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)))], IIC_DEFAULT, 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)))],
80 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
81 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
82 string OpcodeStr, X86MemOperand x86memop,
83 list<dag> pat_rr, list<dag> pat_rm,
85 bit rr_hasSideEffects = 0> {
86 let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
87 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
89 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
90 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
91 pat_rr, IIC_DEFAULT, d>;
92 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
94 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
95 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
96 pat_rm, IIC_DEFAULT, d>;
99 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
100 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
101 string asm, string SSEVer, string FPSizeStr,
102 X86MemOperand x86memop, PatFrag mem_frag,
103 Domain d, bit Is2Addr = 1> {
104 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
106 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
107 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
108 [(set RC:$dst, (!cast<Intrinsic>(
109 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
110 RC:$src1, RC:$src2))], IIC_DEFAULT, d>;
111 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
113 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
114 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
115 [(set RC:$dst, (!cast<Intrinsic>(
116 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
117 RC:$src1, (mem_frag addr:$src2)))], IIC_DEFAULT, d>;
120 //===----------------------------------------------------------------------===//
121 // Non-instruction patterns
122 //===----------------------------------------------------------------------===//
124 // A vector extract of the first f32/f64 position is a subregister copy
125 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
126 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
127 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
128 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
130 // A 128-bit subvector extract from the first 256-bit vector position
131 // is a subregister copy that needs no instruction.
132 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
133 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
134 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
135 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
137 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
138 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
139 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
140 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
142 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
143 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
144 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
145 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
147 // A 128-bit subvector insert to the first 256-bit vector position
148 // is a subregister copy that needs no instruction.
149 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
150 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
151 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
152 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
153 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
154 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
155 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
156 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
157 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
158 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
159 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
160 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
162 // Implicitly promote a 32-bit scalar to a vector.
163 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
164 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
165 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
166 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
167 // Implicitly promote a 64-bit scalar to a vector.
168 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
169 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
170 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
171 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
173 // Bitcasts between 128-bit vector types. Return the original type since
174 // no instruction is needed for the conversion
175 let Predicates = [HasSSE2] in {
176 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
177 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
178 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
179 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
180 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
181 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
182 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
183 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
184 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
185 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
186 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
187 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
188 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
189 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
190 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
191 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
192 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
193 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
194 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
195 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
196 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
197 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
198 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
199 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
200 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
201 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
202 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
203 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
204 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
205 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
208 // Bitcasts between 256-bit vector types. Return the original type since
209 // no instruction is needed for the conversion
210 let Predicates = [HasAVX] in {
211 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
212 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
213 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
214 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
215 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
216 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
217 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
218 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
219 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
220 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
221 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
222 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
223 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
224 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
225 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
226 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
227 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
228 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
229 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
230 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
231 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
232 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
233 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
234 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
235 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
236 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
237 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
238 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
239 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
240 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
243 // Alias instructions that map fld0 to pxor for sse.
244 // This is expanded by ExpandPostRAPseudos.
245 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
247 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
248 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
249 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
250 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
253 //===----------------------------------------------------------------------===//
254 // AVX & SSE - Zero/One Vectors
255 //===----------------------------------------------------------------------===//
257 // Alias instruction that maps zero vector to pxor / xorp* for sse.
258 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
259 // swizzled by ExecutionDepsFix to pxor.
260 // We set canFoldAsLoad because this can be converted to a constant-pool
261 // load of an all-zeros value if folding it would be beneficial.
262 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
263 isPseudo = 1, neverHasSideEffects = 1 in {
264 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
267 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
268 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
269 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
270 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
271 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
272 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
275 // The same as done above but for AVX. The 256-bit ISA does not support PI,
276 // and doesn't need it because on sandy bridge the register is set to zero
277 // at the rename stage without using any execution unit, so SET0PSY
278 // and SET0PDY can be used for vector int instructions without penalty
279 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
280 // JIT implementatioan, it does not expand the instructions below like
281 // X86MCInstLower does.
