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 = [HasXMMInt] 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 // FIXME: Set encoding to pseudo!
244 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
245 canFoldAsLoad = 1 in {
246 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
247 [(set FR32:$dst, fp32imm0)]>,
248 Requires<[HasSSE1]>, TB, OpSize;
249 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
250 [(set FR64:$dst, fpimm0)]>,
251 Requires<[HasSSE2]>, TB, OpSize;
252 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
253 [(set FR32:$dst, fp32imm0)]>,
254 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
255 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
256 [(set FR64:$dst, fpimm0)]>,
257 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
260 //===----------------------------------------------------------------------===//
261 // AVX & SSE - Zero/One Vectors
262 //===----------------------------------------------------------------------===//
264 // Alias instruction that maps zero vector to pxor / xorp* for sse.
265 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
266 // swizzled by ExecutionDepsFix to pxor.
267 // We set canFoldAsLoad because this can be converted to a constant-pool
268 // load of an all-zeros value if folding it would be beneficial.
269 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
270 isPseudo = 1, neverHasSideEffects = 1 in {
271 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
274 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
275 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
276 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
277 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
278 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
279 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
282 // The same as done above but for AVX. The 256-bit ISA does not support PI,
283 // and doesn't need it because on sandy bridge the register is set to zero
284 // at the rename stage without using any execution unit, so SET0PSY
285 // and SET0PDY can be used for vector int instructions without penalty
286 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
287 // JIT implementatioan, it does not expand the instructions below like
288 // X86MCInstLower does.
289 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
290 isCodeGenOnly = 1, Predicates = [HasAVX] in {
291 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
292 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
293 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
294 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
298 // AVX has no support for 256-bit integer instructions, but since the 128-bit
299 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
300 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
301 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
302 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
304 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
305 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
306 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
308 // We set canFoldAsLoad because this can be converted to a constant-pool
309 // load of an all-ones value if folding it would be beneficial.
310 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
311 // JIT implementation, it does not expand the instructions below like
312 // X86MCInstLower does.
313 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
314 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
315 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
316 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
317 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
318 isCodeGenOnly = 1, ExeDomain = SSEPackedInt, Predicates = [HasAVX] in
319 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
320 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
323 //===----------------------------------------------------------------------===//
324 // SSE 1 & 2 - Move FP Scalar Instructions
326 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
327 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
328 // is used instead. Register-to-register movss/movsd is not modeled as an
329 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
330 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
331 //===----------------------------------------------------------------------===//
333 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
334 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
335 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
337 // Loading from memory automatically zeroing upper bits.
338 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
339 PatFrag mem_pat, string OpcodeStr> :
340 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
341 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
342 [(set RC:$dst, (mem_pat addr:$src))]>;
345 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
346 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
348 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
349 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
352 // For the disassembler
353 let isCodeGenOnly = 1 in {
354 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
355 (ins VR128:$src1, FR32:$src2),
356 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
358 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
359 (ins VR128:$src1, FR64:$src2),
360 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
364 let canFoldAsLoad = 1, isReMaterializable = 1 in {
365 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
367 let AddedComplexity = 20 in
368 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
372 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
373 "movss\t{$src, $dst|$dst, $src}",
374 [(store FR32:$src, addr:$dst)]>, XS, VEX, VEX_LIG;
375 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
376 "movsd\t{$src, $dst|$dst, $src}",
377 [(store FR64:$src, addr:$dst)]>, XD, VEX, VEX_LIG;
380 let Constraints = "$src1 = $dst" in {
381 def MOVSSrr : sse12_move_rr<FR32, v4f32,
382 "movss\t{$src2, $dst|$dst, $src2}">, XS;
383 def MOVSDrr : sse12_move_rr<FR64, v2f64,
384 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
386 // For the disassembler
387 let isCodeGenOnly = 1 in {
388 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
389 (ins VR128:$src1, FR32:$src2),
390 "movss\t{$src2, $dst|$dst, $src2}", []>, XS;
391 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
392 (ins VR128:$src1, FR64:$src2),
393 "movsd\t{$src2, $dst|$dst, $src2}", []>, XD;
397 let canFoldAsLoad = 1, isReMaterializable = 1 in {
398 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
400 let AddedComplexity = 20 in
401 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
404 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
405 "movss\t{$src, $dst|$dst, $src}",
406 [(store FR32:$src, addr:$dst)]>;
407 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
408 "movsd\t{$src, $dst|$dst, $src}",
409 [(store FR64:$src, addr:$dst)]>;
412 let Predicates = [HasSSE1] in {
413 let AddedComplexity = 15 in {
414 // Extract the low 32-bit value from one vector and insert it into another.
415 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
416 (MOVSSrr (v4f32 VR128:$src1),
417 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
418 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
419 (MOVSSrr (v4i32 VR128:$src1),
420 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
422 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
423 // MOVSS to the lower bits.
424 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
425 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
426 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
427 (MOVSSrr (v4f32 (V_SET0)),
428 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
429 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
430 (MOVSSrr (v4i32 (V_SET0)),
431 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
434 let AddedComplexity = 20 in {
435 // MOVSSrm zeros the high parts of the register; represent this
436 // with SUBREG_TO_REG.
437 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
438 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
439 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
440 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
441 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
442 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
445 // Extract and store.
446 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
449 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
451 // Shuffle with MOVSS
452 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
453 (MOVSSrr VR128:$src1, FR32:$src2)>;
454 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
455 (MOVSSrr (v4i32 VR128:$src1),
456 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
457 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
458 (MOVSSrr (v4f32 VR128:$src1),
459 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
462 let Predicates = [HasSSE2] in {
463 let AddedComplexity = 15 in {
464 // Extract the low 64-bit value from one vector and insert it into another.
465 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
466 (MOVSDrr (v2f64 VR128:$src1),
467 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
468 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
469 (MOVSDrr (v2i64 VR128:$src1),
470 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
472 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
473 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
474 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
475 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
476 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
478 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
479 // MOVSD to the lower bits.
480 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
481 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
484 let AddedComplexity = 20 in {
485 // MOVSDrm zeros the high parts of the register; represent this
486 // with SUBREG_TO_REG.
487 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
488 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
489 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
490 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
491 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
492 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
493 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
494 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
495 def : Pat<(v2f64 (X86vzload addr:$src)),
496 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
499 // Extract and store.
500 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
503 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
505 // Shuffle with MOVSD
506 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
507 (MOVSDrr VR128:$src1, FR64:$src2)>;
508 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
509 (MOVSDrr (v2i64 VR128:$src1),
510 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
511 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
512 (MOVSDrr (v2f64 VR128:$src1),
513 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
514 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
515 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
516 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
517 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
519 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
520 // is during lowering, where it's not possible to recognize the fold cause
521 // it has two uses through a bitcast. One use disappears at isel time and the
522 // fold opportunity reappears.
523 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
524 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
525 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
526 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
527 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
528 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
531 let Predicates = [HasAVX] in {
532 let AddedComplexity = 15 in {
533 // Extract the low 32-bit value from one vector and insert it into another.
534 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
535 (VMOVSSrr (v4f32 VR128:$src1),
536 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
537 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
538 (VMOVSSrr (v4i32 VR128:$src1),
539 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
541 // Extract the low 64-bit value from one vector and insert it into another.
542 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
543 (VMOVSDrr (v2f64 VR128:$src1),
544 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
545 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
546 (VMOVSDrr (v2i64 VR128:$src1),
547 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
549 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
550 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
551 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
552 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
553 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
555 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
556 // MOVS{S,D} to the lower bits.
557 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
558 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
559 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
560 (VMOVSSrr (v4f32 (V_SET0)),
561 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
562 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
563 (VMOVSSrr (v4i32 (V_SET0)),
564 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
565 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
566 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
569 let AddedComplexity = 20 in {
570 // MOVSSrm zeros the high parts of the register; represent this
571 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
572 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
573 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
574 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
575 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
576 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
577 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
579 // MOVSDrm zeros the high parts of the register; represent this
580 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
581 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
582 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
583 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
584 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
585 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
586 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
587 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
588 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
589 def : Pat<(v2f64 (X86vzload addr:$src)),
590 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
592 // Represent the same patterns above but in the form they appear for
594 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
595 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
596 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
597 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
598 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
599 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
601 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
602 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
603 (SUBREG_TO_REG (i32 0),
604 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
606 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
607 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
608 (SUBREG_TO_REG (i64 0),
609 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
612 // Extract and store.
613 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
616 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
617 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
620 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
622 // Shuffle with VMOVSS
623 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
624 (VMOVSSrr VR128:$src1, FR32:$src2)>;
625 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
626 (VMOVSSrr (v4i32 VR128:$src1),
627 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
628 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
629 (VMOVSSrr (v4f32 VR128:$src1),
630 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
632 // Shuffle with VMOVSD
633 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
634 (VMOVSDrr VR128:$src1, FR64:$src2)>;
635 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
636 (VMOVSDrr (v2i64 VR128:$src1),
637 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
638 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
639 (VMOVSDrr (v2f64 VR128:$src1),
640 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
641 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
642 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
644 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
645 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
648 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
649 // is during lowering, where it's not possible to recognize the fold cause
650 // it has two uses through a bitcast. One use disappears at isel time and the
651 // fold opportunity reappears.
652 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
653 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
655 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
656 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
658 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
659 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
663 //===----------------------------------------------------------------------===//
664 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
665 //===----------------------------------------------------------------------===//
667 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
668 X86MemOperand x86memop, PatFrag ld_frag,
669 string asm, Domain d,
670 bit IsReMaterializable = 1> {
671 let neverHasSideEffects = 1 in
672 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
673 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
674 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
675 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
676 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
677 [(set RC:$dst, (ld_frag addr:$src))], d>;
680 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
681 "movaps", SSEPackedSingle>, TB, VEX;
682 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
683 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
684 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
685 "movups", SSEPackedSingle>, TB, VEX;
686 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
687 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
689 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
690 "movaps", SSEPackedSingle>, TB, VEX;
691 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
692 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
693 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
694 "movups", SSEPackedSingle>, TB, VEX;
695 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
696 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
697 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
698 "movaps", SSEPackedSingle>, TB;
699 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
700 "movapd", SSEPackedDouble>, TB, OpSize;
701 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
702 "movups", SSEPackedSingle>, TB;
703 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
704 "movupd", SSEPackedDouble, 0>, TB, OpSize;
706 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
707 "movaps\t{$src, $dst|$dst, $src}",
708 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
709 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
710 "movapd\t{$src, $dst|$dst, $src}",
711 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
712 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
713 "movups\t{$src, $dst|$dst, $src}",
714 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
715 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
716 "movupd\t{$src, $dst|$dst, $src}",
717 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
718 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
719 "movaps\t{$src, $dst|$dst, $src}",
720 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)]>, VEX;
721 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
722 "movapd\t{$src, $dst|$dst, $src}",
723 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)]>, VEX;
724 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
725 "movups\t{$src, $dst|$dst, $src}",
726 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
727 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
728 "movupd\t{$src, $dst|$dst, $src}",
729 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
732 let isCodeGenOnly = 1 in {
733 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
735 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
736 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
738 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
739 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
741 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
742 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
744 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
745 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
747 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
748 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
750 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
751 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
753 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
754 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
756 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
759 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
760 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
761 (VMOVUPSYmr addr:$dst, VR256:$src)>;
763 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
764 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
765 (VMOVUPDYmr addr:$dst, VR256:$src)>;
767 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
768 "movaps\t{$src, $dst|$dst, $src}",
769 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
770 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
771 "movapd\t{$src, $dst|$dst, $src}",
772 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
773 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
774 "movups\t{$src, $dst|$dst, $src}",
775 [(store (v4f32 VR128:$src), addr:$dst)]>;
776 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
777 "movupd\t{$src, $dst|$dst, $src}",
778 [(store (v2f64 VR128:$src), addr:$dst)]>;
781 let isCodeGenOnly = 1 in {
782 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
783 "movaps\t{$src, $dst|$dst, $src}", []>;
784 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
785 "movapd\t{$src, $dst|$dst, $src}", []>;
786 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
787 "movups\t{$src, $dst|$dst, $src}", []>;
788 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
789 "movupd\t{$src, $dst|$dst, $src}", []>;
792 let Predicates = [HasAVX] in {
793 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
794 (VMOVUPSmr addr:$dst, VR128:$src)>;
795 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
796 (VMOVUPDmr addr:$dst, VR128:$src)>;
799 let Predicates = [HasSSE1] in
800 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
801 (MOVUPSmr addr:$dst, VR128:$src)>;
802 let Predicates = [HasSSE2] in
803 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
804 (MOVUPDmr addr:$dst, VR128:$src)>;
806 // Use movaps / movups for SSE integer load / store (one byte shorter).
807 // The instructions selected below are then converted to MOVDQA/MOVDQU
808 // during the SSE domain pass.
809 let Predicates = [HasSSE1] in {
810 def : Pat<(alignedloadv4i32 addr:$src),
811 (MOVAPSrm addr:$src)>;
812 def : Pat<(loadv4i32 addr:$src),
813 (MOVUPSrm addr:$src)>;
814 def : Pat<(alignedloadv2i64 addr:$src),
815 (MOVAPSrm addr:$src)>;
816 def : Pat<(loadv2i64 addr:$src),
817 (MOVUPSrm addr:$src)>;
819 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
820 (MOVAPSmr addr:$dst, VR128:$src)>;
821 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
822 (MOVAPSmr addr:$dst, VR128:$src)>;
823 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
824 (MOVAPSmr addr:$dst, VR128:$src)>;
825 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
826 (MOVAPSmr addr:$dst, VR128:$src)>;
827 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
828 (MOVUPSmr addr:$dst, VR128:$src)>;
829 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
830 (MOVUPSmr addr:$dst, VR128:$src)>;
831 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
832 (MOVUPSmr addr:$dst, VR128:$src)>;
833 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
834 (MOVUPSmr addr:$dst, VR128:$src)>;
837 // Use vmovaps/vmovups for AVX integer load/store.
838 let Predicates = [HasAVX] in {
839 // 128-bit load/store
840 def : Pat<(alignedloadv4i32 addr:$src),
841 (VMOVAPSrm addr:$src)>;
842 def : Pat<(loadv4i32 addr:$src),
843 (VMOVUPSrm addr:$src)>;
844 def : Pat<(alignedloadv2i64 addr:$src),
845 (VMOVAPSrm addr:$src)>;
846 def : Pat<(loadv2i64 addr:$src),
847 (VMOVUPSrm addr:$src)>;
849 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
850 (VMOVAPSmr addr:$dst, VR128:$src)>;
851 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
852 (VMOVAPSmr addr:$dst, VR128:$src)>;
853 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
854 (VMOVAPSmr addr:$dst, VR128:$src)>;
855 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
856 (VMOVAPSmr addr:$dst, VR128:$src)>;
857 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
858 (VMOVUPSmr addr:$dst, VR128:$src)>;
859 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
860 (VMOVUPSmr addr:$dst, VR128:$src)>;
861 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
862 (VMOVUPSmr addr:$dst, VR128:$src)>;
863 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
864 (VMOVUPSmr addr:$dst, VR128:$src)>;
866 // 256-bit load/store
867 def : Pat<(alignedloadv4i64 addr:$src),
868 (VMOVAPSYrm addr:$src)>;
869 def : Pat<(loadv4i64 addr:$src),
870 (VMOVUPSYrm addr:$src)>;
871 def : Pat<(alignedloadv8i32 addr:$src),
872 (VMOVAPSYrm addr:$src)>;
873 def : Pat<(loadv8i32 addr:$src),
874 (VMOVUPSYrm addr:$src)>;
875 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
876 (VMOVAPSYmr addr:$dst, VR256:$src)>;
877 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
878 (VMOVAPSYmr addr:$dst, VR256:$src)>;
879 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
880 (VMOVAPSYmr addr:$dst, VR256:$src)>;
881 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
882 (VMOVAPSYmr addr:$dst, VR256:$src)>;
883 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
884 (VMOVUPSYmr addr:$dst, VR256:$src)>;
885 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
886 (VMOVUPSYmr addr:$dst, VR256:$src)>;
887 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
888 (VMOVUPSYmr addr:$dst, VR256:$src)>;
889 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
890 (VMOVUPSYmr addr:$dst, VR256:$src)>;
893 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
894 // bits are disregarded. FIXME: Set encoding to pseudo!
895 let neverHasSideEffects = 1 in {
896 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
897 "movaps\t{$src, $dst|$dst, $src}", []>;
898 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
899 "movapd\t{$src, $dst|$dst, $src}", []>;
900 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
901 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
902 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
903 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
906 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
907 // bits are disregarded. FIXME: Set encoding to pseudo!
908 let canFoldAsLoad = 1, isReMaterializable = 1 in {
909 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
910 "movaps\t{$src, $dst|$dst, $src}",
911 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
912 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
913 "movapd\t{$src, $dst|$dst, $src}",
914 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
915 let isCodeGenOnly = 1 in {
916 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
917 "movaps\t{$src, $dst|$dst, $src}",
918 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
919 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
920 "movapd\t{$src, $dst|$dst, $src}",
921 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
925 //===----------------------------------------------------------------------===//
926 // SSE 1 & 2 - Move Low packed FP Instructions
927 //===----------------------------------------------------------------------===//
929 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
930 PatFrag mov_frag, string base_opc,
932 def PSrm : PI<opc, MRMSrcMem,
933 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
934 !strconcat(base_opc, "s", asm_opr),
937 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
938 SSEPackedSingle>, TB;
940 def PDrm : PI<opc, MRMSrcMem,
941 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
942 !strconcat(base_opc, "d", asm_opr),
943 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
944 (scalar_to_vector (loadf64 addr:$src2)))))],
945 SSEPackedDouble>, TB, OpSize;
948 let AddedComplexity = 20 in {
949 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
950 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
952 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
953 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
954 "\t{$src2, $dst|$dst, $src2}">;
957 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
958 "movlps\t{$src, $dst|$dst, $src}",
959 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
960 (iPTR 0))), addr:$dst)]>, VEX;
961 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
962 "movlpd\t{$src, $dst|$dst, $src}",
963 [(store (f64 (vector_extract (v2f64 VR128:$src),
964 (iPTR 0))), addr:$dst)]>, VEX;
965 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
966 "movlps\t{$src, $dst|$dst, $src}",
967 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
968 (iPTR 0))), addr:$dst)]>;
969 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
970 "movlpd\t{$src, $dst|$dst, $src}",
971 [(store (f64 (vector_extract (v2f64 VR128:$src),
972 (iPTR 0))), addr:$dst)]>;
974 let Predicates = [HasAVX] in {
975 let AddedComplexity = 20 in {
976 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
977 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
978 (VMOVLPSrm VR128:$src1, addr:$src2)>;
979 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
980 (VMOVLPSrm VR128:$src1, addr:$src2)>;
981 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
982 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
983 (VMOVLPDrm VR128:$src1, addr:$src2)>;
984 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
985 (VMOVLPDrm VR128:$src1, addr:$src2)>;
988 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
989 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
990 (VMOVLPSmr addr:$src1, VR128:$src2)>;
991 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
992 VR128:$src2)), addr:$src1),
993 (VMOVLPSmr addr:$src1, VR128:$src2)>;
995 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
996 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
997 (VMOVLPDmr addr:$src1, VR128:$src2)>;
998 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
999 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1001 // Shuffle with VMOVLPS
1002 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1003 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1004 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1005 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1006 def : Pat<(X86Movlps VR128:$src1,
1007 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1008 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1010 // Shuffle with VMOVLPD
1011 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1012 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1013 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1014 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1015 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1016 (scalar_to_vector (loadf64 addr:$src2)))),
1017 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1020 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1022 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1023 def : Pat<(store (v4i32 (X86Movlps
1024 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1025 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1026 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1028 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1029 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1031 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1034 let Predicates = [HasSSE1] in {
1035 let AddedComplexity = 20 in {
1036 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1037 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1038 (MOVLPSrm VR128:$src1, addr:$src2)>;
1039 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1040 (MOVLPSrm VR128:$src1, addr:$src2)>;
1043 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1044 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1045 (iPTR 0))), addr:$src1),
1046 (MOVLPSmr addr:$src1, VR128:$src2)>;
1047 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1048 (MOVLPSmr addr:$src1, VR128:$src2)>;
1049 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1050 VR128:$src2)), addr:$src1),
1051 (MOVLPSmr addr:$src1, VR128:$src2)>;
1053 // Shuffle with MOVLPS
1054 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1055 (MOVLPSrm VR128:$src1, addr:$src2)>;
1056 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1057 (MOVLPSrm VR128:$src1, addr:$src2)>;
1058 def : Pat<(X86Movlps VR128:$src1,
1059 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1060 (MOVLPSrm VR128:$src1, addr:$src2)>;
1061 def : Pat<(X86Movlps VR128:$src1,
1062 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1063 (MOVLPSrm VR128:$src1, addr:$src2)>;
1066 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1068 (MOVLPSmr addr:$src1, VR128:$src2)>;
1069 def : Pat<(store (v4i32 (X86Movlps
1070 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1072 (MOVLPSmr addr:$src1, VR128:$src2)>;
1075 let Predicates = [HasSSE2] in {
1076 let AddedComplexity = 20 in {
1077 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1078 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1079 (MOVLPDrm VR128:$src1, addr:$src2)>;
1080 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1081 (MOVLPDrm VR128:$src1, addr:$src2)>;
1084 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1085 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1086 (MOVLPDmr addr:$src1, VR128:$src2)>;
1087 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1088 (MOVLPDmr addr:$src1, VR128:$src2)>;
1090 // Shuffle with MOVLPD
1091 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1092 (MOVLPDrm VR128:$src1, addr:$src2)>;
1093 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1094 (MOVLPDrm VR128:$src1, addr:$src2)>;
1095 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1096 (scalar_to_vector (loadf64 addr:$src2)))),
1097 (MOVLPDrm VR128:$src1, addr:$src2)>;
1100 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1102 (MOVLPDmr addr:$src1, VR128:$src2)>;
1103 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1105 (MOVLPDmr addr:$src1, VR128:$src2)>;
1108 //===----------------------------------------------------------------------===//
1109 // SSE 1 & 2 - Move Hi packed FP Instructions
1110 //===----------------------------------------------------------------------===//
1112 let AddedComplexity = 20 in {
1113 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1114 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1116 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1117 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1118 "\t{$src2, $dst|$dst, $src2}">;
1121 // v2f64 extract element 1 is always custom lowered to unpack high to low
1122 // and extract element 0 so the non-store version isn't too horrible.
