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 let isCommutable = 1 in
85 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
87 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
88 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
90 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
92 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
93 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
97 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
98 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
99 string asm, string SSEVer, string FPSizeStr,
100 X86MemOperand x86memop, PatFrag mem_frag,
101 Domain d, bit Is2Addr = 1> {
102 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
104 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
105 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
106 [(set RC:$dst, (!cast<Intrinsic>(
107 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
108 RC:$src1, RC:$src2))], d>;
109 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
111 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
112 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
113 [(set RC:$dst, (!cast<Intrinsic>(
114 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
115 RC:$src1, (mem_frag addr:$src2)))], d>;
118 //===----------------------------------------------------------------------===//
119 // Non-instruction patterns
120 //===----------------------------------------------------------------------===//
122 // A vector extract of the first f32/f64 position is a subregister copy
123 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
124 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
125 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
126 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
128 // A 128-bit subvector extract from the first 256-bit vector position
129 // is a subregister copy that needs no instruction.
130 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
131 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
132 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
133 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
135 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
136 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
137 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
138 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
140 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
141 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
142 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
143 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
145 // A 128-bit subvector insert to the first 256-bit vector position
146 // is a subregister copy that needs no instruction.
147 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
148 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
149 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
150 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
151 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
152 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
153 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
154 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
155 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
156 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
157 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
158 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
160 // Implicitly promote a 32-bit scalar to a vector.
161 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
162 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
163 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
164 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
165 // Implicitly promote a 64-bit scalar to a vector.
166 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
167 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
168 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
169 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
171 // Bitcasts between 128-bit vector types. Return the original type since
172 // no instruction is needed for the conversion
173 let Predicates = [HasXMMInt] in {
174 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
175 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
176 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
177 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
178 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
179 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
180 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
181 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
182 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
183 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
184 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
185 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
186 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
187 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
188 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
189 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
190 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
191 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
192 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
193 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
194 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
195 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
196 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
197 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
198 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
199 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
200 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
201 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
202 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
203 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
206 // Bitcasts between 256-bit vector types. Return the original type since
207 // no instruction is needed for the conversion
208 let Predicates = [HasAVX] in {
209 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
210 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
211 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
212 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
213 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
214 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
215 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
216 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
217 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
218 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
219 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
220 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
221 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
222 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
223 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
224 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
225 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
226 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
227 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
228 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
229 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
230 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
231 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
232 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
233 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
234 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
235 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
236 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
237 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
238 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
241 // Alias instructions that map fld0 to pxor for sse.
242 // FIXME: Set encoding to pseudo!
243 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
244 canFoldAsLoad = 1 in {
245 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
246 [(set FR32:$dst, fp32imm0)]>,
247 Requires<[HasSSE1]>, TB, OpSize;
248 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
249 [(set FR64:$dst, fpimm0)]>,
250 Requires<[HasSSE2]>, TB, OpSize;
251 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
252 [(set FR32:$dst, fp32imm0)]>,
253 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
254 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
255 [(set FR64:$dst, fpimm0)]>,
256 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
259 //===----------------------------------------------------------------------===//
260 // AVX & SSE - Zero/One Vectors
261 //===----------------------------------------------------------------------===//
263 // Alias instructions that map zero vector to pxor / xorp* for sse.
264 // We set canFoldAsLoad because this can be converted to a constant-pool
265 // load of an all-zeros value if folding it would be beneficial.
266 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
267 // JIT implementation, it does not expand the instructions below like
268 // X86MCInstLower does.
269 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
270 isCodeGenOnly = 1 in {
271 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
272 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
273 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
274 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
275 let ExeDomain = SSEPackedInt in
276 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
277 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
280 // The same as done above but for AVX. The 128-bit versions are the
281 // same, but re-encoded. The 256-bit does not support PI version, and
282 // doesn't need it because on sandy bridge the register is set to zero
283 // at the rename stage without using any execution unit, so SET0PSY
284 // and SET0PDY can be used for vector int instructions without penalty
285 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
286 // JIT implementatioan, it does not expand the instructions below like
287 // X86MCInstLower does.
288 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
289 isCodeGenOnly = 1, Predicates = [HasAVX] in {
290 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
291 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
292 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
293 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
294 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
295 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
296 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
297 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
298 let ExeDomain = SSEPackedInt in
299 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
300 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
303 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
304 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
305 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
307 // AVX has no support for 256-bit integer instructions, but since the 128-bit
308 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
309 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
310 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
311 (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
313 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
314 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
315 (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
317 // We set canFoldAsLoad because this can be converted to a constant-pool
318 // load of an all-ones value if folding it would be beneficial.
319 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
320 // JIT implementation, it does not expand the instructions below like
321 // X86MCInstLower does.
322 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
323 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
324 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
325 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
326 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
327 isCodeGenOnly = 1, ExeDomain = SSEPackedInt, Predicates = [HasAVX] in
328 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
329 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
332 //===----------------------------------------------------------------------===//
333 // SSE 1 & 2 - Move FP Scalar Instructions
335 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
336 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
337 // is used instead. Register-to-register movss/movsd is not modeled as an
338 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
339 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
340 //===----------------------------------------------------------------------===//
342 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
343 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
344 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
346 // Loading from memory automatically zeroing upper bits.
347 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
348 PatFrag mem_pat, string OpcodeStr> :
349 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
350 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
351 [(set RC:$dst, (mem_pat addr:$src))]>;
354 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
355 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
356 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
357 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
359 // For the disassembler
360 let isCodeGenOnly = 1 in {
361 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
362 (ins VR128:$src1, FR32:$src2),
363 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
365 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
366 (ins VR128:$src1, FR64:$src2),
367 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
371 let canFoldAsLoad = 1, isReMaterializable = 1 in {
372 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
373 let AddedComplexity = 20 in
374 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
377 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
378 "movss\t{$src, $dst|$dst, $src}",
379 [(store FR32:$src, addr:$dst)]>, XS, VEX;
380 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
381 "movsd\t{$src, $dst|$dst, $src}",
382 [(store FR64:$src, addr:$dst)]>, XD, VEX;
385 let Constraints = "$src1 = $dst" in {
386 def MOVSSrr : sse12_move_rr<FR32, v4f32,
387 "movss\t{$src2, $dst|$dst, $src2}">, XS;
388 def MOVSDrr : sse12_move_rr<FR64, v2f64,
389 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
391 // For the disassembler
392 let isCodeGenOnly = 1 in {
393 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
394 (ins VR128:$src1, FR32:$src2),
395 "movss\t{$src2, $dst|$dst, $src2}", []>, XS;
396 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
397 (ins VR128:$src1, FR64:$src2),
398 "movsd\t{$src2, $dst|$dst, $src2}", []>, XD;
402 let canFoldAsLoad = 1, isReMaterializable = 1 in {
403 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
405 let AddedComplexity = 20 in
406 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
409 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
410 "movss\t{$src, $dst|$dst, $src}",
411 [(store FR32:$src, addr:$dst)]>;
412 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
413 "movsd\t{$src, $dst|$dst, $src}",
414 [(store FR64:$src, addr:$dst)]>;
417 let Predicates = [HasSSE1] in {
418 let AddedComplexity = 15 in {
419 // Extract the low 32-bit value from one vector and insert it into another.
420 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
421 (MOVSSrr (v4f32 VR128:$src1),
422 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
423 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
424 (MOVSSrr (v4i32 VR128:$src1),
425 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
427 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
428 // MOVSS to the lower bits.
429 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
430 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
431 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
432 (MOVSSrr (v4f32 (V_SET0PS)),
433 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
434 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
435 (MOVSSrr (v4i32 (V_SET0PI)),
436 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
439 let AddedComplexity = 20 in {
440 // MOVSSrm zeros the high parts of the register; represent this
441 // with SUBREG_TO_REG.
442 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
443 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
444 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
445 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
446 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
447 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
450 // Extract and store.
451 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
454 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
456 // Shuffle with MOVSS
457 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
458 (MOVSSrr VR128:$src1, FR32:$src2)>;
459 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
460 (MOVSSrr (v4i32 VR128:$src1),
461 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
462 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
463 (MOVSSrr (v4f32 VR128:$src1),
464 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
467 let Predicates = [HasSSE2] in {
468 let AddedComplexity = 15 in {
469 // Extract the low 64-bit value from one vector and insert it into another.
470 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
471 (MOVSDrr (v2f64 VR128:$src1),
472 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
473 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
474 (MOVSDrr (v2i64 VR128:$src1),
475 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
477 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
478 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
479 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
480 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
481 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
483 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
484 // MOVSD to the lower bits.
485 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
486 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
489 let AddedComplexity = 20 in {
490 // MOVSDrm zeros the high parts of the register; represent this
491 // with SUBREG_TO_REG.
492 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
493 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
494 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
495 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
496 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
497 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
498 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
499 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
500 def : Pat<(v2f64 (X86vzload addr:$src)),
501 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
504 // Extract and store.
505 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
508 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
510 // Shuffle with MOVSD
511 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
512 (MOVSDrr VR128:$src1, FR64:$src2)>;
513 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
514 (MOVSDrr (v2i64 VR128:$src1),
515 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
516 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
517 (MOVSDrr (v2f64 VR128:$src1),
518 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
519 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
520 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
521 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
522 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
524 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
525 // is during lowering, where it's not possible to recognize the fold cause
526 // it has two uses through a bitcast. One use disappears at isel time and the
527 // fold opportunity reappears.
528 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
529 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
530 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
531 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
534 let Predicates = [HasAVX] in {
535 let AddedComplexity = 15 in {
536 // Extract the low 32-bit value from one vector and insert it into another.
537 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
538 (VMOVSSrr (v4f32 VR128:$src1),
539 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
540 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
541 (VMOVSSrr (v4i32 VR128:$src1),
542 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
544 // Extract the low 64-bit value from one vector and insert it into another.
545 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
546 (VMOVSDrr (v2f64 VR128:$src1),
547 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
548 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
549 (VMOVSDrr (v2i64 VR128:$src1),
550 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
552 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
553 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
554 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
555 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
556 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
558 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
559 // MOVS{S,D} to the lower bits.
560 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
561 (VMOVSSrr (v4f32 (AVX_SET0PS)), FR32:$src)>;
562 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
563 (VMOVSSrr (v4f32 (AVX_SET0PS)),
564 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
565 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
566 (VMOVSSrr (v4i32 (AVX_SET0PI)),
567 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
568 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
569 (VMOVSDrr (v2f64 (AVX_SET0PS)), FR64:$src)>;
572 let AddedComplexity = 20 in {
573 // MOVSSrm zeros the high parts of the register; represent this
574 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
575 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
576 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
577 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
578 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
579 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
580 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
582 // MOVSDrm zeros the high parts of the register; represent this
583 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
584 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
585 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
586 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
587 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
588 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
589 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
590 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
591 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
592 def : Pat<(v2f64 (X86vzload addr:$src)),
593 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
595 // Represent the same patterns above but in the form they appear for
597 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
598 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
599 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
600 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
601 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
602 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
604 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
605 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
606 (SUBREG_TO_REG (i32 0),
607 (v4f32 (VMOVSSrr (v4f32 (AVX_SET0PS)), FR32:$src)),
609 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
610 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
611 (SUBREG_TO_REG (i64 0),
612 (v2f64 (VMOVSDrr (v2f64 (AVX_SET0PS)), FR64:$src)),
615 // Extract and store.
616 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
619 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
620 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
623 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
625 // Shuffle with VMOVSS
626 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
627 (VMOVSSrr VR128:$src1, FR32:$src2)>;
628 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
629 (VMOVSSrr (v4i32 VR128:$src1),
630 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
631 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
632 (VMOVSSrr (v4f32 VR128:$src1),
633 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
635 // Shuffle with VMOVSD
636 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
637 (VMOVSDrr VR128:$src1, FR64:$src2)>;
638 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
639 (VMOVSDrr (v2i64 VR128:$src1),
640 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
641 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
642 (VMOVSDrr (v2f64 VR128:$src1),
643 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
644 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
645 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
647 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
648 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
651 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
652 // is during lowering, where it's not possible to recognize the fold cause
653 // it has two uses through a bitcast. One use disappears at isel time and the
654 // fold opportunity reappears.
655 def : Pat<(v4f32 (X86Movlps VR128:$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 (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1045 (MOVLPSmr addr:$src1, VR128:$src2)>;
1046 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1047 VR128:$src2)), addr:$src1),
1048 (MOVLPSmr addr:$src1, VR128:$src2)>;
1050 // Shuffle with MOVLPS
1051 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1052 (MOVLPSrm VR128:$src1, addr:$src2)>;
1053 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1054 (MOVLPSrm VR128:$src1, addr:$src2)>;
1055 def : Pat<(X86Movlps VR128:$src1,
1056 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1057 (MOVLPSrm VR128:$src1, addr:$src2)>;
1060 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1062 (MOVLPSmr addr:$src1, VR128:$src2)>;
1063 def : Pat<(store (v4i32 (X86Movlps
1064 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1066 (MOVLPSmr addr:$src1, VR128:$src2)>;
1069 let Predicates = [HasSSE2] in {
1070 let AddedComplexity = 20 in {
1071 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1072 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1073 (MOVLPDrm VR128:$src1, addr:$src2)>;
1074 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1075 (MOVLPDrm VR128:$src1, addr:$src2)>;
1078 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1079 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1080 (MOVLPDmr addr:$src1, VR128:$src2)>;
1081 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1082 (MOVLPDmr addr:$src1, VR128:$src2)>;
1084 // Shuffle with MOVLPD
1085 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1086 (MOVLPDrm VR128:$src1, addr:$src2)>;
1087 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1088 (MOVLPDrm VR128:$src1, addr:$src2)>;
1089 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1090 (scalar_to_vector (loadf64 addr:$src2)))),
1091 (MOVLPDrm VR128:$src1, addr:$src2)>;
1094 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1096 (MOVLPDmr addr:$src1, VR128:$src2)>;
1097 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1099 (MOVLPDmr addr:$src1, VR128:$src2)>;
1102 //===----------------------------------------------------------------------===//
1103 // SSE 1 & 2 - Move Hi packed FP Instructions
1104 //===----------------------------------------------------------------------===//
1106 let AddedComplexity = 20 in {
1107 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1108 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1110 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1111 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1112 "\t{$src2, $dst|$dst, $src2}">;
1115 // v2f64 extract element 1 is always custom lowered to unpack high to low
1116 // and extract element 0 so the non-store version isn't too horrible.
1117 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1118 "movhps\t{$src, $dst|$dst, $src}",
1119 [(store (f64 (vector_extract
1120 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1121 (undef)), (iPTR 0))), addr:$dst)]>,
1123 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1124 "movhpd\t{$src, $dst|$dst, $src}",
1125 [(store (f64 (vector_extract
1126 (v2f64 (unpckh VR128:$src, (undef))),
1127 (iPTR 0))), addr:$dst)]>,
1129 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1130 "movhps\t{$src, $dst|$dst, $src}",
1131 [(store (f64 (vector_extract
1132 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1133 (undef)), (iPTR 0))), addr:$dst)]>;
1134 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1135 "movhpd\t{$src, $dst|$dst, $src}",
1136 [(store (f64 (vector_extract
1137 (v2f64 (unpckh VR128:$src, (undef))),
1138 (iPTR 0))), addr:$dst)]>;
1140 let Predicates = [HasAVX] in {
1142 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1143 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1144 def : Pat<(X86Movlhps VR128:$src1,
1145 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1146 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1147 def : Pat<(X86Movlhps VR128:$src1,
1148 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1149 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1151 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
1152 // is during lowering, where it's not possible to recognize the load fold cause
1153 // it has two uses through a bitcast. One use disappears at isel time and the
1154 // fold opportunity reappears.
1155 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
1156 (scalar_to_vector (loadf64 addr:$src2)))),
1157 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1159 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1160 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1161 (scalar_to_vector (loadf64 addr:$src2)))),
1162 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1165 def : Pat<(store (f64 (vector_extract
1166 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1167 (VMOVHPSmr addr:$dst, VR128:$src)>;
1168 def : Pat<(store (f64 (vector_extract
1169 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1170 (VMOVHPDmr addr:$dst, VR128:$src)>;
1173 let Predicates = [HasSSE1] in {
1175 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1176 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1177 def : Pat<(X86Movlhps VR128:$src1,
1178 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1179 (MOVHPSrm VR128:$src1, addr:$src2)>;
1180 def : Pat<(X86Movlhps VR128:$src1,
1181 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1182 (MOVHPSrm VR128:$src1, addr:$src2)>;
1185 def : Pat<(store (f64 (vector_extract
1186 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1187 (MOVHPSmr addr:$dst, VR128:$src)>;
1190 let Predicates = [HasSSE2] in {
1191 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
1192 // is during lowering, where it's not possible to recognize the load fold cause
1193 // it has two uses through a bitcast. One use disappears at isel time and the
1194 // fold opportunity reappears.
