2))]>, XS,
1753 Requires<[HasSSE2]>;
1754 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1755 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1756 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1757 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1758 (load addr:$src2)))]>, XS,
1759 Requires<[HasSSE2]>;
1762 // Convert doubleword to packed single/double fp
1763 // SSE2 instructions without OpSize prefix
1764 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1765 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1766 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1767 TB, VEX, Requires<[HasAVX]>;
1768 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1769 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1770 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1771 (bitconvert (memopv2i64 addr:$src))))]>,
1772 TB, VEX, Requires<[HasAVX]>;
1773 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1774 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1775 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1776 TB, Requires<[HasSSE2]>;
1777 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1778 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1779 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1780 (bitconvert (memopv2i64 addr:$src))))]>,
1781 TB, Requires<[HasSSE2]>;
1783 // FIXME: why the non-intrinsic version is described as SSE3?
1784 // SSE2 instructions with XS prefix
1785 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1786 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1787 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1788 XS, VEX, Requires<[HasAVX]>;
1789 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1790 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1791 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1792 (bitconvert (memopv2i64 addr:$src))))]>,
1793 XS, VEX, Requires<[HasAVX]>;
1794 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1795 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1796 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1797 XS, Requires<[HasSSE2]>;
1798 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1799 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1800 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1801 (bitconvert (memopv2i64 addr:$src))))]>,
1802 XS, Requires<[HasSSE2]>;
1805 // Convert packed single/double fp to doubleword
1806 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1807 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1808 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1809 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1810 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1811 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1812 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1813 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1814 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1815 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1816 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1817 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1819 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1820 "cvtps2dq\t{$src, $dst|$dst, $src}",
1821 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1823 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1825 "cvtps2dq\t{$src, $dst|$dst, $src}",
1826 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1827 (memop addr:$src)))]>, VEX;
1828 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1829 "cvtps2dq\t{$src, $dst|$dst, $src}",
1830 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1831 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1832 "cvtps2dq\t{$src, $dst|$dst, $src}",
1833 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1834 (memop addr:$src)))]>;
1836 // SSE2 packed instructions with XD prefix
1837 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1838 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1839 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1840 XD, VEX, Requires<[HasAVX]>;
1841 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1842 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1843 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1844 (memop addr:$src)))]>,
1845 XD, VEX, Requires<[HasAVX]>;
1846 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1847 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1848 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1849 XD, Requires<[HasSSE2]>;
1850 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1851 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1852 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1853 (memop addr:$src)))]>,
1854 XD, Requires<[HasSSE2]>;
1857 // Convert with truncation packed single/double fp to doubleword
1858 // SSE2 packed instructions with XS prefix
1859 let neverHasSideEffects = 1 in {
1860 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1861 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1863 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1864 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1865 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1866 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1868 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1869 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1870 } // neverHasSideEffects = 1
1872 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1873 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1875 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1876 XS, VEX, Requires<[HasAVX]>;
1877 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1878 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1879 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1880 (memop addr:$src)))]>,
1881 XS, VEX, Requires<[HasAVX]>;
1883 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1884 "cvttps2dq\t{$src, $dst|$dst, $src}",
1886 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1887 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1888 "cvttps2dq\t{$src, $dst|$dst, $src}",
1890 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1892 let Predicates = [HasAVX] in {
1893 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1894 (Int_VCVTDQ2PSrr VR128:$src)>;
1895 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1896 (VCVTTPS2DQrr VR128:$src)>;
1897 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1898 (VCVTDQ2PSYrr VR256:$src)>;
1899 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1900 (VCVTTPS2DQYrr VR256:$src)>;
1903 let Predicates = [HasSSE2] in {
1904 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1905 (Int_CVTDQ2PSrr VR128:$src)>;
1906 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1907 (CVTTPS2DQrr VR128:$src)>;
1910 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1911 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1913 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1914 let isCodeGenOnly = 1 in
1915 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1916 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1917 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1918 (memop addr:$src)))]>, VEX;
1919 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1920 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1921 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1922 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1923 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1924 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1925 (memop addr:$src)))]>;
1927 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1928 // register, but the same isn't true when using memory operands instead.
1929 // Provide other assembly rr and rm forms to address this explicitly.
1930 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1931 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1934 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1935 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1936 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1937 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1940 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1941 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1942 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1943 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1945 // Convert packed single to packed double
1946 let Predicates = [HasAVX] in {
1947 // SSE2 instructions without OpSize prefix
1948 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1949 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1950 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1951 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1952 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1953 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1954 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1955 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1957 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1958 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1959 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1960 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1962 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1963 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1964 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1965 TB, VEX, Requires<[HasAVX]>;
1966 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1967 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1968 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1969 (load addr:$src)))]>,
1970 TB, VEX, Requires<[HasAVX]>;
1971 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1972 "cvtps2pd\t{$src, $dst|$dst, $src}",
1973 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1974 TB, Requires<[HasSSE2]>;
1975 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1976 "cvtps2pd\t{$src, $dst|$dst, $src}",
1977 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1978 (load addr:$src)))]>,
1979 TB, Requires<[HasSSE2]>;
1981 // Convert packed double to packed single
1982 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1983 // register, but the same isn't true when using memory operands instead.
1984 // Provide other assembly rr and rm forms to address this explicitly.
1985 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1986 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1987 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1988 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1991 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1992 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1993 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1994 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1997 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1998 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1999 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2000 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
2001 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2002 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
2003 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2004 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
2007 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2008 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2009 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2010 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
2012 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2013 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2014 (memop addr:$src)))]>;
2015 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2016 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2017 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2018 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2019 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2020 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2021 (memop addr:$src)))]>;
2023 // AVX 256-bit register conversion intrinsics
2024 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2025 // whenever possible to avoid declaring two versions of each one.
2026 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2027 (VCVTDQ2PSYrr VR256:$src)>;
2028 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2029 (VCVTDQ2PSYrm addr:$src)>;
2031 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2032 (VCVTPD2PSYrr VR256:$src)>;
2033 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2034 (VCVTPD2PSYrm addr:$src)>;
2036 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2037 (VCVTPS2DQYrr VR256:$src)>;
2038 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2039 (VCVTPS2DQYrm addr:$src)>;
2041 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2042 (VCVTPS2PDYrr VR128:$src)>;
2043 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2044 (VCVTPS2PDYrm addr:$src)>;
2046 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2047 (VCVTTPD2DQYrr VR256:$src)>;
2048 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2049 (VCVTTPD2DQYrm addr:$src)>;
2051 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
2052 (VCVTTPS2DQYrr VR256:$src)>;
2053 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
2054 (VCVTTPS2DQYrm addr:$src)>;
2056 // Match fround and fextend for 128/256-bit conversions
2057 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2058 (VCVTPD2PSYrr VR256:$src)>;
2059 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2060 (VCVTPD2PSYrm addr:$src)>;
2062 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2063 (VCVTPS2PDYrr VR128:$src)>;
2064 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2065 (VCVTPS2PDYrm addr:$src)>;
2067 //===----------------------------------------------------------------------===//
2068 // SSE 1 & 2 - Compare Instructions
2069 //===----------------------------------------------------------------------===//
2071 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2072 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2073 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2074 string asm, string asm_alt> {
2075 def rr : SIi8<0xC2, MRMSrcReg,
2076 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2077 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2078 def rm : SIi8<0xC2, MRMSrcMem,
2079 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2080 [(set RC:$dst, (OpNode (VT RC:$src1),
2081 (ld_frag addr:$src2), imm:$cc))]>;
2083 // Accept explicit immediate argument form instead of comparison code.
2084 let neverHasSideEffects = 1 in {
2085 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2086 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2088 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2089 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2093 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2094 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2095 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2096 XS, VEX_4V, VEX_LIG;
2097 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2098 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2099 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2100 XD, VEX_4V, VEX_LIG;
2102 let Constraints = "$src1 = $dst" in {
2103 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2104 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2105 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2107 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2108 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2109 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2113 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2114 Intrinsic Int, string asm> {
2115 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2116 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2117 [(set VR128:$dst, (Int VR128:$src1,
2118 VR128:$src, imm:$cc))]>;
2119 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2120 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
2121 [(set VR128:$dst, (Int VR128:$src1,
2122 (load addr:$src), imm:$cc))]>;
2125 // Aliases to match intrinsics which expect XMM operand(s).
2126 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2127 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2129 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2130 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2132 let Constraints = "$src1 = $dst" in {
2133 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2134 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2135 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2136 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2140 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2141 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2142 ValueType vt, X86MemOperand x86memop,
2143 PatFrag ld_frag, string OpcodeStr, Domain d> {
2144 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2145 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2146 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2147 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2148 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2149 [(set EFLAGS, (OpNode (vt RC:$src1),
2150 (ld_frag addr:$src2)))], d>;
2153 let Defs = [EFLAGS] in {
2154 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2155 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2156 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2157 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2159 let Pattern = []<dag> in {
2160 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2161 "comiss", SSEPackedSingle>, TB, VEX,
2163 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2164 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2168 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2169 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2170 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2171 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2173 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2174 load, "comiss", SSEPackedSingle>, TB, VEX;
2175 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2176 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2177 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2178 "ucomiss", SSEPackedSingle>, TB;
2179 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2180 "ucomisd", SSEPackedDouble>, TB, OpSize;
2182 let Pattern = []<dag> in {
2183 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2184 "comiss", SSEPackedSingle>, TB;
2185 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2186 "comisd", SSEPackedDouble>, TB, OpSize;
2189 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2190 load, "ucomiss", SSEPackedSingle>, TB;
2191 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2192 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2194 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2195 "comiss", SSEPackedSingle>, TB;
2196 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2197 "comisd", SSEPackedDouble>, TB, OpSize;
2198 } // Defs = [EFLAGS]
2200 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2201 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2202 Intrinsic Int, string asm, string asm_alt,
2204 let isAsmParserOnly = 1 in {
2205 def rri : PIi8<0xC2, MRMSrcReg,
2206 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2207 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2208 def rmi : PIi8<0xC2, MRMSrcMem,
2209 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, SSECC:$cc), asm,
2210 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2213 // Accept explicit immediate argument form instead of comparison code.
2214 def rri_alt : PIi8<0xC2, MRMSrcReg,
2215 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2217 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2218 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, i8imm:$cc),
2222 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2223 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2224 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2225 SSEPackedSingle>, TB, VEX_4V;
2226 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2227 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2228 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2229 SSEPackedDouble>, TB, OpSize, VEX_4V;
2230 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2231 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2232 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2233 SSEPackedSingle>, TB, VEX_4V;
2234 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2235 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2236 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2237 SSEPackedDouble>, TB, OpSize, VEX_4V;
2238 let Constraints = "$src1 = $dst" in {
2239 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2240 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2241 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2242 SSEPackedSingle>, TB;
2243 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2244 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2245 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2246 SSEPackedDouble>, TB, OpSize;
2249 let Predicates = [HasAVX] in {
2250 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2251 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2252 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2253 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2254 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2255 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2256 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2257 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2259 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2260 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2261 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2262 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2263 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2264 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2265 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2266 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2269 let Predicates = [HasSSE1] in {
2270 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2271 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2272 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2273 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2276 let Predicates = [HasSSE2] in {
2277 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2278 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2279 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2280 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2283 //===----------------------------------------------------------------------===//
2284 // SSE 1 & 2 - Shuffle Instructions
2285 //===----------------------------------------------------------------------===//
2287 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2288 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2289 ValueType vt, string asm, PatFrag mem_frag,
2290 Domain d, bit IsConvertibleToThreeAddress = 0> {
2291 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2292 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
2293 [(set RC:$dst, (vt (shufp:$src3
2294 RC:$src1, (mem_frag addr:$src2))))], d>;
2295 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2296 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2297 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2299 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2302 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2303 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2304 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2305 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2306 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2307 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2308 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2309 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2310 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2311 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2312 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2313 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2315 let Constraints = "$src1 = $dst" in {
2316 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2317 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2318 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2320 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2321 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2322 memopv2f64, SSEPackedDouble>, TB, OpSize;
2325 let Predicates = [HasAVX] in {
2326 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2327 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2328 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2329 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2330 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2331 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2332 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2333 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2334 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2335 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2336 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2337 // fall back to this for SSE1)
2338 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2339 (VSHUFPSrri VR128:$src2, VR128:$src1,
2340 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2341 // Special unary SHUFPSrri case.
2342 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2343 (VSHUFPSrri VR128:$src1, VR128:$src1,
2344 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2345 // Special binary v4i32 shuffle cases with SHUFPS.
2346 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2347 (VSHUFPSrri VR128:$src1, VR128:$src2,
2348 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2349 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2350 (bc_v4i32 (memopv2i64 addr:$src2)))),
2351 (VSHUFPSrmi VR128:$src1, addr:$src2,
2352 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2353 // Special unary SHUFPDrri cases.
2354 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2355 (VSHUFPDrri VR128:$src1, VR128:$src1,
2356 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2357 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2358 (VSHUFPDrri VR128:$src1, VR128:$src1,
2359 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2360 // Special binary v2i64 shuffle cases using SHUFPDrri.
2361 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2362 (VSHUFPDrri VR128:$src1, VR128:$src2,
2363 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2365 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2366 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2367 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2368 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2369 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2370 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2371 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2372 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2373 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2374 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2377 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2378 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2379 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2380 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2381 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2383 def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2384 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2385 def : Pat<(v8f32 (X86Shufp VR256:$src1,
2386 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2387 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2389 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2390 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2391 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2392 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2393 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2395 def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2396 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2397 def : Pat<(v4f64 (X86Shufp VR256:$src1,
2398 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2399 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2402 let Predicates = [HasSSE1] in {
2403 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2404 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2405 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2406 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2407 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2408 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2409 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2410 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2411 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2412 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2413 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2414 // fall back to this for SSE1)
2415 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2416 (SHUFPSrri VR128:$src2, VR128:$src1,
2417 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2418 // Special unary SHUFPSrri case.
2419 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2420 (SHUFPSrri VR128:$src1, VR128:$src1,
2421 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2424 let Predicates = [HasSSE2] in {
2425 // Special binary v4i32 shuffle cases with SHUFPS.
2426 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2427 (SHUFPSrri VR128:$src1, VR128:$src2,
2428 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2429 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2430 (bc_v4i32 (memopv2i64 addr:$src2)))),
2431 (SHUFPSrmi VR128:$src1, addr:$src2,
2432 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2433 // Special unary SHUFPDrri cases.
2434 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2435 (SHUFPDrri VR128:$src1, VR128:$src1,
2436 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2437 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2438 (SHUFPDrri VR128:$src1, VR128:$src1,
2439 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2440 // Special binary v2i64 shuffle cases using SHUFPDrri.
2441 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2442 (SHUFPDrri VR128:$src1, VR128:$src2,
2443 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2444 // Generic SHUFPD patterns
2445 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2446 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2447 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2448 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2449 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2450 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2451 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2452 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2453 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2454 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2457 //===----------------------------------------------------------------------===//
2458 // SSE 1 & 2 - Unpack Instructions
2459 //===----------------------------------------------------------------------===//
2461 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2462 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2463 PatFrag mem_frag, RegisterClass RC,
2464 X86MemOperand x86memop, string asm,
2466 def rr : PI<opc, MRMSrcReg,
2467 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2469 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2470 def rm : PI<opc, MRMSrcMem,
2471 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2473 (vt (OpNode RC:$src1,
2474 (mem_frag addr:$src2))))], d>;
2477 let AddedComplexity = 10 in {
2478 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2479 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2480 SSEPackedSingle>, TB, VEX_4V;
2481 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2482 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2483 SSEPackedDouble>, TB, OpSize, VEX_4V;
2484 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2485 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2486 SSEPackedSingle>, TB, VEX_4V;
2487 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2488 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2489 SSEPackedDouble>, TB, OpSize, VEX_4V;
2491 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2492 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2493 SSEPackedSingle>, TB, VEX_4V;
2494 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2495 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2496 SSEPackedDouble>, TB, OpSize, VEX_4V;
2497 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2498 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2499 SSEPackedSingle>, TB, VEX_4V;
2500 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2501 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2502 SSEPackedDouble>, TB, OpSize, VEX_4V;
2504 let Constraints = "$src1 = $dst" in {
2505 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2506 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2507 SSEPackedSingle>, TB;
2508 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2509 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2510 SSEPackedDouble>, TB, OpSize;
2511 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2512 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2513 SSEPackedSingle>, TB;
2514 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2515 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2516 SSEPackedDouble>, TB, OpSize;
2517 } // Constraints = "$src1 = $dst"
2518 } // AddedComplexity
2520 let Predicates = [HasSSE1] in {
2521 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2522 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2523 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2524 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2525 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2526 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2527 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2528 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2531 let Predicates = [HasSSE2] in {
2532 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2533 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2534 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2535 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2536 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2537 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2538 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2539 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2541 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2542 // problem is during lowering, where it's not possible to recognize the load
2543 // fold cause it has two uses through a bitcast. One use disappears at isel
2544 // time and the fold opportunity reappears.