1123 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1124 "movhps\t{$src, $dst|$dst, $src}",
1125 [(store (f64 (vector_extract
1126 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1127 (undef)), (iPTR 0))), addr:$dst)]>,
1129 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1130 "movhpd\t{$src, $dst|$dst, $src}",
1131 [(store (f64 (vector_extract
1132 (v2f64 (unpckh VR128:$src, (undef))),
1133 (iPTR 0))), addr:$dst)]>,
1135 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1136 "movhps\t{$src, $dst|$dst, $src}",
1137 [(store (f64 (vector_extract
1138 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1139 (undef)), (iPTR 0))), addr:$dst)]>;
1140 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1141 "movhpd\t{$src, $dst|$dst, $src}",
1142 [(store (f64 (vector_extract
1143 (v2f64 (unpckh VR128:$src, (undef))),
1144 (iPTR 0))), addr:$dst)]>;
1146 let Predicates = [HasAVX] in {
1148 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1149 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1150 def : Pat<(X86Movlhps VR128:$src1,
1151 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1152 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1153 def : Pat<(X86Movlhps VR128:$src1,
1154 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1155 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1157 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
1158 // is during lowering, where it's not possible to recognize the load fold cause
1159 // it has two uses through a bitcast. One use disappears at isel time and the
1160 // fold opportunity reappears.
1161 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
1162 (scalar_to_vector (loadf64 addr:$src2)))),
1163 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1165 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1166 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1167 (scalar_to_vector (loadf64 addr:$src2)))),
1168 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1171 def : Pat<(store (f64 (vector_extract
1172 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1173 (VMOVHPSmr addr:$dst, VR128:$src)>;
1174 def : Pat<(store (f64 (vector_extract
1175 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1176 (VMOVHPDmr addr:$dst, VR128:$src)>;
1179 let Predicates = [HasSSE1] in {
1181 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1182 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1183 def : Pat<(X86Movlhps VR128:$src1,
1184 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1185 (MOVHPSrm VR128:$src1, addr:$src2)>;
1186 def : Pat<(X86Movlhps VR128:$src1,
1187 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1188 (MOVHPSrm VR128:$src1, addr:$src2)>;
1191 def : Pat<(store (f64 (vector_extract
1192 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1193 (MOVHPSmr addr:$dst, VR128:$src)>;
1196 let Predicates = [HasSSE2] in {
1197 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
1198 // is during lowering, where it's not possible to recognize the load fold cause
1199 // it has two uses through a bitcast. One use disappears at isel time and the
1200 // fold opportunity reappears.
1201 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
1202 (scalar_to_vector (loadf64 addr:$src2)))),
1203 (MOVHPDrm VR128:$src1, addr:$src2)>;
1205 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1206 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1207 (scalar_to_vector (loadf64 addr:$src2)))),
1208 (MOVHPDrm VR128:$src1, addr:$src2)>;
1211 def : Pat<(store (f64 (vector_extract
1212 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
1213 (MOVHPDmr addr:$dst, VR128:$src)>;
1216 //===----------------------------------------------------------------------===//
1217 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1218 //===----------------------------------------------------------------------===//
1220 let AddedComplexity = 20 in {
1221 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1222 (ins VR128:$src1, VR128:$src2),
1223 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1225 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1227 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1228 (ins VR128:$src1, VR128:$src2),
1229 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1231 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1234 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1235 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1236 (ins VR128:$src1, VR128:$src2),
1237 "movlhps\t{$src2, $dst|$dst, $src2}",
1239 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1240 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1241 (ins VR128:$src1, VR128:$src2),
1242 "movhlps\t{$src2, $dst|$dst, $src2}",
1244 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1247 let Predicates = [HasAVX] in {
1249 let AddedComplexity = 20 in {
1250 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1251 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1252 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1253 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1255 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1256 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1257 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1259 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1260 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1261 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1262 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1263 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1264 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1267 let AddedComplexity = 20 in {
1268 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1269 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1270 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1272 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1273 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1274 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1275 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1276 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1279 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1280 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1281 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1282 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1285 let Predicates = [HasSSE1] in {
1287 let AddedComplexity = 20 in {
1288 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1289 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1290 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1291 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1293 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1294 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1295 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1297 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1298 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1299 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1300 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1301 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1302 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1305 let AddedComplexity = 20 in {
1306 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1307 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1308 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1310 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1311 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1312 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1313 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1314 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1317 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1318 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1319 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1320 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1323 //===----------------------------------------------------------------------===//
1324 // SSE 1 & 2 - Conversion Instructions
1325 //===----------------------------------------------------------------------===//
1327 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1328 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1330 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1331 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1332 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1333 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1336 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1337 X86MemOperand x86memop, string asm> {
1338 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, []>;
1340 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, []>;
1343 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1344 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1345 string asm, Domain d> {
1346 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1347 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1348 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1349 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1352 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1353 X86MemOperand x86memop, string asm> {
1354 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1355 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1357 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1358 (ins DstRC:$src1, x86memop:$src),
1359 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1362 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1363 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1365 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1366 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1368 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1369 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1371 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1372 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1373 VEX, VEX_W, VEX_LIG;
1375 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1376 // register, but the same isn't true when only using memory operands,
1377 // provide other assembly "l" and "q" forms to address this explicitly
1378 // where appropriate to do so.
1379 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1381 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1382 VEX_4V, VEX_W, VEX_LIG;
1383 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1385 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1387 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1388 VEX_4V, VEX_W, VEX_LIG;
1390 let Predicates = [HasAVX] in {
1391 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1392 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1393 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1394 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1395 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1396 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1397 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1398 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1400 def : Pat<(f32 (sint_to_fp GR32:$src)),
1401 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1402 def : Pat<(f32 (sint_to_fp GR64:$src)),
1403 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1404 def : Pat<(f64 (sint_to_fp GR32:$src)),
1405 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1406 def : Pat<(f64 (sint_to_fp GR64:$src)),
1407 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1410 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1411 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1412 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1413 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1414 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1415 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1416 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1417 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1418 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1419 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1420 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1421 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1422 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1423 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1424 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1425 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1427 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1428 // and/or XMM operand(s).
1430 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1431 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1433 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1434 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1435 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1436 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1437 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1438 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1441 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1442 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1443 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1444 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1446 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1447 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1448 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1449 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1450 (ins DstRC:$src1, x86memop:$src2),
1452 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1453 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1454 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1457 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1458 f128mem, load, "cvtsd2si">, XD, VEX;
1459 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1460 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1463 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1464 // Get rid of this hack or rename the intrinsics, there are several
1465 // intructions that only match with the intrinsic form, why create duplicates
1466 // to let them be recognized by the assembler?
1467 defm VCVTSD2SI : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
1468 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_LIG;
1469 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
1470 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W,
1473 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1474 f128mem, load, "cvtsd2si{l}">, XD;
1475 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1476 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1479 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1480 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1481 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1482 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1484 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1485 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1486 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1487 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1490 let Constraints = "$src1 = $dst" in {
1491 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1492 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1494 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1495 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1496 "cvtsi2ss{q}">, XS, REX_W;
1497 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1498 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1500 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1501 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1502 "cvtsi2sd">, XD, REX_W;
1507 // Aliases for intrinsics
1508 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1509 f32mem, load, "cvttss2si">, XS, VEX;
1510 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1511 int_x86_sse_cvttss2si64, f32mem, load,
1512 "cvttss2si">, XS, VEX, VEX_W;
1513 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1514 f128mem, load, "cvttsd2si">, XD, VEX;
1515 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1516 int_x86_sse2_cvttsd2si64, f128mem, load,
1517 "cvttsd2si">, XD, VEX, VEX_W;
1518 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1519 f32mem, load, "cvttss2si">, XS;
1520 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1521 int_x86_sse_cvttss2si64, f32mem, load,
1522 "cvttss2si{q}">, XS, REX_W;
1523 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1524 f128mem, load, "cvttsd2si">, XD;
1525 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1526 int_x86_sse2_cvttsd2si64, f128mem, load,
1527 "cvttsd2si{q}">, XD, REX_W;
1529 let Pattern = []<dag> in {
1530 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1531 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1533 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1534 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1536 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1537 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1538 SSEPackedSingle>, TB, VEX;
1539 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1540 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1541 SSEPackedSingle>, TB, VEX;
1544 let Pattern = []<dag> in {
1545 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1546 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1547 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1548 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1549 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1550 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1551 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1554 let Predicates = [HasSSE1] in {
1555 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1556 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1557 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1558 (CVTSS2SIrm addr:$src)>;
1559 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1560 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1561 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1562 (CVTSS2SI64rm addr:$src)>;
1565 let Predicates = [HasAVX] in {
1566 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1567 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1568 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1569 (VCVTSS2SIrm addr:$src)>;
1570 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1571 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1572 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1573 (VCVTSS2SI64rm addr:$src)>;
1578 // Convert scalar double to scalar single
1579 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1580 (ins FR64:$src1, FR64:$src2),
1581 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1584 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1585 (ins FR64:$src1, f64mem:$src2),
1586 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1587 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1589 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1592 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1593 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1594 [(set FR32:$dst, (fround FR64:$src))]>;
1595 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1596 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1597 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1598 Requires<[HasSSE2, OptForSize]>;
1600 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1601 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1603 let Constraints = "$src1 = $dst" in
1604 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1605 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1607 // Convert scalar single to scalar double
1608 // SSE2 instructions with XS prefix
1609 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1610 (ins FR32:$src1, FR32:$src2),
1611 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1612 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1614 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1615 (ins FR32:$src1, f32mem:$src2),
1616 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1617 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1619 let Predicates = [HasAVX] in {
1620 def : Pat<(f64 (fextend FR32:$src)),
1621 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1622 def : Pat<(fextend (loadf32 addr:$src)),
1623 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1624 def : Pat<(extloadf32 addr:$src),
1625 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1628 def : Pat<(extloadf32 addr:$src),
1629 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1630 Requires<[HasAVX, OptForSpeed]>;
1632 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1633 "cvtss2sd\t{$src, $dst|$dst, $src}",
1634 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1635 Requires<[HasSSE2]>;
1636 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1637 "cvtss2sd\t{$src, $dst|$dst, $src}",
1638 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1639 Requires<[HasSSE2, OptForSize]>;
1641 // extload f32 -> f64. This matches load+fextend because we have a hack in
1642 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1644 // Since these loads aren't folded into the fextend, we have to match it
1646 def : Pat<(fextend (loadf32 addr:$src)),
1647 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1648 def : Pat<(extloadf32 addr:$src),
1649 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1651 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1652 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1653 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1654 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1655 VR128:$src2))]>, XS, VEX_4V,
1657 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1658 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1659 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1660 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1661 (load addr:$src2)))]>, XS, VEX_4V,
1663 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1664 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1665 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1666 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1667 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1668 VR128:$src2))]>, XS,
1669 Requires<[HasSSE2]>;
1670 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1671 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1672 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1673 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1674 (load addr:$src2)))]>, XS,
1675 Requires<[HasSSE2]>;
1678 // Convert doubleword to packed single/double fp
1679 // SSE2 instructions without OpSize prefix
1680 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1681 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1682 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1683 TB, VEX, Requires<[HasAVX]>;
1684 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1685 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1686 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1687 (bitconvert (memopv2i64 addr:$src))))]>,
1688 TB, VEX, Requires<[HasAVX]>;
1689 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1690 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1691 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1692 TB, Requires<[HasSSE2]>;
1693 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1694 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1695 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1696 (bitconvert (memopv2i64 addr:$src))))]>,
1697 TB, Requires<[HasSSE2]>;
1699 // FIXME: why the non-intrinsic version is described as SSE3?
1700 // SSE2 instructions with XS prefix
1701 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1702 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1703 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1704 XS, VEX, Requires<[HasAVX]>;
1705 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1706 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1707 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1708 (bitconvert (memopv2i64 addr:$src))))]>,
1709 XS, VEX, Requires<[HasAVX]>;
1710 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1711 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1712 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1713 XS, Requires<[HasSSE2]>;
1714 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1715 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1716 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1717 (bitconvert (memopv2i64 addr:$src))))]>,
1718 XS, Requires<[HasSSE2]>;
1721 // Convert packed single/double fp to doubleword
1722 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1723 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1724 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1725 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1726 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1727 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1728 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1729 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1730 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1731 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1732 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1733 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1735 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1736 "cvtps2dq\t{$src, $dst|$dst, $src}",
1737 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1739 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1741 "cvtps2dq\t{$src, $dst|$dst, $src}",
1742 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1743 (memop addr:$src)))]>, VEX;
1744 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1745 "cvtps2dq\t{$src, $dst|$dst, $src}",
1746 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1747 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1748 "cvtps2dq\t{$src, $dst|$dst, $src}",
1749 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1750 (memop addr:$src)))]>;
1752 // SSE2 packed instructions with XD prefix
1753 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1754 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1755 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1756 XD, VEX, Requires<[HasAVX]>;
1757 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1758 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1759 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1760 (memop addr:$src)))]>,
1761 XD, VEX, Requires<[HasAVX]>;
1762 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1763 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1764 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1765 XD, Requires<[HasSSE2]>;
1766 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1767 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1768 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1769 (memop addr:$src)))]>,
1770 XD, Requires<[HasSSE2]>;
1773 // Convert with truncation packed single/double fp to doubleword
1774 // SSE2 packed instructions with XS prefix
1775 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1776 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1778 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1779 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1780 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1781 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1783 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1784 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1785 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1786 "cvttps2dq\t{$src, $dst|$dst, $src}",
1788 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1789 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1790 "cvttps2dq\t{$src, $dst|$dst, $src}",
1792 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1794 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1795 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1797 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1798 XS, VEX, Requires<[HasAVX]>;
1799 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1800 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1801 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1802 (memop addr:$src)))]>,
1803 XS, VEX, Requires<[HasAVX]>;
1805 let Predicates = [HasSSE2] in {
1806 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1807 (Int_CVTDQ2PSrr VR128:$src)>;
1808 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1809 (CVTTPS2DQrr VR128:$src)>;
1812 let Predicates = [HasAVX] in {
1813 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1814 (Int_VCVTDQ2PSrr VR128:$src)>;
1815 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1816 (VCVTTPS2DQrr VR128:$src)>;
1817 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1818 (VCVTDQ2PSYrr VR256:$src)>;
1819 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1820 (VCVTTPS2DQYrr VR256:$src)>;
1823 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1824 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1826 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1827 let isCodeGenOnly = 1 in
1828 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1829 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1830 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1831 (memop addr:$src)))]>, VEX;
1832 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1833 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1834 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1835 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1836 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1837 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1838 (memop addr:$src)))]>;
1840 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1841 // register, but the same isn't true when using memory operands instead.
1842 // Provide other assembly rr and rm forms to address this explicitly.
1843 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1844 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1847 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1848 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1849 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1850 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1853 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1854 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1855 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1856 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1858 // Convert packed single to packed double
1859 let Predicates = [HasAVX] in {
1860 // SSE2 instructions without OpSize prefix
1861 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1862 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1863 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1864 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1865 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1866 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1867 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1868 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1870 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1871 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1872 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1873 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1875 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1876 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1877 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1878 TB, VEX, Requires<[HasAVX]>;
1879 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1880 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1881 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1882 (load addr:$src)))]>,
1883 TB, VEX, Requires<[HasAVX]>;
1884 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1885 "cvtps2pd\t{$src, $dst|$dst, $src}",
1886 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1887 TB, Requires<[HasSSE2]>;
1888 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1889 "cvtps2pd\t{$src, $dst|$dst, $src}",
1890 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1891 (load addr:$src)))]>,
1892 TB, Requires<[HasSSE2]>;
1894 // Convert packed double to packed single
1895 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1896 // register, but the same isn't true when using memory operands instead.
1897 // Provide other assembly rr and rm forms to address this explicitly.
1898 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1899 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1900 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1901 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1904 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1905 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1906 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1907 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1910 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1911 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1912 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1913 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1914 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1915 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1916 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1917 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1920 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1921 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1922 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1923 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1925 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1926 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1927 (memop addr:$src)))]>;
1928 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1929 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1930 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1931 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1932 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1933 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1934 (memop addr:$src)))]>;
1936 // AVX 256-bit register conversion intrinsics
1937 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1938 // whenever possible to avoid declaring two versions of each one.
1939 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1940 (VCVTDQ2PSYrr VR256:$src)>;
1941 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1942 (VCVTDQ2PSYrm addr:$src)>;
1944 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1945 (VCVTPD2PSYrr VR256:$src)>;
1946 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1947 (VCVTPD2PSYrm addr:$src)>;
1949 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1950 (VCVTPS2DQYrr VR256:$src)>;
1951 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1952 (VCVTPS2DQYrm addr:$src)>;
1954 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1955 (VCVTPS2PDYrr VR128:$src)>;
1956 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1957 (VCVTPS2PDYrm addr:$src)>;
1959 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1960 (VCVTTPD2DQYrr VR256:$src)>;
1961 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1962 (VCVTTPD2DQYrm addr:$src)>;
1964 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1965 (VCVTTPS2DQYrr VR256:$src)>;
1966 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1967 (VCVTTPS2DQYrm addr:$src)>;
1969 // Match fround and fextend for 128/256-bit conversions
1970 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1971 (VCVTPD2PSYrr VR256:$src)>;
1972 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1973 (VCVTPD2PSYrm addr:$src)>;
1975 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1976 (VCVTPS2PDYrr VR128:$src)>;
1977 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1978 (VCVTPS2PDYrm addr:$src)>;
1980 //===----------------------------------------------------------------------===//
1981 // SSE 1 & 2 - Compare Instructions
1982 //===----------------------------------------------------------------------===//
1984 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1985 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1986 SDNode OpNode, ValueType VT, PatFrag ld_frag,
1987 string asm, string asm_alt> {
1988 def rr : SIi8<0xC2, MRMSrcReg,
1989 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
1990 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
1991 def rm : SIi8<0xC2, MRMSrcMem,
1992 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
1993 [(set RC:$dst, (OpNode (VT RC:$src1),
1994 (ld_frag addr:$src2), imm:$cc))]>;
1996 // Accept explicit immediate argument form instead of comparison code.
1997 let neverHasSideEffects = 1 in {
1998 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
1999 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2001 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2002 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2006 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2007 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2008 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2009 XS, VEX_4V, VEX_LIG;
2010 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2011 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2012 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2013 XD, VEX_4V, VEX_LIG;
2015 let Constraints = "$src1 = $dst" in {
2016 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2017 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2018 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2020 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2021 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2022 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2026 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2027 Intrinsic Int, string asm> {
2028 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2029 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2030 [(set VR128:$dst, (Int VR128:$src1,
2031 VR128:$src, imm:$cc))]>;
2032 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2033 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
2034 [(set VR128:$dst, (Int VR128:$src1,
2035 (load addr:$src), imm:$cc))]>;
2038 // Aliases to match intrinsics which expect XMM operand(s).