1195 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
1196 (scalar_to_vector (loadf64 addr:$src2)))),
1197 (MOVHPDrm VR128:$src1, addr:$src2)>;
1199 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1200 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1201 (scalar_to_vector (loadf64 addr:$src2)))),
1202 (MOVHPDrm VR128:$src1, addr:$src2)>;
1205 def : Pat<(store (f64 (vector_extract
1206 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
1207 (MOVHPDmr addr:$dst, VR128:$src)>;
1210 //===----------------------------------------------------------------------===//
1211 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1212 //===----------------------------------------------------------------------===//
1214 let AddedComplexity = 20 in {
1215 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1216 (ins VR128:$src1, VR128:$src2),
1217 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1219 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1221 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1222 (ins VR128:$src1, VR128:$src2),
1223 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1225 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1228 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1229 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1230 (ins VR128:$src1, VR128:$src2),
1231 "movlhps\t{$src2, $dst|$dst, $src2}",
1233 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1234 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1235 (ins VR128:$src1, VR128:$src2),
1236 "movhlps\t{$src2, $dst|$dst, $src2}",
1238 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1241 let Predicates = [HasAVX] in {
1243 let AddedComplexity = 20 in {
1244 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1245 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1246 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1247 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1249 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1250 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1251 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1253 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1254 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1255 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1256 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1257 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1258 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1261 let AddedComplexity = 20 in {
1262 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1263 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1264 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1266 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1267 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1268 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1269 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1270 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1273 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1274 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1275 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1276 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1279 let Predicates = [HasSSE1] in {
1281 let AddedComplexity = 20 in {
1282 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1283 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1284 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1285 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1287 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1288 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1289 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1291 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1292 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1293 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1294 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1295 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1296 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1299 let AddedComplexity = 20 in {
1300 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1301 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1302 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1304 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1305 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1306 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1307 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1308 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1311 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1312 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1313 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1314 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1317 //===----------------------------------------------------------------------===//
1318 // SSE 1 & 2 - Conversion Instructions
1319 //===----------------------------------------------------------------------===//
1321 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1322 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1324 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1325 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1326 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1327 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1330 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1331 X86MemOperand x86memop, string asm> {
1332 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, []>;
1334 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, []>;
1337 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1338 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1339 string asm, Domain d> {
1340 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1341 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1342 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1343 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1346 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1347 X86MemOperand x86memop, string asm> {
1348 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1349 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1351 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1352 (ins DstRC:$src1, x86memop:$src),
1353 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1356 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1357 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
1358 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1359 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1361 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1362 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
1363 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1364 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1367 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1368 // register, but the same isn't true when only using memory operands,
1369 // provide other assembly "l" and "q" forms to address this explicitly
1370 // where appropriate to do so.
1371 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1373 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1375 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1377 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1379 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1382 let Predicates = [HasAVX] in {
1383 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1384 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1385 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1386 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1387 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1388 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1389 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1390 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1392 def : Pat<(f32 (sint_to_fp GR32:$src)),
1393 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1394 def : Pat<(f32 (sint_to_fp GR64:$src)),
1395 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1396 def : Pat<(f64 (sint_to_fp GR32:$src)),
1397 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1398 def : Pat<(f64 (sint_to_fp GR64:$src)),
1399 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1402 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1403 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1404 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1405 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1406 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1407 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1408 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1409 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1410 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1411 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1412 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1413 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1414 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1415 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1416 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1417 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1419 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1420 // and/or XMM operand(s).
1422 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1423 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1425 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1426 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1427 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1428 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1429 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1430 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1433 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1434 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1435 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1436 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1438 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1439 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1440 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1441 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1442 (ins DstRC:$src1, x86memop:$src2),
1444 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1445 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1446 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1449 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1450 f128mem, load, "cvtsd2si">, XD, VEX;
1451 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1452 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1455 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1456 // Get rid of this hack or rename the intrinsics, there are several
1457 // intructions that only match with the intrinsic form, why create duplicates
1458 // to let them be recognized by the assembler?
1459 defm VCVTSD2SI : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
1460 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
1461 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
1462 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
1464 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1465 f128mem, load, "cvtsd2si{l}">, XD;
1466 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1467 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1470 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1471 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1472 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1473 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1475 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1476 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1477 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1478 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1481 let Constraints = "$src1 = $dst" in {
1482 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1483 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1485 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1486 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1487 "cvtsi2ss{q}">, XS, REX_W;
1488 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1489 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1491 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1492 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1493 "cvtsi2sd">, XD, REX_W;
1498 // Aliases for intrinsics
1499 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1500 f32mem, load, "cvttss2si">, XS, VEX;
1501 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1502 int_x86_sse_cvttss2si64, f32mem, load,
1503 "cvttss2si">, XS, VEX, VEX_W;
1504 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1505 f128mem, load, "cvttsd2si">, XD, VEX;
1506 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1507 int_x86_sse2_cvttsd2si64, f128mem, load,
1508 "cvttsd2si">, XD, VEX, VEX_W;
1509 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1510 f32mem, load, "cvttss2si">, XS;
1511 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1512 int_x86_sse_cvttss2si64, f32mem, load,
1513 "cvttss2si{q}">, XS, REX_W;
1514 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1515 f128mem, load, "cvttsd2si">, XD;
1516 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1517 int_x86_sse2_cvttsd2si64, f128mem, load,
1518 "cvttsd2si{q}">, XD, REX_W;
1520 let Pattern = []<dag> in {
1521 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1522 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
1523 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1524 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1526 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1527 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1528 SSEPackedSingle>, TB, VEX;
1529 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1530 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1531 SSEPackedSingle>, TB, VEX;
1534 let Pattern = []<dag> in {
1535 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1536 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1537 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1538 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1539 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1540 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1541 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1544 let Predicates = [HasSSE1] in {
1545 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1546 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1547 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1548 (CVTSS2SIrm addr:$src)>;
1549 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1550 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1551 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1552 (CVTSS2SI64rm addr:$src)>;
1555 let Predicates = [HasAVX] in {
1556 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1557 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1558 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1559 (VCVTSS2SIrm addr:$src)>;
1560 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1561 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1562 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1563 (VCVTSS2SI64rm addr:$src)>;
1568 // Convert scalar double to scalar single
1569 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1570 (ins FR64:$src1, FR64:$src2),
1571 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1574 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1575 (ins FR64:$src1, f64mem:$src2),
1576 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1577 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
1579 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1582 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1583 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1584 [(set FR32:$dst, (fround FR64:$src))]>;
1585 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1586 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1587 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1588 Requires<[HasSSE2, OptForSize]>;
1590 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1591 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1593 let Constraints = "$src1 = $dst" in
1594 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1595 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1597 // Convert scalar single to scalar double
1598 // SSE2 instructions with XS prefix
1599 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1600 (ins FR32:$src1, FR32:$src2),
1601 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1602 []>, XS, Requires<[HasAVX]>, VEX_4V;
1604 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1605 (ins FR32:$src1, f32mem:$src2),
1606 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1607 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
1609 let Predicates = [HasAVX] in {
1610 def : Pat<(f64 (fextend FR32:$src)),
1611 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1612 def : Pat<(fextend (loadf32 addr:$src)),
1613 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1614 def : Pat<(extloadf32 addr:$src),
1615 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1618 def : Pat<(extloadf32 addr:$src),
1619 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1620 Requires<[HasAVX, OptForSpeed]>;
1622 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1623 "cvtss2sd\t{$src, $dst|$dst, $src}",
1624 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1625 Requires<[HasSSE2]>;
1626 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1627 "cvtss2sd\t{$src, $dst|$dst, $src}",
1628 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1629 Requires<[HasSSE2, OptForSize]>;
1631 // extload f32 -> f64. This matches load+fextend because we have a hack in
1632 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1634 // Since these loads aren't folded into the fextend, we have to match it
1636 def : Pat<(fextend (loadf32 addr:$src)),
1637 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1638 def : Pat<(extloadf32 addr:$src),
1639 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1641 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1642 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1643 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1644 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1645 VR128:$src2))]>, XS, VEX_4V,
1647 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1648 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1649 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1650 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1651 (load addr:$src2)))]>, XS, VEX_4V,
1653 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1654 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1655 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1656 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1657 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1658 VR128:$src2))]>, XS,
1659 Requires<[HasSSE2]>;
1660 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1661 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1662 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1663 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1664 (load addr:$src2)))]>, XS,
1665 Requires<[HasSSE2]>;
1668 // Convert doubleword to packed single/double fp
1669 // SSE2 instructions without OpSize prefix
1670 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1671 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1672 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1673 TB, VEX, Requires<[HasAVX]>;
1674 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1675 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1676 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1677 (bitconvert (memopv2i64 addr:$src))))]>,
1678 TB, VEX, Requires<[HasAVX]>;
1679 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1680 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1681 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1682 TB, Requires<[HasSSE2]>;
1683 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1684 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1685 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1686 (bitconvert (memopv2i64 addr:$src))))]>,
1687 TB, Requires<[HasSSE2]>;
1689 // FIXME: why the non-intrinsic version is described as SSE3?
1690 // SSE2 instructions with XS prefix
1691 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1692 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1693 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1694 XS, VEX, Requires<[HasAVX]>;
1695 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1696 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1697 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1698 (bitconvert (memopv2i64 addr:$src))))]>,
1699 XS, VEX, Requires<[HasAVX]>;
1700 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1701 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1702 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1703 XS, Requires<[HasSSE2]>;
1704 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1705 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1706 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1707 (bitconvert (memopv2i64 addr:$src))))]>,
1708 XS, Requires<[HasSSE2]>;
1711 // Convert packed single/double fp to doubleword
1712 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1713 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1714 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1715 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1716 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1717 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1718 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1719 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1720 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1721 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1722 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1723 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1725 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1726 "cvtps2dq\t{$src, $dst|$dst, $src}",
1727 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1729 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1731 "cvtps2dq\t{$src, $dst|$dst, $src}",
1732 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1733 (memop addr:$src)))]>, VEX;
1734 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1735 "cvtps2dq\t{$src, $dst|$dst, $src}",
1736 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1737 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1738 "cvtps2dq\t{$src, $dst|$dst, $src}",
1739 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1740 (memop addr:$src)))]>;
1742 // SSE2 packed instructions with XD prefix
1743 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1744 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1745 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1746 XD, VEX, Requires<[HasAVX]>;
1747 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1748 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1749 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1750 (memop addr:$src)))]>,
1751 XD, VEX, Requires<[HasAVX]>;
1752 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1753 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1754 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1755 XD, Requires<[HasSSE2]>;
1756 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1757 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1758 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1759 (memop addr:$src)))]>,
1760 XD, Requires<[HasSSE2]>;
1763 // Convert with truncation packed single/double fp to doubleword
1764 // SSE2 packed instructions with XS prefix
1765 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1766 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1768 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1769 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1770 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1771 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1773 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1774 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1775 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1776 "cvttps2dq\t{$src, $dst|$dst, $src}",
1778 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1779 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1780 "cvttps2dq\t{$src, $dst|$dst, $src}",
1782 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1784 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1785 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1787 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1788 XS, VEX, Requires<[HasAVX]>;
1789 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1790 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1791 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1792 (memop addr:$src)))]>,
1793 XS, VEX, Requires<[HasAVX]>;
1795 let Predicates = [HasSSE2] in {
1796 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1797 (Int_CVTDQ2PSrr VR128:$src)>;
1798 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1799 (CVTTPS2DQrr VR128:$src)>;
1802 let Predicates = [HasAVX] in {
1803 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1804 (Int_VCVTDQ2PSrr VR128:$src)>;
1805 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1806 (VCVTTPS2DQrr VR128:$src)>;
1807 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1808 (VCVTDQ2PSYrr VR256:$src)>;
1809 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1810 (VCVTTPS2DQYrr VR256:$src)>;
1813 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1814 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1816 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1817 let isCodeGenOnly = 1 in
1818 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1819 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1820 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1821 (memop addr:$src)))]>, VEX;
1822 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1823 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1824 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1825 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1826 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1827 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1828 (memop addr:$src)))]>;
1830 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1831 // register, but the same isn't true when using memory operands instead.
1832 // Provide other assembly rr and rm forms to address this explicitly.
1833 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1834 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1837 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1838 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1839 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1840 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1843 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1844 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1845 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1846 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1848 // Convert packed single to packed double
1849 let Predicates = [HasAVX] in {
1850 // SSE2 instructions without OpSize prefix
1851 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1852 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1853 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1854 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1855 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1856 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1857 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1858 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1860 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1861 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1862 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1863 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1865 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1866 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1867 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1868 TB, VEX, Requires<[HasAVX]>;
1869 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1870 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1871 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1872 (load addr:$src)))]>,
1873 TB, VEX, Requires<[HasAVX]>;
1874 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1875 "cvtps2pd\t{$src, $dst|$dst, $src}",
1876 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1877 TB, Requires<[HasSSE2]>;
1878 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1879 "cvtps2pd\t{$src, $dst|$dst, $src}",
1880 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1881 (load addr:$src)))]>,
1882 TB, Requires<[HasSSE2]>;
1884 // Convert packed double to packed single
1885 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1886 // register, but the same isn't true when using memory operands instead.
1887 // Provide other assembly rr and rm forms to address this explicitly.
1888 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1889 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1890 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1891 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1894 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1895 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1896 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1897 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1900 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1901 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1902 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1903 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1904 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1905 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1906 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1907 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1910 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1911 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1912 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1913 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1915 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1916 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1917 (memop addr:$src)))]>;
1918 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1919 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1920 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1921 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1922 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1923 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1924 (memop addr:$src)))]>;
1926 // AVX 256-bit register conversion intrinsics
1927 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1928 // whenever possible to avoid declaring two versions of each one.
1929 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1930 (VCVTDQ2PSYrr VR256:$src)>;
1931 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1932 (VCVTDQ2PSYrm addr:$src)>;
1934 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1935 (VCVTPD2PSYrr VR256:$src)>;
1936 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1937 (VCVTPD2PSYrm addr:$src)>;
1939 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1940 (VCVTPS2DQYrr VR256:$src)>;
1941 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1942 (VCVTPS2DQYrm addr:$src)>;
1944 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1945 (VCVTPS2PDYrr VR128:$src)>;
1946 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1947 (VCVTPS2PDYrm addr:$src)>;
1949 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1950 (VCVTTPD2DQYrr VR256:$src)>;
1951 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1952 (VCVTTPD2DQYrm addr:$src)>;
1954 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1955 (VCVTTPS2DQYrr VR256:$src)>;
1956 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1957 (VCVTTPS2DQYrm addr:$src)>;
1959 // Match fround and fextend for 128/256-bit conversions
1960 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1961 (VCVTPD2PSYrr VR256:$src)>;
1962 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1963 (VCVTPD2PSYrm addr:$src)>;
1965 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1966 (VCVTPS2PDYrr VR128:$src)>;
1967 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1968 (VCVTPS2PDYrm addr:$src)>;
1970 //===----------------------------------------------------------------------===//
1971 // SSE 1 & 2 - Compare Instructions
1972 //===----------------------------------------------------------------------===//
1974 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1975 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1976 string asm, string asm_alt> {
1977 let isAsmParserOnly = 1 in {
1978 def rr : SIi8<0xC2, MRMSrcReg,
1979 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1982 def rm : SIi8<0xC2, MRMSrcMem,
1983 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1987 // Accept explicit immediate argument form instead of comparison code.
1988 def rr_alt : SIi8<0xC2, MRMSrcReg,
1989 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1992 def rm_alt : SIi8<0xC2, MRMSrcMem,
1993 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1997 let neverHasSideEffects = 1 in {
1998 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1999 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2000 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
2002 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
2003 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2004 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
2008 let Constraints = "$src1 = $dst" in {
2009 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
2010 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
2011 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2012 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2,
2014 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
2015 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
2016 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2017 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1),
2018 (loadf32 addr:$src2), imm:$cc))]>, XS;
2019 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
2020 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
2021 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2022 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2,
2024 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
2025 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
2026 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2027 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2),
2030 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
2031 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
2032 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
2033 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
2034 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
2035 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
2036 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
2037 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
2038 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
2039 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
2040 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
2041 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
2042 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
2045 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2046 Intrinsic Int, string asm> {
2047 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2048 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2049 [(set VR128:$dst, (Int VR128:$src1,
2050 VR128:$src, imm:$cc))]>;
2051 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2052 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
2053 [(set VR128:$dst, (Int VR128:$src1,
2054 (load addr:$src), imm:$cc))]>;
2057 // Aliases to match intrinsics which expect XMM operand(s).