2545 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2546 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2548 let AddedComplexity = 10 in
2549 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2550 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2553 let Predicates = [HasAVX] in {
2554 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2555 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2556 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2557 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2558 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2559 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2560 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2561 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2563 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2564 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2565 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2566 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2567 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2568 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2569 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2570 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2572 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2573 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2574 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2575 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2576 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2577 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2578 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2579 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2581 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2582 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2583 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2584 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2585 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2586 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2587 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2588 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2590 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2591 // problem is during lowering, where it's not possible to recognize the load
2592 // fold cause it has two uses through a bitcast. One use disappears at isel
2593 // time and the fold opportunity reappears.
2594 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2595 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2596 let AddedComplexity = 10 in
2597 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2598 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2601 //===----------------------------------------------------------------------===//
2602 // SSE 1 & 2 - Extract Floating-Point Sign mask
2603 //===----------------------------------------------------------------------===//
2605 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2606 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2608 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2609 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2610 [(set GR32:$dst, (Int RC:$src))], d>;
2611 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2612 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2615 let Predicates = [HasAVX] in {
2616 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2617 "movmskps", SSEPackedSingle>, TB, VEX;
2618 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2619 "movmskpd", SSEPackedDouble>, TB,
2621 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2622 "movmskps", SSEPackedSingle>, TB, VEX;
2623 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2624 "movmskpd", SSEPackedDouble>, TB,
2627 def : Pat<(i32 (X86fgetsign FR32:$src)),
2628 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2630 def : Pat<(i64 (X86fgetsign FR32:$src)),
2631 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2633 def : Pat<(i32 (X86fgetsign FR64:$src)),
2634 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2636 def : Pat<(i64 (X86fgetsign FR64:$src)),
2637 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2641 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2642 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2643 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2644 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2646 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2647 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2648 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2649 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2653 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2654 SSEPackedSingle>, TB;
2655 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2656 SSEPackedDouble>, TB, OpSize;
2658 def : Pat<(i32 (X86fgetsign FR32:$src)),
2659 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2660 sub_ss))>, Requires<[HasSSE1]>;
2661 def : Pat<(i64 (X86fgetsign FR32:$src)),
2662 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2663 sub_ss))>, Requires<[HasSSE1]>;
2664 def : Pat<(i32 (X86fgetsign FR64:$src)),
2665 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2666 sub_sd))>, Requires<[HasSSE2]>;
2667 def : Pat<(i64 (X86fgetsign FR64:$src)),
2668 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2669 sub_sd))>, Requires<[HasSSE2]>;
2671 //===---------------------------------------------------------------------===//
2672 // SSE2 - Packed Integer Logical Instructions
2673 //===---------------------------------------------------------------------===//
2675 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2677 /// PDI_binop_rm - Simple SSE2 binary operator.
2678 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2679 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2680 X86MemOperand x86memop, bit IsCommutable = 0,
2682 let isCommutable = IsCommutable in
2683 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2684 (ins RC:$src1, RC:$src2),
2686 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2687 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2688 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2689 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2690 (ins RC:$src1, x86memop:$src2),
2692 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2693 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2694 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2695 (bitconvert (memop_frag addr:$src2)))))]>;
2697 } // ExeDomain = SSEPackedInt
2699 // These are ordered here for pattern ordering requirements with the fp versions
2701 let Predicates = [HasAVX] in {
2702 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2703 i128mem, 1, 0>, VEX_4V;
2704 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2705 i128mem, 1, 0>, VEX_4V;
2706 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2707 i128mem, 1, 0>, VEX_4V;
2708 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2709 i128mem, 0, 0>, VEX_4V;
2712 let Constraints = "$src1 = $dst" in {
2713 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2715 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2717 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2719 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2721 } // Constraints = "$src1 = $dst"
2723 let Predicates = [HasAVX2] in {
2724 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2725 i256mem, 1, 0>, VEX_4V;
2726 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2727 i256mem, 1, 0>, VEX_4V;
2728 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2729 i256mem, 1, 0>, VEX_4V;
2730 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2731 i256mem, 0, 0>, VEX_4V;
2734 //===----------------------------------------------------------------------===//
2735 // SSE 1 & 2 - Logical Instructions
2736 //===----------------------------------------------------------------------===//
2738 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2740 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2742 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2743 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2745 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2746 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2748 let Constraints = "$src1 = $dst" in {
2749 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2750 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2752 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2753 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2757 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2758 let mayLoad = 0 in {
2759 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2760 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2761 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2764 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2765 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2767 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2769 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2771 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2772 // are all promoted to v2i64, and the patterns are covered by the int
2773 // version. This is needed in SSE only, because v2i64 isn't supported on
2774 // SSE1, but only on SSE2.
2775 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2776 !strconcat(OpcodeStr, "ps"), f128mem, [],
2777 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2778 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2780 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2781 !strconcat(OpcodeStr, "pd"), f128mem,
2782 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2783 (bc_v2i64 (v2f64 VR128:$src2))))],
2784 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2785 (memopv2i64 addr:$src2)))], 0>,
2787 let Constraints = "$src1 = $dst" in {
2788 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2789 !strconcat(OpcodeStr, "ps"), f128mem,
2790 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2791 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2792 (memopv2i64 addr:$src2)))]>, TB;
2794 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2795 !strconcat(OpcodeStr, "pd"), f128mem,
2796 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2797 (bc_v2i64 (v2f64 VR128:$src2))))],
2798 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2799 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2803 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2805 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2807 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2808 !strconcat(OpcodeStr, "ps"), f256mem,
2809 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2810 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2811 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2813 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2814 !strconcat(OpcodeStr, "pd"), f256mem,
2815 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2816 (bc_v4i64 (v4f64 VR256:$src2))))],
2817 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2818 (memopv4i64 addr:$src2)))], 0>,
2822 // AVX 256-bit packed logical ops forms
2823 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2824 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2825 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2826 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2828 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2829 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2830 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2831 let isCommutable = 0 in
2832 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2834 //===----------------------------------------------------------------------===//
2835 // SSE 1 & 2 - Arithmetic Instructions
2836 //===----------------------------------------------------------------------===//
2838 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2841 /// In addition, we also have a special variant of the scalar form here to
2842 /// represent the associated intrinsic operation. This form is unlike the
2843 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2844 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2846 /// These three forms can each be reg+reg or reg+mem.
2849 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2851 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2853 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2854 OpNode, FR32, f32mem, Is2Addr>, XS;
2855 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2856 OpNode, FR64, f64mem, Is2Addr>, XD;
2859 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2861 let mayLoad = 0 in {
2862 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2863 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2864 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2865 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2869 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2871 let mayLoad = 0 in {
2872 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2873 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2874 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2875 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2879 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2881 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2882 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2883 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2884 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2887 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2889 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2890 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2891 SSEPackedSingle, Is2Addr>, TB;
2893 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2894 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2895 SSEPackedDouble, Is2Addr>, TB, OpSize;
2898 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2899 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2900 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2901 SSEPackedSingle, 0>, TB;
2903 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2904 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2905 SSEPackedDouble, 0>, TB, OpSize;
2908 // Binary Arithmetic instructions
2909 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2910 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2911 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2912 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2913 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2914 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2915 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2916 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2918 let isCommutable = 0 in {
2919 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2920 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2921 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2922 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2923 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2924 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2925 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2926 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2927 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2928 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2929 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2930 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2931 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2932 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2933 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2934 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2935 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2936 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2937 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2938 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2941 let Constraints = "$src1 = $dst" in {
2942 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2943 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2944 basic_sse12_fp_binop_s_int<0x58, "add">;
2945 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2946 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2947 basic_sse12_fp_binop_s_int<0x59, "mul">;
2949 let isCommutable = 0 in {
2950 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2951 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2952 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2953 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2954 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2955 basic_sse12_fp_binop_s_int<0x5E, "div">;
2956 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2957 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2958 basic_sse12_fp_binop_s_int<0x5F, "max">,
2959 basic_sse12_fp_binop_p_int<0x5F, "max">;
2960 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2961 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2962 basic_sse12_fp_binop_s_int<0x5D, "min">,
2963 basic_sse12_fp_binop_p_int<0x5D, "min">;
2968 /// In addition, we also have a special variant of the scalar form here to
2969 /// represent the associated intrinsic operation. This form is unlike the
2970 /// plain scalar form, in that it takes an entire vector (instead of a
2971 /// scalar) and leaves the top elements undefined.
2973 /// And, we have a special variant form for a full-vector intrinsic form.
2975 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2976 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2977 SDNode OpNode, Intrinsic F32Int> {
2978 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2979 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2980 [(set FR32:$dst, (OpNode FR32:$src))]>;
2981 // For scalar unary operations, fold a load into the operation
2982 // only in OptForSize mode. It eliminates an instruction, but it also
2983 // eliminates a whole-register clobber (the load), so it introduces a
2984 // partial register update condition.
2985 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2986 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2987 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2988 Requires<[HasSSE1, OptForSize]>;
2989 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2990 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2991 [(set VR128:$dst, (F32Int VR128:$src))]>;
2992 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2993 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2994 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2997 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2998 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2999 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
3000 !strconcat(OpcodeStr,
3001 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3003 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
3004 !strconcat(OpcodeStr,
3005 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3006 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3007 (ins VR128:$src1, ssmem:$src2),
3008 !strconcat(OpcodeStr,
3009 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3012 /// sse1_fp_unop_p - SSE1 unops in packed form.
3013 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3014 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3015 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3016 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
3017 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3018 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3019 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
3022 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3023 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3024 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3025 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3026 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
3027 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3028 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3029 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
3032 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3033 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3034 Intrinsic V4F32Int> {
3035 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3036 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3037 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
3038 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3039 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3040 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
3043 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3044 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3045 Intrinsic V4F32Int> {
3046 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3047 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3048 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
3049 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3050 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3051 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
3054 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3055 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3056 SDNode OpNode, Intrinsic F64Int> {
3057 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3058 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3059 [(set FR64:$dst, (OpNode FR64:$src))]>;
3060 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3061 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3062 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3063 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
3064 Requires<[HasSSE2, OptForSize]>;
3065 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3066 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3067 [(set VR128:$dst, (F64Int VR128:$src))]>;
3068 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3069 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3070 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
3073 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3074 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3075 let neverHasSideEffects = 1 in {
3076 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3077 !strconcat(OpcodeStr,
3078 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3080 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3081 !strconcat(OpcodeStr,
3082 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3084 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3085 (ins VR128:$src1, sdmem:$src2),
3086 !strconcat(OpcodeStr,
3087 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3090 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3091 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3093 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3094 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3095 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3096 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3097 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3098 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3101 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3102 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3103 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3104 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3105 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3106 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3107 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3108 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3111 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3112 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3113 Intrinsic V2F64Int> {
3114 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3115 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3116 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3117 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3118 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3119 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3122 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3123 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3124 Intrinsic V2F64Int> {
3125 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3126 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3127 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3128 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3129 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3130 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3133 let Predicates = [HasAVX] in {
3135 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3136 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3138 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3139 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3140 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3141 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3142 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3143 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3144 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3145 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3148 // Reciprocal approximations. Note that these typically require refinement
3149 // in order to obtain suitable precision.
3150 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3151 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3152 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3153 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3154 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3156 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3157 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3158 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3159 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3160 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3163 let AddedComplexity = 1 in {
3164 def : Pat<(f32 (fsqrt FR32:$src)),
3165 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3166 def : Pat<(f32 (fsqrt (load addr:$src))),
3167 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3168 Requires<[HasAVX, OptForSize]>;
3169 def : Pat<(f64 (fsqrt FR64:$src)),
3170 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3171 def : Pat<(f64 (fsqrt (load addr:$src))),
3172 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3173 Requires<[HasAVX, OptForSize]>;
3175 def : Pat<(f32 (X86frsqrt FR32:$src)),
3176 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3177 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3178 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3179 Requires<[HasAVX, OptForSize]>;
3181 def : Pat<(f32 (X86frcp FR32:$src)),
3182 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3183 def : Pat<(f32 (X86frcp (load addr:$src))),
3184 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3185 Requires<[HasAVX, OptForSize]>;
3188 let Predicates = [HasAVX], AddedComplexity = 1 in {
3189 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3190 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3191 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3192 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3194 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3195 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3197 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3198 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3199 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3200 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3202 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3203 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3205 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3206 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3207 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3208 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3210 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3211 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3213 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3214 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3215 (VRCPSSr (f32 (IMPLICIT_DEF)),
3216 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3218 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3219 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3223 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3224 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3225 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3226 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3227 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3228 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3230 // Reciprocal approximations. Note that these typically require refinement
3231 // in order to obtain suitable precision.
3232 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3233 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3234 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3235 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3236 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3237 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3239 // There is no f64 version of the reciprocal approximation instructions.
3241 //===----------------------------------------------------------------------===//
3242 // SSE 1 & 2 - Non-temporal stores
3243 //===----------------------------------------------------------------------===//
3245 let AddedComplexity = 400 in { // Prefer non-temporal versions
3246 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3247 (ins f128mem:$dst, VR128:$src),
3248 "movntps\t{$src, $dst|$dst, $src}",
3249 [(alignednontemporalstore (v4f32 VR128:$src),
3251 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3252 (ins f128mem:$dst, VR128:$src),
3253 "movntpd\t{$src, $dst|$dst, $src}",
3254 [(alignednontemporalstore (v2f64 VR128:$src),
3257 let ExeDomain = SSEPackedInt in
3258 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3259 (ins f128mem:$dst, VR128:$src),
3260 "movntdq\t{$src, $dst|$dst, $src}",
3261 [(alignednontemporalstore (v2i64 VR128:$src),
3264 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3265 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3267 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3268 (ins f256mem:$dst, VR256:$src),
3269 "movntps\t{$src, $dst|$dst, $src}",
3270 [(alignednontemporalstore (v8f32 VR256:$src),
3272 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3273 (ins f256mem:$dst, VR256:$src),
3274 "movntpd\t{$src, $dst|$dst, $src}",
3275 [(alignednontemporalstore (v4f64 VR256:$src),
3277 let ExeDomain = SSEPackedInt in
3278 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3279 (ins f256mem:$dst, VR256:$src),
3280 "movntdq\t{$src, $dst|$dst, $src}",
3281 [(alignednontemporalstore (v4i64 VR256:$src),
3285 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3286 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3287 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3288 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3289 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3290 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3292 let AddedComplexity = 400 in { // Prefer non-temporal versions
3293 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3294 "movntps\t{$src, $dst|$dst, $src}",
3295 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3296 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3297 "movntpd\t{$src, $dst|$dst, $src}",
3298 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3300 let ExeDomain = SSEPackedInt in
3301 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3302 "movntdq\t{$src, $dst|$dst, $src}",
3303 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)]>;
3305 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3306 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3308 // There is no AVX form for instructions below this point
3309 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3310 "movnti{l}\t{$src, $dst|$dst, $src}",
3311 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3312 TB, Requires<[HasSSE2]>;
3313 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3314 "movnti{q}\t{$src, $dst|$dst, $src}",
3315 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3316 TB, Requires<[HasSSE2]>;
3319 //===----------------------------------------------------------------------===//
3320 // SSE 1 & 2 - Prefetch and memory fence
3321 //===----------------------------------------------------------------------===//
3323 // Prefetch intrinsic.