2039 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2040 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2042 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2043 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2045 let Constraints = "$src1 = $dst" in {
2046 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2047 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2048 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2049 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2053 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2054 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2055 ValueType vt, X86MemOperand x86memop,
2056 PatFrag ld_frag, string OpcodeStr, Domain d> {
2057 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2058 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2059 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2060 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2061 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2062 [(set EFLAGS, (OpNode (vt RC:$src1),
2063 (ld_frag addr:$src2)))], d>;
2066 let Defs = [EFLAGS] in {
2067 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2068 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2069 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2070 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2072 let Pattern = []<dag> in {
2073 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2074 "comiss", SSEPackedSingle>, TB, VEX,
2076 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2077 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2081 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2082 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2083 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2084 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2086 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2087 load, "comiss", SSEPackedSingle>, TB, VEX;
2088 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2089 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2090 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2091 "ucomiss", SSEPackedSingle>, TB;
2092 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2093 "ucomisd", SSEPackedDouble>, TB, OpSize;
2095 let Pattern = []<dag> in {
2096 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2097 "comiss", SSEPackedSingle>, TB;
2098 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2099 "comisd", SSEPackedDouble>, TB, OpSize;
2102 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2103 load, "ucomiss", SSEPackedSingle>, TB;
2104 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2105 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2107 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2108 "comiss", SSEPackedSingle>, TB;
2109 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2110 "comisd", SSEPackedDouble>, TB, OpSize;
2111 } // Defs = [EFLAGS]
2113 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2114 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2115 Intrinsic Int, string asm, string asm_alt,
2117 let isAsmParserOnly = 1 in {
2118 def rri : PIi8<0xC2, MRMSrcReg,
2119 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2120 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2121 def rmi : PIi8<0xC2, MRMSrcMem,
2122 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, SSECC:$cc), asm,
2123 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2126 // Accept explicit immediate argument form instead of comparison code.
2127 def rri_alt : PIi8<0xC2, MRMSrcReg,
2128 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2130 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2131 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, i8imm:$cc),
2135 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2136 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2137 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2138 SSEPackedSingle>, TB, VEX_4V;
2139 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2140 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2141 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2142 SSEPackedDouble>, TB, OpSize, VEX_4V;
2143 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2144 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2145 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2146 SSEPackedSingle>, TB, VEX_4V;
2147 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2148 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2149 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2150 SSEPackedDouble>, TB, OpSize, VEX_4V;
2151 let Constraints = "$src1 = $dst" in {
2152 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2153 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2154 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2155 SSEPackedSingle>, TB;
2156 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2157 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2158 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2159 SSEPackedDouble>, TB, OpSize;
2162 let Predicates = [HasSSE1] in {
2163 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2164 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2165 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2166 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2169 let Predicates = [HasSSE2] in {
2170 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2171 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2172 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2173 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2176 let Predicates = [HasAVX] in {
2177 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2178 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2179 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2180 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2181 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2182 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2183 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2184 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2186 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2187 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2188 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2189 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2190 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2191 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2192 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2193 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2196 //===----------------------------------------------------------------------===//
2197 // SSE 1 & 2 - Shuffle Instructions
2198 //===----------------------------------------------------------------------===//
2200 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2201 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2202 ValueType vt, string asm, PatFrag mem_frag,
2203 Domain d, bit IsConvertibleToThreeAddress = 0> {
2204 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2205 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
2206 [(set RC:$dst, (vt (shufp:$src3
2207 RC:$src1, (mem_frag addr:$src2))))], d>;
2208 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2209 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2210 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2212 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2215 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2216 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2217 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2218 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2219 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2220 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2221 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2222 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2223 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2224 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2225 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2226 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2228 let Constraints = "$src1 = $dst" in {
2229 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2230 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2231 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2233 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2234 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2235 memopv2f64, SSEPackedDouble>, TB, OpSize;
2238 let Predicates = [HasSSE1] in {
2239 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2240 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2241 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2242 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2243 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2244 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2245 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2246 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2247 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2248 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2249 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2250 // fall back to this for SSE1)
2251 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2252 (SHUFPSrri VR128:$src2, VR128:$src1,
2253 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2254 // Special unary SHUFPSrri case.
2255 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2256 (SHUFPSrri VR128:$src1, VR128:$src1,
2257 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2260 let Predicates = [HasSSE2] in {
2261 // Special binary v4i32 shuffle cases with SHUFPS.
2262 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2263 (SHUFPSrri VR128:$src1, VR128:$src2,
2264 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2265 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2266 (bc_v4i32 (memopv2i64 addr:$src2)))),
2267 (SHUFPSrmi VR128:$src1, addr:$src2,
2268 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2269 // Special unary SHUFPDrri cases.
2270 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2271 (SHUFPDrri VR128:$src1, VR128:$src1,
2272 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2273 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2274 (SHUFPDrri VR128:$src1, VR128:$src1,
2275 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2276 // Special binary v2i64 shuffle cases using SHUFPDrri.
2277 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2278 (SHUFPDrri VR128:$src1, VR128:$src2,
2279 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2280 // Generic SHUFPD patterns
2281 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2282 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2283 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2284 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2285 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2286 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2287 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2290 let Predicates = [HasAVX] in {
2291 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2292 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2293 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2294 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2295 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2296 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2297 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2298 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2299 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2300 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2301 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2302 // fall back to this for SSE1)
2303 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2304 (VSHUFPSrri VR128:$src2, VR128:$src1,
2305 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2306 // Special unary SHUFPSrri case.
2307 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2308 (VSHUFPSrri VR128:$src1, VR128:$src1,
2309 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2310 // Special binary v4i32 shuffle cases with SHUFPS.
2311 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2312 (VSHUFPSrri VR128:$src1, VR128:$src2,
2313 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2314 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2315 (bc_v4i32 (memopv2i64 addr:$src2)))),
2316 (VSHUFPSrmi VR128:$src1, addr:$src2,
2317 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2318 // Special unary SHUFPDrri cases.
2319 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2320 (VSHUFPDrri VR128:$src1, VR128:$src1,
2321 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2322 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2323 (VSHUFPDrri VR128:$src1, VR128:$src1,
2324 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2325 // Special binary v2i64 shuffle cases using SHUFPDrri.
2326 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2327 (VSHUFPDrri VR128:$src1, VR128:$src2,
2328 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2330 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2331 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2332 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2333 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2334 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2335 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2336 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2339 def : Pat<(v8i32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2340 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2341 def : Pat<(v8i32 (X86Shufps VR256:$src1,
2342 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2343 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2345 def : Pat<(v8f32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2346 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2347 def : Pat<(v8f32 (X86Shufps VR256:$src1,
2348 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2349 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2351 def : Pat<(v4i64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2352 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2353 def : Pat<(v4i64 (X86Shufpd VR256:$src1,
2354 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2355 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2357 def : Pat<(v4f64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2358 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2359 def : Pat<(v4f64 (X86Shufpd VR256:$src1,
2360 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2361 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2364 //===----------------------------------------------------------------------===//
2365 // SSE 1 & 2 - Unpack Instructions
2366 //===----------------------------------------------------------------------===//
2368 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2369 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2370 PatFrag mem_frag, RegisterClass RC,
2371 X86MemOperand x86memop, string asm,
2373 def rr : PI<opc, MRMSrcReg,
2374 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2376 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2377 def rm : PI<opc, MRMSrcMem,
2378 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2380 (vt (OpNode RC:$src1,
2381 (mem_frag addr:$src2))))], d>;
2384 let AddedComplexity = 10 in {
2385 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2386 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2387 SSEPackedSingle>, TB, VEX_4V;
2388 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2389 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2390 SSEPackedDouble>, TB, OpSize, VEX_4V;
2391 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2392 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2393 SSEPackedSingle>, TB, VEX_4V;
2394 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2395 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2396 SSEPackedDouble>, TB, OpSize, VEX_4V;
2398 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2399 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2400 SSEPackedSingle>, TB, VEX_4V;
2401 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2402 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2403 SSEPackedDouble>, TB, OpSize, VEX_4V;
2404 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2405 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2406 SSEPackedSingle>, TB, VEX_4V;
2407 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2408 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2409 SSEPackedDouble>, TB, OpSize, VEX_4V;
2411 let Constraints = "$src1 = $dst" in {
2412 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2413 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2414 SSEPackedSingle>, TB;
2415 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2416 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2417 SSEPackedDouble>, TB, OpSize;
2418 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2419 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2420 SSEPackedSingle>, TB;
2421 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2422 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2423 SSEPackedDouble>, TB, OpSize;
2424 } // Constraints = "$src1 = $dst"
2425 } // AddedComplexity
2427 let Predicates = [HasSSE1] in {
2428 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
2429 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2430 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
2431 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2432 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
2433 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2434 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
2435 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2438 let Predicates = [HasSSE2] in {
2439 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
2440 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2441 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
2442 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2443 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
2444 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2445 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
2446 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2448 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
2449 // problem is during lowering, where it's not possible to recognize the load
2450 // fold cause it has two uses through a bitcast. One use disappears at isel
2451 // time and the fold opportunity reappears.
2452 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2453 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2455 let AddedComplexity = 10 in
2456 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2457 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2460 let Predicates = [HasAVX] in {
2461 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
2462 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2463 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
2464 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2465 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
2466 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2467 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
2468 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2470 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
2471 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2472 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
2473 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2474 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
2475 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2476 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, (memopv8i32 addr:$src2))),
2477 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2478 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, (memopv8f32 addr:$src2))),
2479 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2480 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
2481 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2482 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, (memopv8i32 addr:$src2))),
2483 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2484 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
2485 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2487 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
2488 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2489 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
2490 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2491 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
2492 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2493 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
2494 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2496 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
2497 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2498 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
2499 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2500 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, (memopv4i64 addr:$src2))),
2501 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2502 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
2503 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2504 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, (memopv4f64 addr:$src2))),
2505 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2506 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
2507 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2508 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, (memopv4i64 addr:$src2))),
2509 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2510 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
2511 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2513 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
2514 // problem is during lowering, where it's not possible to recognize the load
2515 // fold cause it has two uses through a bitcast. One use disappears at isel
2516 // time and the fold opportunity reappears.
2517 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2518 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2519 let AddedComplexity = 10 in
2520 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2521 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2524 //===----------------------------------------------------------------------===//
2525 // SSE 1 & 2 - Extract Floating-Point Sign mask
2526 //===----------------------------------------------------------------------===//
2528 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2529 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2531 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2532 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2533 [(set GR32:$dst, (Int RC:$src))], d>;
2534 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2535 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2538 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2539 SSEPackedSingle>, TB;
2540 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2541 SSEPackedDouble>, TB, OpSize;
2543 def : Pat<(i32 (X86fgetsign FR32:$src)),
2544 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2545 sub_ss))>, Requires<[HasSSE1]>;
2546 def : Pat<(i64 (X86fgetsign FR32:$src)),
2547 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2548 sub_ss))>, Requires<[HasSSE1]>;
2549 def : Pat<(i32 (X86fgetsign FR64:$src)),
2550 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2551 sub_sd))>, Requires<[HasSSE2]>;
2552 def : Pat<(i64 (X86fgetsign FR64:$src)),
2553 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2554 sub_sd))>, Requires<[HasSSE2]>;
2556 let Predicates = [HasAVX] in {
2557 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2558 "movmskps", SSEPackedSingle>, TB, VEX;
2559 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2560 "movmskpd", SSEPackedDouble>, TB,
2562 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2563 "movmskps", SSEPackedSingle>, TB, VEX;
2564 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2565 "movmskpd", SSEPackedDouble>, TB,
2568 def : Pat<(i32 (X86fgetsign FR32:$src)),
2569 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2571 def : Pat<(i64 (X86fgetsign FR32:$src)),
2572 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2574 def : Pat<(i32 (X86fgetsign FR64:$src)),
2575 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2577 def : Pat<(i64 (X86fgetsign FR64:$src)),
2578 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2582 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2583 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2584 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2585 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2587 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2588 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2589 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2590 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2594 //===----------------------------------------------------------------------===//
2595 // SSE 1 & 2 - Logical Instructions
2596 //===----------------------------------------------------------------------===//
2598 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2600 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2602 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2603 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2605 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2606 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2608 let Constraints = "$src1 = $dst" in {
2609 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2610 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2612 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2613 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2617 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2618 let mayLoad = 0 in {
2619 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2620 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2621 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2624 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2625 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2627 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2629 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2631 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2632 // are all promoted to v2i64, and the patterns are covered by the int
2633 // version. This is needed in SSE only, because v2i64 isn't supported on
2634 // SSE1, but only on SSE2.
2635 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2636 !strconcat(OpcodeStr, "ps"), f128mem, [],
2637 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2638 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2640 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2641 !strconcat(OpcodeStr, "pd"), f128mem,
2642 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2643 (bc_v2i64 (v2f64 VR128:$src2))))],
2644 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2645 (memopv2i64 addr:$src2)))], 0>,
2647 let Constraints = "$src1 = $dst" in {
2648 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2649 !strconcat(OpcodeStr, "ps"), f128mem,
2650 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2651 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2652 (memopv2i64 addr:$src2)))]>, TB;
2654 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2655 !strconcat(OpcodeStr, "pd"), f128mem,
2656 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2657 (bc_v2i64 (v2f64 VR128:$src2))))],
2658 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2659 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2663 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2665 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2667 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2668 !strconcat(OpcodeStr, "ps"), f256mem,
2669 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2670 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2671 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2673 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2674 !strconcat(OpcodeStr, "pd"), f256mem,
2675 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2676 (bc_v4i64 (v4f64 VR256:$src2))))],
2677 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2678 (memopv4i64 addr:$src2)))], 0>,
2682 // AVX 256-bit packed logical ops forms
2683 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2684 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2685 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2686 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2688 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2689 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2690 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2691 let isCommutable = 0 in
2692 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2694 //===----------------------------------------------------------------------===//
2695 // SSE 1 & 2 - Arithmetic Instructions
2696 //===----------------------------------------------------------------------===//
2698 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2701 /// In addition, we also have a special variant of the scalar form here to
2702 /// represent the associated intrinsic operation. This form is unlike the
2703 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2704 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2706 /// These three forms can each be reg+reg or reg+mem.
2709 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2711 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2713 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2714 OpNode, FR32, f32mem, Is2Addr>, XS;
2715 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2716 OpNode, FR64, f64mem, Is2Addr>, XD;
2719 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2721 let mayLoad = 0 in {
2722 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2723 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2724 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2725 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2729 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2731 let mayLoad = 0 in {
2732 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2733 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2734 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2735 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2739 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2741 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2742 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2743 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2744 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2747 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2749 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2750 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2751 SSEPackedSingle, Is2Addr>, TB;
2753 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2754 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2755 SSEPackedDouble, Is2Addr>, TB, OpSize;
2758 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2759 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2760 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2761 SSEPackedSingle, 0>, TB;
2763 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2764 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2765 SSEPackedDouble, 0>, TB, OpSize;
2768 // Binary Arithmetic instructions
2769 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2770 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2771 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2772 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2773 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2774 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2775 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2776 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2778 let isCommutable = 0 in {
2779 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2780 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2781 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2782 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2783 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2784 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2785 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2786 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2787 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2788 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2789 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2790 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2791 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2792 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2793 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2794 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2795 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2796 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2797 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2798 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2801 let Constraints = "$src1 = $dst" in {
2802 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2803 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2804 basic_sse12_fp_binop_s_int<0x58, "add">;
2805 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2806 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2807 basic_sse12_fp_binop_s_int<0x59, "mul">;
2809 let isCommutable = 0 in {
2810 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2811 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2812 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2813 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2814 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2815 basic_sse12_fp_binop_s_int<0x5E, "div">;
2816 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2817 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2818 basic_sse12_fp_binop_s_int<0x5F, "max">,
2819 basic_sse12_fp_binop_p_int<0x5F, "max">;
2820 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2821 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2822 basic_sse12_fp_binop_s_int<0x5D, "min">,
2823 basic_sse12_fp_binop_p_int<0x5D, "min">;
2828 /// In addition, we also have a special variant of the scalar form here to
2829 /// represent the associated intrinsic operation. This form is unlike the
2830 /// plain scalar form, in that it takes an entire vector (instead of a
2831 /// scalar) and leaves the top elements undefined.
2833 /// And, we have a special variant form for a full-vector intrinsic form.
2835 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2836 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2837 SDNode OpNode, Intrinsic F32Int> {
2838 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2839 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2840 [(set FR32:$dst, (OpNode FR32:$src))]>;
2841 // For scalar unary operations, fold a load into the operation
2842 // only in OptForSize mode. It eliminates an instruction, but it also
2843 // eliminates a whole-register clobber (the load), so it introduces a
2844 // partial register update condition.
2845 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2846 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2847 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2848 Requires<[HasSSE1, OptForSize]>;
2849 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2850 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2851 [(set VR128:$dst, (F32Int VR128:$src))]>;
2852 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2853 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2854 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2857 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2858 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2859 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2860 !strconcat(OpcodeStr,
2861 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2863 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2864 !strconcat(OpcodeStr,
2865 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2866 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2867 (ins ssmem:$src1, VR128:$src2),
2868 !strconcat(OpcodeStr,
2869 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2872 /// sse1_fp_unop_p - SSE1 unops in packed form.
2873 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2874 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2875 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2876 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2877 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2878 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2879 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2882 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2883 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2884 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2885 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2886 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2887 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2888 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2889 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2892 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2893 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2894 Intrinsic V4F32Int> {
2895 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2896 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2897 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2898 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2899 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2900 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2903 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2904 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2905 Intrinsic V4F32Int> {
2906 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2907 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2908 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2909 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2910 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2911 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2914 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2915 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2916 SDNode OpNode, Intrinsic F64Int> {
2917 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2918 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2919 [(set FR64:$dst, (OpNode FR64:$src))]>;
2920 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2921 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2922 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2923 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2924 Requires<[HasSSE2, OptForSize]>;
2925 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2926 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2927 [(set VR128:$dst, (F64Int VR128:$src))]>;
2928 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2929 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2930 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2933 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2934 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2935 let neverHasSideEffects = 1 in {
2936 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2937 !strconcat(OpcodeStr,
2938 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2940 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2941 !strconcat(OpcodeStr,
2942 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2944 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2945 (ins VR128:$src1, sdmem:$src2),
2946 !strconcat(OpcodeStr,
2947 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2950 /// sse2_fp_unop_p - SSE2 unops in vector forms.
2951 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
2953 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2954 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2955 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
2956 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2957 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2958 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
2961 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
2962 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2963 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2964 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2965 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
2966 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2967 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2968 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
2971 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
2972 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2973 Intrinsic V2F64Int> {
2974 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2975 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2976 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
2977 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2978 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2979 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
2982 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
2983 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2984 Intrinsic V2F64Int> {
2985 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2986 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2987 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
2988 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2989 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2990 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
2993 let Predicates = [HasAVX] in {
2995 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
2996 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
2998 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
2999 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3000 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3001 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3002 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3003 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3004 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3005 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3008 // Reciprocal approximations. Note that these typically require refinement
3009 // in order to obtain suitable precision.
3010 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3011 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3012 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3013 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3014 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3016 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3017 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3018 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3019 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3020 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3023 def : Pat<(f32 (fsqrt FR32:$src)),
3024 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3025 def : Pat<(f32 (fsqrt (load addr:$src))),
3026 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3027 Requires<[HasAVX, OptForSize]>;
3028 def : Pat<(f64 (fsqrt FR64:$src)),
3029 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3030 def : Pat<(f64 (fsqrt (load addr:$src))),
3031 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3032 Requires<[HasAVX, OptForSize]>;
3034 def : Pat<(f32 (X86frsqrt FR32:$src)),
3035 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3036 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3037 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3038 Requires<[HasAVX, OptForSize]>;
3040 def : Pat<(f32 (X86frcp FR32:$src)),
3041 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3042 def : Pat<(f32 (X86frcp (load addr:$src))),
3043 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3044 Requires<[HasAVX, OptForSize]>;
3046 let Predicates = [HasAVX] in {
3047 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3048 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3049 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3050 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3052 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3053 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3055 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3056 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3057 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3058 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3060 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3061 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3063 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3064 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3065 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3066 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3068 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3069 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3071 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3072 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3073 (VRCPSSr (f32 (IMPLICIT_DEF)),
3074 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3076 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3077 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3081 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3082 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3083 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3084 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3085 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3086 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3088 // Reciprocal approximations. Note that these typically require refinement
3089 // in order to obtain suitable precision.
3090 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3091 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3092 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3093 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3094 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3095 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3097 // There is no f64 version of the reciprocal approximation instructions.