2058 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2059 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2061 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2062 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2064 let Constraints = "$src1 = $dst" in {
2065 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2066 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2067 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2068 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2072 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2073 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2074 ValueType vt, X86MemOperand x86memop,
2075 PatFrag ld_frag, string OpcodeStr, Domain d> {
2076 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2077 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2078 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2079 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2080 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2081 [(set EFLAGS, (OpNode (vt RC:$src1),
2082 (ld_frag addr:$src2)))], d>;
2085 let Defs = [EFLAGS] in {
2086 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2087 "ucomiss", SSEPackedSingle>, TB, VEX;
2088 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2089 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2090 let Pattern = []<dag> in {
2091 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2092 "comiss", SSEPackedSingle>, TB, VEX;
2093 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2094 "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2097 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2098 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2099 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2100 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2102 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2103 load, "comiss", SSEPackedSingle>, TB, VEX;
2104 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2105 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2106 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2107 "ucomiss", SSEPackedSingle>, TB;
2108 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2109 "ucomisd", SSEPackedDouble>, TB, OpSize;
2111 let Pattern = []<dag> in {
2112 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2113 "comiss", SSEPackedSingle>, TB;
2114 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2115 "comisd", SSEPackedDouble>, TB, OpSize;
2118 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2119 load, "ucomiss", SSEPackedSingle>, TB;
2120 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2121 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2123 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2124 "comiss", SSEPackedSingle>, TB;
2125 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2126 "comisd", SSEPackedDouble>, TB, OpSize;
2127 } // Defs = [EFLAGS]
2129 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2130 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2131 Intrinsic Int, string asm, string asm_alt,
2133 let isAsmParserOnly = 1 in {
2134 def rri : PIi8<0xC2, MRMSrcReg,
2135 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2136 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2137 def rmi : PIi8<0xC2, MRMSrcMem,
2138 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, SSECC:$cc), asm,
2139 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2142 // Accept explicit immediate argument form instead of comparison code.
2143 def rri_alt : PIi8<0xC2, MRMSrcReg,
2144 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2146 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2147 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, i8imm:$cc),
2151 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2152 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2153 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2154 SSEPackedSingle>, TB, VEX_4V;
2155 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2156 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2157 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2158 SSEPackedDouble>, TB, OpSize, VEX_4V;
2159 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2160 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2161 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2162 SSEPackedSingle>, TB, VEX_4V;
2163 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2164 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2165 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2166 SSEPackedDouble>, TB, OpSize, VEX_4V;
2167 let Constraints = "$src1 = $dst" in {
2168 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2169 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2170 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2171 SSEPackedSingle>, TB;
2172 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2173 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2174 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2175 SSEPackedDouble>, TB, OpSize;
2178 let Predicates = [HasSSE1] in {
2179 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2180 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2181 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2182 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2185 let Predicates = [HasSSE2] in {
2186 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2187 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2188 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2189 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2192 let Predicates = [HasAVX] in {
2193 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2194 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2195 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2196 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2197 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2198 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2199 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2200 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2202 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2203 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2204 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2205 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2206 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2207 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2208 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2209 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2212 //===----------------------------------------------------------------------===//
2213 // SSE 1 & 2 - Shuffle Instructions
2214 //===----------------------------------------------------------------------===//
2216 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2217 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2218 ValueType vt, string asm, PatFrag mem_frag,
2219 Domain d, bit IsConvertibleToThreeAddress = 0> {
2220 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2221 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
2222 [(set RC:$dst, (vt (shufp:$src3
2223 RC:$src1, (mem_frag addr:$src2))))], d>;
2224 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2225 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2226 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2228 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2231 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2232 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2233 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2234 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2235 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2236 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2237 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2238 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2239 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2240 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2241 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2242 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2244 let Constraints = "$src1 = $dst" in {
2245 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2246 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2247 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2249 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2250 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2251 memopv2f64, SSEPackedDouble>, TB, OpSize;
2254 let Predicates = [HasSSE1] in {
2255 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2256 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2257 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2258 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2259 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2260 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2261 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2262 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2263 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2264 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2265 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2266 // fall back to this for SSE1)
2267 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2268 (SHUFPSrri VR128:$src2, VR128:$src1,
2269 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2270 // Special unary SHUFPSrri case.
2271 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2272 (SHUFPSrri VR128:$src1, VR128:$src1,
2273 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2276 let Predicates = [HasSSE2] in {
2277 // Special binary v4i32 shuffle cases with SHUFPS.
2278 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2279 (SHUFPSrri VR128:$src1, VR128:$src2,
2280 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2281 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2282 (bc_v4i32 (memopv2i64 addr:$src2)))),
2283 (SHUFPSrmi VR128:$src1, addr:$src2,
2284 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2285 // Special unary SHUFPDrri cases.
2286 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2287 (SHUFPDrri VR128:$src1, VR128:$src1,
2288 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2289 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2290 (SHUFPDrri VR128:$src1, VR128:$src1,
2291 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2292 // Special binary v2i64 shuffle cases using SHUFPDrri.
2293 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2294 (SHUFPDrri VR128:$src1, VR128:$src2,
2295 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2296 // Generic SHUFPD patterns
2297 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2298 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2299 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2300 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2301 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2302 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2303 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2306 let Predicates = [HasAVX] in {
2307 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2308 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2309 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2310 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2311 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2312 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2313 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2314 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2315 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2316 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2317 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2318 // fall back to this for SSE1)
2319 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2320 (VSHUFPSrri VR128:$src2, VR128:$src1,
2321 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2322 // Special unary SHUFPSrri case.
2323 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2324 (VSHUFPSrri VR128:$src1, VR128:$src1,
2325 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2326 // Special binary v4i32 shuffle cases with SHUFPS.
2327 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2328 (VSHUFPSrri VR128:$src1, VR128:$src2,
2329 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2330 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2331 (bc_v4i32 (memopv2i64 addr:$src2)))),
2332 (VSHUFPSrmi VR128:$src1, addr:$src2,
2333 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2334 // Special unary SHUFPDrri cases.
2335 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2336 (VSHUFPDrri VR128:$src1, VR128:$src1,
2337 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2338 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2339 (VSHUFPDrri VR128:$src1, VR128:$src1,
2340 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2341 // Special binary v2i64 shuffle cases using SHUFPDrri.
2342 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2343 (VSHUFPDrri VR128:$src1, VR128:$src2,
2344 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2346 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2347 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2348 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2349 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2350 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2351 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2352 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2355 def : Pat<(v8i32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2356 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2357 def : Pat<(v8i32 (X86Shufps VR256:$src1,
2358 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2359 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2361 def : Pat<(v8f32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2362 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2363 def : Pat<(v8f32 (X86Shufps VR256:$src1,
2364 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2365 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2367 def : Pat<(v4i64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2368 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2369 def : Pat<(v4i64 (X86Shufpd VR256:$src1,
2370 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2371 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2373 def : Pat<(v4f64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2374 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2375 def : Pat<(v4f64 (X86Shufpd VR256:$src1,
2376 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2377 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2380 //===----------------------------------------------------------------------===//
2381 // SSE 1 & 2 - Unpack Instructions
2382 //===----------------------------------------------------------------------===//
2384 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2385 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2386 PatFrag mem_frag, RegisterClass RC,
2387 X86MemOperand x86memop, string asm,
2389 def rr : PI<opc, MRMSrcReg,
2390 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2392 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2393 def rm : PI<opc, MRMSrcMem,
2394 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2396 (vt (OpNode RC:$src1,
2397 (mem_frag addr:$src2))))], d>;
2400 let AddedComplexity = 10 in {
2401 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2402 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2403 SSEPackedSingle>, TB, VEX_4V;
2404 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2405 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2406 SSEPackedDouble>, TB, OpSize, VEX_4V;
2407 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2408 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2409 SSEPackedSingle>, TB, VEX_4V;
2410 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2411 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2412 SSEPackedDouble>, TB, OpSize, VEX_4V;
2414 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2415 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2416 SSEPackedSingle>, TB, VEX_4V;
2417 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2418 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2419 SSEPackedDouble>, TB, OpSize, VEX_4V;
2420 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2421 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2422 SSEPackedSingle>, TB, VEX_4V;
2423 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2424 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2425 SSEPackedDouble>, TB, OpSize, VEX_4V;
2427 let Constraints = "$src1 = $dst" in {
2428 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2429 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2430 SSEPackedSingle>, TB;
2431 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2432 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2433 SSEPackedDouble>, TB, OpSize;
2434 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2435 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2436 SSEPackedSingle>, TB;
2437 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2438 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2439 SSEPackedDouble>, TB, OpSize;
2440 } // Constraints = "$src1 = $dst"
2441 } // AddedComplexity
2443 let Predicates = [HasSSE1] in {
2444 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
2445 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2446 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
2447 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2448 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
2449 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2450 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
2451 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2454 let Predicates = [HasSSE2] in {
2455 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
2456 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2457 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
2458 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2459 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
2460 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2461 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
2462 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2464 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
2465 // problem is during lowering, where it's not possible to recognize the load
2466 // fold cause it has two uses through a bitcast. One use disappears at isel
2467 // time and the fold opportunity reappears.
2468 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2469 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2471 let AddedComplexity = 10 in
2472 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2473 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2476 let Predicates = [HasAVX] in {
2477 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
2478 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2479 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
2480 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2481 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
2482 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2483 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
2484 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2486 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
2487 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2488 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
2489 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2490 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
2491 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2492 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, (memopv8i32 addr:$src2))),
2493 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2494 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, (memopv8f32 addr:$src2))),
2495 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2496 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
2497 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2498 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, (memopv8i32 addr:$src2))),
2499 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2500 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
2501 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2503 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
2504 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2505 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
2506 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2507 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
2508 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2509 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
2510 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2512 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
2513 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2514 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
2515 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2516 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, (memopv4i64 addr:$src2))),
2517 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2518 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
2519 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2520 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, (memopv4f64 addr:$src2))),
2521 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2522 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
2523 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2524 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, (memopv4i64 addr:$src2))),
2525 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2526 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
2527 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2529 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
2530 // problem is during lowering, where it's not possible to recognize the load
2531 // fold cause it has two uses through a bitcast. One use disappears at isel
2532 // time and the fold opportunity reappears.
2533 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2534 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2535 let AddedComplexity = 10 in
2536 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2537 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2540 //===----------------------------------------------------------------------===//
2541 // SSE 1 & 2 - Extract Floating-Point Sign mask
2542 //===----------------------------------------------------------------------===//
2544 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2545 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2547 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2548 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2549 [(set GR32:$dst, (Int RC:$src))], d>;
2550 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2551 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2554 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2555 SSEPackedSingle>, TB;
2556 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2557 SSEPackedDouble>, TB, OpSize;
2559 def : Pat<(i32 (X86fgetsign FR32:$src)),
2560 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2561 sub_ss))>, Requires<[HasSSE1]>;
2562 def : Pat<(i64 (X86fgetsign FR32:$src)),
2563 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2564 sub_ss))>, Requires<[HasSSE1]>;
2565 def : Pat<(i32 (X86fgetsign FR64:$src)),
2566 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2567 sub_sd))>, Requires<[HasSSE2]>;
2568 def : Pat<(i64 (X86fgetsign FR64:$src)),
2569 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2570 sub_sd))>, Requires<[HasSSE2]>;
2572 let Predicates = [HasAVX] in {
2573 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2574 "movmskps", SSEPackedSingle>, TB, VEX;
2575 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2576 "movmskpd", SSEPackedDouble>, TB,
2578 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2579 "movmskps", SSEPackedSingle>, TB, VEX;
2580 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2581 "movmskpd", SSEPackedDouble>, TB,
2584 def : Pat<(i32 (X86fgetsign FR32:$src)),
2585 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2587 def : Pat<(i64 (X86fgetsign FR32:$src)),
2588 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2590 def : Pat<(i32 (X86fgetsign FR64:$src)),
2591 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2593 def : Pat<(i64 (X86fgetsign FR64:$src)),
2594 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2598 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2599 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2600 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2601 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2603 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2604 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2605 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2606 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2610 //===----------------------------------------------------------------------===//
2611 // SSE 1 & 2 - Logical Instructions
2612 //===----------------------------------------------------------------------===//
2614 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2616 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2618 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2619 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2621 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2622 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2624 let Constraints = "$src1 = $dst" in {
2625 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2626 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2628 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2629 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2633 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2634 let mayLoad = 0 in {
2635 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2636 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2637 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2640 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2641 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2643 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2645 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2647 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2648 // are all promoted to v2i64, and the patterns are covered by the int
2649 // version. This is needed in SSE only, because v2i64 isn't supported on
2650 // SSE1, but only on SSE2.
2651 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2652 !strconcat(OpcodeStr, "ps"), f128mem, [],
2653 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2654 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2656 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2657 !strconcat(OpcodeStr, "pd"), f128mem,
2658 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2659 (bc_v2i64 (v2f64 VR128:$src2))))],
2660 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2661 (memopv2i64 addr:$src2)))], 0>,
2663 let Constraints = "$src1 = $dst" in {
2664 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2665 !strconcat(OpcodeStr, "ps"), f128mem,
2666 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2667 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2668 (memopv2i64 addr:$src2)))]>, TB;
2670 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2671 !strconcat(OpcodeStr, "pd"), f128mem,
2672 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2673 (bc_v2i64 (v2f64 VR128:$src2))))],
2674 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2675 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2679 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2681 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2683 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2684 !strconcat(OpcodeStr, "ps"), f256mem,
2685 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2686 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2687 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2689 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2690 !strconcat(OpcodeStr, "pd"), f256mem,
2691 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2692 (bc_v4i64 (v4f64 VR256:$src2))))],
2693 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2694 (memopv4i64 addr:$src2)))], 0>,
2698 // AVX 256-bit packed logical ops forms
2699 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2700 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2701 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2702 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2704 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2705 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2706 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2707 let isCommutable = 0 in
2708 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2710 //===----------------------------------------------------------------------===//
2711 // SSE 1 & 2 - Arithmetic Instructions
2712 //===----------------------------------------------------------------------===//
2714 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2717 /// In addition, we also have a special variant of the scalar form here to
2718 /// represent the associated intrinsic operation. This form is unlike the
2719 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2720 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2722 /// These three forms can each be reg+reg or reg+mem.
2725 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2727 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2729 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2730 OpNode, FR32, f32mem, Is2Addr>, XS;
2731 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2732 OpNode, FR64, f64mem, Is2Addr>, XD;
2735 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2737 let mayLoad = 0 in {
2738 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2739 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2740 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2741 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2745 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2747 let mayLoad = 0 in {
2748 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2749 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2750 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2751 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2755 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2757 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2758 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2759 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2760 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2763 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2765 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2766 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2767 SSEPackedSingle, Is2Addr>, TB;
2769 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2770 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2771 SSEPackedDouble, Is2Addr>, TB, OpSize;
2774 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2775 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2776 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2777 SSEPackedSingle, 0>, TB;
2779 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2780 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2781 SSEPackedDouble, 0>, TB, OpSize;
2784 // Binary Arithmetic instructions
2785 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2786 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
2787 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2788 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2789 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2790 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
2791 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2792 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2794 let isCommutable = 0 in {
2795 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2796 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
2797 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2798 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2799 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2800 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
2801 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2802 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2803 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2804 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
2805 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2806 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2807 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2808 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2809 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2810 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
2811 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2812 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2813 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2814 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2817 let Constraints = "$src1 = $dst" in {
2818 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2819 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2820 basic_sse12_fp_binop_s_int<0x58, "add">;
2821 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2822 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2823 basic_sse12_fp_binop_s_int<0x59, "mul">;
2825 let isCommutable = 0 in {
2826 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2827 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2828 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2829 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2830 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2831 basic_sse12_fp_binop_s_int<0x5E, "div">;
2832 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2833 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2834 basic_sse12_fp_binop_s_int<0x5F, "max">,
2835 basic_sse12_fp_binop_p_int<0x5F, "max">;
2836 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2837 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2838 basic_sse12_fp_binop_s_int<0x5D, "min">,
2839 basic_sse12_fp_binop_p_int<0x5D, "min">;
2844 /// In addition, we also have a special variant of the scalar form here to
2845 /// represent the associated intrinsic operation. This form is unlike the
2846 /// plain scalar form, in that it takes an entire vector (instead of a
2847 /// scalar) and leaves the top elements undefined.
2849 /// And, we have a special variant form for a full-vector intrinsic form.