3324 let Predicates = [HasSSE1] in {
3325 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3326 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3327 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3328 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3329 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3330 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3331 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3332 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3336 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3337 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3338 TB, Requires<[HasSSE2]>;
3340 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3341 // was introduced with SSE2, it's backward compatible.
3342 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3344 // Load, store, and memory fence
3345 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3346 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3347 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3348 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3349 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3350 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3352 def : Pat<(X86SFence), (SFENCE)>;
3353 def : Pat<(X86LFence), (LFENCE)>;
3354 def : Pat<(X86MFence), (MFENCE)>;
3356 //===----------------------------------------------------------------------===//
3357 // SSE 1 & 2 - Load/Store XCSR register
3358 //===----------------------------------------------------------------------===//
3360 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3361 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3362 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3363 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3365 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3366 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3367 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3368 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3370 //===---------------------------------------------------------------------===//
3371 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3372 //===---------------------------------------------------------------------===//
3374 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3376 let neverHasSideEffects = 1 in {
3377 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3378 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3379 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3380 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3382 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3383 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3384 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3385 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3388 let isCodeGenOnly = 1 in {
3389 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3390 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3391 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3392 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3393 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3394 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3395 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3396 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3399 let canFoldAsLoad = 1, mayLoad = 1 in {
3400 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3401 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3402 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3403 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3404 let Predicates = [HasAVX] in {
3405 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3406 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3407 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3408 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3412 let mayStore = 1 in {
3413 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3414 (ins i128mem:$dst, VR128:$src),
3415 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3416 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3417 (ins i256mem:$dst, VR256:$src),
3418 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3419 let Predicates = [HasAVX] in {
3420 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3421 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3422 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3423 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3427 let neverHasSideEffects = 1 in
3428 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3429 "movdqa\t{$src, $dst|$dst, $src}", []>;
3431 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3432 "movdqu\t{$src, $dst|$dst, $src}",
3433 []>, XS, Requires<[HasSSE2]>;
3436 let isCodeGenOnly = 1 in {
3437 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3438 "movdqa\t{$src, $dst|$dst, $src}", []>;
3440 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3441 "movdqu\t{$src, $dst|$dst, $src}",
3442 []>, XS, Requires<[HasSSE2]>;
3445 let canFoldAsLoad = 1, mayLoad = 1 in {
3446 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3447 "movdqa\t{$src, $dst|$dst, $src}",
3448 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3449 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3450 "movdqu\t{$src, $dst|$dst, $src}",
3451 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3452 XS, Requires<[HasSSE2]>;
3455 let mayStore = 1 in {
3456 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3457 "movdqa\t{$src, $dst|$dst, $src}",
3458 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3459 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3460 "movdqu\t{$src, $dst|$dst, $src}",
3461 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3462 XS, Requires<[HasSSE2]>;
3465 // Intrinsic forms of MOVDQU load and store
3466 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3467 "vmovdqu\t{$src, $dst|$dst, $src}",
3468 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3469 XS, VEX, Requires<[HasAVX]>;
3471 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3472 "movdqu\t{$src, $dst|$dst, $src}",
3473 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3474 XS, Requires<[HasSSE2]>;
3476 } // ExeDomain = SSEPackedInt
3478 let Predicates = [HasAVX] in {
3479 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3480 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3481 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3484 //===---------------------------------------------------------------------===//
3485 // SSE2 - Packed Integer Arithmetic Instructions
3486 //===---------------------------------------------------------------------===//
3488 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3490 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3491 RegisterClass RC, PatFrag memop_frag,
3492 X86MemOperand x86memop, bit IsCommutable = 0,
3494 let isCommutable = IsCommutable in
3495 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3496 (ins RC:$src1, RC:$src2),
3498 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3499 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3500 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3501 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3502 (ins RC:$src1, x86memop:$src2),
3504 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3505 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3506 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3509 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3510 string OpcodeStr, Intrinsic IntId,
3511 Intrinsic IntId2, RegisterClass RC,
3513 // src2 is always 128-bit
3514 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3515 (ins RC:$src1, VR128:$src2),
3517 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3518 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3519 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3520 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3521 (ins RC:$src1, i128mem:$src2),
3523 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3524 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3525 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3526 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3527 (ins RC:$src1, i32i8imm:$src2),
3529 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3530 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3531 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3534 } // ExeDomain = SSEPackedInt
3536 // 128-bit Integer Arithmetic
3538 let Predicates = [HasAVX] in {
3539 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3540 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3541 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3542 i128mem, 1, 0>, VEX_4V;
3543 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3544 i128mem, 1, 0>, VEX_4V;
3545 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3546 i128mem, 1, 0>, VEX_4V;
3547 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3548 i128mem, 1, 0>, VEX_4V;
3549 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3550 i128mem, 0, 0>, VEX_4V;
3551 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3552 i128mem, 0, 0>, VEX_4V;
3553 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3554 i128mem, 0, 0>, VEX_4V;
3555 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3556 i128mem, 0, 0>, VEX_4V;
3559 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3560 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3561 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3562 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3563 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3564 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3565 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3566 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3567 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3568 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3569 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3570 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3571 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3572 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3573 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3574 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3575 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3576 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3577 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3578 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3579 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3580 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3581 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3582 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3583 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3584 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3585 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3586 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3587 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3588 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3589 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3590 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3591 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3592 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3593 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3594 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3595 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3596 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3599 let Predicates = [HasAVX2] in {
3600 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3601 i256mem, 1, 0>, VEX_4V;
3602 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3603 i256mem, 1, 0>, VEX_4V;
3604 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3605 i256mem, 1, 0>, VEX_4V;
3606 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3607 i256mem, 1, 0>, VEX_4V;
3608 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3609 i256mem, 1, 0>, VEX_4V;
3610 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3611 i256mem, 0, 0>, VEX_4V;
3612 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3613 i256mem, 0, 0>, VEX_4V;
3614 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3615 i256mem, 0, 0>, VEX_4V;
3616 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3617 i256mem, 0, 0>, VEX_4V;
3620 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3621 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3622 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3623 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3624 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3625 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3626 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3627 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3628 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3629 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3630 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3631 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3632 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3633 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3634 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3635 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3636 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3637 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3638 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3639 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3640 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3641 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3642 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3643 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3644 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3645 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3646 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3647 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3648 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3649 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3650 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3651 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3652 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3653 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3654 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3655 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3656 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3657 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3660 let Constraints = "$src1 = $dst" in {
3661 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3663 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3665 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3667 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3669 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3671 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3673 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3675 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3677 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3681 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3682 VR128, memopv2i64, i128mem>;
3683 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3684 VR128, memopv2i64, i128mem>;
3685 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3686 VR128, memopv2i64, i128mem>;
3687 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3688 VR128, memopv2i64, i128mem>;
3689 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3690 VR128, memopv2i64, i128mem, 1>;
3691 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3692 VR128, memopv2i64, i128mem, 1>;
3693 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3694 VR128, memopv2i64, i128mem, 1>;
3695 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3696 VR128, memopv2i64, i128mem, 1>;
3697 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3698 VR128, memopv2i64, i128mem, 1>;
3699 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3700 VR128, memopv2i64, i128mem, 1>;
3701 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3702 VR128, memopv2i64, i128mem, 1>;
3703 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3704 VR128, memopv2i64, i128mem, 1>;
3705 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3706 VR128, memopv2i64, i128mem, 1>;
3707 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3708 VR128, memopv2i64, i128mem, 1>;
3709 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3710 VR128, memopv2i64, i128mem, 1>;
3711 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3712 VR128, memopv2i64, i128mem, 1>;
3713 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3714 VR128, memopv2i64, i128mem, 1>;
3715 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3716 VR128, memopv2i64, i128mem, 1>;
3717 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3718 VR128, memopv2i64, i128mem, 1>;
3720 } // Constraints = "$src1 = $dst"
3722 //===---------------------------------------------------------------------===//
3723 // SSE2 - Packed Integer Logical Instructions
3724 //===---------------------------------------------------------------------===//
3726 let Predicates = [HasAVX] in {
3727 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3728 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3730 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3731 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3733 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3734 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3737 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3738 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3740 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3741 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3743 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3744 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3747 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3748 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3750 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3751 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3754 let ExeDomain = SSEPackedInt in {
3755 let neverHasSideEffects = 1 in {
3756 // 128-bit logical shifts.
3757 def VPSLLDQri : PDIi8<0x73, MRM7r,
3758 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3759 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3761 def VPSRLDQri : PDIi8<0x73, MRM3r,
3762 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3763 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3765 // PSRADQri doesn't exist in SSE[1-3].
3770 let Predicates = [HasAVX2] in {
3771 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3772 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3774 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3775 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3777 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3778 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3781 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3782 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3784 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3785 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3787 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3788 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3791 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3792 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3794 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3795 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3798 let ExeDomain = SSEPackedInt in {
3799 let neverHasSideEffects = 1 in {
3800 // 128-bit logical shifts.
3801 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3802 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3803 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3805 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3806 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3807 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3809 // PSRADQYri doesn't exist in SSE[1-3].
3814 let Constraints = "$src1 = $dst" in {
3815 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3816 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3818 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3819 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3821 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3822 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3825 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3826 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3828 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3829 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3831 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3832 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3835 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3836 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3838 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3839 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3842 let ExeDomain = SSEPackedInt in {
3843 let neverHasSideEffects = 1 in {
3844 // 128-bit logical shifts.
3845 def PSLLDQri : PDIi8<0x73, MRM7r,
3846 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3847 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3848 def PSRLDQri : PDIi8<0x73, MRM3r,
3849 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3850 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3851 // PSRADQri doesn't exist in SSE[1-3].
3854 } // Constraints = "$src1 = $dst"
3856 let Predicates = [HasAVX] in {
3857 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3858 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3859 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3860 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3861 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3862 (VPSLLDQri VR128:$src1, imm:$src2)>;
3863 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3864 (VPSRLDQri VR128:$src1, imm:$src2)>;
3865 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3866 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3868 // Shift up / down and insert zero's.
3869 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3870 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3871 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3872 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3875 let Predicates = [HasAVX2] in {
3876 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3877 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3878 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3879 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3880 def : Pat<(int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2),
3881 (VPSLLDQYri VR256:$src1, imm:$src2)>;
3882 def : Pat<(int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2),
3883 (VPSRLDQYri VR256:$src1, imm:$src2)>;
3886 let Predicates = [HasSSE2] in {
3887 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3888 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3889 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3890 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3891 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3892 (PSLLDQri VR128:$src1, imm:$src2)>;
3893 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3894 (PSRLDQri VR128:$src1, imm:$src2)>;
3895 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3896 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3898 // Shift up / down and insert zero's.
3899 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3900 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3901 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3902 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3905 //===---------------------------------------------------------------------===//
3906 // SSE2 - Packed Integer Comparison Instructions
3907 //===---------------------------------------------------------------------===//
3909 let Predicates = [HasAVX] in {
3910 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
3911 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3912 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
3913 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3914 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
3915 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3916 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
3917 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3918 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
3919 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3920 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
3921 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3923 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3924 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3925 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3926 (bc_v16i8 (memopv2i64 addr:$src2)))),
3927 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3928 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3929 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3930 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3931 (bc_v8i16 (memopv2i64 addr:$src2)))),
3932 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3933 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3934 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3935 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3936 (bc_v4i32 (memopv2i64 addr:$src2)))),
3937 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3939 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3940 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3941 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
3942 (bc_v16i8 (memopv2i64 addr:$src2)))),
3943 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3944 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3945 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3946 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
3947 (bc_v8i16 (memopv2i64 addr:$src2)))),
3948 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3949 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3950 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3951 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
3952 (bc_v4i32 (memopv2i64 addr:$src2)))),
3953 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3956 let Predicates = [HasAVX2] in {
3957 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
3958 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3959 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
3960 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3961 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
3962 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3963 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
3964 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3965 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
3966 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3967 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
3968 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3970 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, VR256:$src2)),
3971 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
3972 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1,
3973 (bc_v32i8 (memopv4i64 addr:$src2)))),
3974 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
3975 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, VR256:$src2)),
3976 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
3977 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1,
3978 (bc_v16i16 (memopv4i64 addr:$src2)))),
3979 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
3980 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, VR256:$src2)),
3981 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
3982 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1,
3983 (bc_v8i32 (memopv4i64 addr:$src2)))),
3984 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
3986 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, VR256:$src2)),
3987 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
3988 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1,
3989 (bc_v32i8 (memopv4i64 addr:$src2)))),
3990 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
3991 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, VR256:$src2)),
3992 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
3993 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1,
3994 (bc_v16i16 (memopv4i64 addr:$src2)))),
3995 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
3996 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, VR256:$src2)),
3997 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
3998 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1,
3999 (bc_v8i32 (memopv4i64 addr:$src2)))),
4000 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
4003 let Constraints = "$src1 = $dst" in {
4004 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
4005 VR128, memopv2i64, i128mem, 1>;
4006 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
4007 VR128, memopv2i64, i128mem, 1>;
4008 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
4009 VR128, memopv2i64, i128mem, 1>;
4010 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
4011 VR128, memopv2i64, i128mem>;
4012 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
4013 VR128, memopv2i64, i128mem>;
4014 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
4015 VR128, memopv2i64, i128mem>;
4016 } // Constraints = "$src1 = $dst"
4018 let Predicates = [HasSSE2] in {
4019 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
4020 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
4021 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
4022 (bc_v16i8 (memopv2i64 addr:$src2)))),
4023 (PCMPEQBrm VR128:$src1, addr:$src2)>;
4024 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
4025 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
4026 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
4027 (bc_v8i16 (memopv2i64 addr:$src2)))),
4028 (PCMPEQWrm VR128:$src1, addr:$src2)>;
4029 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
4030 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
4031 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
4032 (bc_v4i32 (memopv2i64 addr:$src2)))),
4033 (PCMPEQDrm VR128:$src1, addr:$src2)>;
4035 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
4036 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
4037 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
4038 (bc_v16i8 (memopv2i64 addr:$src2)))),
4039 (PCMPGTBrm VR128:$src1, addr:$src2)>;
4040 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
4041 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
4042 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
4043 (bc_v8i16 (memopv2i64 addr:$src2)))),
4044 (PCMPGTWrm VR128:$src1, addr:$src2)>;
4045 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
4046 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
4047 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
4048 (bc_v4i32 (memopv2i64 addr:$src2)))),
4049 (PCMPGTDrm VR128:$src1, addr:$src2)>;
4052 //===---------------------------------------------------------------------===//
4053 // SSE2 - Packed Integer Pack Instructions
4054 //===---------------------------------------------------------------------===//
4056 let Predicates = [HasAVX] in {
4057 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4058 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4059 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4060 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4061 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4062 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4065 let Predicates = [HasAVX2] in {
4066 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4067 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4068 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4069 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4070 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4071 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4074 let Constraints = "$src1 = $dst" in {
4075 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4076 VR128, memopv2i64, i128mem>;
4077 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4078 VR128, memopv2i64, i128mem>;
4079 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4080 VR128, memopv2i64, i128mem>;
4081 } // Constraints = "$src1 = $dst"
4083 //===---------------------------------------------------------------------===//
4084 // SSE2 - Packed Integer Shuffle Instructions
4085 //===---------------------------------------------------------------------===//
4087 let ExeDomain = SSEPackedInt in {
4088 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4090 def ri : Ii8<0x70, MRMSrcReg,
4091 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4092 !strconcat(OpcodeStr,
4093 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4094 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4096 def mi : Ii8<0x70, MRMSrcMem,
4097 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4098 !strconcat(OpcodeStr,
4099 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4100 [(set VR128:$dst, (vt (pshuf_frag:$src2
4101 (bc_frag (memopv2i64 addr:$src1)),
4105 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4107 def Yri : Ii8<0x70, MRMSrcReg,
4108 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4109 !strconcat(OpcodeStr,
4110 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4111 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4113 def Ymi : Ii8<0x70, MRMSrcMem,
4114 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4115 !strconcat(OpcodeStr,
4116 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4117 [(set VR256:$dst, (vt (pshuf_frag:$src2
4118 (bc_frag (memopv4i64 addr:$src1)),
4121 } // ExeDomain = SSEPackedInt
4123 let Predicates = [HasAVX] in {
4124 let AddedComplexity = 5 in
4125 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4128 // SSE2 with ImmT == Imm8 and XS prefix.