282 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
283 isCodeGenOnly = 1 in {
284 let Predicates = [HasAVX] in {
285 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
286 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
287 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
288 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
290 let Predicates = [HasAVX2], neverHasSideEffects = 1 in
291 def AVX2_SET0 : PDI<0xef, MRMInitReg, (outs VR256:$dst), (ins), "",
295 let Predicates = [HasAVX2], AddedComplexity = 5 in {
296 def : Pat<(v4i64 immAllZerosV), (AVX2_SET0)>;
297 def : Pat<(v8i32 immAllZerosV), (AVX2_SET0)>;
298 def : Pat<(v16i16 immAllZerosV), (AVX2_SET0)>;
299 def : Pat<(v32i8 immAllZerosV), (AVX2_SET0)>;
302 // AVX has no support for 256-bit integer instructions, but since the 128-bit
303 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
304 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
305 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
306 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
308 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
309 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
310 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
312 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
313 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
314 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
316 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
317 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
318 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
320 // We set canFoldAsLoad because this can be converted to a constant-pool
321 // load of an all-ones value if folding it would be beneficial.
322 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
323 // JIT implementation, it does not expand the instructions below like
324 // X86MCInstLower does.
325 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
326 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
327 let Predicates = [HasAVX] in
328 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
329 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
330 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
331 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
332 let Predicates = [HasAVX2] in
333 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
334 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
338 //===----------------------------------------------------------------------===//
339 // SSE 1 & 2 - Move FP Scalar Instructions
341 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
342 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
343 // is used instead. Register-to-register movss/movsd is not modeled as an
344 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
345 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
346 //===----------------------------------------------------------------------===//
348 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
349 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
350 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
352 // Loading from memory automatically zeroing upper bits.
353 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
354 PatFrag mem_pat, string OpcodeStr> :
355 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
356 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
357 [(set RC:$dst, (mem_pat addr:$src))]>;
360 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
361 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
363 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
364 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
367 // For the disassembler
368 let isCodeGenOnly = 1 in {
369 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
370 (ins VR128:$src1, FR32:$src2),
371 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
373 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
374 (ins VR128:$src1, FR64:$src2),
375 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
379 let canFoldAsLoad = 1, isReMaterializable = 1 in {
380 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
382 let AddedComplexity = 20 in
383 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
387 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
388 "movss\t{$src, $dst|$dst, $src}",
389 [(store FR32:$src, addr:$dst)]>, XS, VEX, VEX_LIG;
390 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
391 "movsd\t{$src, $dst|$dst, $src}",
392 [(store FR64:$src, addr:$dst)]>, XD, VEX, VEX_LIG;
395 let Constraints = "$src1 = $dst" in {
396 def MOVSSrr : sse12_move_rr<FR32, v4f32,
397 "movss\t{$src2, $dst|$dst, $src2}">, XS;
398 def MOVSDrr : sse12_move_rr<FR64, v2f64,
399 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
401 // For the disassembler
402 let isCodeGenOnly = 1 in {
403 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
404 (ins VR128:$src1, FR32:$src2),
405 "movss\t{$src2, $dst|$dst, $src2}", []>, XS;
406 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
407 (ins VR128:$src1, FR64:$src2),
408 "movsd\t{$src2, $dst|$dst, $src2}", []>, XD;
412 let canFoldAsLoad = 1, isReMaterializable = 1 in {
413 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
415 let AddedComplexity = 20 in
416 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
419 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
420 "movss\t{$src, $dst|$dst, $src}",
421 [(store FR32:$src, addr:$dst)]>;
422 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
423 "movsd\t{$src, $dst|$dst, $src}",
424 [(store FR64:$src, addr:$dst)]>;
427 let Predicates = [HasAVX] in {
428 let AddedComplexity = 15 in {
429 // Extract the low 32-bit value from one vector and insert it into another.
430 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
431 (VMOVSSrr (v4f32 VR128:$src1),
432 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
433 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
434 (VMOVSSrr (v4i32 VR128:$src1),
435 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
437 // Extract the low 64-bit value from one vector and insert it into another.