2851 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2852 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2853 SDNode OpNode, Intrinsic F32Int> {
2854 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2855 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2856 [(set FR32:$dst, (OpNode FR32:$src))]>;
2857 // For scalar unary operations, fold a load into the operation
2858 // only in OptForSize mode. It eliminates an instruction, but it also
2859 // eliminates a whole-register clobber (the load), so it introduces a
2860 // partial register update condition.
2861 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2862 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2863 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2864 Requires<[HasSSE1, OptForSize]>;
2865 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2866 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2867 [(set VR128:$dst, (F32Int VR128:$src))]>;
2868 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2869 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2870 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2873 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2874 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2875 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2876 !strconcat(OpcodeStr,
2877 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2879 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2880 !strconcat(OpcodeStr,
2881 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2882 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2883 (ins ssmem:$src1, VR128:$src2),
2884 !strconcat(OpcodeStr,
2885 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2888 /// sse1_fp_unop_p - SSE1 unops in packed form.
2889 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2890 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2891 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2892 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2893 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2894 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2895 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2898 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2899 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2900 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2901 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2902 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2903 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2904 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2905 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2908 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2909 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2910 Intrinsic V4F32Int> {
2911 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2912 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2913 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2914 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2915 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2916 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2919 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2920 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2921 Intrinsic V4F32Int> {
2922 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2923 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2924 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2925 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2926 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2927 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2930 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2931 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2932 SDNode OpNode, Intrinsic F64Int> {
2933 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2934 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2935 [(set FR64:$dst, (OpNode FR64:$src))]>;
2936 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2937 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2938 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2939 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2940 Requires<[HasSSE2, OptForSize]>;
2941 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2942 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2943 [(set VR128:$dst, (F64Int VR128:$src))]>;
2944 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2945 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2946 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2949 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2950 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2951 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2952 !strconcat(OpcodeStr,
2953 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2954 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2955 !strconcat(OpcodeStr,
2956 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2957 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2958 (ins VR128:$src1, sdmem:$src2),
2959 !strconcat(OpcodeStr,
2960 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2963 /// sse2_fp_unop_p - SSE2 unops in vector forms.
2964 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
2966 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2967 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2968 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
2969 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2970 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2971 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
2974 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
2975 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2976 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2977 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2978 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
2979 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2980 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2981 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
2984 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
2985 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2986 Intrinsic V2F64Int> {
2987 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2988 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2989 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
2990 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2991 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2992 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
2995 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
2996 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2997 Intrinsic V2F64Int> {
2998 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2999 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3000 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3001 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3002 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3003 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3006 let Predicates = [HasAVX] in {
3008 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3009 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V;
3011 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3012 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3013 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3014 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3015 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3016 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3017 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3018 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3021 // Reciprocal approximations. Note that these typically require refinement
3022 // in order to obtain suitable precision.
3023 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V;
3024 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3025 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3026 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3027 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3029 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V;
3030 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3031 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3032 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3033 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3036 def : Pat<(f32 (fsqrt FR32:$src)),
3037 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3038 def : Pat<(f32 (fsqrt (load addr:$src))),
3039 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3040 Requires<[HasAVX, OptForSize]>;
3041 def : Pat<(f64 (fsqrt FR64:$src)),
3042 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3043 def : Pat<(f64 (fsqrt (load addr:$src))),
3044 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3045 Requires<[HasAVX, OptForSize]>;
3047 def : Pat<(f32 (X86frsqrt FR32:$src)),
3048 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3049 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3050 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3051 Requires<[HasAVX, OptForSize]>;
3053 def : Pat<(f32 (X86frcp FR32:$src)),
3054 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3055 def : Pat<(f32 (X86frcp (load addr:$src))),
3056 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3057 Requires<[HasAVX, OptForSize]>;
3059 let Predicates = [HasAVX] in {
3060 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3061 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3062 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3063 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3065 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3066 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3068 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3069 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3070 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3071 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3073 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3074 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3076 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3077 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3078 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3079 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3081 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3082 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3084 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3085 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3086 (VRCPSSr (f32 (IMPLICIT_DEF)),
3087 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3089 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3090 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3094 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3095 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3096 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3097 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3098 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3099 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3101 // Reciprocal approximations. Note that these typically require refinement
3102 // in order to obtain suitable precision.
3103 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3104 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3105 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3106 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3107 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3108 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3110 // There is no f64 version of the reciprocal approximation instructions.
3112 //===----------------------------------------------------------------------===//
3113 // SSE 1 & 2 - Non-temporal stores
3114 //===----------------------------------------------------------------------===//
3116 let AddedComplexity = 400 in { // Prefer non-temporal versions
3117 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3118 (ins f128mem:$dst, VR128:$src),
3119 "movntps\t{$src, $dst|$dst, $src}",
3120 [(alignednontemporalstore (v4f32 VR128:$src),
3122 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3123 (ins f128mem:$dst, VR128:$src),
3124 "movntpd\t{$src, $dst|$dst, $src}",
3125 [(alignednontemporalstore (v2f64 VR128:$src),
3127 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
3128 (ins f128mem:$dst, VR128:$src),
3129 "movntdq\t{$src, $dst|$dst, $src}",
3130 [(alignednontemporalstore (v2f64 VR128:$src),
3133 let ExeDomain = SSEPackedInt in
3134 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3135 (ins f128mem:$dst, VR128:$src),
3136 "movntdq\t{$src, $dst|$dst, $src}",
3137 [(alignednontemporalstore (v4f32 VR128:$src),
3140 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3141 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3143 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3144 (ins f256mem:$dst, VR256:$src),
3145 "movntps\t{$src, $dst|$dst, $src}",
3146 [(alignednontemporalstore (v8f32 VR256:$src),
3148 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3149 (ins f256mem:$dst, VR256:$src),
3150 "movntpd\t{$src, $dst|$dst, $src}",
3151 [(alignednontemporalstore (v4f64 VR256:$src),
3153 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
3154 (ins f256mem:$dst, VR256:$src),
3155 "movntdq\t{$src, $dst|$dst, $src}",
3156 [(alignednontemporalstore (v4f64 VR256:$src),
3158 let ExeDomain = SSEPackedInt in
3159 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3160 (ins f256mem:$dst, VR256:$src),
3161 "movntdq\t{$src, $dst|$dst, $src}",
3162 [(alignednontemporalstore (v8f32 VR256:$src),
3166 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3167 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3168 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3169 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3170 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3171 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3173 let AddedComplexity = 400 in { // Prefer non-temporal versions
3174 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3175 "movntps\t{$src, $dst|$dst, $src}",
3176 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3177 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3178 "movntpd\t{$src, $dst|$dst, $src}",
3179 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3181 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3182 "movntdq\t{$src, $dst|$dst, $src}",
3183 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
3185 let ExeDomain = SSEPackedInt in
3186 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3187 "movntdq\t{$src, $dst|$dst, $src}",
3188 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3190 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3191 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3193 // There is no AVX form for instructions below this point
3194 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3195 "movnti{l}\t{$src, $dst|$dst, $src}",
3196 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3197 TB, Requires<[HasSSE2]>;
3198 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3199 "movnti{q}\t{$src, $dst|$dst, $src}",
3200 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3201 TB, Requires<[HasSSE2]>;
3204 //===----------------------------------------------------------------------===//
3205 // SSE 1 & 2 - Prefetch and memory fence
3206 //===----------------------------------------------------------------------===//
3208 // Prefetch intrinsic.
3209 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
3210 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
3211 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
3212 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
3213 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
3214 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
3215 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
3216 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
3219 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3220 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3221 TB, Requires<[HasSSE2]>;
3223 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3224 // was introduced with SSE2, it's backward compatible.
3225 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3227 // Load, store, and memory fence
3228 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3229 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3230 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3231 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3232 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3233 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3235 def : Pat<(X86SFence), (SFENCE)>;
3236 def : Pat<(X86LFence), (LFENCE)>;
3237 def : Pat<(X86MFence), (MFENCE)>;
3239 //===----------------------------------------------------------------------===//
3240 // SSE 1 & 2 - Load/Store XCSR register
3241 //===----------------------------------------------------------------------===//
3243 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3244 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3245 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3246 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3248 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3249 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3250 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3251 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3253 //===---------------------------------------------------------------------===//
3254 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3255 //===---------------------------------------------------------------------===//
3257 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3259 let neverHasSideEffects = 1 in {
3260 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3261 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3262 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3263 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3265 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3266 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3267 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3268 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3271 let isCodeGenOnly = 1 in {
3272 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3273 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3274 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3275 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3276 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3277 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3278 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3279 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3282 let canFoldAsLoad = 1, mayLoad = 1 in {
3283 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3284 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3285 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3286 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3287 let Predicates = [HasAVX] in {
3288 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3289 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3290 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3291 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3295 let mayStore = 1 in {
3296 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3297 (ins i128mem:$dst, VR128:$src),
3298 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3299 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3300 (ins i256mem:$dst, VR256:$src),
3301 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3302 let Predicates = [HasAVX] in {
3303 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3304 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3305 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3306 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3310 let neverHasSideEffects = 1 in
3311 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3312 "movdqa\t{$src, $dst|$dst, $src}", []>;
3314 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3315 "movdqu\t{$src, $dst|$dst, $src}",
3316 []>, XS, Requires<[HasSSE2]>;
3319 let isCodeGenOnly = 1 in {
3320 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3321 "movdqa\t{$src, $dst|$dst, $src}", []>;
3323 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3324 "movdqu\t{$src, $dst|$dst, $src}",
3325 []>, XS, Requires<[HasSSE2]>;
3328 let canFoldAsLoad = 1, mayLoad = 1 in {
3329 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3330 "movdqa\t{$src, $dst|$dst, $src}",
3331 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3332 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3333 "movdqu\t{$src, $dst|$dst, $src}",
3334 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3335 XS, Requires<[HasSSE2]>;
3338 let mayStore = 1 in {
3339 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3340 "movdqa\t{$src, $dst|$dst, $src}",
3341 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3342 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3343 "movdqu\t{$src, $dst|$dst, $src}",
3344 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3345 XS, Requires<[HasSSE2]>;
3348 // Intrinsic forms of MOVDQU load and store
3349 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3350 "vmovdqu\t{$src, $dst|$dst, $src}",
3351 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3352 XS, VEX, Requires<[HasAVX]>;
3354 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3355 "movdqu\t{$src, $dst|$dst, $src}",
3356 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3357 XS, Requires<[HasSSE2]>;
3359 } // ExeDomain = SSEPackedInt
3361 let Predicates = [HasAVX] in {
3362 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3363 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3364 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3367 //===---------------------------------------------------------------------===//
3368 // SSE2 - Packed Integer Arithmetic Instructions
3369 //===---------------------------------------------------------------------===//
3371 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3373 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3374 bit IsCommutable = 0, bit Is2Addr = 1> {
3375 let isCommutable = IsCommutable in
3376 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
3377 (ins VR128:$src1, VR128:$src2),
3379 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3380 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3381 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
3382 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
3383 (ins VR128:$src1, i128mem:$src2),
3385 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3386 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3387 [(set VR128:$dst, (IntId VR128:$src1,
3388 (bitconvert (memopv2i64 addr:$src2))))]>;
3391 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3392 string OpcodeStr, Intrinsic IntId,
3393 Intrinsic IntId2, bit Is2Addr = 1> {
3394 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
3395 (ins VR128:$src1, VR128:$src2),
3397 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3398 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3399 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
3400 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
3401 (ins VR128:$src1, i128mem:$src2),
3403 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3404 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3405 [(set VR128:$dst, (IntId VR128:$src1,
3406 (bitconvert (memopv2i64 addr:$src2))))]>;
3407 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
3408 (ins VR128:$src1, i32i8imm:$src2),
3410 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3411 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3412 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
3415 /// PDI_binop_rm - Simple SSE2 binary operator.
3416 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
3417 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
3418 let isCommutable = IsCommutable in
3419 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
3420 (ins VR128:$src1, VR128:$src2),
3422 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3423 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3424 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
3425 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
3426 (ins VR128:$src1, i128mem:$src2),
3428 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3429 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3430 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
3431 (bitconvert (memopv2i64 addr:$src2)))))]>;
3434 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
3436 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
3437 /// to collapse (bitconvert VT to VT) into its operand.
3439 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
3440 bit IsCommutable = 0, bit Is2Addr = 1> {
3441 let isCommutable = IsCommutable in
3442 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
3443 (ins VR128:$src1, VR128:$src2),
3445 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3446 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3447 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
3448 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
3449 (ins VR128:$src1, i128mem:$src2),
3451 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3452 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3453 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
3456 } // ExeDomain = SSEPackedInt
3458 // 128-bit Integer Arithmetic
3460 let Predicates = [HasAVX] in {
3461 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
3462 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
3463 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
3464 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
3465 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
3466 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
3467 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
3468 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
3469 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
3472 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
3474 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
3476 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
3478 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
3480 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
3482 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
3484 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
3486 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
3488 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
3490 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
3492 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
3494 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
3496 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
3498 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
3500 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
3502 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
3504 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
3506 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
3508 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
3512 let Constraints = "$src1 = $dst" in {
3513 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
3514 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
3515 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
3516 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
3517 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
3518 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
3519 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
3520 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
3521 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
3524 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
3525 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
3526 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
3527 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
3528 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
3529 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
3530 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
3531 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
3532 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
3533 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
3534 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
3535 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
3536 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
3537 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
3538 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
3539 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
3540 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
3541 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
3542 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
3544 } // Constraints = "$src1 = $dst"
3546 //===---------------------------------------------------------------------===//
3547 // SSE2 - Packed Integer Logical Instructions
3548 //===---------------------------------------------------------------------===//
3550 let Predicates = [HasAVX] in {
3551 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3552 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
3554 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3555 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
3557 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3558 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
3561 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3562 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
3564 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3565 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
3567 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3568 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
3571 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3572 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
3574 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3575 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
3578 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
3579 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
3580 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
3582 let ExeDomain = SSEPackedInt in {
3583 let neverHasSideEffects = 1 in {
3584 // 128-bit logical shifts.
3585 def VPSLLDQri : PDIi8<0x73, MRM7r,
3586 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3587 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3589 def VPSRLDQri : PDIi8<0x73, MRM3r,
3590 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3591 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3593 // PSRADQri doesn't exist in SSE[1-3].
3595 def VPANDNrr : PDI<0xDF, MRMSrcReg,
3596 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3597 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3599 (v2i64 (X86andnp VR128:$src1, VR128:$src2)))]>,VEX_4V;
3601 def VPANDNrm : PDI<0xDF, MRMSrcMem,
3602 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3603 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3604 [(set VR128:$dst, (X86andnp VR128:$src1,
3605 (memopv2i64 addr:$src2)))]>, VEX_4V;
3609 let Constraints = "$src1 = $dst" in {
3610 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3611 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
3612 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3613 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
3614 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3615 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
3617 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3618 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
3619 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3620 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
3621 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3622 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
3624 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3625 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
3626 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3627 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
3629 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
3630 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
3631 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
3633 let ExeDomain = SSEPackedInt in {
3634 let neverHasSideEffects = 1 in {
3635 // 128-bit logical shifts.
3636 def PSLLDQri : PDIi8<0x73, MRM7r,
3637 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3638 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3639 def PSRLDQri : PDIi8<0x73, MRM3r,
3640 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3641 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3642 // PSRADQri doesn't exist in SSE[1-3].
3644 def PANDNrr : PDI<0xDF, MRMSrcReg,
3645 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3646 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3648 def PANDNrm : PDI<0xDF, MRMSrcMem,
3649 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3650 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3652 } // Constraints = "$src1 = $dst"
3654 let Predicates = [HasAVX] in {
3655 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3656 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3657 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3658 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3659 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3660 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
3661 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3662 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
3663 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3664 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3666 // Shift up / down and insert zero's.
3667 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3668 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3669 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3670 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3673 let Predicates = [HasSSE2] in {
3674 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3675 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3676 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3677 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3678 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3679 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
3680 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3681 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
3682 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3683 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3685 // Shift up / down and insert zero's.