4129 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4132 // SSE2 with ImmT == Imm8 and XD prefix.
4133 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4136 let AddedComplexity = 5 in
4137 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4138 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4139 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4140 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4141 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4143 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4145 (VPSHUFDmi addr:$src1, imm:$imm)>;
4146 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4148 (VPSHUFDmi addr:$src1, imm:$imm)>;
4149 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4150 (VPSHUFDri VR128:$src1, imm:$imm)>;
4151 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4152 (VPSHUFDri VR128:$src1, imm:$imm)>;
4153 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4154 (VPSHUFHWri VR128:$src, imm:$imm)>;
4155 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4157 (VPSHUFHWmi addr:$src, imm:$imm)>;
4158 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4159 (VPSHUFLWri VR128:$src, imm:$imm)>;
4160 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4162 (VPSHUFLWmi addr:$src, imm:$imm)>;
4165 let Predicates = [HasAVX2] in {
4166 let AddedComplexity = 5 in
4167 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4170 // SSE2 with ImmT == Imm8 and XS prefix.
4171 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4174 // SSE2 with ImmT == Imm8 and XD prefix.
4175 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4179 let Predicates = [HasSSE2] in {
4180 let AddedComplexity = 5 in
4181 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4183 // SSE2 with ImmT == Imm8 and XS prefix.
4184 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4186 // SSE2 with ImmT == Imm8 and XD prefix.
4187 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4189 let AddedComplexity = 5 in
4190 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4191 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4192 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4193 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4194 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4196 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4198 (PSHUFDmi addr:$src1, imm:$imm)>;
4199 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4201 (PSHUFDmi addr:$src1, imm:$imm)>;
4202 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4203 (PSHUFDri VR128:$src1, imm:$imm)>;
4204 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4205 (PSHUFDri VR128:$src1, imm:$imm)>;
4206 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4207 (PSHUFHWri VR128:$src, imm:$imm)>;
4208 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4210 (PSHUFHWmi addr:$src, imm:$imm)>;
4211 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4212 (PSHUFLWri VR128:$src, imm:$imm)>;
4213 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4215 (PSHUFLWmi addr:$src, imm:$imm)>;
4218 //===---------------------------------------------------------------------===//
4219 // SSE2 - Packed Integer Unpack Instructions
4220 //===---------------------------------------------------------------------===//
4222 let ExeDomain = SSEPackedInt in {
4223 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4224 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4225 def rr : PDI<opc, MRMSrcReg,
4226 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4228 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4229 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4230 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4231 def rm : PDI<opc, MRMSrcMem,
4232 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4234 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4235 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4236 [(set VR128:$dst, (OpNode VR128:$src1,
4237 (bc_frag (memopv2i64
4241 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4242 SDNode OpNode, PatFrag bc_frag> {
4243 def Yrr : PDI<opc, MRMSrcReg,
4244 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4245 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4246 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4247 def Yrm : PDI<opc, MRMSrcMem,
4248 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4249 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4250 [(set VR256:$dst, (OpNode VR256:$src1,
4251 (bc_frag (memopv4i64 addr:$src2))))]>;
4254 let Predicates = [HasAVX] in {
4255 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4256 bc_v16i8, 0>, VEX_4V;
4257 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4258 bc_v8i16, 0>, VEX_4V;
4259 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4260 bc_v4i32, 0>, VEX_4V;
4261 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4262 bc_v2i64, 0>, VEX_4V;
4264 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4265 bc_v16i8, 0>, VEX_4V;
4266 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4267 bc_v8i16, 0>, VEX_4V;
4268 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4269 bc_v4i32, 0>, VEX_4V;
4270 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4271 bc_v2i64, 0>, VEX_4V;
4274 let Predicates = [HasAVX2] in {
4275 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4277 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4279 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4281 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4284 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4286 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4288 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4290 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4294 let Constraints = "$src1 = $dst" in {
4295 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4297 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4299 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4301 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4304 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4306 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4308 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4310 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4313 } // ExeDomain = SSEPackedInt
4315 // Patterns for using AVX1 instructions with integer vectors
4316 // Here to give AVX2 priority
4317 let Predicates = [HasAVX] in {
4318 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4319 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4320 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4321 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4322 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4323 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4324 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4325 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4327 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4328 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4329 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4330 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4331 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4332 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4333 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4334 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4337 // Splat v2f64 / v2i64
4338 let AddedComplexity = 10 in {
4339 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4340 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4341 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4342 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4345 //===---------------------------------------------------------------------===//
4346 // SSE2 - Packed Integer Extract and Insert
4347 //===---------------------------------------------------------------------===//
4349 let ExeDomain = SSEPackedInt in {
4350 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4351 def rri : Ii8<0xC4, MRMSrcReg,
4352 (outs VR128:$dst), (ins VR128:$src1,
4353 GR32:$src2, i32i8imm:$src3),
4355 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4356 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4358 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4359 def rmi : Ii8<0xC4, MRMSrcMem,
4360 (outs VR128:$dst), (ins VR128:$src1,
4361 i16mem:$src2, i32i8imm:$src3),
4363 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4364 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4366 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4371 let Predicates = [HasAVX] in
4372 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4373 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4374 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4375 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4376 imm:$src2))]>, TB, OpSize, VEX;
4377 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4378 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4379 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4380 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4384 let Predicates = [HasAVX] in {
4385 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4386 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4387 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4388 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4389 []>, TB, OpSize, VEX_4V;
4392 let Constraints = "$src1 = $dst" in
4393 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4395 } // ExeDomain = SSEPackedInt
4397 //===---------------------------------------------------------------------===//
4398 // SSE2 - Packed Mask Creation
4399 //===---------------------------------------------------------------------===//
4401 let ExeDomain = SSEPackedInt in {
4403 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4404 "pmovmskb\t{$src, $dst|$dst, $src}",
4405 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4406 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4407 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4409 let Predicates = [HasAVX2] in {
4410 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4411 "pmovmskb\t{$src, $dst|$dst, $src}",
4412 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4413 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4414 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4417 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4418 "pmovmskb\t{$src, $dst|$dst, $src}",
4419 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4421 } // ExeDomain = SSEPackedInt
4423 //===---------------------------------------------------------------------===//
4424 // SSE2 - Conditional Store
4425 //===---------------------------------------------------------------------===//
4427 let ExeDomain = SSEPackedInt in {
4430 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4431 (ins VR128:$src, VR128:$mask),
4432 "maskmovdqu\t{$mask, $src|$src, $mask}",
4433 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4435 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4436 (ins VR128:$src, VR128:$mask),
4437 "maskmovdqu\t{$mask, $src|$src, $mask}",
4438 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4441 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4442 "maskmovdqu\t{$mask, $src|$src, $mask}",
4443 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4445 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4446 "maskmovdqu\t{$mask, $src|$src, $mask}",
4447 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4449 } // ExeDomain = SSEPackedInt
4451 //===---------------------------------------------------------------------===//
4452 // SSE2 - Move Doubleword
4453 //===---------------------------------------------------------------------===//
4455 //===---------------------------------------------------------------------===//
4456 // Move Int Doubleword to Packed Double Int
4458 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4459 "movd\t{$src, $dst|$dst, $src}",
4461 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4462 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4463 "movd\t{$src, $dst|$dst, $src}",
4465 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4467 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4468 "mov{d|q}\t{$src, $dst|$dst, $src}",
4470 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4471 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4472 "mov{d|q}\t{$src, $dst|$dst, $src}",
4473 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4475 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4476 "movd\t{$src, $dst|$dst, $src}",
4478 (v4i32 (scalar_to_vector GR32:$src)))]>;
4479 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4480 "movd\t{$src, $dst|$dst, $src}",
4482 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4483 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4484 "mov{d|q}\t{$src, $dst|$dst, $src}",
4486 (v2i64 (scalar_to_vector GR64:$src)))]>;
4487 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4488 "mov{d|q}\t{$src, $dst|$dst, $src}",
4489 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4491 //===---------------------------------------------------------------------===//
4492 // Move Int Doubleword to Single Scalar
4494 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4495 "movd\t{$src, $dst|$dst, $src}",
4496 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4498 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4499 "movd\t{$src, $dst|$dst, $src}",
4500 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4502 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4503 "movd\t{$src, $dst|$dst, $src}",
4504 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4506 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4507 "movd\t{$src, $dst|$dst, $src}",
4508 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4510 //===---------------------------------------------------------------------===//
4511 // Move Packed Doubleword Int to Packed Double Int
4513 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4514 "movd\t{$src, $dst|$dst, $src}",
4515 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4517 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4518 (ins i32mem:$dst, VR128:$src),
4519 "movd\t{$src, $dst|$dst, $src}",
4520 [(store (i32 (vector_extract (v4i32 VR128:$src),
4521 (iPTR 0))), addr:$dst)]>, VEX;
4522 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4523 "movd\t{$src, $dst|$dst, $src}",
4524 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4526 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4527 "movd\t{$src, $dst|$dst, $src}",
4528 [(store (i32 (vector_extract (v4i32 VR128:$src),
4529 (iPTR 0))), addr:$dst)]>;
4531 //===---------------------------------------------------------------------===//
4532 // Move Packed Doubleword Int first element to Doubleword Int
4534 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4535 "mov{d|q}\t{$src, $dst|$dst, $src}",
4536 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4538 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4540 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4541 "mov{d|q}\t{$src, $dst|$dst, $src}",
4542 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4545 //===---------------------------------------------------------------------===//
4546 // Bitcast FR64 <-> GR64
4548 let Predicates = [HasAVX] in
4549 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4550 "vmovq\t{$src, $dst|$dst, $src}",
4551 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4553 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4554 "mov{d|q}\t{$src, $dst|$dst, $src}",
4555 [(set GR64:$dst, (bitconvert FR64:$src))]>, VEX;
4556 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4557 "movq\t{$src, $dst|$dst, $src}",
4558 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>,
4561 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4562 "movq\t{$src, $dst|$dst, $src}",
4563 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4564 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4565 "mov{d|q}\t{$src, $dst|$dst, $src}",
4566 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4567 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4568 "movq\t{$src, $dst|$dst, $src}",
4569 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4571 //===---------------------------------------------------------------------===//
4572 // Move Scalar Single to Double Int
4574 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4575 "movd\t{$src, $dst|$dst, $src}",
4576 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4577 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4578 "movd\t{$src, $dst|$dst, $src}",
4579 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4580 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4581 "movd\t{$src, $dst|$dst, $src}",
4582 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4583 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4584 "movd\t{$src, $dst|$dst, $src}",
4585 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4587 //===---------------------------------------------------------------------===//
4588 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4590 let AddedComplexity = 15 in {
4591 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4592 "movd\t{$src, $dst|$dst, $src}",
4593 [(set VR128:$dst, (v4i32 (X86vzmovl
4594 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4596 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4597 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4598 [(set VR128:$dst, (v2i64 (X86vzmovl
4599 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4602 let AddedComplexity = 15 in {
4603 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4604 "movd\t{$src, $dst|$dst, $src}",
4605 [(set VR128:$dst, (v4i32 (X86vzmovl
4606 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4607 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$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:$src)))))]>;
4613 let AddedComplexity = 20 in {
4614 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4615 "movd\t{$src, $dst|$dst, $src}",
4617 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4618 (loadi32 addr:$src))))))]>,
4620 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4621 "movd\t{$src, $dst|$dst, $src}",
4623 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4624 (loadi32 addr:$src))))))]>;
4627 let Predicates = [HasAVX] in {
4628 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4629 let AddedComplexity = 20 in {
4630 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4631 (VMOVZDI2PDIrm addr:$src)>;
4632 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4633 (VMOVZDI2PDIrm addr:$src)>;
4634 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4635 (VMOVZDI2PDIrm addr:$src)>;
4637 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4638 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4639 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4640 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4641 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4642 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4643 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4646 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4647 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4648 (MOVZDI2PDIrm addr:$src)>;
4649 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4650 (MOVZDI2PDIrm addr:$src)>;
4651 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4652 (MOVZDI2PDIrm addr:$src)>;
4655 // These are the correct encodings of the instructions so that we know how to
4656 // read correct assembly, even though we continue to emit the wrong ones for
4657 // compatibility with Darwin's buggy assembler.
4658 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4659 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4660 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4661 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4662 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4663 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4664 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4665 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4666 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4667 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4668 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4669 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4671 //===---------------------------------------------------------------------===//
4672 // SSE2 - Move Quadword
4673 //===---------------------------------------------------------------------===//
4675 //===---------------------------------------------------------------------===//
4676 // Move Quadword Int to Packed Quadword Int
4678 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4679 "vmovq\t{$src, $dst|$dst, $src}",
4681 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4682 VEX, Requires<[HasAVX]>;
4683 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4684 "movq\t{$src, $dst|$dst, $src}",
4686 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4687 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4689 //===---------------------------------------------------------------------===//
4690 // Move Packed Quadword Int to Quadword Int
4692 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4693 "movq\t{$src, $dst|$dst, $src}",
4694 [(store (i64 (vector_extract (v2i64 VR128:$src),
4695 (iPTR 0))), addr:$dst)]>, VEX;
4696 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4697 "movq\t{$src, $dst|$dst, $src}",
4698 [(store (i64 (vector_extract (v2i64 VR128:$src),
4699 (iPTR 0))), addr:$dst)]>;
4701 //===---------------------------------------------------------------------===//
4702 // Store / copy lower 64-bits of a XMM register.
4704 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4705 "movq\t{$src, $dst|$dst, $src}",
4706 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4707 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4708 "movq\t{$src, $dst|$dst, $src}",
4709 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4711 let AddedComplexity = 20 in
4712 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4713 "vmovq\t{$src, $dst|$dst, $src}",
4715 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4716 (loadi64 addr:$src))))))]>,
4717 XS, VEX, Requires<[HasAVX]>;
4719 let AddedComplexity = 20 in
4720 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4721 "movq\t{$src, $dst|$dst, $src}",
4723 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4724 (loadi64 addr:$src))))))]>,
4725 XS, Requires<[HasSSE2]>;
4727 let Predicates = [HasAVX], AddedComplexity = 20 in {
4728 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4729 (VMOVZQI2PQIrm addr:$src)>;
4730 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4731 (VMOVZQI2PQIrm addr:$src)>;
4732 def : Pat<(v2i64 (X86vzload addr:$src)),
4733 (VMOVZQI2PQIrm addr:$src)>;
4736 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4737 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4738 (MOVZQI2PQIrm addr:$src)>;
4739 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4740 (MOVZQI2PQIrm addr:$src)>;
4741 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4744 let Predicates = [HasAVX] in {
4745 def : Pat<(v4i64 (X86vzload addr:$src)),
4746 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4749 //===---------------------------------------------------------------------===//
4750 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4751 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4753 let AddedComplexity = 15 in
4754 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4755 "vmovq\t{$src, $dst|$dst, $src}",
4756 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4757 XS, VEX, Requires<[HasAVX]>;
4758 let AddedComplexity = 15 in
4759 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4760 "movq\t{$src, $dst|$dst, $src}",
4761 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4762 XS, Requires<[HasSSE2]>;
4764 let AddedComplexity = 20 in
4765 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4766 "vmovq\t{$src, $dst|$dst, $src}",
4767 [(set VR128:$dst, (v2i64 (X86vzmovl
4768 (loadv2i64 addr:$src))))]>,
4769 XS, VEX, Requires<[HasAVX]>;
4770 let AddedComplexity = 20 in {
4771 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4772 "movq\t{$src, $dst|$dst, $src}",
4773 [(set VR128:$dst, (v2i64 (X86vzmovl
4774 (loadv2i64 addr:$src))))]>,
4775 XS, Requires<[HasSSE2]>;
4778 let AddedComplexity = 20 in {
4779 let Predicates = [HasAVX] in {
4780 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4781 (VMOVZPQILo2PQIrm addr:$src)>;
4782 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4783 (VMOVZPQILo2PQIrr VR128:$src)>;
4785 let Predicates = [HasSSE2] in {
4786 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4787 (MOVZPQILo2PQIrm addr:$src)>;
4788 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4789 (MOVZPQILo2PQIrr VR128:$src)>;
4793 // Instructions to match in the assembler
4794 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4795 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4796 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4797 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4798 // Recognize "movd" with GR64 destination, but encode as a "movq"
4799 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4800 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4802 // Instructions for the disassembler
4803 // xr = XMM register
4806 let Predicates = [HasAVX] in
4807 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4808 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4809 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4810 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4812 //===---------------------------------------------------------------------===//
4813 // SSE3 - Conversion Instructions
4814 //===---------------------------------------------------------------------===//
4816 // Convert Packed Double FP to Packed DW Integers
4817 let Predicates = [HasAVX] in {
4818 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4819 // register, but the same isn't true when using memory operands instead.