3099 //===----------------------------------------------------------------------===//
3100 // SSE 1 & 2 - Non-temporal stores
3101 //===----------------------------------------------------------------------===//
3103 let AddedComplexity = 400 in { // Prefer non-temporal versions
3104 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3105 (ins f128mem:$dst, VR128:$src),
3106 "movntps\t{$src, $dst|$dst, $src}",
3107 [(alignednontemporalstore (v4f32 VR128:$src),
3109 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3110 (ins f128mem:$dst, VR128:$src),
3111 "movntpd\t{$src, $dst|$dst, $src}",
3112 [(alignednontemporalstore (v2f64 VR128:$src),
3114 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
3115 (ins f128mem:$dst, VR128:$src),
3116 "movntdq\t{$src, $dst|$dst, $src}",
3117 [(alignednontemporalstore (v2f64 VR128:$src),
3120 let ExeDomain = SSEPackedInt in
3121 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3122 (ins f128mem:$dst, VR128:$src),
3123 "movntdq\t{$src, $dst|$dst, $src}",
3124 [(alignednontemporalstore (v4f32 VR128:$src),
3127 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3128 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3130 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3131 (ins f256mem:$dst, VR256:$src),
3132 "movntps\t{$src, $dst|$dst, $src}",
3133 [(alignednontemporalstore (v8f32 VR256:$src),
3135 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3136 (ins f256mem:$dst, VR256:$src),
3137 "movntpd\t{$src, $dst|$dst, $src}",
3138 [(alignednontemporalstore (v4f64 VR256:$src),
3140 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
3141 (ins f256mem:$dst, VR256:$src),
3142 "movntdq\t{$src, $dst|$dst, $src}",
3143 [(alignednontemporalstore (v4f64 VR256:$src),
3145 let ExeDomain = SSEPackedInt in
3146 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3147 (ins f256mem:$dst, VR256:$src),
3148 "movntdq\t{$src, $dst|$dst, $src}",
3149 [(alignednontemporalstore (v8f32 VR256:$src),
3153 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3154 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3155 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3156 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3157 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3158 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3160 let AddedComplexity = 400 in { // Prefer non-temporal versions
3161 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3162 "movntps\t{$src, $dst|$dst, $src}",
3163 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3164 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3165 "movntpd\t{$src, $dst|$dst, $src}",
3166 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3168 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3169 "movntdq\t{$src, $dst|$dst, $src}",
3170 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
3172 let ExeDomain = SSEPackedInt in
3173 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3174 "movntdq\t{$src, $dst|$dst, $src}",
3175 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3177 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3178 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3180 // There is no AVX form for instructions below this point
3181 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3182 "movnti{l}\t{$src, $dst|$dst, $src}",
3183 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3184 TB, Requires<[HasSSE2]>;
3185 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3186 "movnti{q}\t{$src, $dst|$dst, $src}",
3187 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3188 TB, Requires<[HasSSE2]>;
3191 //===----------------------------------------------------------------------===//
3192 // SSE 1 & 2 - Prefetch and memory fence
3193 //===----------------------------------------------------------------------===//
3195 // Prefetch intrinsic.
3196 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
3197 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
3198 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
3199 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
3200 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
3201 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
3202 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
3203 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
3206 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3207 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3208 TB, Requires<[HasSSE2]>;
3210 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3211 // was introduced with SSE2, it's backward compatible.
3212 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3214 // Load, store, and memory fence
3215 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3216 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3217 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3218 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3219 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3220 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3222 def : Pat<(X86SFence), (SFENCE)>;
3223 def : Pat<(X86LFence), (LFENCE)>;
3224 def : Pat<(X86MFence), (MFENCE)>;
3226 //===----------------------------------------------------------------------===//
3227 // SSE 1 & 2 - Load/Store XCSR register
3228 //===----------------------------------------------------------------------===//
3230 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3231 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3232 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3233 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3235 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3236 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3237 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3238 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3240 //===---------------------------------------------------------------------===//
3241 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3242 //===---------------------------------------------------------------------===//
3244 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3246 let neverHasSideEffects = 1 in {
3247 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3248 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3249 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3250 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3252 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3253 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3254 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3255 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3258 let isCodeGenOnly = 1 in {
3259 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3260 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3261 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3262 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3263 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3264 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3265 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3266 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3269 let canFoldAsLoad = 1, mayLoad = 1 in {
3270 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3271 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3272 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3273 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3274 let Predicates = [HasAVX] in {
3275 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3276 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3277 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3278 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3282 let mayStore = 1 in {
3283 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3284 (ins i128mem:$dst, VR128:$src),
3285 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3286 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3287 (ins i256mem:$dst, VR256:$src),
3288 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3289 let Predicates = [HasAVX] in {
3290 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3291 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3292 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3293 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3297 let neverHasSideEffects = 1 in
3298 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3299 "movdqa\t{$src, $dst|$dst, $src}", []>;
3301 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3302 "movdqu\t{$src, $dst|$dst, $src}",
3303 []>, XS, Requires<[HasSSE2]>;
3306 let isCodeGenOnly = 1 in {
3307 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3308 "movdqa\t{$src, $dst|$dst, $src}", []>;
3310 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3311 "movdqu\t{$src, $dst|$dst, $src}",
3312 []>, XS, Requires<[HasSSE2]>;
3315 let canFoldAsLoad = 1, mayLoad = 1 in {
3316 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3317 "movdqa\t{$src, $dst|$dst, $src}",
3318 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3319 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3320 "movdqu\t{$src, $dst|$dst, $src}",
3321 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3322 XS, Requires<[HasSSE2]>;
3325 let mayStore = 1 in {
3326 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3327 "movdqa\t{$src, $dst|$dst, $src}",
3328 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3329 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3330 "movdqu\t{$src, $dst|$dst, $src}",
3331 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3332 XS, Requires<[HasSSE2]>;
3335 // Intrinsic forms of MOVDQU load and store
3336 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3337 "vmovdqu\t{$src, $dst|$dst, $src}",
3338 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3339 XS, VEX, Requires<[HasAVX]>;
3341 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3342 "movdqu\t{$src, $dst|$dst, $src}",
3343 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3344 XS, Requires<[HasSSE2]>;
3346 } // ExeDomain = SSEPackedInt
3348 let Predicates = [HasAVX] in {
3349 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3350 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3351 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3354 //===---------------------------------------------------------------------===//
3355 // SSE2 - Packed Integer Arithmetic Instructions
3356 //===---------------------------------------------------------------------===//
3358 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3360 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3361 RegisterClass RC, PatFrag memop_frag,
3362 X86MemOperand x86memop, bit IsCommutable = 0,
3364 let isCommutable = IsCommutable in
3365 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3366 (ins RC:$src1, RC:$src2),
3368 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3369 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3370 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3371 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3372 (ins RC:$src1, x86memop:$src2),
3374 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3375 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3376 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3379 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3380 string OpcodeStr, Intrinsic IntId,
3381 Intrinsic IntId2, RegisterClass RC,
3383 // src2 is always 128-bit
3384 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3385 (ins RC:$src1, VR128:$src2),
3387 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3388 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3389 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3390 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3391 (ins RC:$src1, i128mem:$src2),
3393 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3394 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3395 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3396 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3397 (ins RC:$src1, i32i8imm:$src2),
3399 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3400 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3401 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3404 /// PDI_binop_rm - Simple SSE2 binary operator.
3405 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
3406 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
3407 X86MemOperand x86memop, bit IsCommutable = 0,
3409 let isCommutable = IsCommutable in
3410 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3411 (ins RC:$src1, RC:$src2),
3413 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3414 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3415 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
3416 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3417 (ins RC:$src1, x86memop:$src2),
3419 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3420 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3421 [(set RC:$dst, (OpVT (OpNode RC:$src1,
3422 (bitconvert (memop_frag addr:$src2)))))]>;
3425 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
3427 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
3428 /// to collapse (bitconvert VT to VT) into its operand.
3430 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
3431 bit IsCommutable = 0, bit Is2Addr = 1> {
3432 let isCommutable = IsCommutable in
3433 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
3434 (ins VR128:$src1, VR128:$src2),
3436 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3437 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3438 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
3439 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
3440 (ins VR128:$src1, i128mem:$src2),
3442 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3443 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3444 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
3447 /// PDI_binop_rm_v4i64 - Simple AVX2 binary operator whose type is v4i64.
3449 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
3450 /// to collapse (bitconvert VT to VT) into its operand.
3452 multiclass PDI_binop_rm_v4i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
3453 bit IsCommutable = 0> {
3454 let isCommutable = IsCommutable in
3455 def rr : PDI<opc, MRMSrcReg, (outs VR256:$dst),
3456 (ins VR256:$src1, VR256:$src2),
3457 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3458 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))]>;
3459 def rm : PDI<opc, MRMSrcMem, (outs VR256:$dst),
3460 (ins VR256:$src1, i256mem:$src2),
3461 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3462 [(set VR256:$dst, (OpNode VR256:$src1, (memopv4i64 addr:$src2)))]>;
3465 } // ExeDomain = SSEPackedInt
3467 // 128-bit Integer Arithmetic
3469 let Predicates = [HasAVX] in {
3470 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3471 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3472 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3473 i128mem, 1, 0>, VEX_4V;
3474 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3475 i128mem, 1, 0>, VEX_4V;
3476 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
3477 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3478 i128mem, 1, 0>, VEX_4V;
3479 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3480 i128mem, 0, 0>, VEX_4V;
3481 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3482 i128mem, 0, 0>, VEX_4V;
3483 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3484 i128mem, 0, 0>, VEX_4V;
3485 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
3488 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3489 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3490 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3491 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3492 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3493 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3494 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3495 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3496 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3497 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3498 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3499 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3500 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3501 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3502 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3503 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3504 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3505 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3506 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3507 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3508 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3509 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3510 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3511 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3512 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3513 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3514 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3515 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3516 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3517 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3518 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3519 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3520 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3521 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3522 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3523 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3524 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3525 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3528 let Predicates = [HasAVX2] in {
3529 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3530 i256mem, 1, 0>, VEX_4V;
3531 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3532 i256mem, 1, 0>, VEX_4V;
3533 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3534 i256mem, 1, 0>, VEX_4V;
3535 defm VPADDQY : PDI_binop_rm_v4i64<0xD4, "vpaddq", add, 1>, VEX_4V;
3536 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3537 i256mem, 1, 0>, VEX_4V;
3538 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3539 i256mem, 0, 0>, VEX_4V;
3540 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3541 i256mem, 0, 0>, VEX_4V;
3542 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3543 i256mem, 0, 0>, VEX_4V;
3544 defm VPSUBQY : PDI_binop_rm_v4i64<0xFB, "vpsubq", sub, 0>, VEX_4V;
3547 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3548 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3549 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3550 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3551 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3552 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3553 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3554 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3555 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3556 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3557 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3558 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3559 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3560 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3561 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3562 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3563 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3564 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3565 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3566 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3567 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3568 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3569 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3570 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3571 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3572 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3573 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3574 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3575 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3576 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3577 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3578 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3579 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3580 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3581 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3582 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3583 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3584 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3587 let Constraints = "$src1 = $dst" in {
3588 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3590 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3592 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3594 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
3595 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3597 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3599 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3601 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3603 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
3606 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3607 VR128, memopv2i64, i128mem>;
3608 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3609 VR128, memopv2i64, i128mem>;
3610 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3611 VR128, memopv2i64, i128mem>;
3612 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3613 VR128, memopv2i64, i128mem>;
3614 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3615 VR128, memopv2i64, i128mem, 1>;
3616 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3617 VR128, memopv2i64, i128mem, 1>;
3618 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3619 VR128, memopv2i64, i128mem, 1>;
3620 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3621 VR128, memopv2i64, i128mem, 1>;
3622 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3623 VR128, memopv2i64, i128mem, 1>;
3624 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3625 VR128, memopv2i64, i128mem, 1>;
3626 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3627 VR128, memopv2i64, i128mem, 1>;
3628 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3629 VR128, memopv2i64, i128mem, 1>;
3630 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3631 VR128, memopv2i64, i128mem, 1>;
3632 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3633 VR128, memopv2i64, i128mem, 1>;
3634 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3635 VR128, memopv2i64, i128mem, 1>;
3636 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3637 VR128, memopv2i64, i128mem, 1>;
3638 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3639 VR128, memopv2i64, i128mem, 1>;
3640 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3641 VR128, memopv2i64, i128mem, 1>;
3642 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3643 VR128, memopv2i64, i128mem, 1>;
3645 } // Constraints = "$src1 = $dst"
3647 //===---------------------------------------------------------------------===//
3648 // SSE2 - Packed Integer Logical Instructions
3649 //===---------------------------------------------------------------------===//
3651 let Predicates = [HasAVX] in {
3652 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3653 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3655 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3656 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3658 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3659 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3662 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3663 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3665 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3666 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3668 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3669 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3672 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3673 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3675 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3676 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3679 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
3680 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
3681 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
3683 let ExeDomain = SSEPackedInt in {
3684 let neverHasSideEffects = 1 in {
3685 // 128-bit logical shifts.
3686 def VPSLLDQri : PDIi8<0x73, MRM7r,
3687 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3688 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3690 def VPSRLDQri : PDIi8<0x73, MRM3r,
3691 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3692 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3694 // PSRADQri doesn't exist in SSE[1-3].
3696 def VPANDNrr : PDI<0xDF, MRMSrcReg,
3697 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3698 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3700 (v2i64 (X86andnp VR128:$src1, VR128:$src2)))]>,VEX_4V;
3702 def VPANDNrm : PDI<0xDF, MRMSrcMem,
3703 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3704 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3705 [(set VR128:$dst, (X86andnp VR128:$src1,
3706 (memopv2i64 addr:$src2)))]>, VEX_4V;
3710 let Predicates = [HasAVX2] in {
3711 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3712 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3714 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3715 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3717 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3718 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3721 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3722 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3724 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3725 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3727 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3728 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3731 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3732 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3734 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3735 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3738 defm VPANDY : PDI_binop_rm_v4i64<0xDB, "vpand", and, 1>, VEX_4V;
3739 defm VPORY : PDI_binop_rm_v4i64<0xEB, "vpor" , or, 1>, VEX_4V;
3740 defm VPXORY : PDI_binop_rm_v4i64<0xEF, "vpxor", xor, 1>, VEX_4V;
3742 let ExeDomain = SSEPackedInt in {
3743 let neverHasSideEffects = 1 in {
3744 // 128-bit logical shifts.
3745 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3746 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3747 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3749 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3750 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3751 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3753 // PSRADQYri doesn't exist in SSE[1-3].
3755 def VPANDNYrr : PDI<0xDF, MRMSrcReg,
3756 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
3757 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3759 (v4i64 (X86andnp VR256:$src1, VR256:$src2)))]>,VEX_4V;
3761 def VPANDNYrm : PDI<0xDF, MRMSrcMem,
3762 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
3763 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3764 [(set VR256:$dst, (X86andnp VR256:$src1,
3765 (memopv4i64 addr:$src2)))]>, VEX_4V;
3769 let Constraints = "$src1 = $dst" in {
3770 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3771 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3773 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3774 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3776 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3777 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3780 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3781 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3783 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3784 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3786 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3787 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3790 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3791 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3793 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3794 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3797 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
3798 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
3799 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
3801 let ExeDomain = SSEPackedInt in {
3802 let neverHasSideEffects = 1 in {
3803 // 128-bit logical shifts.
3804 def PSLLDQri : PDIi8<0x73, MRM7r,
3805 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3806 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3807 def PSRLDQri : PDIi8<0x73, MRM3r,
3808 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3809 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3810 // PSRADQri doesn't exist in SSE[1-3].
3811 def PANDNrr : PDI<0xDF, MRMSrcReg,
3812 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3813 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3816 def PANDNrm : PDI<0xDF, MRMSrcMem,
3817 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3818 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3821 } // Constraints = "$src1 = $dst"
3823 let Predicates = [HasAVX] in {
3824 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3825 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3826 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3827 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3828 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3829 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
3830 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3831 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
3832 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3833 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3835 // Shift up / down and insert zero's.
3836 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3837 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3838 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3839 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3842 let Predicates = [HasAVX2] in {
3843 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3844 (v4i64 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2)))>;
3845 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3846 (v4i64 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2)))>;
3847 def : Pat<(int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2),
3848 (v4i64 (VPSLLDQYri VR256:$src1, imm:$src2))>;
3849 def : Pat<(int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2),
3850 (v4i64 (VPSRLDQYri VR256:$src1, imm:$src2))>;
3853 let Predicates = [HasSSE2] in {
3854 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3855 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3856 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3857 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3858 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3859 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
3860 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3861 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
3862 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3863 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3865 // Shift up / down and insert zero's.
3866 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3867 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3868 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3869 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3872 //===---------------------------------------------------------------------===//
3873 // SSE2 - Packed Integer Comparison Instructions
3874 //===---------------------------------------------------------------------===//
3876 let Predicates = [HasAVX] in {
3877 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
3878 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3879 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
3880 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3881 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
3882 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3883 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
3884 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3885 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
3886 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3887 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
3888 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3890 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3891 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3892 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3893 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3894 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3895 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3896 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3897 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3898 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3899 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3900 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3901 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3903 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3904 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3905 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3906 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3907 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3908 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3909 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3910 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3911 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3912 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3913 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3914 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3917 let Predicates = [HasAVX2] in {
3918 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
3919 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3920 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
3921 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3922 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
3923 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3924 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
3925 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3926 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
3927 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3928 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
3929 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3931 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, VR256:$src2)),
3932 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
3933 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, (memop addr:$src2))),
3934 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
3935 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, VR256:$src2)),
3936 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
3937 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, (memop addr:$src2))),
3938 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
3939 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, VR256:$src2)),
3940 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
3941 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, (memop addr:$src2))),
3942 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
3944 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, VR256:$src2)),
3945 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
3946 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, (memop addr:$src2))),
3947 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
3948 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, VR256:$src2)),
3949 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
3950 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, (memop addr:$src2))),
3951 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
3952 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, VR256:$src2)),
3953 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
3954 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, (memop addr:$src2))),
3955 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
3958 let Constraints = "$src1 = $dst" in {
3959 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
3960 VR128, memopv2i64, i128mem, 1>;
3961 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
3962 VR128, memopv2i64, i128mem, 1>;
3963 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
3964 VR128, memopv2i64, i128mem, 1>;
3965 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
3966 VR128, memopv2i64, i128mem>;
3967 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
3968 VR128, memopv2i64, i128mem>;
3969 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
3970 VR128, memopv2i64, i128mem>;
3971 } // Constraints = "$src1 = $dst"
3973 let Predicates = [HasSSE2] in {
3974 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3975 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
3976 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3977 (PCMPEQBrm VR128:$src1, addr:$src2)>;
3978 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3979 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
3980 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3981 (PCMPEQWrm VR128:$src1, addr:$src2)>;
3982 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3983 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
3984 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3985 (PCMPEQDrm VR128:$src1, addr:$src2)>;
3987 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3988 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
3989 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3990 (PCMPGTBrm VR128:$src1, addr:$src2)>;
3991 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3992 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
3993 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3994 (PCMPGTWrm VR128:$src1, addr:$src2)>;
3995 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3996 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
3997 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3998 (PCMPGTDrm VR128:$src1, addr:$src2)>;
4001 //===---------------------------------------------------------------------===//
4002 // SSE2 - Packed Integer Pack Instructions
4003 //===---------------------------------------------------------------------===//
4005 let Predicates = [HasAVX] in {
4006 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4007 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4008 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4009 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4010 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4011 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4014 let Predicates = [HasAVX2] in {
4015 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4016 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4017 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4018 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4019 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4020 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4023 let Constraints = "$src1 = $dst" in {
4024 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4025 VR128, memopv2i64, i128mem>;
4026 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4027 VR128, memopv2i64, i128mem>;
4028 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4029 VR128, memopv2i64, i128mem>;
4030 } // Constraints = "$src1 = $dst"
4032 //===---------------------------------------------------------------------===//
4033 // SSE2 - Packed Integer Shuffle Instructions
4034 //===---------------------------------------------------------------------===//
4036 let ExeDomain = SSEPackedInt in {
4037 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4039 def ri : Ii8<0x70, MRMSrcReg,
4040 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4041 !strconcat(OpcodeStr,
4042 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4043 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4045 def mi : Ii8<0x70, MRMSrcMem,
4046 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4047 !strconcat(OpcodeStr,
4048 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4049 [(set VR128:$dst, (vt (pshuf_frag:$src2
4050 (bc_frag (memopv2i64 addr:$src1)),
4054 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4056 def Yri : Ii8<0x70, MRMSrcReg,
4057 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4058 !strconcat(OpcodeStr,
4059 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4060 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4062 def Ymi : Ii8<0x70, MRMSrcMem,
4063 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4064 !strconcat(OpcodeStr,
4065 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4066 [(set VR256:$dst, (vt (pshuf_frag:$src2
4067 (bc_frag (memopv4i64 addr:$src1)),
4070 } // ExeDomain = SSEPackedInt
4072 let Predicates = [HasAVX] in {
4073 let AddedComplexity = 5 in
4074 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4077 // SSE2 with ImmT == Imm8 and XS prefix.
4078 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4081 // SSE2 with ImmT == Imm8 and XD prefix.
4082 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4085 let AddedComplexity = 5 in
4086 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4087 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4088 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4089 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4090 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4092 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4094 (VPSHUFDmi addr:$src1, imm:$imm)>;
4095 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4097 (VPSHUFDmi addr:$src1, imm:$imm)>;
4098 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4099 (VPSHUFDri VR128:$src1, imm:$imm)>;
4100 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4101 (VPSHUFDri VR128:$src1, imm:$imm)>;
4102 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4103 (VPSHUFHWri VR128:$src, imm:$imm)>;
4104 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4106 (VPSHUFHWmi addr:$src, imm:$imm)>;
4107 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4108 (VPSHUFLWri VR128:$src, imm:$imm)>;
4109 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4111 (VPSHUFLWmi addr:$src, imm:$imm)>;
4114 let Predicates = [HasAVX2] in {
4115 let AddedComplexity = 5 in
4116 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4119 // SSE2 with ImmT == Imm8 and XS prefix.