3686 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3687 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3688 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3689 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3692 //===---------------------------------------------------------------------===//
3693 // SSE2 - Packed Integer Comparison Instructions
3694 //===---------------------------------------------------------------------===//
3696 let Predicates = [HasAVX] in {
3697 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
3699 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
3701 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
3703 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
3705 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
3707 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
3710 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3711 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3712 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3713 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3714 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3715 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3716 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3717 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3718 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3719 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3720 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3721 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3723 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3724 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3725 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3726 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3727 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3728 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3729 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3730 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3731 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3732 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3733 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3734 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3737 let Constraints = "$src1 = $dst" in {
3738 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
3739 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
3740 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
3741 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
3742 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
3743 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
3744 } // Constraints = "$src1 = $dst"
3746 let Predicates = [HasSSE2] in {
3747 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3748 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
3749 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3750 (PCMPEQBrm VR128:$src1, addr:$src2)>;
3751 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3752 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
3753 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3754 (PCMPEQWrm VR128:$src1, addr:$src2)>;
3755 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3756 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
3757 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3758 (PCMPEQDrm VR128:$src1, addr:$src2)>;
3760 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3761 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
3762 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3763 (PCMPGTBrm VR128:$src1, addr:$src2)>;
3764 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3765 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
3766 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3767 (PCMPGTWrm VR128:$src1, addr:$src2)>;
3768 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3769 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
3770 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3771 (PCMPGTDrm VR128:$src1, addr:$src2)>;
3774 //===---------------------------------------------------------------------===//
3775 // SSE2 - Packed Integer Pack Instructions
3776 //===---------------------------------------------------------------------===//
3778 let Predicates = [HasAVX] in {
3779 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3781 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3783 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3787 let Constraints = "$src1 = $dst" in {
3788 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
3789 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
3790 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
3791 } // Constraints = "$src1 = $dst"
3793 //===---------------------------------------------------------------------===//
3794 // SSE2 - Packed Integer Shuffle Instructions
3795 //===---------------------------------------------------------------------===//
3797 let ExeDomain = SSEPackedInt in {
3798 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
3800 def ri : Ii8<0x70, MRMSrcReg,
3801 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3802 !strconcat(OpcodeStr,
3803 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3804 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
3806 def mi : Ii8<0x70, MRMSrcMem,
3807 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3808 !strconcat(OpcodeStr,
3809 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3810 [(set VR128:$dst, (vt (pshuf_frag:$src2
3811 (bc_frag (memopv2i64 addr:$src1)),
3814 } // ExeDomain = SSEPackedInt
3816 let Predicates = [HasAVX] in {
3817 let AddedComplexity = 5 in
3818 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
3821 // SSE2 with ImmT == Imm8 and XS prefix.
3822 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
3825 // SSE2 with ImmT == Imm8 and XD prefix.
3826 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
3829 let AddedComplexity = 5 in
3830 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3831 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3832 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
3833 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3834 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3836 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3838 (VPSHUFDmi addr:$src1, imm:$imm)>;
3839 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
3841 (VPSHUFDmi addr:$src1, imm:$imm)>;
3842 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3843 (VPSHUFDri VR128:$src1, imm:$imm)>;
3844 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3845 (VPSHUFDri VR128:$src1, imm:$imm)>;
3846 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3847 (VPSHUFHWri VR128:$src, imm:$imm)>;
3848 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3850 (VPSHUFHWmi addr:$src, imm:$imm)>;
3851 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3852 (VPSHUFLWri VR128:$src, imm:$imm)>;
3853 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3855 (VPSHUFLWmi addr:$src, imm:$imm)>;
3858 let Predicates = [HasSSE2] in {
3859 let AddedComplexity = 5 in
3860 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
3862 // SSE2 with ImmT == Imm8 and XS prefix.
3863 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
3865 // SSE2 with ImmT == Imm8 and XD prefix.
3866 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
3868 let AddedComplexity = 5 in
3869 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3870 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3871 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3872 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3873 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3875 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3877 (PSHUFDmi addr:$src1, imm:$imm)>;
3878 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
3880 (PSHUFDmi addr:$src1, imm:$imm)>;
3881 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3882 (PSHUFDri VR128:$src1, imm:$imm)>;
3883 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3884 (PSHUFDri VR128:$src1, imm:$imm)>;
3885 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3886 (PSHUFHWri VR128:$src, imm:$imm)>;
3887 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3889 (PSHUFHWmi addr:$src, imm:$imm)>;
3890 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3891 (PSHUFLWri VR128:$src, imm:$imm)>;
3892 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3894 (PSHUFLWmi addr:$src, imm:$imm)>;
3897 //===---------------------------------------------------------------------===//
3898 // SSE2 - Packed Integer Unpack Instructions
3899 //===---------------------------------------------------------------------===//
3901 let ExeDomain = SSEPackedInt in {
3902 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
3903 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
3904 def rr : PDI<opc, MRMSrcReg,
3905 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3907 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3908 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3909 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
3910 def rm : PDI<opc, MRMSrcMem,
3911 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3913 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3914 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3915 [(set VR128:$dst, (OpNode VR128:$src1,
3916 (bc_frag (memopv2i64
3920 let Predicates = [HasAVX] in {
3921 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Punpcklbw,
3922 bc_v16i8, 0>, VEX_4V;
3923 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Punpcklwd,
3924 bc_v8i16, 0>, VEX_4V;
3925 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Punpckldq,
3926 bc_v4i32, 0>, VEX_4V;
3928 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3929 /// knew to collapse (bitconvert VT to VT) into its operand.
3930 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
3931 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3932 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3933 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
3934 VR128:$src2)))]>, VEX_4V;
3935 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
3936 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3937 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3938 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
3939 (memopv2i64 addr:$src2))))]>, VEX_4V;
3941 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Punpckhbw,
3942 bc_v16i8, 0>, VEX_4V;
3943 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Punpckhwd,
3944 bc_v8i16, 0>, VEX_4V;
3945 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Punpckhdq,
3946 bc_v4i32, 0>, VEX_4V;
3948 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3949 /// knew to collapse (bitconvert VT to VT) into its operand.
3950 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
3951 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3952 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3953 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
3954 VR128:$src2)))]>, VEX_4V;
3955 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
3956 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3957 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3958 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
3959 (memopv2i64 addr:$src2))))]>, VEX_4V;
3962 let Constraints = "$src1 = $dst" in {
3963 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Punpcklbw, bc_v16i8>;
3964 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Punpcklwd, bc_v8i16>;
3965 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Punpckldq, bc_v4i32>;
3967 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3968 /// knew to collapse (bitconvert VT to VT) into its operand.
3969 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
3970 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3971 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
3973 (v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)))]>;
3974 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
3975 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3976 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
3978 (v2i64 (X86Punpcklqdq VR128:$src1,
3979 (memopv2i64 addr:$src2))))]>;
3981 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Punpckhbw, bc_v16i8>;
3982 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Punpckhwd, bc_v8i16>;
3983 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Punpckhdq, bc_v4i32>;
3985 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3986 /// knew to collapse (bitconvert VT to VT) into its operand.
3987 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
3988 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3989 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
3991 (v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)))]>;
3992 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
3993 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3994 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
3996 (v2i64 (X86Punpckhqdq VR128:$src1,
3997 (memopv2i64 addr:$src2))))]>;
3999 } // ExeDomain = SSEPackedInt
4001 // Splat v2f64 / v2i64
4002 let AddedComplexity = 10 in {
4003 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4004 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4005 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4006 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4009 //===---------------------------------------------------------------------===//
4010 // SSE2 - Packed Integer Extract and Insert
4011 //===---------------------------------------------------------------------===//
4013 let ExeDomain = SSEPackedInt in {
4014 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4015 def rri : Ii8<0xC4, MRMSrcReg,
4016 (outs VR128:$dst), (ins VR128:$src1,
4017 GR32:$src2, i32i8imm:$src3),
4019 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4020 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4022 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4023 def rmi : Ii8<0xC4, MRMSrcMem,
4024 (outs VR128:$dst), (ins VR128:$src1,
4025 i16mem:$src2, i32i8imm:$src3),
4027 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4028 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4030 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4035 let Predicates = [HasAVX] in
4036 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4037 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4038 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4039 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4040 imm:$src2))]>, TB, OpSize, VEX;
4041 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4042 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4043 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4044 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4048 let Predicates = [HasAVX] in {
4049 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4050 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4051 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4052 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4053 []>, TB, OpSize, VEX_4V;
4056 let Constraints = "$src1 = $dst" in
4057 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4059 } // ExeDomain = SSEPackedInt
4061 //===---------------------------------------------------------------------===//
4062 // SSE2 - Packed Mask Creation
4063 //===---------------------------------------------------------------------===//
4065 let ExeDomain = SSEPackedInt in {
4067 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4068 "pmovmskb\t{$src, $dst|$dst, $src}",
4069 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4070 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4071 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4072 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4073 "pmovmskb\t{$src, $dst|$dst, $src}",
4074 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4076 } // ExeDomain = SSEPackedInt
4078 //===---------------------------------------------------------------------===//
4079 // SSE2 - Conditional Store
4080 //===---------------------------------------------------------------------===//
4082 let ExeDomain = SSEPackedInt in {
4085 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4086 (ins VR128:$src, VR128:$mask),
4087 "maskmovdqu\t{$mask, $src|$src, $mask}",
4088 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4090 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4091 (ins VR128:$src, VR128:$mask),
4092 "maskmovdqu\t{$mask, $src|$src, $mask}",
4093 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4096 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4097 "maskmovdqu\t{$mask, $src|$src, $mask}",
4098 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4100 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4101 "maskmovdqu\t{$mask, $src|$src, $mask}",
4102 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4104 } // ExeDomain = SSEPackedInt
4106 //===---------------------------------------------------------------------===//
4107 // SSE2 - Move Doubleword
4108 //===---------------------------------------------------------------------===//
4110 //===---------------------------------------------------------------------===//
4111 // Move Int Doubleword to Packed Double Int
4113 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4114 "movd\t{$src, $dst|$dst, $src}",
4116 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4117 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4118 "movd\t{$src, $dst|$dst, $src}",
4120 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4122 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4123 "mov{d|q}\t{$src, $dst|$dst, $src}",
4125 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4126 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4127 "mov{d|q}\t{$src, $dst|$dst, $src}",
4128 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4130 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4131 "movd\t{$src, $dst|$dst, $src}",
4133 (v4i32 (scalar_to_vector GR32:$src)))]>;
4134 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4135 "movd\t{$src, $dst|$dst, $src}",
4137 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4138 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4139 "mov{d|q}\t{$src, $dst|$dst, $src}",
4141 (v2i64 (scalar_to_vector GR64:$src)))]>;
4142 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4143 "mov{d|q}\t{$src, $dst|$dst, $src}",
4144 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4146 //===---------------------------------------------------------------------===//
4147 // Move Int Doubleword to Single Scalar
4149 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4150 "movd\t{$src, $dst|$dst, $src}",
4151 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4153 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4154 "movd\t{$src, $dst|$dst, $src}",
4155 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4157 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4158 "movd\t{$src, $dst|$dst, $src}",
4159 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4161 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4162 "movd\t{$src, $dst|$dst, $src}",
4163 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4165 //===---------------------------------------------------------------------===//
4166 // Move Packed Doubleword Int to Packed Double Int
4168 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4169 "movd\t{$src, $dst|$dst, $src}",
4170 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4172 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4173 (ins i32mem:$dst, VR128:$src),
4174 "movd\t{$src, $dst|$dst, $src}",
4175 [(store (i32 (vector_extract (v4i32 VR128:$src),
4176 (iPTR 0))), addr:$dst)]>, VEX;
4177 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4178 "movd\t{$src, $dst|$dst, $src}",
4179 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4181 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4182 "movd\t{$src, $dst|$dst, $src}",
4183 [(store (i32 (vector_extract (v4i32 VR128:$src),
4184 (iPTR 0))), addr:$dst)]>;
4186 //===---------------------------------------------------------------------===//
4187 // Move Packed Doubleword Int first element to Doubleword Int
4189 let isCodeGenOnly = 1 in
4190 def VMOVPQIto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4191 "mov{d|q}\t{$src, $dst|$dst, $src}",
4192 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4195 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4196 "mov{d|q}\t{$src, $dst|$dst, $src}",
4197 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4200 //===---------------------------------------------------------------------===//
4201 // Bitcast FR64 <-> GR64
4203 let Predicates = [HasAVX] in
4204 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4205 "vmovq\t{$src, $dst|$dst, $src}",
4206 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4208 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4209 "mov{d|q}\t{$src, $dst|$dst, $src}",
4210 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4211 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4212 "movq\t{$src, $dst|$dst, $src}",
4213 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4215 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4216 "movq\t{$src, $dst|$dst, $src}",
4217 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4218 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4219 "mov{d|q}\t{$src, $dst|$dst, $src}",
4220 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4221 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4222 "movq\t{$src, $dst|$dst, $src}",
4223 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4225 //===---------------------------------------------------------------------===//
4226 // Move Scalar Single to Double Int
4228 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4229 "movd\t{$src, $dst|$dst, $src}",
4230 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4231 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4232 "movd\t{$src, $dst|$dst, $src}",
4233 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4234 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4235 "movd\t{$src, $dst|$dst, $src}",
4236 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4237 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4238 "movd\t{$src, $dst|$dst, $src}",
4239 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4241 //===---------------------------------------------------------------------===//
4242 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4244 let AddedComplexity = 15 in {
4245 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4246 "movd\t{$src, $dst|$dst, $src}",
4247 [(set VR128:$dst, (v4i32 (X86vzmovl
4248 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4250 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4251 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4252 [(set VR128:$dst, (v2i64 (X86vzmovl
4253 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4256 let AddedComplexity = 15 in {
4257 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4258 "movd\t{$src, $dst|$dst, $src}",
4259 [(set VR128:$dst, (v4i32 (X86vzmovl
4260 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4261 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4262 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4263 [(set VR128:$dst, (v2i64 (X86vzmovl
4264 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4267 let AddedComplexity = 20 in {
4268 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4269 "movd\t{$src, $dst|$dst, $src}",
4271 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4272 (loadi32 addr:$src))))))]>,
4274 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4275 "movd\t{$src, $dst|$dst, $src}",
4277 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4278 (loadi32 addr:$src))))))]>;
4281 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4282 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4283 (MOVZDI2PDIrm addr:$src)>;
4284 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4285 (MOVZDI2PDIrm addr:$src)>;
4286 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4287 (MOVZDI2PDIrm addr:$src)>;
4290 let Predicates = [HasAVX] in {
4291 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4292 let AddedComplexity = 20 in {
4293 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4294 (VMOVZDI2PDIrm addr:$src)>;
4295 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4296 (VMOVZDI2PDIrm addr:$src)>;
4297 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4298 (VMOVZDI2PDIrm addr:$src)>;
4300 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4301 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4302 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4303 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4304 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4305 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4306 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4309 // These are the correct encodings of the instructions so that we know how to
4310 // read correct assembly, even though we continue to emit the wrong ones for
4311 // compatibility with Darwin's buggy assembler.
4312 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4313 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4314 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4315 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4316 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4317 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4318 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4319 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4320 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4321 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4322 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4323 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4325 //===---------------------------------------------------------------------===//
4326 // SSE2 - Move Quadword
4327 //===---------------------------------------------------------------------===//
4329 //===---------------------------------------------------------------------===//
4330 // Move Quadword Int to Packed Quadword Int
4332 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4333 "vmovq\t{$src, $dst|$dst, $src}",
4335 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4336 VEX, Requires<[HasAVX]>;
4337 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4338 "movq\t{$src, $dst|$dst, $src}",
4340 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4341 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4343 //===---------------------------------------------------------------------===//
4344 // Move Packed Quadword Int to Quadword Int
4346 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4347 "movq\t{$src, $dst|$dst, $src}",
4348 [(store (i64 (vector_extract (v2i64 VR128:$src),
4349 (iPTR 0))), addr:$dst)]>, VEX;
4350 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4351 "movq\t{$src, $dst|$dst, $src}",
4352 [(store (i64 (vector_extract (v2i64 VR128:$src),
4353 (iPTR 0))), addr:$dst)]>;
4355 //===---------------------------------------------------------------------===//
4356 // Store / copy lower 64-bits of a XMM register.