438 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
439 (VMOVSDrr (v2f64 VR128:$src1),
440 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
441 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
442 (VMOVSDrr (v2i64 VR128:$src1),
443 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
445 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
446 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
447 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
448 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
449 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
451 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
452 // MOVS{S,D} to the lower bits.
453 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
454 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
455 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
456 (VMOVSSrr (v4f32 (V_SET0)),
457 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
458 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
459 (VMOVSSrr (v4i32 (V_SET0)),
460 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
461 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
462 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
464 // Move low f32 and clear high bits.
465 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
466 (SUBREG_TO_REG (i32 0),
467 (VMOVSSrr (v4f32 (V_SET0)),
468 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
469 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
470 (SUBREG_TO_REG (i32 0),
471 (VMOVSSrr (v4i32 (V_SET0)),
472 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
475 let AddedComplexity = 20 in {
476 // MOVSSrm zeros the high parts of the register; represent this
477 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
478 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
479 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
480 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
481 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
482 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
483 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
485 // MOVSDrm zeros the high parts of the register; represent this
486 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
487 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
488 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
489 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
490 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
491 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
492 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
493 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
494 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
495 def : Pat<(v2f64 (X86vzload addr:$src)),
496 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
498 // Represent the same patterns above but in the form they appear for
500 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
501 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
502 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
503 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
504 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
505 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
506 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
507 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
508 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
510 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
511 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
512 (SUBREG_TO_REG (i32 0),
513 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
515 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
516 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
517 (SUBREG_TO_REG (i64 0),
518 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
520 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
521 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
522 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
524 // Move low f64 and clear high bits.
525 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
526 (SUBREG_TO_REG (i32 0),
527 (VMOVSDrr (v2f64 (V_SET0)),
528 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
530 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
531 (SUBREG_TO_REG (i32 0),
532 (VMOVSDrr (v2i64 (V_SET0)),
533 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_sd)), sub_xmm)>;
535 // Extract and store.
536 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
539 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
540 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
543 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
545 // Shuffle with VMOVSS
546 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
547 (VMOVSSrr VR128:$src1, FR32:$src2)>;
548 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
549 (VMOVSSrr (v4i32 VR128:$src1),
550 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
551 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
552 (VMOVSSrr (v4f32 VR128:$src1),
553 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
556 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
557 (SUBREG_TO_REG (i32 0),
558 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
559 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
560 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
561 (SUBREG_TO_REG (i32 0),
562 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
563 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
565 // Shuffle with VMOVSD
566 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
567 (VMOVSDrr VR128:$src1, FR64:$src2)>;
568 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
569 (VMOVSDrr (v2i64 VR128:$src1),
570 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
571 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
572 (VMOVSDrr (v2f64 VR128:$src1),
573 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
574 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
575 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
577 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
578 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
582 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
583 (SUBREG_TO_REG (i32 0),
584 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
585 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
586 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
587 (SUBREG_TO_REG (i32 0),
588 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
589 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
592 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
593 // is during lowering, where it's not possible to recognize the fold cause
594 // it has two uses through a bitcast. One use disappears at isel time and the
595 // fold opportunity reappears.
596 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
597 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
599 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
600 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
602 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
603 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
605 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
606 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
610 let Predicates = [HasSSE1] in {
611 let AddedComplexity = 15 in {
612 // Extract the low 32-bit value from one vector and insert it into another.
613 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
614 (MOVSSrr (v4f32 VR128:$src1),
615 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
616 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
617 (MOVSSrr (v4i32 VR128:$src1),
618 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
620 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
621 // MOVSS to the lower bits.
622 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
623 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
624 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
625 (MOVSSrr (v4f32 (V_SET0)),
626 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
627 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
628 (MOVSSrr (v4i32 (V_SET0)),
629 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
632 let AddedComplexity = 20 in {
633 // MOVSSrm zeros the high parts of the register; represent this
634 // with SUBREG_TO_REG.
635 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
636 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
637 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
638 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
639 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
640 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
643 // Extract and store.