4120 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4123 // SSE2 with ImmT == Imm8 and XD prefix.
4124 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4128 let Predicates = [HasSSE2] in {
4129 let AddedComplexity = 5 in
4130 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4132 // SSE2 with ImmT == Imm8 and XS prefix.
4133 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4135 // SSE2 with ImmT == Imm8 and XD prefix.
4136 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4138 let AddedComplexity = 5 in
4139 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4140 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4141 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4142 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4143 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4145 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4147 (PSHUFDmi addr:$src1, imm:$imm)>;
4148 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4150 (PSHUFDmi addr:$src1, imm:$imm)>;
4151 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4152 (PSHUFDri VR128:$src1, imm:$imm)>;
4153 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4154 (PSHUFDri VR128:$src1, imm:$imm)>;
4155 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4156 (PSHUFHWri VR128:$src, imm:$imm)>;
4157 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4159 (PSHUFHWmi addr:$src, imm:$imm)>;
4160 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4161 (PSHUFLWri VR128:$src, imm:$imm)>;
4162 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4164 (PSHUFLWmi addr:$src, imm:$imm)>;
4167 //===---------------------------------------------------------------------===//
4168 // SSE2 - Packed Integer Unpack Instructions
4169 //===---------------------------------------------------------------------===//
4171 let ExeDomain = SSEPackedInt in {
4172 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4173 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4174 def rr : PDI<opc, MRMSrcReg,
4175 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4177 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4178 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4179 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4180 def rm : PDI<opc, MRMSrcMem,
4181 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4183 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4184 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4185 [(set VR128:$dst, (OpNode VR128:$src1,
4186 (bc_frag (memopv2i64
4190 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4191 SDNode OpNode, PatFrag bc_frag> {
4192 def Yrr : PDI<opc, MRMSrcReg,
4193 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4194 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4195 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4196 def Yrm : PDI<opc, MRMSrcMem,
4197 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4198 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4199 [(set VR256:$dst, (OpNode VR256:$src1,
4200 (bc_frag (memopv4i64 addr:$src2))))]>;
4203 let Predicates = [HasAVX] in {
4204 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Punpcklbw,
4205 bc_v16i8, 0>, VEX_4V;
4206 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Punpcklwd,
4207 bc_v8i16, 0>, VEX_4V;
4208 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Punpckldq,
4209 bc_v4i32, 0>, VEX_4V;
4211 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
4212 /// knew to collapse (bitconvert VT to VT) into its operand.
4213 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
4214 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4215 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4216 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
4217 VR128:$src2)))]>, VEX_4V;
4218 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
4219 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4220 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4221 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
4222 (memopv2i64 addr:$src2))))]>, VEX_4V;
4224 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Punpckhbw,
4225 bc_v16i8, 0>, VEX_4V;
4226 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Punpckhwd,
4227 bc_v8i16, 0>, VEX_4V;
4228 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Punpckhdq,
4229 bc_v4i32, 0>, VEX_4V;
4231 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
4232 /// knew to collapse (bitconvert VT to VT) into its operand.
4233 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
4234 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4235 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4236 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
4237 VR128:$src2)))]>, VEX_4V;
4238 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
4239 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4240 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4241 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
4242 (memopv2i64 addr:$src2))))]>, VEX_4V;
4245 let Predicates = [HasAVX2] in {
4246 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Punpcklbw,
4248 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Punpcklwd,
4250 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Punpckldq,
4253 /// FIXME: we could eliminate this and use sse2_unpack_y instead if tblgen
4254 /// knew to collapse (bitconvert VT to VT) into its operand.
4255 def VPUNPCKLQDQYrr : PDI<0x6C, MRMSrcReg,
4256 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4257 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4258 [(set VR256:$dst, (v4i64 (X86Punpcklqdq VR256:$src1,
4259 VR256:$src2)))]>, VEX_4V;
4260 def VPUNPCKLQDQYrm : PDI<0x6C, MRMSrcMem,
4261 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4262 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4263 [(set VR256:$dst, (v4i64 (X86Punpcklqdq VR256:$src1,
4264 (memopv4i64 addr:$src2))))]>, VEX_4V;
4266 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Punpckhbw,
4268 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Punpckhwd,
4270 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Punpckhdq,
4273 /// FIXME: we could eliminate this and use sse2_unpack_y instead if tblgen
4274 /// knew to collapse (bitconvert VT to VT) into its operand.
4275 def VPUNPCKHQDQYrr : PDI<0x6D, MRMSrcReg,
4276 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4277 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4278 [(set VR256:$dst, (v4i64 (X86Punpckhqdq VR256:$src1,
4279 VR256:$src2)))]>, VEX_4V;
4280 def VPUNPCKHQDQYrm : PDI<0x6D, MRMSrcMem,
4281 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4282 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4283 [(set VR256:$dst, (v4i64 (X86Punpckhqdq VR256:$src1,
4284 (memopv4i64 addr:$src2))))]>, VEX_4V;
4287 let Constraints = "$src1 = $dst" in {
4288 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Punpcklbw, bc_v16i8>;
4289 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Punpcklwd, bc_v8i16>;
4290 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Punpckldq, bc_v4i32>;
4292 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
4293 /// knew to collapse (bitconvert VT to VT) into its operand.
4294 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
4295 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4296 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
4298 (v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)))]>;
4299 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
4300 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4301 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
4303 (v2i64 (X86Punpcklqdq VR128:$src1,
4304 (memopv2i64 addr:$src2))))]>;
4306 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Punpckhbw, bc_v16i8>;
4307 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Punpckhwd, bc_v8i16>;
4308 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Punpckhdq, bc_v4i32>;
4310 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
4311 /// knew to collapse (bitconvert VT to VT) into its operand.
4312 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
4313 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4314 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
4316 (v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)))]>;
4317 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
4318 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4319 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
4321 (v2i64 (X86Punpckhqdq VR128:$src1,
4322 (memopv2i64 addr:$src2))))]>;
4324 } // ExeDomain = SSEPackedInt
4326 // Splat v2f64 / v2i64
4327 let AddedComplexity = 10 in {
4328 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4329 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4330 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4331 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4334 //===---------------------------------------------------------------------===//
4335 // SSE2 - Packed Integer Extract and Insert
4336 //===---------------------------------------------------------------------===//
4338 let ExeDomain = SSEPackedInt in {
4339 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4340 def rri : Ii8<0xC4, MRMSrcReg,
4341 (outs VR128:$dst), (ins VR128:$src1,
4342 GR32:$src2, i32i8imm:$src3),
4344 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4345 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4347 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4348 def rmi : Ii8<0xC4, MRMSrcMem,
4349 (outs VR128:$dst), (ins VR128:$src1,
4350 i16mem:$src2, i32i8imm:$src3),
4352 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4353 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4355 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4360 let Predicates = [HasAVX] in
4361 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4362 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4363 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4364 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4365 imm:$src2))]>, TB, OpSize, VEX;
4366 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4367 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4368 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4369 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4373 let Predicates = [HasAVX] in {
4374 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4375 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4376 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4377 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4378 []>, TB, OpSize, VEX_4V;
4381 let Constraints = "$src1 = $dst" in
4382 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4384 } // ExeDomain = SSEPackedInt
4386 //===---------------------------------------------------------------------===//
4387 // SSE2 - Packed Mask Creation
4388 //===---------------------------------------------------------------------===//
4390 let ExeDomain = SSEPackedInt in {
4392 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4393 "pmovmskb\t{$src, $dst|$dst, $src}",
4394 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4395 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4396 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4398 let Predicates = [HasAVX2] in {
4399 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4400 "pmovmskb\t{$src, $dst|$dst, $src}",
4401 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4402 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4403 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4406 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4407 "pmovmskb\t{$src, $dst|$dst, $src}",
4408 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4410 } // ExeDomain = SSEPackedInt
4412 //===---------------------------------------------------------------------===//
4413 // SSE2 - Conditional Store
4414 //===---------------------------------------------------------------------===//
4416 let ExeDomain = SSEPackedInt in {
4419 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4420 (ins VR128:$src, VR128:$mask),
4421 "maskmovdqu\t{$mask, $src|$src, $mask}",
4422 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4424 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4425 (ins VR128:$src, VR128:$mask),
4426 "maskmovdqu\t{$mask, $src|$src, $mask}",
4427 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4430 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4431 "maskmovdqu\t{$mask, $src|$src, $mask}",
4432 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4434 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4435 "maskmovdqu\t{$mask, $src|$src, $mask}",
4436 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4438 } // ExeDomain = SSEPackedInt
4440 //===---------------------------------------------------------------------===//
4441 // SSE2 - Move Doubleword
4442 //===---------------------------------------------------------------------===//
4444 //===---------------------------------------------------------------------===//
4445 // Move Int Doubleword to Packed Double Int
4447 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4448 "movd\t{$src, $dst|$dst, $src}",
4450 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4451 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4452 "movd\t{$src, $dst|$dst, $src}",
4454 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4456 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4457 "mov{d|q}\t{$src, $dst|$dst, $src}",
4459 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4460 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4461 "mov{d|q}\t{$src, $dst|$dst, $src}",
4462 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4464 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4465 "movd\t{$src, $dst|$dst, $src}",
4467 (v4i32 (scalar_to_vector GR32:$src)))]>;
4468 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4469 "movd\t{$src, $dst|$dst, $src}",
4471 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4472 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4473 "mov{d|q}\t{$src, $dst|$dst, $src}",
4475 (v2i64 (scalar_to_vector GR64:$src)))]>;
4476 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4477 "mov{d|q}\t{$src, $dst|$dst, $src}",
4478 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4480 //===---------------------------------------------------------------------===//
4481 // Move Int Doubleword to Single Scalar
4483 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4484 "movd\t{$src, $dst|$dst, $src}",
4485 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4487 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4488 "movd\t{$src, $dst|$dst, $src}",
4489 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4491 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4492 "movd\t{$src, $dst|$dst, $src}",
4493 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4495 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4496 "movd\t{$src, $dst|$dst, $src}",
4497 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4499 //===---------------------------------------------------------------------===//
4500 // Move Packed Doubleword Int to Packed Double Int
4502 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4503 "movd\t{$src, $dst|$dst, $src}",
4504 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4506 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4507 (ins i32mem:$dst, VR128:$src),
4508 "movd\t{$src, $dst|$dst, $src}",
4509 [(store (i32 (vector_extract (v4i32 VR128:$src),
4510 (iPTR 0))), addr:$dst)]>, VEX;
4511 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4512 "movd\t{$src, $dst|$dst, $src}",
4513 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4515 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4516 "movd\t{$src, $dst|$dst, $src}",
4517 [(store (i32 (vector_extract (v4i32 VR128:$src),
4518 (iPTR 0))), addr:$dst)]>;
4520 //===---------------------------------------------------------------------===//
4521 // Move Packed Doubleword Int first element to Doubleword Int
4523 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4524 "mov{d|q}\t{$src, $dst|$dst, $src}",
4525 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4527 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4529 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4530 "mov{d|q}\t{$src, $dst|$dst, $src}",
4531 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4534 //===---------------------------------------------------------------------===//
4535 // Bitcast FR64 <-> GR64
4537 let Predicates = [HasAVX] in
4538 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4539 "vmovq\t{$src, $dst|$dst, $src}",
4540 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4542 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4543 "mov{d|q}\t{$src, $dst|$dst, $src}",
4544 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4545 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4546 "movq\t{$src, $dst|$dst, $src}",
4547 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4549 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4550 "movq\t{$src, $dst|$dst, $src}",
4551 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4552 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4553 "mov{d|q}\t{$src, $dst|$dst, $src}",
4554 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4555 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4556 "movq\t{$src, $dst|$dst, $src}",
4557 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4559 //===---------------------------------------------------------------------===//
4560 // Move Scalar Single to Double Int
4562 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4563 "movd\t{$src, $dst|$dst, $src}",
4564 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4565 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4566 "movd\t{$src, $dst|$dst, $src}",
4567 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4568 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4569 "movd\t{$src, $dst|$dst, $src}",
4570 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4571 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4572 "movd\t{$src, $dst|$dst, $src}",
4573 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4575 //===---------------------------------------------------------------------===//
4576 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4578 let AddedComplexity = 15 in {
4579 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4580 "movd\t{$src, $dst|$dst, $src}",
4581 [(set VR128:$dst, (v4i32 (X86vzmovl
4582 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4584 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4585 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4586 [(set VR128:$dst, (v2i64 (X86vzmovl
4587 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4590 let AddedComplexity = 15 in {
4591 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4592 "movd\t{$src, $dst|$dst, $src}",
4593 [(set VR128:$dst, (v4i32 (X86vzmovl
4594 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4595 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4596 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4597 [(set VR128:$dst, (v2i64 (X86vzmovl
4598 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4601 let AddedComplexity = 20 in {
4602 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4603 "movd\t{$src, $dst|$dst, $src}",
4605 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4606 (loadi32 addr:$src))))))]>,
4608 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4609 "movd\t{$src, $dst|$dst, $src}",
4611 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4612 (loadi32 addr:$src))))))]>;
4615 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4616 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4617 (MOVZDI2PDIrm addr:$src)>;
4618 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4619 (MOVZDI2PDIrm addr:$src)>;
4620 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4621 (MOVZDI2PDIrm addr:$src)>;
4624 let Predicates = [HasAVX] in {
4625 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4626 let AddedComplexity = 20 in {
4627 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4628 (VMOVZDI2PDIrm addr:$src)>;
4629 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4630 (VMOVZDI2PDIrm addr:$src)>;
4631 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4632 (VMOVZDI2PDIrm addr:$src)>;
4634 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4635 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4636 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4637 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4638 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4639 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4640 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4643 // These are the correct encodings of the instructions so that we know how to
4644 // read correct assembly, even though we continue to emit the wrong ones for
4645 // compatibility with Darwin's buggy assembler.
4646 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4647 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4648 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4649 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4650 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4651 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4652 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4653 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4654 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4655 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4656 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4657 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4659 //===---------------------------------------------------------------------===//
4660 // SSE2 - Move Quadword
4661 //===---------------------------------------------------------------------===//
4663 //===---------------------------------------------------------------------===//
4664 // Move Quadword Int to Packed Quadword Int
4666 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4667 "vmovq\t{$src, $dst|$dst, $src}",
4669 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4670 VEX, Requires<[HasAVX]>;
4671 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4672 "movq\t{$src, $dst|$dst, $src}",
4674 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4675 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4677 //===---------------------------------------------------------------------===//
4678 // Move Packed Quadword Int to Quadword Int
4680 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4681 "movq\t{$src, $dst|$dst, $src}",
4682 [(store (i64 (vector_extract (v2i64 VR128:$src),
4683 (iPTR 0))), addr:$dst)]>, VEX;
4684 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4685 "movq\t{$src, $dst|$dst, $src}",
4686 [(store (i64 (vector_extract (v2i64 VR128:$src),
4687 (iPTR 0))), addr:$dst)]>;
4689 //===---------------------------------------------------------------------===//
4690 // Store / copy lower 64-bits of a XMM register.
4692 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4693 "movq\t{$src, $dst|$dst, $src}",
4694 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4695 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4696 "movq\t{$src, $dst|$dst, $src}",
4697 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4699 let AddedComplexity = 20 in
4700 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4701 "vmovq\t{$src, $dst|$dst, $src}",
4703 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4704 (loadi64 addr:$src))))))]>,
4705 XS, VEX, Requires<[HasAVX]>;
4707 let AddedComplexity = 20 in
4708 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4709 "movq\t{$src, $dst|$dst, $src}",
4711 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4712 (loadi64 addr:$src))))))]>,
4713 XS, Requires<[HasSSE2]>;
4715 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4716 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4717 (MOVZQI2PQIrm addr:$src)>;
4718 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4719 (MOVZQI2PQIrm addr:$src)>;
4720 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4723 let Predicates = [HasAVX], AddedComplexity = 20 in {
4724 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4725 (VMOVZQI2PQIrm addr:$src)>;
4726 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4727 (VMOVZQI2PQIrm addr:$src)>;
4728 def : Pat<(v2i64 (X86vzload addr:$src)),
4729 (VMOVZQI2PQIrm addr:$src)>;
4732 //===---------------------------------------------------------------------===//
4733 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4734 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4736 let AddedComplexity = 15 in
4737 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4738 "vmovq\t{$src, $dst|$dst, $src}",
4739 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4740 XS, VEX, Requires<[HasAVX]>;
4741 let AddedComplexity = 15 in
4742 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4743 "movq\t{$src, $dst|$dst, $src}",
4744 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4745 XS, Requires<[HasSSE2]>;
4747 let AddedComplexity = 20 in
4748 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4749 "vmovq\t{$src, $dst|$dst, $src}",
4750 [(set VR128:$dst, (v2i64 (X86vzmovl
4751 (loadv2i64 addr:$src))))]>,
4752 XS, VEX, Requires<[HasAVX]>;
4753 let AddedComplexity = 20 in {
4754 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4755 "movq\t{$src, $dst|$dst, $src}",
4756 [(set VR128:$dst, (v2i64 (X86vzmovl
4757 (loadv2i64 addr:$src))))]>,
4758 XS, Requires<[HasSSE2]>;
4761 let AddedComplexity = 20 in {
4762 let Predicates = [HasSSE2] in {
4763 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4764 (MOVZPQILo2PQIrm addr:$src)>;
4765 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4766 (MOVZPQILo2PQIrr VR128:$src)>;
4768 let Predicates = [HasAVX] in {
4769 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4770 (VMOVZPQILo2PQIrm addr:$src)>;
4771 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4772 (VMOVZPQILo2PQIrr VR128:$src)>;
4776 // Instructions to match in the assembler
4777 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4778 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4779 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4780 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4781 // Recognize "movd" with GR64 destination, but encode as a "movq"
4782 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4783 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4785 // Instructions for the disassembler
4786 // xr = XMM register
4789 let Predicates = [HasAVX] in
4790 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4791 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4792 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4793 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4795 //===---------------------------------------------------------------------===//
4796 // SSE3 - Conversion Instructions
4797 //===---------------------------------------------------------------------===//
4799 // Convert Packed Double FP to Packed DW Integers
4800 let Predicates = [HasAVX] in {
4801 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4802 // register, but the same isn't true when using memory operands instead.
4803 // Provide other assembly rr and rm forms to address this explicitly.