4820 // Provide other assembly rr and rm forms to address this explicitly.
4821 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4822 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4823 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4824 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4827 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4828 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4829 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4830 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4833 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4834 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4835 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4836 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4839 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4840 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4841 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4842 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4844 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4845 (VCVTPD2DQYrr VR256:$src)>;
4846 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4847 (VCVTPD2DQYrm addr:$src)>;
4849 // Convert Packed DW Integers to Packed Double FP
4850 let Predicates = [HasAVX] in {
4851 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4852 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4853 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4854 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4855 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4856 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4857 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4858 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4861 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4862 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4863 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4864 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4866 // AVX 256-bit register conversion intrinsics
4867 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4868 (VCVTDQ2PDYrr VR128:$src)>;
4869 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4870 (VCVTDQ2PDYrm addr:$src)>;
4872 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4873 (VCVTPD2DQYrr VR256:$src)>;
4874 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4875 (VCVTPD2DQYrm addr:$src)>;
4877 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4878 (VCVTDQ2PDYrr VR128:$src)>;
4879 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4880 (VCVTDQ2PDYrm addr:$src)>;
4882 //===---------------------------------------------------------------------===//
4883 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4884 //===---------------------------------------------------------------------===//
4885 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4886 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4887 X86MemOperand x86memop> {
4888 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4889 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4890 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4891 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4892 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4893 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4896 let Predicates = [HasAVX] in {
4897 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4898 v4f32, VR128, memopv4f32, f128mem>, VEX;
4899 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4900 v4f32, VR128, memopv4f32, f128mem>, VEX;
4901 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4902 v8f32, VR256, memopv8f32, f256mem>, VEX;
4903 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4904 v8f32, VR256, memopv8f32, f256mem>, VEX;
4906 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4907 memopv4f32, f128mem>;
4908 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4909 memopv4f32, f128mem>;
4911 let Predicates = [HasAVX] in {
4912 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4913 (VMOVSHDUPrr VR128:$src)>;
4914 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4915 (VMOVSHDUPrm addr:$src)>;
4916 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4917 (VMOVSLDUPrr VR128:$src)>;
4918 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4919 (VMOVSLDUPrm addr:$src)>;
4920 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4921 (VMOVSHDUPYrr VR256:$src)>;
4922 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4923 (VMOVSHDUPYrm addr:$src)>;
4924 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4925 (VMOVSLDUPYrr VR256:$src)>;
4926 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4927 (VMOVSLDUPYrm addr:$src)>;
4930 let Predicates = [HasSSE3] in {
4931 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4932 (MOVSHDUPrr VR128:$src)>;
4933 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4934 (MOVSHDUPrm addr:$src)>;
4935 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4936 (MOVSLDUPrr VR128:$src)>;
4937 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4938 (MOVSLDUPrm addr:$src)>;
4941 //===---------------------------------------------------------------------===//
4942 // SSE3 - Replicate Double FP - MOVDDUP
4943 //===---------------------------------------------------------------------===//
4945 multiclass sse3_replicate_dfp<string OpcodeStr> {
4946 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4947 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4948 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4949 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4950 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4952 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4956 // FIXME: Merge with above classe when there're patterns for the ymm version
4957 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4958 let Predicates = [HasAVX] in {
4959 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4960 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4962 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4963 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4968 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4969 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4970 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4972 let Predicates = [HasAVX] in {
4973 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4975 (VMOVDDUPrm addr:$src)>;
4976 let AddedComplexity = 5 in {
4977 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4978 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4979 (VMOVDDUPrm addr:$src)>;
4980 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4981 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4982 (VMOVDDUPrm addr:$src)>;
4984 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4985 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4986 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4987 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4988 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4989 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4990 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4991 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4992 def : Pat<(X86Movddup (bc_v2f64
4993 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4994 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4997 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4998 (VMOVDDUPYrm addr:$src)>;
4999 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
5000 (VMOVDDUPYrm addr:$src)>;
5001 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
5002 (VMOVDDUPYrm addr:$src)>;
5003 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
5004 (VMOVDDUPYrm addr:$src)>;
5005 def : Pat<(X86Movddup (v4f64 VR256:$src)),
5006 (VMOVDDUPYrr VR256:$src)>;
5007 def : Pat<(X86Movddup (v4i64 VR256:$src)),
5008 (VMOVDDUPYrr VR256:$src)>;
5011 let Predicates = [HasSSE3] in {
5012 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
5014 (MOVDDUPrm addr:$src)>;
5015 let AddedComplexity = 5 in {
5016 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
5017 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
5018 (MOVDDUPrm addr:$src)>;
5019 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
5020 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
5021 (MOVDDUPrm addr:$src)>;
5023 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5024 (MOVDDUPrm addr:$src)>;
5025 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5026 (MOVDDUPrm addr:$src)>;
5027 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5028 (MOVDDUPrm addr:$src)>;
5029 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5030 (MOVDDUPrm addr:$src)>;
5031 def : Pat<(X86Movddup (bc_v2f64
5032 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5033 (MOVDDUPrm addr:$src)>;
5036 //===---------------------------------------------------------------------===//
5037 // SSE3 - Move Unaligned Integer
5038 //===---------------------------------------------------------------------===//
5040 let Predicates = [HasAVX] in {
5041 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5042 "vlddqu\t{$src, $dst|$dst, $src}",
5043 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5044 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5045 "vlddqu\t{$src, $dst|$dst, $src}",
5046 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5048 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5049 "lddqu\t{$src, $dst|$dst, $src}",
5050 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
5052 //===---------------------------------------------------------------------===//
5053 // SSE3 - Arithmetic
5054 //===---------------------------------------------------------------------===//
5056 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5057 X86MemOperand x86memop, bit Is2Addr = 1> {
5058 def rr : I<0xD0, MRMSrcReg,
5059 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5061 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5062 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5063 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
5064 def rm : I<0xD0, MRMSrcMem,
5065 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5067 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5068 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5069 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
5072 let Predicates = [HasAVX] in {
5073 let ExeDomain = SSEPackedSingle in {
5074 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5075 f128mem, 0>, TB, XD, VEX_4V;
5076 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5077 f256mem, 0>, TB, XD, VEX_4V;
5079 let ExeDomain = SSEPackedDouble in {
5080 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5081 f128mem, 0>, TB, OpSize, VEX_4V;
5082 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5083 f256mem, 0>, TB, OpSize, VEX_4V;
5086 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5087 let ExeDomain = SSEPackedSingle in
5088 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5090 let ExeDomain = SSEPackedDouble in
5091 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5092 f128mem>, TB, OpSize;
5095 //===---------------------------------------------------------------------===//
5096 // SSE3 Instructions
5097 //===---------------------------------------------------------------------===//
5100 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5101 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5102 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5104 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5105 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5106 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5108 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5110 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5111 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5112 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5114 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5115 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5116 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5118 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5119 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5120 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5122 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5124 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5125 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5126 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5129 let Predicates = [HasAVX] in {
5130 let ExeDomain = SSEPackedSingle in {
5131 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5132 X86fhadd, 0>, VEX_4V;
5133 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5134 X86fhsub, 0>, VEX_4V;
5135 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5136 X86fhadd, 0>, VEX_4V;
5137 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5138 X86fhsub, 0>, VEX_4V;
5140 let ExeDomain = SSEPackedDouble in {
5141 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5142 X86fhadd, 0>, VEX_4V;
5143 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5144 X86fhsub, 0>, VEX_4V;
5145 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5146 X86fhadd, 0>, VEX_4V;
5147 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5148 X86fhsub, 0>, VEX_4V;
5152 let Constraints = "$src1 = $dst" in {
5153 let ExeDomain = SSEPackedSingle in {
5154 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5155 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5157 let ExeDomain = SSEPackedDouble in {
5158 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5159 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5163 //===---------------------------------------------------------------------===//
5164 // SSSE3 - Packed Absolute Instructions
5165 //===---------------------------------------------------------------------===//
5168 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5169 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5170 Intrinsic IntId128> {
5171 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5173 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5174 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5177 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5179 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5182 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5185 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5186 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5187 Intrinsic IntId256> {
5188 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5190 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5191 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5194 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5196 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5199 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5202 let Predicates = [HasAVX] in {
5203 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5204 int_x86_ssse3_pabs_b_128>, VEX;
5205 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5206 int_x86_ssse3_pabs_w_128>, VEX;
5207 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5208 int_x86_ssse3_pabs_d_128>, VEX;
5211 let Predicates = [HasAVX2] in {
5212 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5213 int_x86_avx2_pabs_b>, VEX;
5214 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5215 int_x86_avx2_pabs_w>, VEX;
5216 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5217 int_x86_avx2_pabs_d>, VEX;
5220 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5221 int_x86_ssse3_pabs_b_128>;
5222 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5223 int_x86_ssse3_pabs_w_128>;
5224 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5225 int_x86_ssse3_pabs_d_128>;
5227 //===---------------------------------------------------------------------===//
5228 // SSSE3 - Packed Binary Operator Instructions
5229 //===---------------------------------------------------------------------===//
5231 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5232 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5233 Intrinsic IntId128, bit Is2Addr = 1> {
5234 let isCommutable = 1 in
5235 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5236 (ins VR128:$src1, VR128:$src2),
5238 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5239 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5240 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5242 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5243 (ins VR128:$src1, i128mem:$src2),
5245 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5246 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5248 (IntId128 VR128:$src1,
5249 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5252 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5253 Intrinsic IntId256> {
5254 let isCommutable = 1 in
5255 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5256 (ins VR256:$src1, VR256:$src2),
5257 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5258 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5260 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5261 (ins VR256:$src1, i256mem:$src2),
5262 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5264 (IntId256 VR256:$src1,
5265 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5268 let ImmT = NoImm, Predicates = [HasAVX] in {
5269 let isCommutable = 0 in {
5270 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw",
5271 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5272 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd",
5273 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5274 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5275 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5276 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw",
5277 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5278 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd",
5279 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5280 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5281 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5282 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5283 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5284 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb",
5285 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5286 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb",
5287 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5288 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw",
5289 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5290 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd",
5291 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5293 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5294 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5297 let ImmT = NoImm, Predicates = [HasAVX2] in {
5298 let isCommutable = 0 in {
5299 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw",
5300 int_x86_avx2_phadd_w>, VEX_4V;
5301 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd",
5302 int_x86_avx2_phadd_d>, VEX_4V;
5303 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5304 int_x86_avx2_phadd_sw>, VEX_4V;
5305 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw",
5306 int_x86_avx2_phsub_w>, VEX_4V;
5307 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd",
5308 int_x86_avx2_phsub_d>, VEX_4V;
5309 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5310 int_x86_avx2_phsub_sw>, VEX_4V;
5311 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5312 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5313 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb",
5314 int_x86_avx2_pshuf_b>, VEX_4V;
5315 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb",
5316 int_x86_avx2_psign_b>, VEX_4V;
5317 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw",
5318 int_x86_avx2_psign_w>, VEX_4V;
5319 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd",
5320 int_x86_avx2_psign_d>, VEX_4V;
5322 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5323 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5326 // None of these have i8 immediate fields.
5327 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5328 let isCommutable = 0 in {
5329 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw",
5330 int_x86_ssse3_phadd_w_128>;
5331 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd",
5332 int_x86_ssse3_phadd_d_128>;
5333 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5334 int_x86_ssse3_phadd_sw_128>;
5335 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw",
5336 int_x86_ssse3_phsub_w_128>;
5337 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd",
5338 int_x86_ssse3_phsub_d_128>;
5339 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5340 int_x86_ssse3_phsub_sw_128>;
5341 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5342 int_x86_ssse3_pmadd_ub_sw_128>;
5343 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb",
5344 int_x86_ssse3_pshuf_b_128>;
5345 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb",
5346 int_x86_ssse3_psign_b_128>;
5347 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw",
5348 int_x86_ssse3_psign_w_128>;
5349 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd",
5350 int_x86_ssse3_psign_d_128>;
5352 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5353 int_x86_ssse3_pmul_hr_sw_128>;
5356 let Predicates = [HasAVX] in {
5357 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5358 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5359 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5360 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5362 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5363 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5364 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5365 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5366 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5367 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5369 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5370 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5371 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5372 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5373 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5374 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5375 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5376 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5379 let Predicates = [HasAVX2] in {
5380 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5381 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5382 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5383 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5384 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5385 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5387 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5388 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5389 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5390 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5391 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5392 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5393 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5394 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5397 let Predicates = [HasSSSE3] in {
5398 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5399 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5400 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5401 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5403 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5404 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5405 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5406 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5407 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5408 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5410 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5411 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5412 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5413 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5414 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5415 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5416 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5417 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5420 //===---------------------------------------------------------------------===//
5421 // SSSE3 - Packed Align Instruction Patterns
5422 //===---------------------------------------------------------------------===//
5424 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5425 let neverHasSideEffects = 1 in {
5426 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5427 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5429 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5431 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5434 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5435 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5437 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5439 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5444 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5445 let neverHasSideEffects = 1 in {
5446 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5447 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5449 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5452 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5453 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5455 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5460 let Predicates = [HasAVX] in
5461 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5462 let Predicates = [HasAVX2] in
5463 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5464 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5465 defm PALIGN : ssse3_palign<"palignr">;
5467 let Predicates = [HasAVX] in {
5468 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5469 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5470 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5471 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5472 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5473 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5474 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5475 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5478 let Predicates = [HasSSSE3] in {
5479 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5480 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5481 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5482 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5483 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5484 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5485 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5486 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5489 //===---------------------------------------------------------------------===//
5490 // SSSE3 - Thread synchronization
5491 //===---------------------------------------------------------------------===//
5493 let usesCustomInserter = 1 in {
5494 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5495 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5496 Requires<[HasSSE3]>;
5497 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5498 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5499 Requires<[HasSSE3]>;
5502 let Uses = [EAX, ECX, EDX] in
5503 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5504 Requires<[HasSSE3]>;
5505 let Uses = [ECX, EAX] in
5506 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5507 Requires<[HasSSE3]>;
5509 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5510 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5512 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5513 Requires<[In32BitMode]>;
5514 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5515 Requires<[In64BitMode]>;
5517 //===----------------------------------------------------------------------===//
5518 // SSE4.1 - Packed Move with Sign/Zero Extend
5519 //===----------------------------------------------------------------------===//
5521 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5522 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5523 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5524 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5526 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5527 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5529 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5533 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5535 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5536 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5537 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5539 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5540 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5541 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5544 let Predicates = [HasAVX] in {
5545 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5547 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5549 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5551 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5553 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5555 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5559 let Predicates = [HasAVX2] in {
5560 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5561 int_x86_avx2_pmovsxbw>, VEX;
5562 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5563 int_x86_avx2_pmovsxwd>, VEX;
5564 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5565 int_x86_avx2_pmovsxdq>, VEX;
5566 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5567 int_x86_avx2_pmovzxbw>, VEX;
5568 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5569 int_x86_avx2_pmovzxwd>, VEX;
5570 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5571 int_x86_avx2_pmovzxdq>, VEX;
5574 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5575 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5576 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5577 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5578 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5579 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5581 let Predicates = [HasAVX] in {
5582 // Common patterns involving scalar load.