4358 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4359 "movq\t{$src, $dst|$dst, $src}",
4360 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4361 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4362 "movq\t{$src, $dst|$dst, $src}",
4363 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4365 let AddedComplexity = 20 in
4366 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4367 "vmovq\t{$src, $dst|$dst, $src}",
4369 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4370 (loadi64 addr:$src))))))]>,
4371 XS, VEX, Requires<[HasAVX]>;
4373 let AddedComplexity = 20 in
4374 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4375 "movq\t{$src, $dst|$dst, $src}",
4377 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4378 (loadi64 addr:$src))))))]>,
4379 XS, Requires<[HasSSE2]>;
4381 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4382 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4383 (MOVZQI2PQIrm addr:$src)>;
4384 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4385 (MOVZQI2PQIrm addr:$src)>;
4386 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4389 let Predicates = [HasAVX], AddedComplexity = 20 in {
4390 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4391 (VMOVZQI2PQIrm addr:$src)>;
4392 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4393 (VMOVZQI2PQIrm addr:$src)>;
4394 def : Pat<(v2i64 (X86vzload addr:$src)),
4395 (VMOVZQI2PQIrm addr:$src)>;
4398 //===---------------------------------------------------------------------===//
4399 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4400 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4402 let AddedComplexity = 15 in
4403 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4404 "vmovq\t{$src, $dst|$dst, $src}",
4405 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4406 XS, VEX, Requires<[HasAVX]>;
4407 let AddedComplexity = 15 in
4408 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4409 "movq\t{$src, $dst|$dst, $src}",
4410 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4411 XS, Requires<[HasSSE2]>;
4413 let AddedComplexity = 20 in
4414 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4415 "vmovq\t{$src, $dst|$dst, $src}",
4416 [(set VR128:$dst, (v2i64 (X86vzmovl
4417 (loadv2i64 addr:$src))))]>,
4418 XS, VEX, Requires<[HasAVX]>;
4419 let AddedComplexity = 20 in {
4420 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4421 "movq\t{$src, $dst|$dst, $src}",
4422 [(set VR128:$dst, (v2i64 (X86vzmovl
4423 (loadv2i64 addr:$src))))]>,
4424 XS, Requires<[HasSSE2]>;
4427 let AddedComplexity = 20 in {
4428 let Predicates = [HasSSE2] in {
4429 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4430 (MOVZPQILo2PQIrm addr:$src)>;
4431 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4432 (MOVZPQILo2PQIrr VR128:$src)>;
4434 let Predicates = [HasAVX] in {
4435 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4436 (VMOVZPQILo2PQIrm addr:$src)>;
4437 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4438 (VMOVZPQILo2PQIrr VR128:$src)>;
4442 // Instructions to match in the assembler
4443 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4444 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4445 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4446 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4447 // Recognize "movd" with GR64 destination, but encode as a "movq"
4448 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4449 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4451 // Instructions for the disassembler
4452 // xr = XMM register
4455 let Predicates = [HasAVX] in
4456 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4457 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4458 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4459 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4461 //===---------------------------------------------------------------------===//
4462 // SSE3 - Conversion Instructions
4463 //===---------------------------------------------------------------------===//
4465 // Convert Packed Double FP to Packed DW Integers
4466 let Predicates = [HasAVX] in {
4467 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4468 // register, but the same isn't true when using memory operands instead.
4469 // Provide other assembly rr and rm forms to address this explicitly.
4470 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4471 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4472 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4473 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4476 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4477 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4478 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4479 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4482 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4483 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4484 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4485 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4488 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4489 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4490 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4491 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4493 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4494 (VCVTPD2DQYrr VR256:$src)>;
4495 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4496 (VCVTPD2DQYrm addr:$src)>;
4498 // Convert Packed DW Integers to Packed Double FP
4499 let Predicates = [HasAVX] in {
4500 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4501 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4502 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4503 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4504 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4505 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4506 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4507 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4510 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4511 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4512 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4513 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4515 // AVX 256-bit register conversion intrinsics
4516 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4517 (VCVTDQ2PDYrr VR128:$src)>;
4518 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
4519 (VCVTDQ2PDYrm addr:$src)>;
4521 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4522 (VCVTPD2DQYrr VR256:$src)>;
4523 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4524 (VCVTPD2DQYrm addr:$src)>;
4526 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4527 (VCVTDQ2PDYrr VR128:$src)>;
4528 def : Pat<(v4f64 (sint_to_fp (memopv4i32 addr:$src))),
4529 (VCVTDQ2PDYrm addr:$src)>;
4531 //===---------------------------------------------------------------------===//
4532 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4533 //===---------------------------------------------------------------------===//
4534 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4535 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4536 X86MemOperand x86memop> {
4537 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4538 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4539 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4540 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4541 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4542 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4545 let Predicates = [HasAVX] in {
4546 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4547 v4f32, VR128, memopv4f32, f128mem>, VEX;
4548 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4549 v4f32, VR128, memopv4f32, f128mem>, VEX;
4550 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4551 v8f32, VR256, memopv8f32, f256mem>, VEX;
4552 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4553 v8f32, VR256, memopv8f32, f256mem>, VEX;
4555 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4556 memopv4f32, f128mem>;
4557 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4558 memopv4f32, f128mem>;
4560 let Predicates = [HasSSE3] in {
4561 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4562 (MOVSHDUPrr VR128:$src)>;
4563 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4564 (MOVSHDUPrm addr:$src)>;
4565 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4566 (MOVSLDUPrr VR128:$src)>;
4567 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4568 (MOVSLDUPrm addr:$src)>;
4571 let Predicates = [HasAVX] in {
4572 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4573 (VMOVSHDUPrr VR128:$src)>;
4574 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4575 (VMOVSHDUPrm addr:$src)>;
4576 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4577 (VMOVSLDUPrr VR128:$src)>;
4578 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4579 (VMOVSLDUPrm addr:$src)>;
4580 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4581 (VMOVSHDUPYrr VR256:$src)>;
4582 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4583 (VMOVSHDUPYrm addr:$src)>;
4584 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4585 (VMOVSLDUPYrr VR256:$src)>;
4586 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4587 (VMOVSLDUPYrm addr:$src)>;
4590 //===---------------------------------------------------------------------===//
4591 // SSE3 - Replicate Double FP - MOVDDUP
4592 //===---------------------------------------------------------------------===//
4594 multiclass sse3_replicate_dfp<string OpcodeStr> {
4595 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4596 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4597 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4598 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4599 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4601 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4605 // FIXME: Merge with above classe when there're patterns for the ymm version
4606 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4607 let Predicates = [HasAVX] in {
4608 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4609 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4611 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4612 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4617 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4618 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4619 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4621 let Predicates = [HasSSE3] in {
4622 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4624 (MOVDDUPrm addr:$src)>;
4625 let AddedComplexity = 5 in {
4626 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4627 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4628 (MOVDDUPrm addr:$src)>;
4629 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4630 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4631 (MOVDDUPrm addr:$src)>;
4633 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4634 (MOVDDUPrm addr:$src)>;
4635 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4636 (MOVDDUPrm addr:$src)>;
4637 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4638 (MOVDDUPrm addr:$src)>;
4639 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4640 (MOVDDUPrm addr:$src)>;
4641 def : Pat<(X86Movddup (bc_v2f64
4642 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4643 (MOVDDUPrm addr:$src)>;
4646 let Predicates = [HasAVX] in {
4647 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4649 (VMOVDDUPrm addr:$src)>;
4650 let AddedComplexity = 5 in {
4651 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4652 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4653 (VMOVDDUPrm addr:$src)>;
4654 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4655 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4656 (VMOVDDUPrm addr:$src)>;
4658 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4659 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4660 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4661 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4662 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4663 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4664 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4665 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4666 def : Pat<(X86Movddup (bc_v2f64
4667 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4668 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4671 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4672 (VMOVDDUPYrm addr:$src)>;
4673 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4674 (VMOVDDUPYrm addr:$src)>;
4675 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
4676 (VMOVDDUPYrm addr:$src)>;
4677 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4678 (VMOVDDUPYrm addr:$src)>;
4679 def : Pat<(X86Movddup (v4f64 VR256:$src)),
4680 (VMOVDDUPYrr VR256:$src)>;
4681 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4682 (VMOVDDUPYrr VR256:$src)>;
4685 //===---------------------------------------------------------------------===//
4686 // SSE3 - Move Unaligned Integer
4687 //===---------------------------------------------------------------------===//
4689 let Predicates = [HasAVX] in {
4690 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4691 "vlddqu\t{$src, $dst|$dst, $src}",
4692 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4693 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4694 "vlddqu\t{$src, $dst|$dst, $src}",
4695 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4697 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4698 "lddqu\t{$src, $dst|$dst, $src}",
4699 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
4701 //===---------------------------------------------------------------------===//
4702 // SSE3 - Arithmetic
4703 //===---------------------------------------------------------------------===//
4705 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4706 X86MemOperand x86memop, bit Is2Addr = 1> {
4707 def rr : I<0xD0, MRMSrcReg,
4708 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4710 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4711 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4712 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
4713 def rm : I<0xD0, MRMSrcMem,
4714 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4716 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4717 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4718 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
4721 let Predicates = [HasAVX],
4722 ExeDomain = SSEPackedDouble in {
4723 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4724 f128mem, 0>, TB, XD, VEX_4V;
4725 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4726 f128mem, 0>, TB, OpSize, VEX_4V;
4727 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4728 f256mem, 0>, TB, XD, VEX_4V;
4729 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4730 f256mem, 0>, TB, OpSize, VEX_4V;
4732 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
4733 ExeDomain = SSEPackedDouble in {
4734 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4736 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4737 f128mem>, TB, OpSize;
4740 //===---------------------------------------------------------------------===//
4741 // SSE3 Instructions
4742 //===---------------------------------------------------------------------===//
4745 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4746 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
4747 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4749 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4750 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4751 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
4753 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4755 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4756 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4757 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
4759 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4760 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
4761 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4763 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4764 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4765 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
4767 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4769 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4770 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4771 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
4774 let Predicates = [HasAVX] in {
4775 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4776 int_x86_sse3_hadd_ps, 0>, VEX_4V;
4777 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4778 int_x86_sse3_hadd_pd, 0>, VEX_4V;
4779 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4780 int_x86_sse3_hsub_ps, 0>, VEX_4V;
4781 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4782 int_x86_sse3_hsub_pd, 0>, VEX_4V;
4783 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4784 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
4785 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4786 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
4787 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4788 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
4789 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4790 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
4793 let Constraints = "$src1 = $dst" in {
4794 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
4795 int_x86_sse3_hadd_ps>;
4796 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
4797 int_x86_sse3_hadd_pd>;
4798 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
4799 int_x86_sse3_hsub_ps>;
4800 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
4801 int_x86_sse3_hsub_pd>;
4804 //===---------------------------------------------------------------------===//
4805 // SSSE3 - Packed Absolute Instructions
4806 //===---------------------------------------------------------------------===//
4809 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4810 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4811 PatFrag mem_frag128, Intrinsic IntId128> {
4812 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4814 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4815 [(set VR128:$dst, (IntId128 VR128:$src))]>,
4818 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4820 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4823 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
4826 let Predicates = [HasAVX] in {
4827 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
4828 int_x86_ssse3_pabs_b_128>, VEX;
4829 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
4830 int_x86_ssse3_pabs_w_128>, VEX;
4831 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
4832 int_x86_ssse3_pabs_d_128>, VEX;
4835 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
4836 int_x86_ssse3_pabs_b_128>;
4837 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
4838 int_x86_ssse3_pabs_w_128>;
4839 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
4840 int_x86_ssse3_pabs_d_128>;
4842 //===---------------------------------------------------------------------===//
4843 // SSSE3 - Packed Binary Operator Instructions
4844 //===---------------------------------------------------------------------===//
4846 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
4847 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
4848 PatFrag mem_frag128, Intrinsic IntId128,
4850 let isCommutable = 1 in
4851 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4852 (ins VR128:$src1, VR128:$src2),
4854 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4855 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4856 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4858 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4859 (ins VR128:$src1, i128mem:$src2),
4861 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4862 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4864 (IntId128 VR128:$src1,
4865 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4868 let Predicates = [HasAVX] in {
4869 let isCommutable = 0 in {
4870 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
4871 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
4872 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
4873 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
4874 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
4875 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
4876 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
4877 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
4878 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
4879 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
4880 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
4881 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
4882 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
4883 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
4884 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
4885 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
4886 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
4887 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
4888 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
4889 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
4890 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
4891 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
4893 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
4894 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
4897 // None of these have i8 immediate fields.
4898 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
4899 let isCommutable = 0 in {
4900 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
4901 int_x86_ssse3_phadd_w_128>;
4902 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
4903 int_x86_ssse3_phadd_d_128>;
4904 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
4905 int_x86_ssse3_phadd_sw_128>;
4906 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
4907 int_x86_ssse3_phsub_w_128>;
4908 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
4909 int_x86_ssse3_phsub_d_128>;
4910 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
4911 int_x86_ssse3_phsub_sw_128>;
4912 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
4913 int_x86_ssse3_pmadd_ub_sw_128>;
4914 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
4915 int_x86_ssse3_pshuf_b_128>;
4916 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
4917 int_x86_ssse3_psign_b_128>;
4918 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
4919 int_x86_ssse3_psign_w_128>;
4920 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
4921 int_x86_ssse3_psign_d_128>;
4923 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
4924 int_x86_ssse3_pmul_hr_sw_128>;
4927 let Predicates = [HasSSSE3] in {
4928 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
4929 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
4930 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
4931 (PSHUFBrm128 VR128:$src, addr:$mask)>;
4933 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
4934 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
4935 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
4936 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
4937 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
4938 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
4941 let Predicates = [HasAVX] in {
4942 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
4943 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
4944 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
4945 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
4947 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
4948 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
4949 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
4950 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
4951 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
4952 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
4955 //===---------------------------------------------------------------------===//
4956 // SSSE3 - Packed Align Instruction Patterns
4957 //===---------------------------------------------------------------------===//
4959 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
4960 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
4961 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4963 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4965 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4967 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
4968 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4970 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4972 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4976 let Predicates = [HasAVX] in
4977 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
4978 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
4979 defm PALIGN : ssse3_palign<"palignr">;
4981 let Predicates = [HasSSSE3] in {
4982 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4983 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4984 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4985 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4986 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4987 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4988 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4989 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4992 let Predicates = [HasAVX] in {
4993 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4994 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4995 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4996 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4997 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4998 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4999 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5000 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5003 //===---------------------------------------------------------------------===//
5004 // SSSE3 - Thread synchronization
5005 //===---------------------------------------------------------------------===//
5007 let usesCustomInserter = 1 in {
5008 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5009 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
5010 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5011 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
5014 let Uses = [EAX, ECX, EDX] in
5015 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5016 Requires<[HasSSE3]>;
5017 let Uses = [ECX, EAX] in
5018 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5019 Requires<[HasSSE3]>;
5021 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5022 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5024 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5025 Requires<[In32BitMode]>;
5026 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5027 Requires<[In64BitMode]>;
5029 //===----------------------------------------------------------------------===//
5030 // SSE4.1 - Packed Move with Sign/Zero Extend
5031 //===----------------------------------------------------------------------===//
5033 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5034 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5035 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5036 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5038 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5039 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5041 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5045 let Predicates = [HasAVX] in {
5046 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5048 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5050 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5052 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5054 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5056 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5060 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5061 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5062 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5063 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5064 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5065 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5067 let Predicates = [HasSSE41] in {
5068 // Common patterns involving scalar load.
5069 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5070 (PMOVSXBWrm addr:$src)>;
5071 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5072 (PMOVSXBWrm addr:$src)>;
5074 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5075 (PMOVSXWDrm addr:$src)>;
5076 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5077 (PMOVSXWDrm addr:$src)>;
5079 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5080 (PMOVSXDQrm addr:$src)>;
5081 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5082 (PMOVSXDQrm addr:$src)>;
5084 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5085 (PMOVZXBWrm addr:$src)>;
5086 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5087 (PMOVZXBWrm addr:$src)>;
5089 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5090 (PMOVZXWDrm addr:$src)>;
5091 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5092 (PMOVZXWDrm addr:$src)>;
5094 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5095 (PMOVZXDQrm addr:$src)>;
5096 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5097 (PMOVZXDQrm addr:$src)>;
5100 let Predicates = [HasAVX] in {
5101 // Common patterns involving scalar load.
5102 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5103 (VPMOVSXBWrm addr:$src)>;
5104 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5105 (VPMOVSXBWrm addr:$src)>;
5107 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5108 (VPMOVSXWDrm addr:$src)>;
5109 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5110 (VPMOVSXWDrm addr:$src)>;
5112 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5113 (VPMOVSXDQrm addr:$src)>;
5114 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5115 (VPMOVSXDQrm addr:$src)>;
5117 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5118 (VPMOVZXBWrm addr:$src)>;
5119 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5120 (VPMOVZXBWrm addr:$src)>;
5122 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5123 (VPMOVZXWDrm addr:$src)>;
5124 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5125 (VPMOVZXWDrm addr:$src)>;
5127 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5128 (VPMOVZXDQrm addr:$src)>;
5129 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5130 (VPMOVZXDQrm addr:$src)>;
5134 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5135 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5136 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5137 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5139 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5140 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5142 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5146 let Predicates = [HasAVX] in {
5147 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5149 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5151 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5153 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5157 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5158 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5159 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5160 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5162 let Predicates = [HasSSE41] in {
5163 // Common patterns involving scalar load
5164 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5165 (PMOVSXBDrm addr:$src)>;
5166 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5167 (PMOVSXWQrm addr:$src)>;
5169 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5170 (PMOVZXBDrm addr:$src)>;
5171 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5172 (PMOVZXWQrm addr:$src)>;
5175 let Predicates = [HasAVX] in {
5176 // Common patterns involving scalar load
5177 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5178 (VPMOVSXBDrm addr:$src)>;
5179 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5180 (VPMOVSXWQrm addr:$src)>;
5182 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5183 (VPMOVZXBDrm addr:$src)>;
5184 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5185 (VPMOVZXWQrm addr:$src)>;
5188 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5189 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5190 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5191 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5193 // Expecting a i16 load any extended to i32 value.