4804 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4805 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4806 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4807 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4810 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4811 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4812 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4813 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4816 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4817 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4818 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4819 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4822 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4823 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4824 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4825 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4827 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4828 (VCVTPD2DQYrr VR256:$src)>;
4829 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4830 (VCVTPD2DQYrm addr:$src)>;
4832 // Convert Packed DW Integers to Packed Double FP
4833 let Predicates = [HasAVX] in {
4834 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4835 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4836 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4837 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4838 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4839 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4840 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4841 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4844 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4845 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4846 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4847 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4849 // AVX 256-bit register conversion intrinsics
4850 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4851 (VCVTDQ2PDYrr VR128:$src)>;
4852 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
4853 (VCVTDQ2PDYrm addr:$src)>;
4855 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4856 (VCVTPD2DQYrr VR256:$src)>;
4857 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4858 (VCVTPD2DQYrm addr:$src)>;
4860 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4861 (VCVTDQ2PDYrr VR128:$src)>;
4862 def : Pat<(v4f64 (sint_to_fp (memopv4i32 addr:$src))),
4863 (VCVTDQ2PDYrm addr:$src)>;
4865 //===---------------------------------------------------------------------===//
4866 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4867 //===---------------------------------------------------------------------===//
4868 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4869 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4870 X86MemOperand x86memop> {
4871 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4872 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4873 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4874 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4875 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4876 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4879 let Predicates = [HasAVX] in {
4880 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4881 v4f32, VR128, memopv4f32, f128mem>, VEX;
4882 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4883 v4f32, VR128, memopv4f32, f128mem>, VEX;
4884 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4885 v8f32, VR256, memopv8f32, f256mem>, VEX;
4886 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4887 v8f32, VR256, memopv8f32, f256mem>, VEX;
4889 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4890 memopv4f32, f128mem>;
4891 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4892 memopv4f32, f128mem>;
4894 let Predicates = [HasSSE3] in {
4895 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4896 (MOVSHDUPrr VR128:$src)>;
4897 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4898 (MOVSHDUPrm addr:$src)>;
4899 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4900 (MOVSLDUPrr VR128:$src)>;
4901 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4902 (MOVSLDUPrm addr:$src)>;
4905 let Predicates = [HasAVX] in {
4906 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4907 (VMOVSHDUPrr VR128:$src)>;
4908 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4909 (VMOVSHDUPrm addr:$src)>;
4910 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4911 (VMOVSLDUPrr VR128:$src)>;
4912 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4913 (VMOVSLDUPrm addr:$src)>;
4914 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4915 (VMOVSHDUPYrr VR256:$src)>;
4916 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4917 (VMOVSHDUPYrm addr:$src)>;
4918 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4919 (VMOVSLDUPYrr VR256:$src)>;
4920 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4921 (VMOVSLDUPYrm addr:$src)>;
4924 //===---------------------------------------------------------------------===//
4925 // SSE3 - Replicate Double FP - MOVDDUP
4926 //===---------------------------------------------------------------------===//
4928 multiclass sse3_replicate_dfp<string OpcodeStr> {
4929 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4930 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4931 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4932 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4933 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4935 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4939 // FIXME: Merge with above classe when there're patterns for the ymm version
4940 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4941 let Predicates = [HasAVX] in {
4942 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4943 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4945 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4946 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4951 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4952 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4953 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4955 let Predicates = [HasSSE3] in {
4956 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4958 (MOVDDUPrm addr:$src)>;
4959 let AddedComplexity = 5 in {
4960 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4961 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4962 (MOVDDUPrm addr:$src)>;
4963 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4964 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4965 (MOVDDUPrm addr:$src)>;
4967 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4968 (MOVDDUPrm addr:$src)>;
4969 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4970 (MOVDDUPrm addr:$src)>;
4971 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4972 (MOVDDUPrm addr:$src)>;
4973 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4974 (MOVDDUPrm addr:$src)>;
4975 def : Pat<(X86Movddup (bc_v2f64
4976 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4977 (MOVDDUPrm addr:$src)>;
4980 let Predicates = [HasAVX] in {
4981 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4983 (VMOVDDUPrm addr:$src)>;
4984 let AddedComplexity = 5 in {
4985 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4986 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4987 (VMOVDDUPrm addr:$src)>;
4988 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4989 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4990 (VMOVDDUPrm addr:$src)>;
4992 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4993 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4994 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4995 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4996 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4997 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4998 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4999 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5000 def : Pat<(X86Movddup (bc_v2f64
5001 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5002 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5005 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
5006 (VMOVDDUPYrm addr:$src)>;
5007 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
5008 (VMOVDDUPYrm addr:$src)>;
5009 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
5010 (VMOVDDUPYrm addr:$src)>;
5011 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
5012 (VMOVDDUPYrm addr:$src)>;
5013 def : Pat<(X86Movddup (v4f64 VR256:$src)),
5014 (VMOVDDUPYrr VR256:$src)>;
5015 def : Pat<(X86Movddup (v4i64 VR256:$src)),
5016 (VMOVDDUPYrr VR256:$src)>;
5019 //===---------------------------------------------------------------------===//
5020 // SSE3 - Move Unaligned Integer
5021 //===---------------------------------------------------------------------===//
5023 let Predicates = [HasAVX] in {
5024 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5025 "vlddqu\t{$src, $dst|$dst, $src}",
5026 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5027 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5028 "vlddqu\t{$src, $dst|$dst, $src}",
5029 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5031 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5032 "lddqu\t{$src, $dst|$dst, $src}",
5033 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
5035 //===---------------------------------------------------------------------===//
5036 // SSE3 - Arithmetic
5037 //===---------------------------------------------------------------------===//
5039 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5040 X86MemOperand x86memop, bit Is2Addr = 1> {
5041 def rr : I<0xD0, MRMSrcReg,
5042 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5044 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5045 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5046 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
5047 def rm : I<0xD0, MRMSrcMem,
5048 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5050 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5051 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5052 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
5055 let Predicates = [HasAVX],
5056 ExeDomain = SSEPackedDouble in {
5057 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5058 f128mem, 0>, TB, XD, VEX_4V;
5059 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5060 f128mem, 0>, TB, OpSize, VEX_4V;
5061 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5062 f256mem, 0>, TB, XD, VEX_4V;
5063 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5064 f256mem, 0>, TB, OpSize, VEX_4V;
5066 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
5067 ExeDomain = SSEPackedDouble in {
5068 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5070 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5071 f128mem>, TB, OpSize;
5074 //===---------------------------------------------------------------------===//
5075 // SSE3 Instructions
5076 //===---------------------------------------------------------------------===//
5079 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5080 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5081 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5083 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5084 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5085 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5087 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5089 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5090 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5091 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5093 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5094 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5095 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5097 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5098 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5099 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5101 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5103 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5104 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5105 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5108 let Predicates = [HasAVX] in {
5109 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5110 X86fhadd, 0>, VEX_4V;
5111 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5112 X86fhadd, 0>, VEX_4V;
5113 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5114 X86fhsub, 0>, VEX_4V;
5115 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5116 X86fhsub, 0>, VEX_4V;
5117 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5118 X86fhadd, 0>, VEX_4V;
5119 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5120 X86fhadd, 0>, VEX_4V;
5121 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5122 X86fhsub, 0>, VEX_4V;
5123 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5124 X86fhsub, 0>, VEX_4V;
5127 let Constraints = "$src1 = $dst" in {
5128 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5129 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5130 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5131 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5134 //===---------------------------------------------------------------------===//
5135 // SSSE3 - Packed Absolute Instructions
5136 //===---------------------------------------------------------------------===//
5139 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5140 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5141 PatFrag mem_frag128, Intrinsic IntId128> {
5142 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5144 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5145 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5148 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5150 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5153 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
5156 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5157 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5158 PatFrag mem_frag256, Intrinsic IntId256> {
5159 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5161 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5162 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5165 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5167 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5170 (bitconvert (mem_frag256 addr:$src))))]>, OpSize;
5173 let Predicates = [HasAVX] in {
5174 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
5175 int_x86_ssse3_pabs_b_128>, VEX;
5176 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
5177 int_x86_ssse3_pabs_w_128>, VEX;
5178 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
5179 int_x86_ssse3_pabs_d_128>, VEX;
5182 let Predicates = [HasAVX2] in {
5183 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb", memopv32i8,
5184 int_x86_avx2_pabs_b>, VEX;
5185 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw", memopv16i16,
5186 int_x86_avx2_pabs_w>, VEX;
5187 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd", memopv8i32,
5188 int_x86_avx2_pabs_d>, VEX;
5191 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
5192 int_x86_ssse3_pabs_b_128>;
5193 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
5194 int_x86_ssse3_pabs_w_128>;
5195 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
5196 int_x86_ssse3_pabs_d_128>;
5198 //===---------------------------------------------------------------------===//
5199 // SSSE3 - Packed Binary Operator Instructions
5200 //===---------------------------------------------------------------------===//
5202 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5203 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5204 PatFrag mem_frag128, Intrinsic IntId128,
5206 let isCommutable = 1 in
5207 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5208 (ins VR128:$src1, VR128:$src2),
5210 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5211 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5212 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5214 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5215 (ins VR128:$src1, i128mem:$src2),
5217 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5218 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5220 (IntId128 VR128:$src1,
5221 (bitconvert (mem_frag128 addr:$src2))))]>, OpSize;
5224 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5225 PatFrag mem_frag256, Intrinsic IntId256> {
5226 let isCommutable = 1 in
5227 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5228 (ins VR256:$src1, VR256:$src2),
5229 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5230 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5232 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5233 (ins VR256:$src1, i256mem:$src2),
5234 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5236 (IntId256 VR256:$src1,
5237 (bitconvert (mem_frag256 addr:$src2))))]>, OpSize;
5240 let ImmT = NoImm, Predicates = [HasAVX] in {
5241 let isCommutable = 0 in {
5242 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
5243 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5244 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
5245 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5246 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
5247 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5248 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
5249 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5250 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
5251 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5252 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
5253 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5254 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
5255 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5256 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
5257 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5258 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
5259 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5260 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
5261 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5262 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
5263 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5265 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
5266 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5269 let ImmT = NoImm, Predicates = [HasAVX2] in {
5270 let isCommutable = 0 in {
5271 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw", memopv16i16,
5272 int_x86_avx2_phadd_w>, VEX_4V;
5273 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd", memopv8i32,
5274 int_x86_avx2_phadd_d>, VEX_4V;
5275 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw", memopv16i16,
5276 int_x86_avx2_phadd_sw>, VEX_4V;
5277 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw", memopv16i16,
5278 int_x86_avx2_phsub_w>, VEX_4V;
5279 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd", memopv8i32,
5280 int_x86_avx2_phsub_d>, VEX_4V;
5281 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw", memopv16i16,
5282 int_x86_avx2_phsub_sw>, VEX_4V;
5283 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw", memopv32i8,
5284 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5285 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb", memopv32i8,
5286 int_x86_avx2_pshuf_b>, VEX_4V;
5287 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb", memopv16i8,
5288 int_x86_avx2_psign_b>, VEX_4V;
5289 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw", memopv8i16,
5290 int_x86_avx2_psign_w>, VEX_4V;
5291 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd", memopv4i32,
5292 int_x86_avx2_psign_d>, VEX_4V;
5294 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw", memopv16i16,
5295 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5298 // None of these have i8 immediate fields.
5299 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5300 let isCommutable = 0 in {
5301 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
5302 int_x86_ssse3_phadd_w_128>;
5303 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
5304 int_x86_ssse3_phadd_d_128>;
5305 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
5306 int_x86_ssse3_phadd_sw_128>;
5307 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
5308 int_x86_ssse3_phsub_w_128>;
5309 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
5310 int_x86_ssse3_phsub_d_128>;
5311 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
5312 int_x86_ssse3_phsub_sw_128>;
5313 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
5314 int_x86_ssse3_pmadd_ub_sw_128>;
5315 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
5316 int_x86_ssse3_pshuf_b_128>;
5317 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
5318 int_x86_ssse3_psign_b_128>;
5319 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
5320 int_x86_ssse3_psign_w_128>;
5321 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
5322 int_x86_ssse3_psign_d_128>;
5324 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
5325 int_x86_ssse3_pmul_hr_sw_128>;
5328 let Predicates = [HasSSSE3] in {
5329 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5330 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5331 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5332 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5334 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
5335 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5336 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
5337 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5338 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
5339 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5342 let Predicates = [HasAVX] in {
5343 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5344 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5345 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5346 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5348 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
5349 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5350 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
5351 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5352 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
5353 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5356 //===---------------------------------------------------------------------===//
5357 // SSSE3 - Packed Align Instruction Patterns
5358 //===---------------------------------------------------------------------===//
5360 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5361 let neverHasSideEffects = 1 in {
5362 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5363 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5365 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5367 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5370 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5371 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5373 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5375 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5380 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5381 let neverHasSideEffects = 1 in {
5382 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5383 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5385 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5388 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5389 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5391 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5396 let Predicates = [HasAVX] in
5397 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5398 let Predicates = [HasAVX2] in
5399 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5400 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5401 defm PALIGN : ssse3_palign<"palignr">;
5403 let Predicates = [HasSSSE3] in {
5404 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5405 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5406 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5407 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5408 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5409 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5410 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5411 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5414 let Predicates = [HasAVX] in {
5415 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5416 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5417 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5418 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5419 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5420 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5421 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5422 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5425 //===---------------------------------------------------------------------===//
5426 // SSSE3 - Thread synchronization
5427 //===---------------------------------------------------------------------===//
5429 let usesCustomInserter = 1 in {
5430 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5431 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
5432 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5433 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
5436 let Uses = [EAX, ECX, EDX] in
5437 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5438 Requires<[HasSSE3]>;
5439 let Uses = [ECX, EAX] in
5440 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5441 Requires<[HasSSE3]>;
5443 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5444 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5446 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5447 Requires<[In32BitMode]>;
5448 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5449 Requires<[In64BitMode]>;
5451 //===----------------------------------------------------------------------===//
5452 // SSE4.1 - Packed Move with Sign/Zero Extend
5453 //===----------------------------------------------------------------------===//
5455 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5456 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5457 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5458 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5460 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5461 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5463 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5467 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5469 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5470 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5471 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5473 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5474 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5475 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5478 let Predicates = [HasAVX] in {
5479 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5481 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5483 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5485 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5487 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5489 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5493 let Predicates = [HasAVX2] in {
5494 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5495 int_x86_avx2_pmovsxbw>, VEX;
5496 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5497 int_x86_avx2_pmovsxwd>, VEX;
5498 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5499 int_x86_avx2_pmovsxdq>, VEX;
5500 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5501 int_x86_avx2_pmovzxbw>, VEX;
5502 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5503 int_x86_avx2_pmovzxwd>, VEX;
5504 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5505 int_x86_avx2_pmovzxdq>, VEX;
5508 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5509 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5510 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5511 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5512 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5513 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5515 let Predicates = [HasSSE41] in {
5516 // Common patterns involving scalar load.
5517 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5518 (PMOVSXBWrm addr:$src)>;
5519 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5520 (PMOVSXBWrm addr:$src)>;
5522 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5523 (PMOVSXWDrm addr:$src)>;
5524 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5525 (PMOVSXWDrm addr:$src)>;
5527 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5528 (PMOVSXDQrm addr:$src)>;
5529 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5530 (PMOVSXDQrm addr:$src)>;
5532 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5533 (PMOVZXBWrm addr:$src)>;
5534 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5535 (PMOVZXBWrm addr:$src)>;
5537 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5538 (PMOVZXWDrm addr:$src)>;
5539 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5540 (PMOVZXWDrm addr:$src)>;
5542 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5543 (PMOVZXDQrm addr:$src)>;
5544 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5545 (PMOVZXDQrm addr:$src)>;
5548 let Predicates = [HasAVX] in {
5549 // Common patterns involving scalar load.
5550 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5551 (VPMOVSXBWrm addr:$src)>;
5552 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5553 (VPMOVSXBWrm addr:$src)>;
5555 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5556 (VPMOVSXWDrm addr:$src)>;
5557 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5558 (VPMOVSXWDrm addr:$src)>;
5560 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5561 (VPMOVSXDQrm addr:$src)>;
5562 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5563 (VPMOVSXDQrm addr:$src)>;
5565 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5566 (VPMOVZXBWrm addr:$src)>;
5567 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5568 (VPMOVZXBWrm addr:$src)>;
5570 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5571 (VPMOVZXWDrm addr:$src)>;
5572 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5573 (VPMOVZXWDrm addr:$src)>;
5575 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5576 (VPMOVZXDQrm addr:$src)>;
5577 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5578 (VPMOVZXDQrm addr:$src)>;
5582 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5583 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5584 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5585 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5587 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5588 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5590 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5594 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5596 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5597 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5598 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5600 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5601 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5603 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5607 let Predicates = [HasAVX] in {
5608 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5610 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5612 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5614 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5618 let Predicates = [HasAVX2] in {
5619 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5620 int_x86_avx2_pmovsxbd>, VEX;
5621 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5622 int_x86_avx2_pmovsxwq>, VEX;
5623 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5624 int_x86_avx2_pmovzxbd>, VEX;
5625 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5626 int_x86_avx2_pmovzxwq>, VEX;
5629 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5630 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5631 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5632 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5634 let Predicates = [HasSSE41] in {
5635 // Common patterns involving scalar load
5636 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5637 (PMOVSXBDrm addr:$src)>;
5638 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5639 (PMOVSXWQrm addr:$src)>;
5641 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5642 (PMOVZXBDrm addr:$src)>;
5643 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5644 (PMOVZXWQrm addr:$src)>;
5647 let Predicates = [HasAVX] in {
5648 // Common patterns involving scalar load
5649 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5650 (VPMOVSXBDrm addr:$src)>;
5651 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5652 (VPMOVSXWQrm addr:$src)>;
5654 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5655 (VPMOVZXBDrm addr:$src)>;
5656 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5657 (VPMOVZXWQrm addr:$src)>;
5660 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5661 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5662 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5663 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5665 // Expecting a i16 load any extended to i32 value.
5666 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5667 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5668 [(set VR128:$dst, (IntId (bitconvert
5669 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5673 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5675 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5676 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5677 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5679 // Expecting a i16 load any extended to i32 value.
5680 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5681 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5682 [(set VR256:$dst, (IntId (bitconvert
5683 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5687 let Predicates = [HasAVX] in {
5688 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5690 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5693 let Predicates = [HasAVX2] in {
5694 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5695 int_x86_avx2_pmovsxbq>, VEX;
5696 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5697 int_x86_avx2_pmovzxbq>, VEX;
5699 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5700 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5702 let Predicates = [HasSSE41] in {
5703 // Common patterns involving scalar load
5704 def : Pat<(int_x86_sse41_pmovsxbq
5705 (bitconvert (v4i32 (X86vzmovl
5706 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5707 (PMOVSXBQrm addr:$src)>;
5709 def : Pat<(int_x86_sse41_pmovzxbq
5710 (bitconvert (v4i32 (X86vzmovl
5711 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5712 (PMOVZXBQrm addr:$src)>;
5715 let Predicates = [HasAVX] in {
5716 // Common patterns involving scalar load
5717 def : Pat<(int_x86_sse41_pmovsxbq
5718 (bitconvert (v4i32 (X86vzmovl
5719 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5720 (VPMOVSXBQrm addr:$src)>;
5722 def : Pat<(int_x86_sse41_pmovzxbq
5723 (bitconvert (v4i32 (X86vzmovl
5724 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5725 (VPMOVZXBQrm addr:$src)>;
5728 //===----------------------------------------------------------------------===//
5729 // SSE4.1 - Extract Instructions
5730 //===----------------------------------------------------------------------===//
5732 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5733 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5734 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5735 (ins VR128:$src1, i32i8imm:$src2),
5736 !strconcat(OpcodeStr,
5737 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5738 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5740 let neverHasSideEffects = 1, mayStore = 1 in
5741 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5742 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5743 !strconcat(OpcodeStr,
5744 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5747 // There's an AssertZext in the way of writing the store pattern
5748 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5751 let Predicates = [HasAVX] in {
5752 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5753 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5754 (ins VR128:$src1, i32i8imm:$src2),
5755 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5758 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5761 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5762 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5763 let neverHasSideEffects = 1, mayStore = 1 in
5764 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5765 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5766 !strconcat(OpcodeStr,
5767 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5770 // There's an AssertZext in the way of writing the store pattern
5771 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5774 let Predicates = [HasAVX] in
5775 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5777 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5780 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5781 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5782 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5783 (ins VR128:$src1, i32i8imm:$src2),
5784 !strconcat(OpcodeStr,
5785 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5787 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5788 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5789 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5790 !strconcat(OpcodeStr,
5791 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5792 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5793 addr:$dst)]>, OpSize;
5796 let Predicates = [HasAVX] in
5797 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5799 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5801 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5802 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5803 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5804 (ins VR128:$src1, i32i8imm:$src2),
5805 !strconcat(OpcodeStr,
5806 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5808 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5809 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5810 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5811 !strconcat(OpcodeStr,
5812 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5813 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5814 addr:$dst)]>, OpSize, REX_W;
5817 let Predicates = [HasAVX] in
5818 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5820 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5822 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5824 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5825 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5826 (ins VR128:$src1, i32i8imm:$src2),
5827 !strconcat(OpcodeStr,
5828 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5830 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5832 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5833 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5834 !strconcat(OpcodeStr,
5835 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5836 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5837 addr:$dst)]>, OpSize;
5840 let Predicates = [HasAVX] in {
5841 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5842 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5843 (ins VR128:$src1, i32i8imm:$src2),
5844 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5847 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5849 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5850 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5853 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5854 Requires<[HasSSE41]>;
5855 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5858 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5861 //===----------------------------------------------------------------------===//
5862 // SSE4.1 - Insert Instructions
5863 //===----------------------------------------------------------------------===//
5865 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5866 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5867 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5869 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5871 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5873 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5874 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5875 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5877 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5879 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5881 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5882 imm:$src3))]>, OpSize;
5885 let Predicates = [HasAVX] in
5886 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5887 let Constraints = "$src1 = $dst" in
5888 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5890 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5891 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5892 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5894 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5896 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5898 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5900 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5901 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5903 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5905 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5907 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5908 imm:$src3)))]>, OpSize;
5911 let Predicates = [HasAVX] in
5912 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5913 let Constraints = "$src1 = $dst" in
5914 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5916 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5917 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5918 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5920 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5922 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5924 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5926 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5927 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5929 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5931 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5933 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5934 imm:$src3)))]>, OpSize;
5937 let Predicates = [HasAVX] in
5938 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5939 let Constraints = "$src1 = $dst" in
5940 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5942 // insertps has a few different modes, there's the first two here below which
5943 // are optimized inserts that won't zero arbitrary elements in the destination
5944 // vector. The next one matches the intrinsic and could zero arbitrary elements
5945 // in the target vector.
5946 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5947 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5948 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5950 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5952 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5954 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5956 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5957 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5959 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5961 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5963 (X86insrtps VR128:$src1,
5964 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5965 imm:$src3))]>, OpSize;
5968 let Constraints = "$src1 = $dst" in
5969 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5970 let Predicates = [HasAVX] in
5971 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5973 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5974 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5976 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5977 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5978 Requires<[HasSSE41]>;
5980 //===----------------------------------------------------------------------===//
5981 // SSE4.1 - Round Instructions
5982 //===----------------------------------------------------------------------===//
5984 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5985 X86MemOperand x86memop, RegisterClass RC,
5986 PatFrag mem_frag32, PatFrag mem_frag64,
5987 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5988 // Intrinsic operation, reg.