5583 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5584 (VPMOVSXBWrm addr:$src)>;
5585 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5586 (VPMOVSXBWrm addr:$src)>;
5588 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5589 (VPMOVSXWDrm addr:$src)>;
5590 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5591 (VPMOVSXWDrm addr:$src)>;
5593 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5594 (VPMOVSXDQrm addr:$src)>;
5595 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5596 (VPMOVSXDQrm addr:$src)>;
5598 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5599 (VPMOVZXBWrm addr:$src)>;
5600 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5601 (VPMOVZXBWrm addr:$src)>;
5603 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5604 (VPMOVZXWDrm addr:$src)>;
5605 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5606 (VPMOVZXWDrm addr:$src)>;
5608 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5609 (VPMOVZXDQrm addr:$src)>;
5610 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5611 (VPMOVZXDQrm addr:$src)>;
5614 let Predicates = [HasSSE41] in {
5615 // Common patterns involving scalar load.
5616 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5617 (PMOVSXBWrm addr:$src)>;
5618 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5619 (PMOVSXBWrm addr:$src)>;
5621 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5622 (PMOVSXWDrm addr:$src)>;
5623 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5624 (PMOVSXWDrm addr:$src)>;
5626 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5627 (PMOVSXDQrm addr:$src)>;
5628 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5629 (PMOVSXDQrm addr:$src)>;
5631 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5632 (PMOVZXBWrm addr:$src)>;
5633 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5634 (PMOVZXBWrm addr:$src)>;
5636 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5637 (PMOVZXWDrm addr:$src)>;
5638 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5639 (PMOVZXWDrm addr:$src)>;
5641 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5642 (PMOVZXDQrm addr:$src)>;
5643 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5644 (PMOVZXDQrm addr:$src)>;
5648 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5649 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5650 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5651 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5653 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5654 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5656 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5660 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5662 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5663 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5664 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5666 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5667 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5669 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5673 let Predicates = [HasAVX] in {
5674 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5676 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5678 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5680 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5684 let Predicates = [HasAVX2] in {
5685 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5686 int_x86_avx2_pmovsxbd>, VEX;
5687 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5688 int_x86_avx2_pmovsxwq>, VEX;
5689 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5690 int_x86_avx2_pmovzxbd>, VEX;
5691 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5692 int_x86_avx2_pmovzxwq>, VEX;
5695 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5696 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5697 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5698 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5700 let Predicates = [HasAVX] in {
5701 // Common patterns involving scalar load
5702 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5703 (VPMOVSXBDrm addr:$src)>;
5704 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5705 (VPMOVSXWQrm addr:$src)>;
5707 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5708 (VPMOVZXBDrm addr:$src)>;
5709 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5710 (VPMOVZXWQrm addr:$src)>;
5713 let Predicates = [HasSSE41] in {
5714 // Common patterns involving scalar load
5715 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5716 (PMOVSXBDrm addr:$src)>;
5717 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5718 (PMOVSXWQrm addr:$src)>;
5720 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5721 (PMOVZXBDrm addr:$src)>;
5722 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5723 (PMOVZXWQrm addr:$src)>;
5726 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5727 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5728 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5729 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5731 // Expecting a i16 load any extended to i32 value.
5732 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5733 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5734 [(set VR128:$dst, (IntId (bitconvert
5735 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5739 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5741 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5742 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5743 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5745 // Expecting a i16 load any extended to i32 value.
5746 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5747 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5748 [(set VR256:$dst, (IntId (bitconvert
5749 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5753 let Predicates = [HasAVX] in {
5754 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5756 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5759 let Predicates = [HasAVX2] in {
5760 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5761 int_x86_avx2_pmovsxbq>, VEX;
5762 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5763 int_x86_avx2_pmovzxbq>, VEX;
5765 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5766 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5768 let Predicates = [HasAVX] in {
5769 // Common patterns involving scalar load
5770 def : Pat<(int_x86_sse41_pmovsxbq
5771 (bitconvert (v4i32 (X86vzmovl
5772 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5773 (VPMOVSXBQrm addr:$src)>;
5775 def : Pat<(int_x86_sse41_pmovzxbq
5776 (bitconvert (v4i32 (X86vzmovl
5777 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5778 (VPMOVZXBQrm addr:$src)>;
5781 let Predicates = [HasSSE41] in {
5782 // Common patterns involving scalar load
5783 def : Pat<(int_x86_sse41_pmovsxbq
5784 (bitconvert (v4i32 (X86vzmovl
5785 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5786 (PMOVSXBQrm addr:$src)>;
5788 def : Pat<(int_x86_sse41_pmovzxbq
5789 (bitconvert (v4i32 (X86vzmovl
5790 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5791 (PMOVZXBQrm addr:$src)>;
5794 //===----------------------------------------------------------------------===//
5795 // SSE4.1 - Extract Instructions
5796 //===----------------------------------------------------------------------===//
5798 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5799 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5800 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5801 (ins VR128:$src1, i32i8imm:$src2),
5802 !strconcat(OpcodeStr,
5803 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5804 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5806 let neverHasSideEffects = 1, mayStore = 1 in
5807 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5808 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5809 !strconcat(OpcodeStr,
5810 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5813 // There's an AssertZext in the way of writing the store pattern
5814 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5817 let Predicates = [HasAVX] in {
5818 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5819 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5820 (ins VR128:$src1, i32i8imm:$src2),
5821 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5824 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5827 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5828 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5829 let neverHasSideEffects = 1, mayStore = 1 in
5830 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5831 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5832 !strconcat(OpcodeStr,
5833 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5836 // There's an AssertZext in the way of writing the store pattern
5837 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5840 let Predicates = [HasAVX] in
5841 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5843 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5846 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5847 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5848 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5849 (ins VR128:$src1, i32i8imm:$src2),
5850 !strconcat(OpcodeStr,
5851 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5853 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5854 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5855 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5856 !strconcat(OpcodeStr,
5857 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5858 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5859 addr:$dst)]>, OpSize;
5862 let Predicates = [HasAVX] in
5863 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5865 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5867 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5868 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5869 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5870 (ins VR128:$src1, i32i8imm:$src2),
5871 !strconcat(OpcodeStr,
5872 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5874 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5875 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5876 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5877 !strconcat(OpcodeStr,
5878 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5879 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5880 addr:$dst)]>, OpSize, REX_W;
5883 let Predicates = [HasAVX] in
5884 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5886 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5888 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5890 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5891 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5892 (ins VR128:$src1, i32i8imm:$src2),
5893 !strconcat(OpcodeStr,
5894 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5896 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5898 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5899 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5900 !strconcat(OpcodeStr,
5901 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5902 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5903 addr:$dst)]>, OpSize;
5906 let ExeDomain = SSEPackedSingle in {
5907 let Predicates = [HasAVX] in {
5908 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5909 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5910 (ins VR128:$src1, i32i8imm:$src2),
5911 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5914 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5917 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5918 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5921 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5923 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5926 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5927 Requires<[HasSSE41]>;
5929 //===----------------------------------------------------------------------===//
5930 // SSE4.1 - Insert Instructions
5931 //===----------------------------------------------------------------------===//
5933 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5934 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5935 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5937 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5939 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5941 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5942 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5943 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5945 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5947 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5949 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5950 imm:$src3))]>, OpSize;
5953 let Predicates = [HasAVX] in
5954 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5955 let Constraints = "$src1 = $dst" in
5956 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5958 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5959 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5960 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5962 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5964 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5966 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5968 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5969 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5971 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5973 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5975 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5976 imm:$src3)))]>, OpSize;
5979 let Predicates = [HasAVX] in
5980 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5981 let Constraints = "$src1 = $dst" in
5982 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5984 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5985 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5986 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5988 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5990 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5992 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5994 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5995 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5997 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5999 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6001 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6002 imm:$src3)))]>, OpSize;
6005 let Predicates = [HasAVX] in
6006 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6007 let Constraints = "$src1 = $dst" in
6008 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6010 // insertps has a few different modes, there's the first two here below which
6011 // are optimized inserts that won't zero arbitrary elements in the destination
6012 // vector. The next one matches the intrinsic and could zero arbitrary elements
6013 // in the target vector.
6014 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6015 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6016 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6018 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6020 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6022 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6024 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6025 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6027 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6029 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6031 (X86insrtps VR128:$src1,
6032 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6033 imm:$src3))]>, OpSize;
6036 let ExeDomain = SSEPackedSingle in {
6037 let Predicates = [HasAVX] in
6038 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6039 let Constraints = "$src1 = $dst" in
6040 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6043 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6044 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6046 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6047 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6048 Requires<[HasSSE41]>;
6050 //===----------------------------------------------------------------------===//
6051 // SSE4.1 - Round Instructions
6052 //===----------------------------------------------------------------------===//
6054 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6055 X86MemOperand x86memop, RegisterClass RC,
6056 PatFrag mem_frag32, PatFrag mem_frag64,
6057 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6058 let ExeDomain = SSEPackedSingle in {
6059 // Intrinsic operation, reg.
6060 // Vector intrinsic operation, reg
6061 def PSr : SS4AIi8<opcps, MRMSrcReg,
6062 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6063 !strconcat(OpcodeStr,
6064 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6065 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6068 // Vector intrinsic operation, mem
6069 def PSm : SS4AIi8<opcps, MRMSrcMem,
6070 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6071 !strconcat(OpcodeStr,
6072 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6074 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6076 } // ExeDomain = SSEPackedSingle
6078 let ExeDomain = SSEPackedDouble in {
6079 // Vector intrinsic operation, reg
6080 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6081 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6082 !strconcat(OpcodeStr,
6083 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6084 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6087 // Vector intrinsic operation, mem
6088 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6089 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6090 !strconcat(OpcodeStr,
6091 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6093 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6095 } // ExeDomain = SSEPackedDouble
6098 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6101 Intrinsic F64Int, bit Is2Addr = 1> {
6102 let ExeDomain = GenericDomain in {
6104 def SSr : SS4AIi8<opcss, MRMSrcReg,
6105 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6107 !strconcat(OpcodeStr,
6108 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6109 !strconcat(OpcodeStr,
6110 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6113 // Intrinsic operation, reg.
6114 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6115 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6117 !strconcat(OpcodeStr,
6118 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6119 !strconcat(OpcodeStr,
6120 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6121 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6124 // Intrinsic operation, mem.
6125 def SSm : SS4AIi8<opcss, MRMSrcMem,
6126 (outs VR128:$dst), (ins VR128:$src1, ssmem:$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, $src3}")),
6133 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6137 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6138 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6140 !strconcat(OpcodeStr,
6141 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6142 !strconcat(OpcodeStr,
6143 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6146 // Intrinsic operation, reg.
6147 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6148 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6150 !strconcat(OpcodeStr,
6151 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6152 !strconcat(OpcodeStr,
6153 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6154 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6157 // Intrinsic operation, mem.
6158 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6159 (outs VR128:$dst), (ins VR128:$src1, sdmem:$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, $src3}")),
6166 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6168 } // ExeDomain = GenericDomain
6171 // FP round - roundss, roundps, roundsd, roundpd
6172 let Predicates = [HasAVX] in {
6174 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6175 memopv4f32, memopv2f64,
6176 int_x86_sse41_round_ps,
6177 int_x86_sse41_round_pd>, VEX;
6178 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6179 memopv8f32, memopv4f64,
6180 int_x86_avx_round_ps_256,
6181 int_x86_avx_round_pd_256>, VEX;
6182 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6183 int_x86_sse41_round_ss,
6184 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6186 def : Pat<(ffloor FR32:$src),
6187 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6188 def : Pat<(f64 (ffloor FR64:$src)),
6189 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6190 def : Pat<(f32 (fnearbyint FR32:$src)),
6191 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6192 def : Pat<(f64 (fnearbyint FR64:$src)),
6193 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6194 def : Pat<(f32 (fceil FR32:$src)),
6195 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6196 def : Pat<(f64 (fceil FR64:$src)),
6197 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6198 def : Pat<(f32 (frint FR32:$src)),
6199 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6200 def : Pat<(f64 (frint FR64:$src)),
6201 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6202 def : Pat<(f32 (ftrunc FR32:$src)),
6203 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6204 def : Pat<(f64 (ftrunc FR64:$src)),
6205 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6208 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6209 memopv4f32, memopv2f64,
6210 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6211 let Constraints = "$src1 = $dst" in
6212 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6213 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6215 def : Pat<(ffloor FR32:$src),
6216 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6217 def : Pat<(f64 (ffloor FR64:$src)),
6218 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6219 def : Pat<(f32 (fnearbyint FR32:$src)),
6220 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6221 def : Pat<(f64 (fnearbyint FR64:$src)),
6222 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6223 def : Pat<(f32 (fceil FR32:$src)),
6224 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6225 def : Pat<(f64 (fceil FR64:$src)),
6226 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6227 def : Pat<(f32 (frint FR32:$src)),
6228 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6229 def : Pat<(f64 (frint FR64:$src)),
6230 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6231 def : Pat<(f32 (ftrunc FR32:$src)),
6232 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6233 def : Pat<(f64 (ftrunc FR64:$src)),
6234 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6236 //===----------------------------------------------------------------------===//
6237 // SSE4.1 - Packed Bit Test
6238 //===----------------------------------------------------------------------===//
6240 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6241 // the intel intrinsic that corresponds to this.