5194 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5195 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5196 [(set VR128:$dst, (IntId (bitconvert
5197 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5201 let Predicates = [HasAVX] in {
5202 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5204 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5207 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5208 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5210 let Predicates = [HasSSE41] in {
5211 // Common patterns involving scalar load
5212 def : Pat<(int_x86_sse41_pmovsxbq
5213 (bitconvert (v4i32 (X86vzmovl
5214 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5215 (PMOVSXBQrm addr:$src)>;
5217 def : Pat<(int_x86_sse41_pmovzxbq
5218 (bitconvert (v4i32 (X86vzmovl
5219 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5220 (PMOVZXBQrm addr:$src)>;
5223 let Predicates = [HasAVX] in {
5224 // Common patterns involving scalar load
5225 def : Pat<(int_x86_sse41_pmovsxbq
5226 (bitconvert (v4i32 (X86vzmovl
5227 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5228 (VPMOVSXBQrm addr:$src)>;
5230 def : Pat<(int_x86_sse41_pmovzxbq
5231 (bitconvert (v4i32 (X86vzmovl
5232 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5233 (VPMOVZXBQrm addr:$src)>;
5236 //===----------------------------------------------------------------------===//
5237 // SSE4.1 - Extract Instructions
5238 //===----------------------------------------------------------------------===//
5240 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5241 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5242 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5243 (ins VR128:$src1, i32i8imm:$src2),
5244 !strconcat(OpcodeStr,
5245 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5246 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5248 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5249 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5250 !strconcat(OpcodeStr,
5251 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5254 // There's an AssertZext in the way of writing the store pattern
5255 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5258 let Predicates = [HasAVX] in {
5259 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5260 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5261 (ins VR128:$src1, i32i8imm:$src2),
5262 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5265 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5268 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5269 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5270 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5271 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5272 !strconcat(OpcodeStr,
5273 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5276 // There's an AssertZext in the way of writing the store pattern
5277 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5280 let Predicates = [HasAVX] in
5281 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5283 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5286 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5287 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5288 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5289 (ins VR128:$src1, i32i8imm:$src2),
5290 !strconcat(OpcodeStr,
5291 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5293 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5294 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5295 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5296 !strconcat(OpcodeStr,
5297 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5298 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5299 addr:$dst)]>, OpSize;
5302 let Predicates = [HasAVX] in
5303 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5305 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5307 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5308 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5309 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5310 (ins VR128:$src1, i32i8imm:$src2),
5311 !strconcat(OpcodeStr,
5312 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5314 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5315 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5316 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5317 !strconcat(OpcodeStr,
5318 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5319 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5320 addr:$dst)]>, OpSize, REX_W;
5323 let Predicates = [HasAVX] in
5324 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5326 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5328 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5330 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5331 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5332 (ins VR128:$src1, i32i8imm:$src2),
5333 !strconcat(OpcodeStr,
5334 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5336 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5338 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5339 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5340 !strconcat(OpcodeStr,
5341 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5342 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5343 addr:$dst)]>, OpSize;
5346 let Predicates = [HasAVX] in {
5347 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5348 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5349 (ins VR128:$src1, i32i8imm:$src2),
5350 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5353 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5355 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5356 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5359 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5360 Requires<[HasSSE41]>;
5361 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5364 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5367 //===----------------------------------------------------------------------===//
5368 // SSE4.1 - Insert Instructions
5369 //===----------------------------------------------------------------------===//
5371 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5372 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5373 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5375 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5377 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5379 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5380 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5381 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5383 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5385 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5387 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5388 imm:$src3))]>, OpSize;
5391 let Predicates = [HasAVX] in
5392 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5393 let Constraints = "$src1 = $dst" in
5394 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5396 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5397 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5398 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5400 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5402 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5404 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5406 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5407 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5409 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5411 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5413 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5414 imm:$src3)))]>, OpSize;
5417 let Predicates = [HasAVX] in
5418 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5419 let Constraints = "$src1 = $dst" in
5420 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5422 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5423 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5424 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5426 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5428 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5430 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5432 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5433 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5435 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5437 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5439 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5440 imm:$src3)))]>, OpSize;
5443 let Predicates = [HasAVX] in
5444 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5445 let Constraints = "$src1 = $dst" in
5446 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5448 // insertps has a few different modes, there's the first two here below which
5449 // are optimized inserts that won't zero arbitrary elements in the destination
5450 // vector. The next one matches the intrinsic and could zero arbitrary elements
5451 // in the target vector.
5452 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5453 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5454 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5456 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5458 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5460 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5462 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5463 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5465 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5467 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5469 (X86insrtps VR128:$src1,
5470 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5471 imm:$src3))]>, OpSize;
5474 let Constraints = "$src1 = $dst" in
5475 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5476 let Predicates = [HasAVX] in
5477 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5479 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5480 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5482 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5483 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5484 Requires<[HasSSE41]>;
5486 //===----------------------------------------------------------------------===//
5487 // SSE4.1 - Round Instructions
5488 //===----------------------------------------------------------------------===//
5490 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5491 X86MemOperand x86memop, RegisterClass RC,
5492 PatFrag mem_frag32, PatFrag mem_frag64,
5493 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5494 // Intrinsic operation, reg.
5495 // Vector intrinsic operation, reg
5496 def PSr : SS4AIi8<opcps, MRMSrcReg,
5497 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5498 !strconcat(OpcodeStr,
5499 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5500 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5503 // Vector intrinsic operation, mem
5504 def PSm : Ii8<opcps, MRMSrcMem,
5505 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5506 !strconcat(OpcodeStr,
5507 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5509 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
5511 Requires<[HasSSE41]>;
5513 // Vector intrinsic operation, reg
5514 def PDr : SS4AIi8<opcpd, MRMSrcReg,
5515 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5516 !strconcat(OpcodeStr,
5517 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5518 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
5521 // Vector intrinsic operation, mem
5522 def PDm : SS4AIi8<opcpd, MRMSrcMem,
5523 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5524 !strconcat(OpcodeStr,
5525 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5527 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
5531 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
5532 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
5533 // Intrinsic operation, reg.
5534 // Vector intrinsic operation, reg
5535 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
5536 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5537 !strconcat(OpcodeStr,
5538 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5541 // Vector intrinsic operation, mem
5542 def PSm_AVX : Ii8<opcps, MRMSrcMem,
5543 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5544 !strconcat(OpcodeStr,
5545 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5546 []>, TA, OpSize, Requires<[HasSSE41]>;
5548 // Vector intrinsic operation, reg
5549 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
5550 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5551 !strconcat(OpcodeStr,
5552 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5555 // Vector intrinsic operation, mem
5556 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
5557 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5558 !strconcat(OpcodeStr,
5559 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5563 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
5566 Intrinsic F64Int, bit Is2Addr = 1> {
5567 // Intrinsic operation, reg.
5568 def SSr : SS4AIi8<opcss, MRMSrcReg,
5569 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5571 !strconcat(OpcodeStr,
5572 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5573 !strconcat(OpcodeStr,
5574 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5575 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5578 // Intrinsic operation, mem.
5579 def SSm : SS4AIi8<opcss, MRMSrcMem,
5580 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5582 !strconcat(OpcodeStr,
5583 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5584 !strconcat(OpcodeStr,
5585 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5587 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
5590 // Intrinsic operation, reg.
5591 def SDr : SS4AIi8<opcsd, MRMSrcReg,
5592 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5594 !strconcat(OpcodeStr,
5595 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5596 !strconcat(OpcodeStr,
5597 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5598 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5601 // Intrinsic operation, mem.
5602 def SDm : SS4AIi8<opcsd, MRMSrcMem,
5603 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5605 !strconcat(OpcodeStr,
5606 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5607 !strconcat(OpcodeStr,
5608 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5610 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
5614 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
5616 // Intrinsic operation, reg.
5617 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
5618 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5619 !strconcat(OpcodeStr,
5620 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5623 // Intrinsic operation, mem.
5624 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
5625 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5626 !strconcat(OpcodeStr,
5627 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5630 // Intrinsic operation, reg.
5631 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
5632 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5633 !strconcat(OpcodeStr,
5634 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5637 // Intrinsic operation, mem.
5638 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
5639 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5640 !strconcat(OpcodeStr,
5641 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5645 // FP round - roundss, roundps, roundsd, roundpd
5646 let Predicates = [HasAVX] in {
5648 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
5649 memopv4f32, memopv2f64,
5650 int_x86_sse41_round_ps,
5651 int_x86_sse41_round_pd>, VEX;
5652 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
5653 memopv8f32, memopv4f64,
5654 int_x86_avx_round_ps_256,
5655 int_x86_avx_round_pd_256>, VEX;
5656 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
5657 int_x86_sse41_round_ss,
5658 int_x86_sse41_round_sd, 0>, VEX_4V;
5660 // Instructions for the assembler
5661 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
5663 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
5665 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
5668 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
5669 memopv4f32, memopv2f64,
5670 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
5671 let Constraints = "$src1 = $dst" in
5672 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
5673 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
5675 //===----------------------------------------------------------------------===//
5676 // SSE4.1 - Packed Bit Test
5677 //===----------------------------------------------------------------------===//
5679 // ptest instruction we'll lower to this in X86ISelLowering primarily from
5680 // the intel intrinsic that corresponds to this.
5681 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5682 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5683 "vptest\t{$src2, $src1|$src1, $src2}",
5684 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5686 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5687 "vptest\t{$src2, $src1|$src1, $src2}",
5688 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5691 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
5692 "vptest\t{$src2, $src1|$src1, $src2}",
5693 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
5695 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
5696 "vptest\t{$src2, $src1|$src1, $src2}",
5697 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
5701 let Defs = [EFLAGS] in {
5702 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5703 "ptest \t{$src2, $src1|$src1, $src2}",
5704 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5706 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5707 "ptest \t{$src2, $src1|$src1, $src2}",
5708 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5712 // The bit test instructions below are AVX only
5713 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
5714 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
5715 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
5716 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5717 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
5718 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
5719 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5720 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
5724 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5725 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
5726 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
5727 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
5728 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
5731 //===----------------------------------------------------------------------===//
5732 // SSE4.1 - Misc Instructions
5733 //===----------------------------------------------------------------------===//
5735 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
5736 "popcnt{w}\t{$src, $dst|$dst, $src}",
5737 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
5738 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
5739 "popcnt{w}\t{$src, $dst|$dst, $src}",
5740 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
5742 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
5743 "popcnt{l}\t{$src, $dst|$dst, $src}",
5744 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
5745 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
5746 "popcnt{l}\t{$src, $dst|$dst, $src}",
5747 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
5749 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
5750 "popcnt{q}\t{$src, $dst|$dst, $src}",
5751 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
5752 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
5753 "popcnt{q}\t{$src, $dst|$dst, $src}",
5754 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
5758 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
5759 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
5760 Intrinsic IntId128> {
5761 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5763 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5764 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
5765 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5767 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5770 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
5773 let Predicates = [HasAVX] in
5774 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
5775 int_x86_sse41_phminposuw>, VEX;
5776 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
5777 int_x86_sse41_phminposuw>;
5779 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
5780 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
5781 Intrinsic IntId128, bit Is2Addr = 1> {
5782 let isCommutable = 1 in
5783 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5784 (ins VR128:$src1, VR128:$src2),
5786 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5787 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5788 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
5789 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5790 (ins VR128:$src1, i128mem:$src2),
5792 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5793 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5795 (IntId128 VR128:$src1,
5796 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5799 let Predicates = [HasAVX] in {
5800 let isCommutable = 0 in
5801 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
5803 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
5805 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
5807 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
5809 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
5811 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
5813 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
5815 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
5817 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
5819 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
5821 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
5824 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
5825 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
5826 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
5827 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
5830 let Constraints = "$src1 = $dst" in {
5831 let isCommutable = 0 in
5832 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
5833 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
5834 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
5835 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
5836 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
5837 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
5838 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
5839 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
5840 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
5841 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
5842 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
5845 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
5846 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
5847 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
5848 (PCMPEQQrm VR128:$src1, addr:$src2)>;
5850 /// SS48I_binop_rm - Simple SSE41 binary operator.