5989 // Vector intrinsic operation, reg
5990 def PSr : SS4AIi8<opcps, MRMSrcReg,
5991 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5992 !strconcat(OpcodeStr,
5993 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5994 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5997 // Vector intrinsic operation, mem
5998 def PSm : Ii8<opcps, MRMSrcMem,
5999 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6000 !strconcat(OpcodeStr,
6001 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6003 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6005 Requires<[HasSSE41]>;
6007 // Vector intrinsic operation, reg
6008 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6009 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6010 !strconcat(OpcodeStr,
6011 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6012 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6015 // Vector intrinsic operation, mem
6016 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6017 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6018 !strconcat(OpcodeStr,
6019 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6021 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6025 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
6026 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
6027 // Intrinsic operation, reg.
6028 // Vector intrinsic operation, reg
6029 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
6030 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6031 !strconcat(OpcodeStr,
6032 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6035 // Vector intrinsic operation, mem
6036 def PSm_AVX : Ii8<opcps, MRMSrcMem,
6037 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6038 !strconcat(OpcodeStr,
6039 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6040 []>, TA, OpSize, Requires<[HasSSE41]>;
6042 // Vector intrinsic operation, reg
6043 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
6044 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6045 !strconcat(OpcodeStr,
6046 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6049 // Vector intrinsic operation, mem
6050 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
6051 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6052 !strconcat(OpcodeStr,
6053 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6057 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6060 Intrinsic F64Int, bit Is2Addr = 1> {
6061 // Intrinsic operation, reg.
6062 def SSr : SS4AIi8<opcss, MRMSrcReg,
6063 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6065 !strconcat(OpcodeStr,
6066 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6067 !strconcat(OpcodeStr,
6068 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6069 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6072 // Intrinsic operation, mem.
6073 def SSm : SS4AIi8<opcss, MRMSrcMem,
6074 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6076 !strconcat(OpcodeStr,
6077 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6078 !strconcat(OpcodeStr,
6079 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6081 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6084 // Intrinsic operation, reg.
6085 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6086 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6088 !strconcat(OpcodeStr,
6089 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6090 !strconcat(OpcodeStr,
6091 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6092 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6095 // Intrinsic operation, mem.
6096 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6097 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6099 !strconcat(OpcodeStr,
6100 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6101 !strconcat(OpcodeStr,
6102 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6104 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6108 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
6110 // Intrinsic operation, reg.
6111 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
6112 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6113 !strconcat(OpcodeStr,
6114 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6117 // Intrinsic operation, mem.
6118 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
6119 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6120 !strconcat(OpcodeStr,
6121 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6124 // Intrinsic operation, reg.
6125 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
6126 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6127 !strconcat(OpcodeStr,
6128 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6131 // Intrinsic operation, mem.
6132 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
6133 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6134 !strconcat(OpcodeStr,
6135 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6139 // FP round - roundss, roundps, roundsd, roundpd
6140 let Predicates = [HasAVX] in {
6142 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6143 memopv4f32, memopv2f64,
6144 int_x86_sse41_round_ps,
6145 int_x86_sse41_round_pd>, VEX;
6146 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6147 memopv8f32, memopv4f64,
6148 int_x86_avx_round_ps_256,
6149 int_x86_avx_round_pd_256>, VEX;
6150 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6151 int_x86_sse41_round_ss,
6152 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6154 // Instructions for the assembler
6155 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
6157 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
6159 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V, VEX_LIG;
6162 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6163 memopv4f32, memopv2f64,
6164 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6165 let Constraints = "$src1 = $dst" in
6166 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6167 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6169 //===----------------------------------------------------------------------===//
6170 // SSE4.1 - Packed Bit Test
6171 //===----------------------------------------------------------------------===//
6173 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6174 // the intel intrinsic that corresponds to this.
6175 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6176 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6177 "vptest\t{$src2, $src1|$src1, $src2}",
6178 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6180 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6181 "vptest\t{$src2, $src1|$src1, $src2}",
6182 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6185 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6186 "vptest\t{$src2, $src1|$src1, $src2}",
6187 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6189 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6190 "vptest\t{$src2, $src1|$src1, $src2}",
6191 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6195 let Defs = [EFLAGS] in {
6196 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6197 "ptest \t{$src2, $src1|$src1, $src2}",
6198 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6200 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6201 "ptest \t{$src2, $src1|$src1, $src2}",
6202 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6206 // The bit test instructions below are AVX only
6207 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6208 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6209 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6210 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6211 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6212 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6213 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6214 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6218 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6219 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6220 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6221 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6222 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6225 //===----------------------------------------------------------------------===//
6226 // SSE4.1 - Misc Instructions
6227 //===----------------------------------------------------------------------===//
6229 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6230 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6231 "popcnt{w}\t{$src, $dst|$dst, $src}",
6232 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6234 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6235 "popcnt{w}\t{$src, $dst|$dst, $src}",
6236 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6237 (implicit EFLAGS)]>, OpSize, XS;
6239 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6240 "popcnt{l}\t{$src, $dst|$dst, $src}",
6241 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6243 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6244 "popcnt{l}\t{$src, $dst|$dst, $src}",
6245 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6246 (implicit EFLAGS)]>, XS;
6248 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6249 "popcnt{q}\t{$src, $dst|$dst, $src}",
6250 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6252 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6253 "popcnt{q}\t{$src, $dst|$dst, $src}",
6254 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6255 (implicit EFLAGS)]>, XS;
6260 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6261 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6262 Intrinsic IntId128> {
6263 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6265 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6266 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6267 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6269 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6272 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
6275 let Predicates = [HasAVX] in
6276 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6277 int_x86_sse41_phminposuw>, VEX;
6278 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6279 int_x86_sse41_phminposuw>;
6281 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6282 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6283 Intrinsic IntId128, bit Is2Addr = 1> {
6284 let isCommutable = 1 in
6285 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6286 (ins VR128:$src1, VR128:$src2),
6288 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6289 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6290 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6291 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6292 (ins VR128:$src1, i128mem:$src2),
6294 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6295 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6297 (IntId128 VR128:$src1,
6298 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
6301 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6302 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6303 Intrinsic IntId256> {
6304 let isCommutable = 1 in
6305 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6306 (ins VR256:$src1, VR256:$src2),
6307 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6308 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6309 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6310 (ins VR256:$src1, i256mem:$src2),
6311 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6313 (IntId256 VR256:$src1,
6314 (bitconvert (memopv32i8 addr:$src2))))]>, OpSize;
6317 let Predicates = [HasAVX] in {
6318 let isCommutable = 0 in
6319 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6321 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6323 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6325 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6327 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6329 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6331 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6333 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6335 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6337 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6339 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6342 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6343 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6344 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6345 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6348 let Predicates = [HasAVX2] in {
6349 let isCommutable = 0 in
6350 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6351 int_x86_avx2_packusdw>, VEX_4V;
6352 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6353 int_x86_avx2_pcmpeq_q>, VEX_4V;
6354 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6355 int_x86_avx2_pmins_b>, VEX_4V;
6356 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6357 int_x86_avx2_pmins_d>, VEX_4V;
6358 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6359 int_x86_avx2_pminu_d>, VEX_4V;
6360 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6361 int_x86_avx2_pminu_w>, VEX_4V;
6362 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6363 int_x86_avx2_pmaxs_b>, VEX_4V;
6364 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6365 int_x86_avx2_pmaxs_d>, VEX_4V;
6366 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6367 int_x86_avx2_pmaxu_d>, VEX_4V;
6368 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6369 int_x86_avx2_pmaxu_w>, VEX_4V;
6370 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6371 int_x86_avx2_pmul_dq>, VEX_4V;
6373 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, VR256:$src2)),
6374 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6375 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, (memop addr:$src2))),
6376 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6379 let Constraints = "$src1 = $dst" in {
6380 let isCommutable = 0 in
6381 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6382 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6383 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6384 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6385 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6386 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6387 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6388 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6389 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6390 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6391 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6394 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6395 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6396 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6397 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6399 /// SS48I_binop_rm - Simple SSE41 binary operator.
6400 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6401 ValueType OpVT, bit Is2Addr = 1> {
6402 let isCommutable = 1 in
6403 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6404 (ins VR128:$src1, VR128:$src2),
6406 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6407 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6408 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6410 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6411 (ins VR128:$src1, i128mem:$src2),
6413 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6414 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6415 [(set VR128:$dst, (OpNode VR128:$src1,
6416 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6420 /// SS48I_binop_rm - Simple SSE41 binary operator.
6421 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6423 let isCommutable = 1 in
6424 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6425 (ins VR256:$src1, VR256:$src2),
6426 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6427 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6429 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6430 (ins VR256:$src1, i256mem:$src2),
6431 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6432 [(set VR256:$dst, (OpNode VR256:$src1,
6433 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6437 let Predicates = [HasAVX] in
6438 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6439 let Predicates = [HasAVX2] in
6440 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6441 let Constraints = "$src1 = $dst" in
6442 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6444 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6445 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6446 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6447 X86MemOperand x86memop, bit Is2Addr = 1> {
6448 let isCommutable = 1 in
6449 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6450 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6452 !strconcat(OpcodeStr,
6453 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6454 !strconcat(OpcodeStr,
6455 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6456 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6458 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6459 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6461 !strconcat(OpcodeStr,
6462 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6463 !strconcat(OpcodeStr,
6464 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6467 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6471 let Predicates = [HasAVX] in {
6472 let isCommutable = 0 in {
6473 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6474 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6475 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6476 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6477 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6478 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
6479 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6480 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
6481 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6482 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6483 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6484 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6486 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6487 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6488 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6489 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6490 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6491 VR256, memopv32i8, i256mem, 0>, VEX_4V;
6494 let Predicates = [HasAVX2] in {
6495 let isCommutable = 0 in {
6496 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6497 VR256, memopv32i8, i256mem, 0>, VEX_4V;
6498 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6499 VR256, memopv32i8, i256mem, 0>, VEX_4V;
6503 let Constraints = "$src1 = $dst" in {
6504 let isCommutable = 0 in {
6505 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6506 VR128, memopv16i8, i128mem>;
6507 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6508 VR128, memopv16i8, i128mem>;
6509 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6510 VR128, memopv16i8, i128mem>;
6511 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6512 VR128, memopv16i8, i128mem>;
6514 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6515 VR128, memopv16i8, i128mem>;
6516 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6517 VR128, memopv16i8, i128mem>;
6520 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6521 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6522 RegisterClass RC, X86MemOperand x86memop,
6523 PatFrag mem_frag, Intrinsic IntId> {
6524 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
6525 (ins RC:$src1, RC:$src2, RC:$src3),
6526 !strconcat(OpcodeStr,
6527 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6528 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6529 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6531 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
6532 (ins RC:$src1, x86memop:$src2, RC:$src3),
6533 !strconcat(OpcodeStr,
6534 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6536 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6538 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6541 let Predicates = [HasAVX] in {
6542 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6543 memopv16i8, int_x86_sse41_blendvpd>;
6544 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6545 memopv16i8, int_x86_sse41_blendvps>;
6546 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6547 memopv16i8, int_x86_sse41_pblendvb>;
6548 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6549 memopv32i8, int_x86_avx_blendv_pd_256>;
6550 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6551 memopv32i8, int_x86_avx_blendv_ps_256>;
6554 let Predicates = [HasAVX2] in {
6555 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6556 memopv32i8, int_x86_avx2_pblendvb>;
6559 let Predicates = [HasAVX] in {
6560 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6561 (v16i8 VR128:$src2))),
6562 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6563 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6564 (v4i32 VR128:$src2))),
6565 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6566 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6567 (v4f32 VR128:$src2))),
6568 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6569 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6570 (v2i64 VR128:$src2))),
6571 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6572 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6573 (v2f64 VR128:$src2))),
6574 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6575 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6576 (v8i32 VR256:$src2))),
6577 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6578 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6579 (v8f32 VR256:$src2))),
6580 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6581 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6582 (v4i64 VR256:$src2))),
6583 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6584 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6585 (v4f64 VR256:$src2))),
6586 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6589 let Predicates = [HasAVX2] in {
6590 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6591 (v32i8 VR256:$src2))),
6592 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6595 /// SS41I_ternary_int - SSE 4.1 ternary operator
6596 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6597 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
6598 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6599 (ins VR128:$src1, VR128:$src2),
6600 !strconcat(OpcodeStr,
6601 "\t{$src2, $dst|$dst, $src2}"),
6602 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6605 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6606 (ins VR128:$src1, i128mem:$src2),
6607 !strconcat(OpcodeStr,
6608 "\t{$src2, $dst|$dst, $src2}"),
6611 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
6615 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
6616 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
6617 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
6619 let Predicates = [HasSSE41] in {
6620 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6621 (v16i8 VR128:$src2))),
6622 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6623 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6624 (v4i32 VR128:$src2))),
6625 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6626 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6627 (v4f32 VR128:$src2))),
6628 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6629 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6630 (v2i64 VR128:$src2))),
6631 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6632 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6633 (v2f64 VR128:$src2))),
6634 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6637 let Predicates = [HasAVX] in
6638 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6639 "vmovntdqa\t{$src, $dst|$dst, $src}",
6640 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6642 let Predicates = [HasAVX2] in
6643 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6644 "vmovntdqa\t{$src, $dst|$dst, $src}",
6645 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6647 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6648 "movntdqa\t{$src, $dst|$dst, $src}",
6649 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6652 //===----------------------------------------------------------------------===//
6653 // SSE4.2 - Compare Instructions
6654 //===----------------------------------------------------------------------===//
6656 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6657 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6658 Intrinsic IntId128, bit Is2Addr = 1> {
6659 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6660 (ins VR128:$src1, VR128:$src2),
6662 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6663 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6664 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6666 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6667 (ins VR128:$src1, i128mem:$src2),
6669 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6670 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6672 (IntId128 VR128:$src1,
6673 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
6676 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6677 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6678 Intrinsic IntId256> {
6679 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6680 (ins VR256:$src1, VR256:$src2),
6681 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6682 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6684 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6685 (ins VR256:$src1, i256mem:$src2),
6686 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6688 (IntId256 VR256:$src1,
6689 (bitconvert (memopv32i8 addr:$src2))))]>, OpSize;
6692 let Predicates = [HasAVX] in {
6693 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6696 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6697 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6698 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6699 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6702 let Predicates = [HasAVX2] in {
6703 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6706 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, VR256:$src2)),
6707 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6708 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, (memop addr:$src2))),
6709 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6712 let Constraints = "$src1 = $dst" in
6713 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6715 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6716 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6717 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6718 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6720 //===----------------------------------------------------------------------===//
6721 // SSE4.2 - String/text Processing Instructions
6722 //===----------------------------------------------------------------------===//
6724 // Packed Compare Implicit Length Strings, Return Mask
6725 multiclass pseudo_pcmpistrm<string asm> {
6726 def REG : PseudoI<(outs VR128:$dst),
6727 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6728 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6730 def MEM : PseudoI<(outs VR128:$dst),
6731 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6732 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6733 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6736 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6737 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6738 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6741 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6742 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6743 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6744 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6746 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6747 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6748 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6751 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6752 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6753 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6754 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6756 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6757 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6758 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6761 // Packed Compare Explicit Length Strings, Return Mask
6762 multiclass pseudo_pcmpestrm<string asm> {
6763 def REG : PseudoI<(outs VR128:$dst),
6764 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6765 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6766 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6767 def MEM : PseudoI<(outs VR128:$dst),
6768 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6769 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6770 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6773 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6774 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6775 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6778 let Predicates = [HasAVX],
6779 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6780 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6781 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6782 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6784 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6785 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6786 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6789 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6790 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6791 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6792 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6794 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6795 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6796 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6799 // Packed Compare Implicit Length Strings, Return Index
6800 let Defs = [ECX, EFLAGS] in {
6801 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6802 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6803 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6804 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6805 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6806 (implicit EFLAGS)]>, OpSize;
6807 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6808 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6809 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6810 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6811 (implicit EFLAGS)]>, OpSize;
6815 let Predicates = [HasAVX] in {
6816 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6818 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6820 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6822 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6824 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6826 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6830 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6831 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6832 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6833 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6834 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6835 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6837 // Packed Compare Explicit Length Strings, Return Index
6838 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6839 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6840 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6841 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6842 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6843 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6844 (implicit EFLAGS)]>, OpSize;
6845 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6846 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6847 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6849 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6850 (implicit EFLAGS)]>, OpSize;
6854 let Predicates = [HasAVX] in {
6855 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6857 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6859 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6861 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6863 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6865 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6869 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6870 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6871 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6872 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6873 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6874 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6876 //===----------------------------------------------------------------------===//
6877 // SSE4.2 - CRC Instructions
6878 //===----------------------------------------------------------------------===//
6880 // No CRC instructions have AVX equivalents
6882 // crc intrinsic instruction
6883 // This set of instructions are only rm, the only difference is the size
6885 let Constraints = "$src1 = $dst" in {
6886 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6887 (ins GR32:$src1, i8mem:$src2),
6888 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6890 (int_x86_sse42_crc32_32_8 GR32:$src1,
6891 (load addr:$src2)))]>;
6892 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6893 (ins GR32:$src1, GR8:$src2),
6894 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6896 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6897 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6898 (ins GR32:$src1, i16mem:$src2),
6899 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6901 (int_x86_sse42_crc32_32_16 GR32:$src1,
6902 (load addr:$src2)))]>,
6904 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6905 (ins GR32:$src1, GR16:$src2),
6906 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6908 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6910 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6911 (ins GR32:$src1, i32mem:$src2),
6912 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6914 (int_x86_sse42_crc32_32_32 GR32:$src1,
6915 (load addr:$src2)))]>;
6916 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6917 (ins GR32:$src1, GR32:$src2),
6918 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6920 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6921 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6922 (ins GR64:$src1, i8mem:$src2),