6242 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6243 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6244 "vptest\t{$src2, $src1|$src1, $src2}",
6245 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6247 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6248 "vptest\t{$src2, $src1|$src1, $src2}",
6249 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6252 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6253 "vptest\t{$src2, $src1|$src1, $src2}",
6254 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6256 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6257 "vptest\t{$src2, $src1|$src1, $src2}",
6258 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6262 let Defs = [EFLAGS] in {
6263 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6264 "ptest\t{$src2, $src1|$src1, $src2}",
6265 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6267 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6268 "ptest\t{$src2, $src1|$src1, $src2}",
6269 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6273 // The bit test instructions below are AVX only
6274 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6275 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6276 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6277 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6278 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6279 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6280 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6281 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6285 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6286 let ExeDomain = SSEPackedSingle in {
6287 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6288 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6290 let ExeDomain = SSEPackedDouble in {
6291 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6292 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6296 //===----------------------------------------------------------------------===//
6297 // SSE4.1 - Misc Instructions
6298 //===----------------------------------------------------------------------===//
6300 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6301 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6302 "popcnt{w}\t{$src, $dst|$dst, $src}",
6303 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6305 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6306 "popcnt{w}\t{$src, $dst|$dst, $src}",
6307 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6308 (implicit EFLAGS)]>, OpSize, XS;
6310 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6311 "popcnt{l}\t{$src, $dst|$dst, $src}",
6312 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6314 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6315 "popcnt{l}\t{$src, $dst|$dst, $src}",
6316 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6317 (implicit EFLAGS)]>, XS;
6319 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6320 "popcnt{q}\t{$src, $dst|$dst, $src}",
6321 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6323 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6324 "popcnt{q}\t{$src, $dst|$dst, $src}",
6325 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6326 (implicit EFLAGS)]>, XS;
6331 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6332 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6333 Intrinsic IntId128> {
6334 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6336 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6337 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6338 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6340 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6343 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6346 let Predicates = [HasAVX] in
6347 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6348 int_x86_sse41_phminposuw>, VEX;
6349 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6350 int_x86_sse41_phminposuw>;
6352 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6353 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6354 Intrinsic IntId128, bit Is2Addr = 1> {
6355 let isCommutable = 1 in
6356 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6357 (ins VR128:$src1, VR128:$src2),
6359 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6360 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6361 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6362 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6363 (ins VR128:$src1, i128mem:$src2),
6365 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6366 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6368 (IntId128 VR128:$src1,
6369 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6372 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6373 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6374 Intrinsic IntId256> {
6375 let isCommutable = 1 in
6376 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6377 (ins VR256:$src1, VR256:$src2),
6378 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6379 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6380 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6381 (ins VR256:$src1, i256mem:$src2),
6382 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6384 (IntId256 VR256:$src1,
6385 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6388 let Predicates = [HasAVX] in {
6389 let isCommutable = 0 in
6390 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6392 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6394 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6396 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6398 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6400 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6402 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6404 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6406 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6408 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6410 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6413 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6414 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6415 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6416 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6419 let Predicates = [HasAVX2] in {
6420 let isCommutable = 0 in
6421 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6422 int_x86_avx2_packusdw>, VEX_4V;
6423 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6424 int_x86_avx2_pcmpeq_q>, VEX_4V;
6425 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6426 int_x86_avx2_pmins_b>, VEX_4V;
6427 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6428 int_x86_avx2_pmins_d>, VEX_4V;
6429 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6430 int_x86_avx2_pminu_d>, VEX_4V;
6431 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6432 int_x86_avx2_pminu_w>, VEX_4V;
6433 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6434 int_x86_avx2_pmaxs_b>, VEX_4V;
6435 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6436 int_x86_avx2_pmaxs_d>, VEX_4V;
6437 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6438 int_x86_avx2_pmaxu_d>, VEX_4V;
6439 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6440 int_x86_avx2_pmaxu_w>, VEX_4V;
6441 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6442 int_x86_avx2_pmul_dq>, VEX_4V;
6444 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, VR256:$src2)),
6445 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6446 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, (memop addr:$src2))),
6447 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6450 let Constraints = "$src1 = $dst" in {
6451 let isCommutable = 0 in
6452 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6453 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6454 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6455 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6456 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6457 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6458 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6459 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6460 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6461 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6462 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6465 let Predicates = [HasSSE41] in {
6466 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6467 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6468 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6469 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6472 /// SS48I_binop_rm - Simple SSE41 binary operator.
6473 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6474 ValueType OpVT, bit Is2Addr = 1> {
6475 let isCommutable = 1 in
6476 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6477 (ins VR128:$src1, VR128:$src2),
6479 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6480 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6481 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6483 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6484 (ins VR128:$src1, i128mem:$src2),
6486 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6487 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6488 [(set VR128:$dst, (OpNode VR128:$src1,
6489 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6493 /// SS48I_binop_rm - Simple SSE41 binary operator.
6494 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6496 let isCommutable = 1 in
6497 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6498 (ins VR256:$src1, VR256:$src2),
6499 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6500 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6502 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6503 (ins VR256:$src1, i256mem:$src2),
6504 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6505 [(set VR256:$dst, (OpNode VR256:$src1,
6506 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6510 let Predicates = [HasAVX] in
6511 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6512 let Predicates = [HasAVX2] in
6513 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6514 let Constraints = "$src1 = $dst" in
6515 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6517 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6518 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6519 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6520 X86MemOperand x86memop, bit Is2Addr = 1> {
6521 let isCommutable = 1 in
6522 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6523 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6525 !strconcat(OpcodeStr,
6526 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6527 !strconcat(OpcodeStr,
6528 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6529 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6531 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6532 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6534 !strconcat(OpcodeStr,
6535 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6536 !strconcat(OpcodeStr,
6537 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6540 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6544 let Predicates = [HasAVX] in {
6545 let isCommutable = 0 in {
6546 let ExeDomain = SSEPackedSingle in {
6547 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6548 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6549 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6550 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6552 let ExeDomain = SSEPackedDouble in {
6553 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6554 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6555 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6556 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6558 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6559 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6560 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6561 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6563 let ExeDomain = SSEPackedSingle in
6564 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6565 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6566 let ExeDomain = SSEPackedDouble in
6567 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6568 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6569 let ExeDomain = SSEPackedSingle in
6570 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6571 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6574 let Predicates = [HasAVX2] in {
6575 let isCommutable = 0 in {
6576 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6577 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6578 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6579 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6583 let Constraints = "$src1 = $dst" in {
6584 let isCommutable = 0 in {
6585 let ExeDomain = SSEPackedSingle in
6586 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6587 VR128, memopv4f32, i128mem>;
6588 let ExeDomain = SSEPackedDouble in
6589 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6590 VR128, memopv2f64, i128mem>;
6591 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6592 VR128, memopv2i64, i128mem>;
6593 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6594 VR128, memopv2i64, i128mem>;
6596 let ExeDomain = SSEPackedSingle in
6597 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6598 VR128, memopv4f32, i128mem>;
6599 let ExeDomain = SSEPackedDouble in
6600 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6601 VR128, memopv2f64, i128mem>;
6604 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6605 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6606 RegisterClass RC, X86MemOperand x86memop,
6607 PatFrag mem_frag, Intrinsic IntId> {
6608 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6609 (ins RC:$src1, RC:$src2, RC:$src3),
6610 !strconcat(OpcodeStr,
6611 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6612 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6613 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6615 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6616 (ins RC:$src1, x86memop:$src2, RC:$src3),
6617 !strconcat(OpcodeStr,
6618 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6620 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6622 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6625 let Predicates = [HasAVX] in {
6626 let ExeDomain = SSEPackedDouble in {
6627 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6628 memopv2f64, int_x86_sse41_blendvpd>;
6629 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6630 memopv4f64, int_x86_avx_blendv_pd_256>;
6631 } // ExeDomain = SSEPackedDouble
6632 let ExeDomain = SSEPackedSingle in {
6633 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6634 memopv4f32, int_x86_sse41_blendvps>;
6635 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6636 memopv8f32, int_x86_avx_blendv_ps_256>;
6637 } // ExeDomain = SSEPackedSingle
6638 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6639 memopv2i64, int_x86_sse41_pblendvb>;
6642 let Predicates = [HasAVX2] in {
6643 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6644 memopv4i64, int_x86_avx2_pblendvb>;
6647 let Predicates = [HasAVX] in {
6648 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6649 (v16i8 VR128:$src2))),
6650 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6651 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6652 (v4i32 VR128:$src2))),
6653 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6654 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6655 (v4f32 VR128:$src2))),
6656 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6657 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6658 (v2i64 VR128:$src2))),
6659 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6660 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6661 (v2f64 VR128:$src2))),
6662 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6663 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6664 (v8i32 VR256:$src2))),
6665 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6666 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6667 (v8f32 VR256:$src2))),
6668 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6669 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6670 (v4i64 VR256:$src2))),
6671 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6672 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6673 (v4f64 VR256:$src2))),
6674 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6677 let Predicates = [HasAVX2] in {
6678 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6679 (v32i8 VR256:$src2))),
6680 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6683 /// SS41I_ternary_int - SSE 4.1 ternary operator
6684 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6685 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6687 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6688 (ins VR128:$src1, VR128:$src2),
6689 !strconcat(OpcodeStr,
6690 "\t{$src2, $dst|$dst, $src2}"),
6691 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6694 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6695 (ins VR128:$src1, i128mem:$src2),
6696 !strconcat(OpcodeStr,
6697 "\t{$src2, $dst|$dst, $src2}"),
6700 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6704 let ExeDomain = SSEPackedDouble in
6705 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6706 int_x86_sse41_blendvpd>;
6707 let ExeDomain = SSEPackedSingle in
6708 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6709 int_x86_sse41_blendvps>;
6710 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6711 int_x86_sse41_pblendvb>;
6713 let Predicates = [HasSSE41] in {
6714 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6715 (v16i8 VR128:$src2))),
6716 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6717 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6718 (v4i32 VR128:$src2))),
6719 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6720 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6721 (v4f32 VR128:$src2))),
6722 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6723 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6724 (v2i64 VR128:$src2))),
6725 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6726 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6727 (v2f64 VR128:$src2))),
6728 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6731 let Predicates = [HasAVX] in
6732 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6733 "vmovntdqa\t{$src, $dst|$dst, $src}",
6734 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6736 let Predicates = [HasAVX2] in
6737 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6738 "vmovntdqa\t{$src, $dst|$dst, $src}",
6739 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6741 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6742 "movntdqa\t{$src, $dst|$dst, $src}",
6743 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6746 //===----------------------------------------------------------------------===//
6747 // SSE4.2 - Compare Instructions
6748 //===----------------------------------------------------------------------===//
6750 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6751 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6752 Intrinsic IntId128, bit Is2Addr = 1> {
6753 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6754 (ins VR128:$src1, VR128:$src2),
6756 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6757 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6758 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6760 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6761 (ins VR128:$src1, i128mem:$src2),
6763 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6764 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6766 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6769 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6770 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6771 Intrinsic IntId256> {
6772 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6773 (ins VR256:$src1, VR256:$src2),
6774 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6775 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6777 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6778 (ins VR256:$src1, i256mem:$src2),
6779 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6781 (IntId256 VR256:$src1, (memopv4i64 addr:$src2)))]>, OpSize;
6784 let Predicates = [HasAVX] in {
6785 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6788 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6789 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6790 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6791 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6794 let Predicates = [HasAVX2] in {
6795 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6798 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, VR256:$src2)),
6799 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6800 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, (memop addr:$src2))),
6801 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6804 let Constraints = "$src1 = $dst" in
6805 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6807 let Predicates = [HasSSE42] in {
6808 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6809 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6810 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6811 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6814 //===----------------------------------------------------------------------===//
6815 // SSE4.2 - String/text Processing Instructions
6816 //===----------------------------------------------------------------------===//
6818 // Packed Compare Implicit Length Strings, Return Mask
6819 multiclass pseudo_pcmpistrm<string asm> {
6820 def REG : PseudoI<(outs VR128:$dst),
6821 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6822 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6824 def MEM : PseudoI<(outs VR128:$dst),
6825 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6826 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6827 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6830 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6831 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6832 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6835 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6836 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6837 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6838 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6840 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6841 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6842 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6845 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6846 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6847 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6848 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6850 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6851 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6852 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6855 // Packed Compare Explicit Length Strings, Return Mask
6856 multiclass pseudo_pcmpestrm<string asm> {
6857 def REG : PseudoI<(outs VR128:$dst),
6858 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6859 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6860 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6861 def MEM : PseudoI<(outs VR128:$dst),
6862 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6863 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6864 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6867 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6868 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6869 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6872 let Predicates = [HasAVX],
6873 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6874 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6875 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6876 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6878 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6879 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6880 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6883 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6884 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6885 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6886 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6888 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6889 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6890 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6893 // Packed Compare Implicit Length Strings, Return Index
6894 let Defs = [ECX, EFLAGS] in {
6895 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6896 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6897 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6898 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6899 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6900 (implicit EFLAGS)]>, OpSize;
6901 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6902 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6903 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6904 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6905 (implicit EFLAGS)]>, OpSize;
6909 let Predicates = [HasAVX] in {
6910 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6912 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6914 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6916 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6918 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6920 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6924 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6925 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6926 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6927 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6928 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6929 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6931 // Packed Compare Explicit Length Strings, Return Index
6932 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6933 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6934 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6935 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6936 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6937 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6938 (implicit EFLAGS)]>, OpSize;
6939 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6940 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6941 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6943 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6944 (implicit EFLAGS)]>, OpSize;
6948 let Predicates = [HasAVX] in {
6949 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6951 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6953 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6955 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6957 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6959 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6963 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6964 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6965 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6966 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6967 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6968 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6970 //===----------------------------------------------------------------------===//
6971 // SSE4.2 - CRC Instructions
6972 //===----------------------------------------------------------------------===//
6974 // No CRC instructions have AVX equivalents
6976 // crc intrinsic instruction
6977 // This set of instructions are only rm, the only difference is the size
6979 let Constraints = "$src1 = $dst" in {
6980 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6981 (ins GR32:$src1, i8mem:$src2),
6982 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6984 (int_x86_sse42_crc32_32_8 GR32:$src1,
6985 (load addr:$src2)))]>;
6986 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6987 (ins GR32:$src1, GR8:$src2),
6988 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6990 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6991 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6992 (ins GR32:$src1, i16mem:$src2),
6993 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6995 (int_x86_sse42_crc32_32_16 GR32:$src1,
6996 (load addr:$src2)))]>,
6998 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6999 (ins GR32:$src1, GR16:$src2),
7000 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7002 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7004 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7005 (ins GR32:$src1, i32mem:$src2),
7006 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7008 (int_x86_sse42_crc32_32_32 GR32:$src1,
7009 (load addr:$src2)))]>;
7010 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7011 (ins GR32:$src1, GR32:$src2),
7012 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7014 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7015 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7016 (ins GR64:$src1, i8mem:$src2),
7017 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7019 (int_x86_sse42_crc32_64_8 GR64:$src1,
7020 (load addr:$src2)))]>,
7022 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7023 (ins GR64:$src1, GR8:$src2),
7024 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7026 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7028 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7029 (ins GR64:$src1, i64mem:$src2),
7030 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7032 (int_x86_sse42_crc32_64_64 GR64:$src1,
7033 (load addr:$src2)))]>,
7035 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7036 (ins GR64:$src1, GR64:$src2),
7037 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7039 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7043 //===----------------------------------------------------------------------===//
7044 // AES-NI Instructions
7045 //===----------------------------------------------------------------------===//
7047 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7048 Intrinsic IntId128, bit Is2Addr = 1> {
7049 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7050 (ins VR128:$src1, VR128:$src2),
7052 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7053 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7054 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7056 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7057 (ins VR128:$src1, i128mem:$src2),
7059 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7060 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7062 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7065 // Perform One Round of an AES Encryption/Decryption Flow
7066 let Predicates = [HasAVX, HasAES] in {
7067 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7068 int_x86_aesni_aesenc, 0>, VEX_4V;
7069 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7070 int_x86_aesni_aesenclast, 0>, VEX_4V;
7071 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7072 int_x86_aesni_aesdec, 0>, VEX_4V;
7073 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7074 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7077 let Constraints = "$src1 = $dst" in {
7078 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7079 int_x86_aesni_aesenc>;
7080 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7081 int_x86_aesni_aesenclast>;
7082 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7083 int_x86_aesni_aesdec>;
7084 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7085 int_x86_aesni_aesdeclast>;