5851 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5852 ValueType OpVT, bit Is2Addr = 1> {
5853 let isCommutable = 1 in
5854 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5855 (ins VR128:$src1, VR128:$src2),
5857 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5858 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5859 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
5861 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5862 (ins VR128:$src1, i128mem:$src2),
5864 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5865 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5866 [(set VR128:$dst, (OpNode VR128:$src1,
5867 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
5871 let Predicates = [HasAVX] in
5872 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
5873 let Constraints = "$src1 = $dst" in
5874 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
5876 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
5877 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
5878 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
5879 X86MemOperand x86memop, bit Is2Addr = 1> {
5880 let isCommutable = 1 in
5881 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
5882 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
5884 !strconcat(OpcodeStr,
5885 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5886 !strconcat(OpcodeStr,
5887 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5888 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
5890 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
5891 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
5893 !strconcat(OpcodeStr,
5894 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5895 !strconcat(OpcodeStr,
5896 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5899 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
5903 let Predicates = [HasAVX] in {
5904 let isCommutable = 0 in {
5905 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
5906 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5907 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
5908 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5909 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
5910 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
5911 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
5912 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
5913 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
5914 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5915 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
5916 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5918 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
5919 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5920 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
5921 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5922 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
5923 VR256, memopv32i8, i256mem, 0>, VEX_4V;
5926 let Constraints = "$src1 = $dst" in {
5927 let isCommutable = 0 in {
5928 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
5929 VR128, memopv16i8, i128mem>;
5930 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
5931 VR128, memopv16i8, i128mem>;
5932 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
5933 VR128, memopv16i8, i128mem>;
5934 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
5935 VR128, memopv16i8, i128mem>;
5937 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
5938 VR128, memopv16i8, i128mem>;
5939 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
5940 VR128, memopv16i8, i128mem>;
5943 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
5944 let Predicates = [HasAVX] in {
5945 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
5946 RegisterClass RC, X86MemOperand x86memop,
5947 PatFrag mem_frag, Intrinsic IntId> {
5948 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
5949 (ins RC:$src1, RC:$src2, RC:$src3),
5950 !strconcat(OpcodeStr,
5951 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5952 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
5953 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5955 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
5956 (ins RC:$src1, x86memop:$src2, RC:$src3),
5957 !strconcat(OpcodeStr,
5958 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5960 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
5962 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5966 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
5967 memopv16i8, int_x86_sse41_blendvpd>;
5968 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
5969 memopv16i8, int_x86_sse41_blendvps>;
5970 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
5971 memopv16i8, int_x86_sse41_pblendvb>;
5972 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
5973 memopv32i8, int_x86_avx_blendv_pd_256>;
5974 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
5975 memopv32i8, int_x86_avx_blendv_ps_256>;
5977 let Predicates = [HasAVX] in {
5978 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
5979 (v16i8 VR128:$src2))),
5980 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
5981 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
5982 (v4i32 VR128:$src2))),
5983 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
5984 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
5985 (v4f32 VR128:$src2))),
5986 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
5987 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
5988 (v2i64 VR128:$src2))),
5989 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
5990 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
5991 (v2f64 VR128:$src2))),
5992 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
5993 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
5994 (v8i32 VR256:$src2))),
5995 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
5996 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
5997 (v8f32 VR256:$src2))),
5998 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
5999 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6000 (v4i64 VR256:$src2))),
6001 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6002 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6003 (v4f64 VR256:$src2))),
6004 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6007 /// SS41I_ternary_int - SSE 4.1 ternary operator
6008 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6009 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
6010 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6011 (ins VR128:$src1, VR128:$src2),
6012 !strconcat(OpcodeStr,
6013 "\t{$src2, $dst|$dst, $src2}"),
6014 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6017 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6018 (ins VR128:$src1, i128mem:$src2),
6019 !strconcat(OpcodeStr,
6020 "\t{$src2, $dst|$dst, $src2}"),
6023 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
6027 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
6028 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
6029 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
6031 let Predicates = [HasSSE41] in {
6032 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6033 (v16i8 VR128:$src2))),
6034 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6035 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6036 (v4i32 VR128:$src2))),
6037 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6038 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6039 (v4f32 VR128:$src2))),
6040 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6041 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6042 (v2i64 VR128:$src2))),
6043 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6044 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6045 (v2f64 VR128:$src2))),
6046 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6049 let Predicates = [HasAVX] in
6050 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6051 "vmovntdqa\t{$src, $dst|$dst, $src}",
6052 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6054 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6055 "movntdqa\t{$src, $dst|$dst, $src}",
6056 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6059 //===----------------------------------------------------------------------===//
6060 // SSE4.2 - Compare Instructions
6061 //===----------------------------------------------------------------------===//
6063 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6064 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6065 Intrinsic IntId128, bit Is2Addr = 1> {
6066 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6067 (ins VR128:$src1, VR128:$src2),
6069 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6070 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6071 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6073 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6074 (ins VR128:$src1, i128mem:$src2),
6076 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6077 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6079 (IntId128 VR128:$src1,
6080 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
6083 let Predicates = [HasAVX] in {
6084 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6087 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6088 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6089 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6090 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6093 let Constraints = "$src1 = $dst" in
6094 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6096 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6097 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6098 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6099 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6101 //===----------------------------------------------------------------------===//
6102 // SSE4.2 - String/text Processing Instructions
6103 //===----------------------------------------------------------------------===//
6105 // Packed Compare Implicit Length Strings, Return Mask
6106 multiclass pseudo_pcmpistrm<string asm> {
6107 def REG : PseudoI<(outs VR128:$dst),
6108 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6109 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6111 def MEM : PseudoI<(outs VR128:$dst),
6112 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6113 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6114 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6117 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6118 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6119 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6122 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
6123 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6124 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6125 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6126 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6127 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6128 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6131 let Defs = [XMM0, EFLAGS] in {
6132 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6133 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6134 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6135 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6136 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6137 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6140 // Packed Compare Explicit Length Strings, Return Mask
6141 multiclass pseudo_pcmpestrm<string asm> {
6142 def REG : PseudoI<(outs VR128:$dst),
6143 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6144 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6145 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6146 def MEM : PseudoI<(outs VR128:$dst),
6147 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6148 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6149 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6152 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6153 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6154 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6157 let Predicates = [HasAVX],
6158 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
6159 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6160 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6161 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6162 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6163 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6164 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6167 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
6168 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6169 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6170 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6171 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6172 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6173 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6176 // Packed Compare Implicit Length Strings, Return Index
6177 let Defs = [ECX, EFLAGS] in {
6178 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6179 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6180 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6181 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6182 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6183 (implicit EFLAGS)]>, OpSize;
6184 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6185 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6186 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6187 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6188 (implicit EFLAGS)]>, OpSize;
6192 let Predicates = [HasAVX] in {
6193 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6195 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6197 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6199 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6201 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6203 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6207 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6208 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6209 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6210 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6211 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6212 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6214 // Packed Compare Explicit Length Strings, Return Index
6215 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6216 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6217 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6218 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6219 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6220 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6221 (implicit EFLAGS)]>, OpSize;
6222 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6223 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6224 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6226 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6227 (implicit EFLAGS)]>, OpSize;
6231 let Predicates = [HasAVX] in {
6232 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6234 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6236 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6238 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6240 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6242 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6246 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6247 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6248 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6249 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6250 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6251 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6253 //===----------------------------------------------------------------------===//
6254 // SSE4.2 - CRC Instructions
6255 //===----------------------------------------------------------------------===//
6257 // No CRC instructions have AVX equivalents
6259 // crc intrinsic instruction
6260 // This set of instructions are only rm, the only difference is the size
6262 let Constraints = "$src1 = $dst" in {
6263 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6264 (ins GR32:$src1, i8mem:$src2),
6265 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6267 (int_x86_sse42_crc32_32_8 GR32:$src1,
6268 (load addr:$src2)))]>;
6269 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6270 (ins GR32:$src1, GR8:$src2),
6271 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6273 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6274 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6275 (ins GR32:$src1, i16mem:$src2),
6276 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6278 (int_x86_sse42_crc32_32_16 GR32:$src1,
6279 (load addr:$src2)))]>,
6281 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6282 (ins GR32:$src1, GR16:$src2),
6283 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6285 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6287 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6288 (ins GR32:$src1, i32mem:$src2),
6289 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6291 (int_x86_sse42_crc32_32_32 GR32:$src1,
6292 (load addr:$src2)))]>;
6293 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6294 (ins GR32:$src1, GR32:$src2),
6295 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6297 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6298 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6299 (ins GR64:$src1, i8mem:$src2),
6300 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6302 (int_x86_sse42_crc32_64_8 GR64:$src1,
6303 (load addr:$src2)))]>,
6305 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6306 (ins GR64:$src1, GR8:$src2),
6307 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6309 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6311 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6312 (ins GR64:$src1, i64mem:$src2),
6313 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6315 (int_x86_sse42_crc32_64_64 GR64:$src1,
6316 (load addr:$src2)))]>,
6318 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6319 (ins GR64:$src1, GR64:$src2),
6320 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6322 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6326 //===----------------------------------------------------------------------===//
6327 // AES-NI Instructions
6328 //===----------------------------------------------------------------------===//
6330 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6331 Intrinsic IntId128, bit Is2Addr = 1> {
6332 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6333 (ins VR128:$src1, VR128:$src2),
6335 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6336 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6337 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6339 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6340 (ins VR128:$src1, i128mem:$src2),
6342 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6343 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6345 (IntId128 VR128:$src1,
6346 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
6349 // Perform One Round of an AES Encryption/Decryption Flow
6350 let Predicates = [HasAVX, HasAES] in {
6351 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
6352 int_x86_aesni_aesenc, 0>, VEX_4V;
6353 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
6354 int_x86_aesni_aesenclast, 0>, VEX_4V;
6355 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
6356 int_x86_aesni_aesdec, 0>, VEX_4V;
6357 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
6358 int_x86_aesni_aesdeclast, 0>, VEX_4V;
6361 let Constraints = "$src1 = $dst" in {
6362 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
6363 int_x86_aesni_aesenc>;
6364 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
6365 int_x86_aesni_aesenclast>;
6366 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
6367 int_x86_aesni_aesdec>;
6368 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
6369 int_x86_aesni_aesdeclast>;
6372 let Predicates = [HasAES] in {
6373 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
6374 (AESENCrr VR128:$src1, VR128:$src2)>;
6375 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
6376 (AESENCrm VR128:$src1, addr:$src2)>;
6377 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
6378 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
6379 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
6380 (AESENCLASTrm VR128:$src1, addr:$src2)>;
6381 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
6382 (AESDECrr VR128:$src1, VR128:$src2)>;
6383 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
6384 (AESDECrm VR128:$src1, addr:$src2)>;
6385 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
6386 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
6387 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
6388 (AESDECLASTrm VR128:$src1, addr:$src2)>;
6391 let Predicates = [HasAVX, HasAES], AddedComplexity = 20 in {
6392 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
6393 (VAESENCrr VR128:$src1, VR128:$src2)>;
6394 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
6395 (VAESENCrm VR128:$src1, addr:$src2)>;
6396 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
6397 (VAESENCLASTrr VR128:$src1, VR128:$src2)>;
6398 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
6399 (VAESENCLASTrm VR128:$src1, addr:$src2)>;
6400 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
6401 (VAESDECrr VR128:$src1, VR128:$src2)>;
6402 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
6403 (VAESDECrm VR128:$src1, addr:$src2)>;
6404 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
6405 (VAESDECLASTrr VR128:$src1, VR128:$src2)>;
6406 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
6407 (VAESDECLASTrm VR128:$src1, addr:$src2)>;
6410 // Perform the AES InvMixColumn Transformation
6411 let Predicates = [HasAVX, HasAES] in {
6412 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6414 "vaesimc\t{$src1, $dst|$dst, $src1}",
6416 (int_x86_aesni_aesimc VR128:$src1))]>,
6418 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6419 (ins i128mem:$src1),
6420 "vaesimc\t{$src1, $dst|$dst, $src1}",
6422 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
6425 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6427 "aesimc\t{$src1, $dst|$dst, $src1}",
6429 (int_x86_aesni_aesimc VR128:$src1))]>,
6431 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6432 (ins i128mem:$src1),
6433 "aesimc\t{$src1, $dst|$dst, $src1}",
6435 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
6438 // AES Round Key Generation Assist
6439 let Predicates = [HasAVX, HasAES] in {
6440 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6441 (ins VR128:$src1, i8imm:$src2),
6442 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6444 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6446 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6447 (ins i128mem:$src1, i8imm:$src2),
6448 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6450 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
6454 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6455 (ins VR128:$src1, i8imm:$src2),
6456 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6458 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6460 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6461 (ins i128mem:$src1, i8imm:$src2),
6462 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6464 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
6468 //===----------------------------------------------------------------------===//
6469 // CLMUL Instructions
6470 //===----------------------------------------------------------------------===//
6472 // Carry-less Multiplication instructions
6473 let Constraints = "$src1 = $dst" in {
6474 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6475 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6476 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6479 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6480 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6481 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6485 // AVX carry-less Multiplication instructions
6486 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6487 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6488 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6491 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6492 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6493 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6497 multiclass pclmul_alias<string asm, int immop> {
6498 def : InstAlias<!strconcat("pclmul", asm,
6499 "dq {$src, $dst|$dst, $src}"),
6500 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
6502 def : InstAlias<!strconcat("pclmul", asm,
6503 "dq {$src, $dst|$dst, $src}"),
6504 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
6506 def : InstAlias<!strconcat("vpclmul", asm,
6507 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6508 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
6510 def : InstAlias<!strconcat("vpclmul", asm,
6511 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6512 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
6514 defm : pclmul_alias<"hqhq", 0x11>;
6515 defm : pclmul_alias<"hqlq", 0x01>;
6516 defm : pclmul_alias<"lqhq", 0x10>;
6517 defm : pclmul_alias<"lqlq", 0x00>;
6519 //===----------------------------------------------------------------------===//
6521 //===----------------------------------------------------------------------===//
6523 //===----------------------------------------------------------------------===//
6524 // VBROADCAST - Load from memory and broadcast to all elements of the
6525 // destination operand
6527 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
6528 X86MemOperand x86memop, Intrinsic Int> :
6529 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
6530 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6531 [(set RC:$dst, (Int addr:$src))]>, VEX;
6533 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
6534 int_x86_avx_vbroadcastss>;
6535 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
6536 int_x86_avx_vbroadcastss_256>;
6537 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
6538 int_x86_avx_vbroadcast_sd_256>;
6539 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
6540 int_x86_avx_vbroadcastf128_pd_256>;
6542 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
6543 (VBROADCASTF128 addr:$src)>;
6545 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
6546 (VBROADCASTSSY addr:$src)>;
6547 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
6548 (VBROADCASTSD addr:$src)>;
6549 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
6550 (VBROADCASTSSY addr:$src)>;
6551 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
6552 (VBROADCASTSD addr:$src)>;
6554 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
6555 (VBROADCASTSS addr:$src)>;
6556 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
6557 (VBROADCASTSS addr:$src)>;
6559 //===----------------------------------------------------------------------===//
6560 // VINSERTF128 - Insert packed floating-point values
6562 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
6563 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
6564 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6566 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
6567 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
6568 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6571 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
6572 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6573 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
6574 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6575 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
6576 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6578 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
6580 (VINSERTF128rr VR256:$src1, VR128:$src2,
6581 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6582 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
6584 (VINSERTF128rr VR256:$src1, VR128:$src2,
6585 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6586 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
6588 (VINSERTF128rr VR256:$src1, VR128:$src2,
6589 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6590 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
6592 (VINSERTF128rr VR256:$src1, VR128:$src2,
6593 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6594 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
6596 (VINSERTF128rr VR256:$src1, VR128:$src2,
6597 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6598 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
6600 (VINSERTF128rr VR256:$src1, VR128:$src2,
6601 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6603 //===----------------------------------------------------------------------===//
6604 // VEXTRACTF128 - Extract packed floating-point values
6606 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
6607 (ins VR256:$src1, i8imm:$src2),
6608 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6610 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
6611 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
6612 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6615 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
6616 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6617 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
6618 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6619 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
6620 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6622 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6623 (v4f32 (VEXTRACTF128rr
6624 (v8f32 VR256:$src1),
6625 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6626 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6627 (v2f64 (VEXTRACTF128rr
6628 (v4f64 VR256:$src1),
6629 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6630 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6631 (v4i32 (VEXTRACTF128rr
6632 (v8i32 VR256:$src1),
6633 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6634 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6635 (v2i64 (VEXTRACTF128rr
6636 (v4i64 VR256:$src1),
6637 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6638 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6639 (v8i16 (VEXTRACTF128rr
6640 (v16i16 VR256:$src1),
6641 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6642 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6643 (v16i8 (VEXTRACTF128rr
6644 (v32i8 VR256:$src1),
6645 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6647 //===----------------------------------------------------------------------===//
6648 // VMASKMOV - Conditional SIMD Packed Loads and Stores
6650 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
6651 Intrinsic IntLd, Intrinsic IntLd256,
6652 Intrinsic IntSt, Intrinsic IntSt256,
6653 PatFrag pf128, PatFrag pf256> {
6654 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
6655 (ins VR128:$src1, f128mem:$src2),
6656 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6657 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
6659 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
6660 (ins VR256:$src1, f256mem:$src2),
6661 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6662 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
6664 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
6665 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
6666 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6667 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
6668 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
6669 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
6670 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6671 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
6674 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
6675 int_x86_avx_maskload_ps,
6676 int_x86_avx_maskload_ps_256,
6677 int_x86_avx_maskstore_ps,
6678 int_x86_avx_maskstore_ps_256,
6679 memopv4f32, memopv8f32>;
6680 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
6681 int_x86_avx_maskload_pd,
6682 int_x86_avx_maskload_pd_256,
6683 int_x86_avx_maskstore_pd,
6684 int_x86_avx_maskstore_pd_256,
6685 memopv2f64, memopv4f64>;
6687 //===----------------------------------------------------------------------===//
6688 // VPERMIL - Permute Single and Double Floating-Point Values
6690 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
6691 RegisterClass RC, X86MemOperand x86memop_f,
6692 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
6693 Intrinsic IntVar, Intrinsic IntImm> {
6694 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
6695 (ins RC:$src1, RC:$src2),
6696 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6697 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
6698 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
6699 (ins RC:$src1, x86memop_i:$src2),
6700 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6701 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
6703 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
6704 (ins RC:$src1, i8imm:$src2),
6705 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6706 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
6707 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
6708 (ins x86memop_f:$src1, i8imm:$src2),
6709 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6710 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
6713 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
6714 memopv4f32, memopv4i32,
6715 int_x86_avx_vpermilvar_ps,
6716 int_x86_avx_vpermil_ps>;
6717 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
6718 memopv8f32, memopv8i32,
6719 int_x86_avx_vpermilvar_ps_256,
6720 int_x86_avx_vpermil_ps_256>;
6721 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
6722 memopv2f64, memopv2i64,
6723 int_x86_avx_vpermilvar_pd,
6724 int_x86_avx_vpermil_pd>;
6725 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
6726 memopv4f64, memopv4i64,
6727 int_x86_avx_vpermilvar_pd_256,
6728 int_x86_avx_vpermil_pd_256>;
6730 def : Pat<(v8f32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
6731 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6732 def : Pat<(v4f64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
6733 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6734 def : Pat<(v8i32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
6735 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6736 def : Pat<(v4i64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
6737 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6739 //===----------------------------------------------------------------------===//
6740 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
6742 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
6743 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
6744 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6746 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
6747 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
6748 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6751 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
6752 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6753 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
6754 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6755 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
6756 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6758 def : Pat<(int_x86_avx_vperm2f128_ps_256
6759 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
6760 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6761 def : Pat<(int_x86_avx_vperm2f128_pd_256
6762 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
6763 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6764 def : Pat<(int_x86_avx_vperm2f128_si_256
6765 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
6766 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6768 def : Pat<(v8f32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6769 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6770 def : Pat<(v8i32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6771 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6772 def : Pat<(v4i64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6773 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6774 def : Pat<(v4f64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6775 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6776 def : Pat<(v32i8 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6777 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6778 def : Pat<(v16i16 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6779 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6781 //===----------------------------------------------------------------------===//
6782 // VZERO - Zero YMM registers
6784 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
6785 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
6786 // Zero All YMM registers
6787 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
6788 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
6790 // Zero Upper bits of YMM registers
6791 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
6792 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;