6923 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6925 (int_x86_sse42_crc32_64_8 GR64:$src1,
6926 (load addr:$src2)))]>,
6928 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6929 (ins GR64:$src1, GR8:$src2),
6930 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6932 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6934 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6935 (ins GR64:$src1, i64mem:$src2),
6936 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6938 (int_x86_sse42_crc32_64_64 GR64:$src1,
6939 (load addr:$src2)))]>,
6941 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6942 (ins GR64:$src1, GR64:$src2),
6943 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6945 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6949 //===----------------------------------------------------------------------===//
6950 // AES-NI Instructions
6951 //===----------------------------------------------------------------------===//
6953 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6954 Intrinsic IntId128, bit Is2Addr = 1> {
6955 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6956 (ins VR128:$src1, VR128:$src2),
6958 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6959 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6960 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6962 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6963 (ins VR128:$src1, i128mem:$src2),
6965 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6966 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6968 (IntId128 VR128:$src1,
6969 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
6972 // Perform One Round of an AES Encryption/Decryption Flow
6973 let Predicates = [HasAVX, HasAES] in {
6974 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
6975 int_x86_aesni_aesenc, 0>, VEX_4V;
6976 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
6977 int_x86_aesni_aesenclast, 0>, VEX_4V;
6978 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
6979 int_x86_aesni_aesdec, 0>, VEX_4V;
6980 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
6981 int_x86_aesni_aesdeclast, 0>, VEX_4V;
6984 let Constraints = "$src1 = $dst" in {
6985 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
6986 int_x86_aesni_aesenc>;
6987 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
6988 int_x86_aesni_aesenclast>;
6989 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
6990 int_x86_aesni_aesdec>;
6991 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
6992 int_x86_aesni_aesdeclast>;
6995 let Predicates = [HasAES] in {
6996 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
6997 (AESENCrr VR128:$src1, VR128:$src2)>;
6998 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
6999 (AESENCrm VR128:$src1, addr:$src2)>;
7000 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
7001 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
7002 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
7003 (AESENCLASTrm VR128:$src1, addr:$src2)>;
7004 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
7005 (AESDECrr VR128:$src1, VR128:$src2)>;
7006 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
7007 (AESDECrm VR128:$src1, addr:$src2)>;
7008 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
7009 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
7010 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
7011 (AESDECLASTrm VR128:$src1, addr:$src2)>;
7014 let Predicates = [HasAVX, HasAES], AddedComplexity = 20 in {
7015 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
7016 (VAESENCrr VR128:$src1, VR128:$src2)>;
7017 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
7018 (VAESENCrm VR128:$src1, addr:$src2)>;
7019 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
7020 (VAESENCLASTrr VR128:$src1, VR128:$src2)>;
7021 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
7022 (VAESENCLASTrm VR128:$src1, addr:$src2)>;
7023 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
7024 (VAESDECrr VR128:$src1, VR128:$src2)>;
7025 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
7026 (VAESDECrm VR128:$src1, addr:$src2)>;
7027 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
7028 (VAESDECLASTrr VR128:$src1, VR128:$src2)>;
7029 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
7030 (VAESDECLASTrm VR128:$src1, addr:$src2)>;
7033 // Perform the AES InvMixColumn Transformation
7034 let Predicates = [HasAVX, HasAES] in {
7035 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7037 "vaesimc\t{$src1, $dst|$dst, $src1}",
7039 (int_x86_aesni_aesimc VR128:$src1))]>,
7041 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7042 (ins i128mem:$src1),
7043 "vaesimc\t{$src1, $dst|$dst, $src1}",
7045 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
7048 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7050 "aesimc\t{$src1, $dst|$dst, $src1}",
7052 (int_x86_aesni_aesimc VR128:$src1))]>,
7054 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7055 (ins i128mem:$src1),
7056 "aesimc\t{$src1, $dst|$dst, $src1}",
7058 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
7061 // AES Round Key Generation Assist
7062 let Predicates = [HasAVX, HasAES] in {
7063 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7064 (ins VR128:$src1, i8imm:$src2),
7065 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7067 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7069 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7070 (ins i128mem:$src1, i8imm:$src2),
7071 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7073 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
7077 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7078 (ins VR128:$src1, i8imm:$src2),
7079 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7081 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7083 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7084 (ins i128mem:$src1, i8imm:$src2),
7085 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7087 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
7091 //===----------------------------------------------------------------------===//
7092 // CLMUL Instructions
7093 //===----------------------------------------------------------------------===//
7095 // Carry-less Multiplication instructions
7096 let neverHasSideEffects = 1 in {
7097 let Constraints = "$src1 = $dst" in {
7098 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7099 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7100 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7104 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7105 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7106 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7110 // AVX carry-less Multiplication instructions
7111 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7112 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7113 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7117 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7118 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7119 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7124 multiclass pclmul_alias<string asm, int immop> {
7125 def : InstAlias<!strconcat("pclmul", asm,
7126 "dq {$src, $dst|$dst, $src}"),
7127 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7129 def : InstAlias<!strconcat("pclmul", asm,
7130 "dq {$src, $dst|$dst, $src}"),
7131 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7133 def : InstAlias<!strconcat("vpclmul", asm,
7134 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7135 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7137 def : InstAlias<!strconcat("vpclmul", asm,
7138 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7139 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7141 defm : pclmul_alias<"hqhq", 0x11>;
7142 defm : pclmul_alias<"hqlq", 0x01>;
7143 defm : pclmul_alias<"lqhq", 0x10>;
7144 defm : pclmul_alias<"lqlq", 0x00>;
7146 //===----------------------------------------------------------------------===//
7148 //===----------------------------------------------------------------------===//
7150 //===----------------------------------------------------------------------===//
7151 // VBROADCAST - Load from memory and broadcast to all elements of the
7152 // destination operand
7154 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7155 X86MemOperand x86memop, Intrinsic Int> :
7156 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7157 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7158 [(set RC:$dst, (Int addr:$src))]>, VEX;
7160 // AVX2 adds register forms
7161 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7163 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7164 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7165 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7167 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7168 int_x86_avx_vbroadcast_ss>;
7169 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7170 int_x86_avx_vbroadcast_ss_256>;
7171 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7172 int_x86_avx_vbroadcast_sd_256>;
7173 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7174 int_x86_avx_vbroadcastf128_pd_256>;
7176 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7177 int_x86_avx2_vbroadcast_ss_ps>;
7178 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7179 int_x86_avx2_vbroadcast_ss_ps_256>;
7180 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7181 int_x86_avx2_vbroadcast_sd_pd_256>;
7183 let Predicates = [HasAVX2] in
7184 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7185 int_x86_avx2_vbroadcasti128>;
7187 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7188 (VBROADCASTF128 addr:$src)>;
7190 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7191 (VBROADCASTSSYrm addr:$src)>;
7192 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7193 (VBROADCASTSDrm addr:$src)>;
7194 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7195 (VBROADCASTSSYrm addr:$src)>;
7196 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7197 (VBROADCASTSDrm addr:$src)>;
7199 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7200 (VBROADCASTSSrm addr:$src)>;
7201 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7202 (VBROADCASTSSrm addr:$src)>;
7204 //===----------------------------------------------------------------------===//
7205 // VINSERTF128 - Insert packed floating-point values
7207 let neverHasSideEffects = 1 in {
7208 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7209 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7210 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7213 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7214 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7215 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7219 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7220 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7221 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7222 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7223 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7224 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7226 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7228 (VINSERTF128rr VR256:$src1, VR128:$src2,
7229 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7230 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7232 (VINSERTF128rr VR256:$src1, VR128:$src2,
7233 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7234 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7236 (VINSERTF128rr VR256:$src1, VR128:$src2,
7237 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7238 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7240 (VINSERTF128rr VR256:$src1, VR128:$src2,
7241 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7242 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7244 (VINSERTF128rr VR256:$src1, VR128:$src2,
7245 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7246 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7248 (VINSERTF128rr VR256:$src1, VR128:$src2,
7249 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7251 //===----------------------------------------------------------------------===//
7252 // VEXTRACTF128 - Extract packed floating-point values
7254 let neverHasSideEffects = 1 in {
7255 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7256 (ins VR256:$src1, i8imm:$src2),
7257 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7260 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7261 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7262 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7266 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7267 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7268 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7269 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7270 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7271 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7273 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7274 (v4f32 (VEXTRACTF128rr
7275 (v8f32 VR256:$src1),
7276 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7277 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7278 (v2f64 (VEXTRACTF128rr
7279 (v4f64 VR256:$src1),
7280 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7281 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7282 (v4i32 (VEXTRACTF128rr
7283 (v8i32 VR256:$src1),
7284 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7285 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7286 (v2i64 (VEXTRACTF128rr
7287 (v4i64 VR256:$src1),
7288 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7289 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7290 (v8i16 (VEXTRACTF128rr
7291 (v16i16 VR256:$src1),
7292 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7293 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7294 (v16i8 (VEXTRACTF128rr
7295 (v32i8 VR256:$src1),
7296 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7298 //===----------------------------------------------------------------------===//
7299 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7301 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7302 Intrinsic IntLd, Intrinsic IntLd256,
7303 Intrinsic IntSt, Intrinsic IntSt256,
7304 PatFrag pf128, PatFrag pf256> {
7305 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7306 (ins VR128:$src1, f128mem:$src2),
7307 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7308 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7310 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7311 (ins VR256:$src1, f256mem:$src2),
7312 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7313 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7315 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7316 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7317 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7318 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7319 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7320 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7321 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7322 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7325 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7326 int_x86_avx_maskload_ps,
7327 int_x86_avx_maskload_ps_256,
7328 int_x86_avx_maskstore_ps,
7329 int_x86_avx_maskstore_ps_256,
7330 memopv4f32, memopv8f32>;
7331 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7332 int_x86_avx_maskload_pd,
7333 int_x86_avx_maskload_pd_256,
7334 int_x86_avx_maskstore_pd,
7335 int_x86_avx_maskstore_pd_256,
7336 memopv2f64, memopv4f64>;
7338 //===----------------------------------------------------------------------===//
7339 // VPERMIL - Permute Single and Double Floating-Point Values
7341 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7342 RegisterClass RC, X86MemOperand x86memop_f,
7343 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7344 Intrinsic IntVar, Intrinsic IntImm> {
7345 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7346 (ins RC:$src1, RC:$src2),
7347 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7348 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7349 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7350 (ins RC:$src1, x86memop_i:$src2),
7351 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7352 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
7354 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7355 (ins RC:$src1, i8imm:$src2),
7356 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7357 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7358 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7359 (ins x86memop_f:$src1, i8imm:$src2),
7360 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7361 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7364 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7365 memopv4f32, memopv4i32,
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, memopv8i32,
7370 int_x86_avx_vpermilvar_ps_256,
7371 int_x86_avx_vpermil_ps_256>;
7372 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7373 memopv2f64, memopv2i64,
7374 int_x86_avx_vpermilvar_pd,
7375 int_x86_avx_vpermil_pd>;
7376 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7377 memopv4f64, memopv4i64,
7378 int_x86_avx_vpermilvar_pd_256,
7379 int_x86_avx_vpermil_pd_256>;
7381 def : Pat<(v8f32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
7382 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7383 def : Pat<(v4f64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
7384 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7385 def : Pat<(v8i32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
7386 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7387 def : Pat<(v4i64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
7388 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7390 //===----------------------------------------------------------------------===//
7391 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7393 let neverHasSideEffects = 1 in {
7394 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7395 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7396 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7399 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7400 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7401 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7405 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7406 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7407 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7408 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7409 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7410 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7412 def : Pat<(int_x86_avx_vperm2f128_ps_256
7413 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7414 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7415 def : Pat<(int_x86_avx_vperm2f128_pd_256
7416 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7417 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7418 def : Pat<(int_x86_avx_vperm2f128_si_256
7419 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
7420 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7422 def : Pat<(v8f32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7423 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7424 def : Pat<(v8i32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7425 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7426 def : Pat<(v4i64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7427 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7428 def : Pat<(v4f64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7429 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7430 def : Pat<(v32i8 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7431 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7432 def : Pat<(v16i16 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7433 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7435 //===----------------------------------------------------------------------===//
7436 // VZERO - Zero YMM registers
7438 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7439 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7440 // Zero All YMM registers
7441 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7442 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7444 // Zero Upper bits of YMM registers
7445 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7446 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7449 //===----------------------------------------------------------------------===//
7450 // Half precision conversion instructions
7451 //===----------------------------------------------------------------------===//
7452 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7453 let Predicates = [HasAVX, HasF16C] in {
7454 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7455 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7456 [(set RC:$dst, (Int VR128:$src))]>,
7458 let neverHasSideEffects = 1, mayLoad = 1 in
7459 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7460 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7464 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7465 let Predicates = [HasAVX, HasF16C] in {
7466 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7467 (ins RC:$src1, i32i8imm:$src2),
7468 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7469 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7471 let neverHasSideEffects = 1, mayLoad = 1 in
7472 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7473 (ins RC:$src1, i32i8imm:$src2),
7474 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7479 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7480 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7481 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7482 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7484 //===----------------------------------------------------------------------===//
7485 // AVX2 Instructions
7486 //===----------------------------------------------------------------------===//
7488 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7489 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7490 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7491 X86MemOperand x86memop> {
7492 let isCommutable = 1 in
7493 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7494 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7495 !strconcat(OpcodeStr,
7496 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7497 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7499 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7500 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7501 !strconcat(OpcodeStr,
7502 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7505 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7509 let isCommutable = 0 in {
7510 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7511 VR128, memopv16i8, i128mem>;
7512 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7513 VR256, memopv32i8, i256mem>;
7516 //===----------------------------------------------------------------------===//
7517 // VPBROADCAST - Load from memory and broadcast to all elements of the
7518 // destination operand
7520 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7521 X86MemOperand x86memop, PatFrag ld_frag,
7522 Intrinsic Int128, Intrinsic Int256> {
7523 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7524 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7525 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7526 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7527 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7529 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7530 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7531 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7532 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7533 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7534 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7536 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7539 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7540 int_x86_avx2_pbroadcastb_128,
7541 int_x86_avx2_pbroadcastb_256>;
7542 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7543 int_x86_avx2_pbroadcastw_128,
7544 int_x86_avx2_pbroadcastw_256>;
7545 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7546 int_x86_avx2_pbroadcastd_128,
7547 int_x86_avx2_pbroadcastd_256>;
7548 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7549 int_x86_avx2_pbroadcastq_128,
7550 int_x86_avx2_pbroadcastq_256>;
7552 //===----------------------------------------------------------------------===//
7553 // VPERM - Permute instructions
7556 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7558 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7559 (ins VR256:$src1, VR256:$src2),
7560 !strconcat(OpcodeStr,
7561 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7562 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7563 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7564 (ins VR256:$src1, i256mem:$src2),
7565 !strconcat(OpcodeStr,
7566 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7567 [(set VR256:$dst, (Int VR256:$src1, (mem_frag addr:$src2)))]>,
7571 defm VPERMD : avx2_perm<0x36, "vpermd", memopv8i32, int_x86_avx2_permd>;
7572 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7574 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7576 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7577 (ins VR256:$src1, i8imm:$src2),
7578 !strconcat(OpcodeStr,
7579 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7580 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7581 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7582 (ins i256mem:$src1, i8imm:$src2),
7583 !strconcat(OpcodeStr,
7584 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7585 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7589 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7591 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7594 //===----------------------------------------------------------------------===//
7595 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7597 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7598 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7599 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7601 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7603 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7604 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7605 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7607 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7611 //===----------------------------------------------------------------------===//
7612 // VINSERTI128 - Insert packed integer values
7614 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7615 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7616 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7618 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7620 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7621 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7622 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7624 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7625 imm:$src3))]>, VEX_4V;
7627 //===----------------------------------------------------------------------===//
7628 // VEXTRACTI128 - Extract packed integer values
7630 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7631 (ins VR256:$src1, i8imm:$src2),
7632 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7634 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7636 let neverHasSideEffects = 1, mayStore = 1 in
7637 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7638 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7639 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7641 //===----------------------------------------------------------------------===//
7642 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7644 multiclass avx2_pmovmask<string OpcodeStr,
7645 Intrinsic IntLd128, Intrinsic IntLd256,
7646 Intrinsic IntSt128, Intrinsic IntSt256,
7647 PatFrag pf128, PatFrag pf256> {
7648 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7649 (ins VR128:$src1, i128mem:$src2),
7650 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7651 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7652 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7653 (ins VR256:$src1, i256mem:$src2),
7654 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7655 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7656 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7657 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7658 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7659 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7660 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7661 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7662 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7663 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7666 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7667 int_x86_avx2_maskload_d,
7668 int_x86_avx2_maskload_d_256,
7669 int_x86_avx2_maskstore_d,
7670 int_x86_avx2_maskstore_d_256,
7671 memopv4i32, memopv8i32>;
7672 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7673 int_x86_avx2_maskload_q,
7674 int_x86_avx2_maskload_q_256,
7675 int_x86_avx2_maskstore_q,
7676 int_x86_avx2_maskstore_q_256,
7677 memopv2i64, memopv4i64>, VEX_W;
7680 //===----------------------------------------------------------------------===//
7681 // Variable Bit Shifts
7683 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr,
7684 Intrinsic Int128, Intrinsic Int256> {
7685 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7686 (ins VR128:$src1, VR128:$src2),
7687 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7688 [(set VR128:$dst, (Int128 VR128:$src1, VR128:$src2))]>, VEX_4V;
7689 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7690 (ins VR128:$src1, i128mem:$src2),
7691 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7693 (Int128 VR128:$src1, (bitconvert (memopv2i64 addr:$src2))))]>,
7695 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7696 (ins VR256:$src1, VR256:$src2),
7697 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7698 [(set VR256:$dst, (Int256 VR256:$src1, VR256:$src2))]>, VEX_4V;
7699 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7700 (ins VR256:$src1, i256mem:$src2),
7701 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7703 (Int256 VR256:$src1, (bitconvert (memopv4i64 addr:$src2))))]>,
7707 multiclass avx2_var_shift_i64<bits<8> opc, string OpcodeStr,
7708 Intrinsic Int128, Intrinsic Int256> {
7709 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7710 (ins VR128:$src1, VR128:$src2),
7711 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7712 [(set VR128:$dst, (Int128 VR128:$src1, VR128:$src2))]>, VEX_4V;
7713 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7714 (ins VR128:$src1, i128mem:$src2),
7715 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7717 (Int128 VR128:$src1, (memopv2i64 addr:$src2)))]>,
7719 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7720 (ins VR256:$src1, VR256:$src2),
7721 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7722 [(set VR256:$dst, (Int256 VR256:$src1, VR256:$src2))]>, VEX_4V;
7723 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7724 (ins VR256:$src1, i256mem:$src2),
7725 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7727 (Int256 VR256:$src1, (memopv4i64 addr:$src2)))]>,
7731 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", int_x86_avx2_psllv_d,
7732 int_x86_avx2_psllv_d_256>;
7733 defm VPSLLVQ : avx2_var_shift_i64<0x47, "vpsllvq", int_x86_avx2_psllv_q,
7734 int_x86_avx2_psllv_q_256>, VEX_W;
7735 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", int_x86_avx2_psrlv_d,
7736 int_x86_avx2_psrlv_d_256>;
7737 defm VPSRLVQ : avx2_var_shift_i64<0x45, "vpsrlvq", int_x86_avx2_psrlv_q,
7738 int_x86_avx2_psrlv_q_256>, VEX_W;
7739 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", int_x86_avx2_psrav_d,
7740 int_x86_avx2_psrav_d_256>;
7742 let Predicates = [HasAVX2] in {
7743 def : Pat<(v4i32 (shl (v4i32 VR128:$src1), (v4i32 VR128:$src2))),
7744 (VPSLLVDrr VR128:$src1, VR128:$src2)>;
7745 def : Pat<(v2i64 (shl (v2i64 VR128:$src1), (v2i64 VR128:$src2))),
7746 (VPSLLVQrr VR128:$src1, VR128:$src2)>;
7747 def : Pat<(v4i32 (srl (v4i32 VR128:$src1), (v4i32 VR128:$src2))),
7748 (VPSRLVDrr VR128:$src1, VR128:$src2)>;
7749 def : Pat<(v2i64 (srl (v2i64 VR128:$src1), (v2i64 VR128:$src2))),
7750 (VPSRLVQrr VR128:$src1, VR128:$src2)>;
7751 def : Pat<(v4i32 (sra (v4i32 VR128:$src1), (v4i32 VR128:$src2))),
7752 (VPSRAVDrr VR128:$src1, VR128:$src2)>;
7753 def : Pat<(v8i32 (shl (v8i32 VR256:$src1), (v8i32 VR256:$src2))),
7754 (VPSLLVDYrr VR256:$src1, VR256:$src2)>;
7755 def : Pat<(v4i64 (shl (v4i64 VR256:$src1), (v4i64 VR256:$src2))),
7756 (VPSLLVQYrr VR256:$src1, VR256:$src2)>;
7757 def : Pat<(v8i32 (srl (v8i32 VR256:$src1), (v8i32 VR256:$src2))),
7758 (VPSRLVDYrr VR256:$src1, VR256:$src2)>;
7759 def : Pat<(v4i64 (srl (v4i64 VR256:$src1), (v4i64 VR256:$src2))),
7760 (VPSRLVQYrr VR256:$src1, VR256:$src2)>;
7761 def : Pat<(v8i32 (sra (v8i32 VR256:$src1), (v8i32 VR256:$src2))),
7762 (VPSRAVDYrr VR256:$src1, VR256:$src2)>;
7764 def : Pat<(v4i32 (shl (v4i32 VR128:$src1),
7765 (v4i32 (bitconvert (memopv2i64 addr:$src2))))),
7766 (VPSLLVDrm VR128:$src1, addr:$src2)>;
7767 def : Pat<(v2i64 (shl (v2i64 VR128:$src1), (memopv2i64 addr:$src2))),
7768 (VPSLLVQrm VR128:$src1, addr:$src2)>;
7769 def : Pat<(v4i32 (srl (v4i32 VR128:$src1),
7770 (v4i32 (bitconvert (memopv2i64 addr:$src2))))),
7771 (VPSRLVDrm VR128:$src1, addr:$src2)>;
7772 def : Pat<(v2i64 (srl (v2i64 VR128:$src1), (memopv2i64 addr:$src2))),
7773 (VPSRLVQrm VR128:$src1, addr:$src2)>;
7774 def : Pat<(v4i32 (sra (v4i32 VR128:$src1),
7775 (v4i32 (bitconvert (memopv2i64 addr:$src2))))),
7776 (VPSRAVDrm VR128:$src1, addr:$src2)>;
7777 def : Pat<(v8i32 (shl (v8i32 VR256:$src1),
7778 (v8i32 (bitconvert (memopv4i64 addr:$src2))))),
7779 (VPSLLVDYrm VR256:$src1, addr:$src2)>;
7780 def : Pat<(v4i64 (shl (v4i64 VR256:$src1), (memopv4i64 addr:$src2))),
7781 (VPSLLVQYrm VR256:$src1, addr:$src2)>;
7782 def : Pat<(v8i32 (srl (v8i32 VR256:$src1),
7783 (v8i32 (bitconvert (memopv4i64 addr:$src2))))),
7784 (VPSRLVDYrm VR256:$src1, addr:$src2)>;
7785 def : Pat<(v4i64 (srl (v4i64 VR256:$src1), (memopv4i64 addr:$src2))),
7786 (VPSRLVQYrm VR256:$src1, addr:$src2)>;
7787 def : Pat<(v8i32 (sra (v8i32 VR256:$src1),
7788 (v8i32 (bitconvert (memopv4i64 addr:$src2))))),
7789 (VPSRAVDYrm VR256:$src1, addr:$src2)>;