7088 // Perform the AES InvMixColumn Transformation
7089 let Predicates = [HasAVX, HasAES] in {
7090 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7092 "vaesimc\t{$src1, $dst|$dst, $src1}",
7094 (int_x86_aesni_aesimc VR128:$src1))]>,
7096 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7097 (ins i128mem:$src1),
7098 "vaesimc\t{$src1, $dst|$dst, $src1}",
7099 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7102 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7104 "aesimc\t{$src1, $dst|$dst, $src1}",
7106 (int_x86_aesni_aesimc VR128:$src1))]>,
7108 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7109 (ins i128mem:$src1),
7110 "aesimc\t{$src1, $dst|$dst, $src1}",
7111 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7114 // AES Round Key Generation Assist
7115 let Predicates = [HasAVX, HasAES] in {
7116 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7117 (ins VR128:$src1, i8imm:$src2),
7118 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7120 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7122 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7123 (ins i128mem:$src1, i8imm:$src2),
7124 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7126 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7129 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7130 (ins VR128:$src1, i8imm:$src2),
7131 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7133 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7135 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7136 (ins i128mem:$src1, i8imm:$src2),
7137 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7139 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7142 //===----------------------------------------------------------------------===//
7143 // CLMUL Instructions
7144 //===----------------------------------------------------------------------===//
7146 // Carry-less Multiplication instructions
7147 let neverHasSideEffects = 1 in {
7148 // AVX carry-less Multiplication instructions
7149 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7150 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7151 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7155 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7156 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7157 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7160 let Constraints = "$src1 = $dst" in {
7161 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7162 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7163 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7167 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7168 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7169 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7171 } // Constraints = "$src1 = $dst"
7172 } // neverHasSideEffects = 1
7175 multiclass pclmul_alias<string asm, int immop> {
7176 def : InstAlias<!strconcat("pclmul", asm,
7177 "dq {$src, $dst|$dst, $src}"),
7178 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7180 def : InstAlias<!strconcat("pclmul", asm,
7181 "dq {$src, $dst|$dst, $src}"),
7182 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7184 def : InstAlias<!strconcat("vpclmul", asm,
7185 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7186 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7188 def : InstAlias<!strconcat("vpclmul", asm,
7189 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7190 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7192 defm : pclmul_alias<"hqhq", 0x11>;
7193 defm : pclmul_alias<"hqlq", 0x01>;
7194 defm : pclmul_alias<"lqhq", 0x10>;
7195 defm : pclmul_alias<"lqlq", 0x00>;
7197 //===----------------------------------------------------------------------===//
7199 //===----------------------------------------------------------------------===//
7201 //===----------------------------------------------------------------------===//
7202 // VBROADCAST - Load from memory and broadcast to all elements of the
7203 // destination operand
7205 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7206 X86MemOperand x86memop, Intrinsic Int> :
7207 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7208 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7209 [(set RC:$dst, (Int addr:$src))]>, VEX;
7211 // AVX2 adds register forms
7212 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7214 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7215 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7216 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7218 let ExeDomain = SSEPackedSingle in {
7219 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7220 int_x86_avx_vbroadcast_ss>;
7221 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7222 int_x86_avx_vbroadcast_ss_256>;
7224 let ExeDomain = SSEPackedDouble in
7225 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7226 int_x86_avx_vbroadcast_sd_256>;
7227 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7228 int_x86_avx_vbroadcastf128_pd_256>;
7230 let ExeDomain = SSEPackedSingle in {
7231 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7232 int_x86_avx2_vbroadcast_ss_ps>;
7233 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7234 int_x86_avx2_vbroadcast_ss_ps_256>;
7236 let ExeDomain = SSEPackedDouble in
7237 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7238 int_x86_avx2_vbroadcast_sd_pd_256>;
7240 let Predicates = [HasAVX2] in
7241 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7242 int_x86_avx2_vbroadcasti128>;
7244 let Predicates = [HasAVX] in
7245 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7246 (VBROADCASTF128 addr:$src)>;
7249 //===----------------------------------------------------------------------===//
7250 // VINSERTF128 - Insert packed floating-point values
7252 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7253 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7254 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7255 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7258 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7259 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7260 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7264 let Predicates = [HasAVX] in {
7265 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7266 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7267 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7268 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7269 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7270 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7273 //===----------------------------------------------------------------------===//
7274 // VEXTRACTF128 - Extract packed floating-point values
7276 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7277 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7278 (ins VR256:$src1, i8imm:$src2),
7279 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7282 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7283 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7284 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7288 let Predicates = [HasAVX] in {
7289 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7290 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7291 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7292 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7293 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7294 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7297 //===----------------------------------------------------------------------===//
7298 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7300 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7301 Intrinsic IntLd, Intrinsic IntLd256,
7302 Intrinsic IntSt, Intrinsic IntSt256> {
7303 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7304 (ins VR128:$src1, f128mem:$src2),
7305 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7306 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7308 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7309 (ins VR256:$src1, f256mem:$src2),
7310 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7311 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7313 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7314 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7315 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7316 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7317 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7318 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7319 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7320 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7323 let ExeDomain = SSEPackedSingle in
7324 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7325 int_x86_avx_maskload_ps,
7326 int_x86_avx_maskload_ps_256,
7327 int_x86_avx_maskstore_ps,
7328 int_x86_avx_maskstore_ps_256>;
7329 let ExeDomain = SSEPackedDouble in
7330 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7331 int_x86_avx_maskload_pd,
7332 int_x86_avx_maskload_pd_256,
7333 int_x86_avx_maskstore_pd,
7334 int_x86_avx_maskstore_pd_256>;
7336 //===----------------------------------------------------------------------===//
7337 // VPERMIL - Permute Single and Double Floating-Point Values
7339 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7340 RegisterClass RC, X86MemOperand x86memop_f,
7341 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7342 Intrinsic IntVar, Intrinsic IntImm> {
7343 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7344 (ins RC:$src1, RC:$src2),
7345 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7346 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7347 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7348 (ins RC:$src1, x86memop_i:$src2),
7349 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7350 [(set RC:$dst, (IntVar RC:$src1,
7351 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7353 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7354 (ins RC:$src1, i8imm:$src2),
7355 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7356 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7357 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7358 (ins x86memop_f:$src1, i8imm:$src2),
7359 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7360 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7363 let ExeDomain = SSEPackedSingle in {
7364 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7365 memopv4f32, memopv2i64,
7366 int_x86_avx_vpermilvar_ps,
7367 int_x86_avx_vpermil_ps>;
7368 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7369 memopv8f32, memopv4i64,
7370 int_x86_avx_vpermilvar_ps_256,
7371 int_x86_avx_vpermil_ps_256>;
7373 let ExeDomain = SSEPackedDouble in {
7374 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7375 memopv2f64, memopv2i64,
7376 int_x86_avx_vpermilvar_pd,
7377 int_x86_avx_vpermil_pd>;
7378 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7379 memopv4f64, memopv4i64,
7380 int_x86_avx_vpermilvar_pd_256,
7381 int_x86_avx_vpermil_pd_256>;
7384 let Predicates = [HasAVX] in {
7385 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7386 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7387 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7388 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7389 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7390 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7391 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7392 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7393 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7394 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7395 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7396 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7397 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7399 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7400 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7401 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7404 //===----------------------------------------------------------------------===//
7405 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7407 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7408 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7409 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7410 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7413 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7414 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7415 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7419 let Predicates = [HasAVX] in {
7420 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7421 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7422 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7423 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7424 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7425 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7427 def : Pat<(int_x86_avx_vperm2f128_ps_256
7428 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7429 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7430 def : Pat<(int_x86_avx_vperm2f128_pd_256
7431 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7432 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7433 def : Pat<(int_x86_avx_vperm2f128_si_256
7434 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7435 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7438 //===----------------------------------------------------------------------===//
7439 // VZERO - Zero YMM registers
7441 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7442 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7443 // Zero All YMM registers
7444 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7445 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7447 // Zero Upper bits of YMM registers
7448 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7449 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7452 //===----------------------------------------------------------------------===//
7453 // Half precision conversion instructions
7454 //===----------------------------------------------------------------------===//
7455 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7456 let Predicates = [HasAVX, HasF16C] in {
7457 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7458 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7459 [(set RC:$dst, (Int VR128:$src))]>,
7461 let neverHasSideEffects = 1, mayLoad = 1 in
7462 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7463 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7467 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7468 let Predicates = [HasAVX, HasF16C] in {
7469 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7470 (ins RC:$src1, i32i8imm:$src2),
7471 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7472 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7474 let neverHasSideEffects = 1, mayLoad = 1 in
7475 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7476 (ins RC:$src1, i32i8imm:$src2),
7477 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7482 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7483 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7484 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7485 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7487 //===----------------------------------------------------------------------===//
7488 // AVX2 Instructions
7489 //===----------------------------------------------------------------------===//
7491 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7492 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7493 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7494 X86MemOperand x86memop> {
7495 let isCommutable = 1 in
7496 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7497 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7498 !strconcat(OpcodeStr,
7499 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7500 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7502 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7503 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7504 !strconcat(OpcodeStr,
7505 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7508 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7512 let isCommutable = 0 in {
7513 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7514 VR128, memopv2i64, i128mem>;
7515 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7516 VR256, memopv4i64, i256mem>;
7519 //===----------------------------------------------------------------------===//
7520 // VPBROADCAST - Load from memory and broadcast to all elements of the
7521 // destination operand
7523 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7524 X86MemOperand x86memop, PatFrag ld_frag,
7525 Intrinsic Int128, Intrinsic Int256> {
7526 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7527 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7528 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7529 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7530 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7532 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7533 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7534 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7535 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7536 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7537 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7539 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7542 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7543 int_x86_avx2_pbroadcastb_128,
7544 int_x86_avx2_pbroadcastb_256>;
7545 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7546 int_x86_avx2_pbroadcastw_128,
7547 int_x86_avx2_pbroadcastw_256>;
7548 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7549 int_x86_avx2_pbroadcastd_128,
7550 int_x86_avx2_pbroadcastd_256>;
7551 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7552 int_x86_avx2_pbroadcastq_128,
7553 int_x86_avx2_pbroadcastq_256>;
7555 let Predicates = [HasAVX2] in {
7556 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7557 (VPBROADCASTBrm addr:$src)>;
7558 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7559 (VPBROADCASTBYrm addr:$src)>;
7560 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7561 (VPBROADCASTWrm addr:$src)>;
7562 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7563 (VPBROADCASTWYrm addr:$src)>;
7564 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7565 (VPBROADCASTDrm addr:$src)>;
7566 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7567 (VPBROADCASTDYrm addr:$src)>;
7568 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7569 (VPBROADCASTQrm addr:$src)>;
7570 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7571 (VPBROADCASTQYrm addr:$src)>;
7574 // AVX1 broadcast patterns
7575 let Predicates = [HasAVX] in {
7576 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7577 (VBROADCASTSSYrm addr:$src)>;
7578 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7579 (VBROADCASTSDrm addr:$src)>;
7580 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7581 (VBROADCASTSSYrm addr:$src)>;
7582 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7583 (VBROADCASTSDrm addr:$src)>;
7585 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7586 (VBROADCASTSSrm addr:$src)>;
7587 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7588 (VBROADCASTSSrm addr:$src)>;
7591 //===----------------------------------------------------------------------===//
7592 // VPERM - Permute instructions
7595 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7597 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7598 (ins VR256:$src1, VR256:$src2),
7599 !strconcat(OpcodeStr,
7600 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7601 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7602 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7603 (ins VR256:$src1, i256mem:$src2),
7604 !strconcat(OpcodeStr,
7605 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7606 [(set VR256:$dst, (Int VR256:$src1,
7607 (bitconvert (mem_frag addr:$src2))))]>,
7611 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7612 let ExeDomain = SSEPackedSingle in
7613 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7615 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7617 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7618 (ins VR256:$src1, i8imm:$src2),
7619 !strconcat(OpcodeStr,
7620 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7621 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7622 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7623 (ins i256mem:$src1, i8imm:$src2),
7624 !strconcat(OpcodeStr,
7625 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7626 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7630 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7632 let ExeDomain = SSEPackedDouble in
7633 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7636 //===----------------------------------------------------------------------===//
7637 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7639 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7640 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7641 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7643 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7645 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7646 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7647 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7649 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7653 let Predicates = [HasAVX2] in {
7654 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7655 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7656 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7657 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7658 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7659 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7660 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7661 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7663 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7665 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7666 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7667 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7668 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7669 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7671 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7672 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7674 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7678 let Predicates = [HasAVX] in {
7679 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7680 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7681 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7682 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7683 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7684 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7685 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7686 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7687 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7688 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7689 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7690 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7692 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7693 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7694 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7695 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7696 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7697 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7698 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7699 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7700 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7701 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7702 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7703 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7704 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7705 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7706 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7707 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7708 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7709 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7713 //===----------------------------------------------------------------------===//
7714 // VINSERTI128 - Insert packed integer values
7716 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7717 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7718 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7720 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7722 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7723 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7724 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7726 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7727 imm:$src3))]>, VEX_4V;
7729 let Predicates = [HasAVX2] in {
7730 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7732 (VINSERTI128rr VR256:$src1, VR128:$src2,
7733 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7734 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7736 (VINSERTI128rr VR256:$src1, VR128:$src2,
7737 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7738 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7740 (VINSERTI128rr VR256:$src1, VR128:$src2,
7741 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7742 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7744 (VINSERTI128rr VR256:$src1, VR128:$src2,
7745 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7749 let Predicates = [HasAVX] in {
7750 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7752 (VINSERTF128rr VR256:$src1, VR128:$src2,
7753 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7754 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7756 (VINSERTF128rr VR256:$src1, VR128:$src2,
7757 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7758 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7760 (VINSERTF128rr VR256:$src1, VR128:$src2,
7761 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7762 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7764 (VINSERTF128rr VR256:$src1, VR128:$src2,
7765 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7766 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7768 (VINSERTF128rr VR256:$src1, VR128:$src2,
7769 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7770 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7772 (VINSERTF128rr VR256:$src1, VR128:$src2,
7773 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7776 //===----------------------------------------------------------------------===//
7777 // VEXTRACTI128 - Extract packed integer values
7779 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7780 (ins VR256:$src1, i8imm:$src2),
7781 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7783 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7785 let neverHasSideEffects = 1, mayStore = 1 in
7786 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7787 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7788 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7790 let Predicates = [HasAVX2] in {
7791 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7792 (v2i64 (VEXTRACTI128rr
7793 (v4i64 VR256:$src1),
7794 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7795 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7796 (v4i32 (VEXTRACTI128rr
7797 (v8i32 VR256:$src1),
7798 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7799 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7800 (v8i16 (VEXTRACTI128rr
7801 (v16i16 VR256:$src1),
7802 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7803 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7804 (v16i8 (VEXTRACTI128rr
7805 (v32i8 VR256:$src1),
7806 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7810 let Predicates = [HasAVX] in {
7811 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7812 (v4f32 (VEXTRACTF128rr
7813 (v8f32 VR256:$src1),
7814 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7815 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7816 (v2f64 (VEXTRACTF128rr
7817 (v4f64 VR256:$src1),
7818 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7819 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7820 (v2i64 (VEXTRACTF128rr
7821 (v4i64 VR256:$src1),
7822 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7823 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7824 (v4i32 (VEXTRACTF128rr
7825 (v8i32 VR256:$src1),
7826 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7827 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7828 (v8i16 (VEXTRACTF128rr
7829 (v16i16 VR256:$src1),
7830 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7831 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7832 (v16i8 (VEXTRACTF128rr
7833 (v32i8 VR256:$src1),
7834 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7837 //===----------------------------------------------------------------------===//
7838 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7840 multiclass avx2_pmovmask<string OpcodeStr,
7841 Intrinsic IntLd128, Intrinsic IntLd256,
7842 Intrinsic IntSt128, Intrinsic IntSt256> {
7843 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7844 (ins VR128:$src1, i128mem:$src2),
7845 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7846 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7847 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7848 (ins VR256:$src1, i256mem:$src2),
7849 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7850 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7851 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7852 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7853 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7854 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7855 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7856 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7857 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7858 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7861 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7862 int_x86_avx2_maskload_d,
7863 int_x86_avx2_maskload_d_256,
7864 int_x86_avx2_maskstore_d,
7865 int_x86_avx2_maskstore_d_256>;
7866 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7867 int_x86_avx2_maskload_q,
7868 int_x86_avx2_maskload_q_256,
7869 int_x86_avx2_maskstore_q,
7870 int_x86_avx2_maskstore_q_256>, VEX_W;
7873 //===----------------------------------------------------------------------===//
7874 // Variable Bit Shifts
7876 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7877 ValueType vt128, ValueType vt256> {
7878 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7879 (ins VR128:$src1, VR128:$src2),
7880 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7882 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7884 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7885 (ins VR128:$src1, i128mem:$src2),
7886 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7888 (vt128 (OpNode VR128:$src1,
7889 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7891 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7892 (ins VR256:$src1, VR256:$src2),
7893 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7895 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7897 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7898 (ins VR256:$src1, i256mem:$src2),
7899 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7901 (vt256 (OpNode VR256:$src1,
7902 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7906 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7907 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7908 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7909 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7910 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;