644 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
647 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
649 // Shuffle with MOVSS
650 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
651 (MOVSSrr VR128:$src1, FR32:$src2)>;
652 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
653 (MOVSSrr (v4i32 VR128:$src1),
654 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
655 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
656 (MOVSSrr (v4f32 VR128:$src1),
657 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
660 let Predicates = [HasSSE2] in {
661 let AddedComplexity = 15 in {
662 // Extract the low 64-bit value from one vector and insert it into another.
663 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
664 (MOVSDrr (v2f64 VR128:$src1),
665 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
666 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
667 (MOVSDrr (v2i64 VR128:$src1),
668 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
670 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
671 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
672 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
673 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
674 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
676 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
677 // MOVSD to the lower bits.
678 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
679 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
682 let AddedComplexity = 20 in {
683 // MOVSDrm zeros the high parts of the register; represent this
684 // with SUBREG_TO_REG.
685 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
686 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
687 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
688 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
689 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
690 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
691 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
692 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
693 def : Pat<(v2f64 (X86vzload addr:$src)),
694 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
697 // Extract and store.
698 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
701 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
703 // Shuffle with MOVSD
704 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
705 (MOVSDrr VR128:$src1, FR64:$src2)>;
706 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
707 (MOVSDrr (v2i64 VR128:$src1),
708 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
709 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
710 (MOVSDrr (v2f64 VR128:$src1),
711 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
712 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
713 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
714 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
715 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
717 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
718 // is during lowering, where it's not possible to recognize the fold cause
719 // it has two uses through a bitcast. One use disappears at isel time and the
720 // fold opportunity reappears.
721 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
722 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
723 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
724 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
725 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
726 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
727 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
728 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
731 //===----------------------------------------------------------------------===//
732 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
733 //===----------------------------------------------------------------------===//
735 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
736 X86MemOperand x86memop, PatFrag ld_frag,
737 string asm, Domain d,
738 bit IsReMaterializable = 1> {
739 let neverHasSideEffects = 1 in
740 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
741 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], IIC_DEFAULT, d>;
742 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
743 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
744 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
745 [(set RC:$dst, (ld_frag addr:$src))], IIC_DEFAULT, d>;
748 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
749 "movaps", SSEPackedSingle>, TB, VEX;
750 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
751 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
752 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
753 "movups", SSEPackedSingle>, TB, VEX;
754 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
755 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
757 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
758 "movaps", SSEPackedSingle>, TB, VEX;
759 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
760 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
761 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
762 "movups", SSEPackedSingle>, TB, VEX;
763 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
764 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
765 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
766 "movaps", SSEPackedSingle>, TB;
767 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
768 "movapd", SSEPackedDouble>, TB, OpSize;
769 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
770 "movups", SSEPackedSingle>, TB;
771 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
772 "movupd", SSEPackedDouble, 0>, TB, OpSize;
774 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
775 "movaps\t{$src, $dst|$dst, $src}",
776 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
777 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
778 "movapd\t{$src, $dst|$dst, $src}",
779 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
780 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
781 "movups\t{$src, $dst|$dst, $src}",
782 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
783 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
784 "movupd\t{$src, $dst|$dst, $src}",
785 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
786 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
787 "movaps\t{$src, $dst|$dst, $src}",
788 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)]>, VEX;
789 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
790 "movapd\t{$src, $dst|$dst, $src}",
791 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)]>, VEX;
792 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
793 "movups\t{$src, $dst|$dst, $src}",
794 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
795 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
796 "movupd\t{$src, $dst|$dst, $src}",
797 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
800 let isCodeGenOnly = 1 in {
801 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
803 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
804 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
806 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
807 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
809 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
810 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
812 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
813 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
815 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
816 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
818 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
819 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
821 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
822 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
824 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
827 let Predicates = [HasAVX] in {
828 def : Pat<(v8i32 (X86vzmovl
829 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
830 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
831 def : Pat<(v4i64 (X86vzmovl
832 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
833 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
834 def : Pat<(v8f32 (X86vzmovl
835 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
836 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
837 def : Pat<(v4f64 (X86vzmovl
838 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
839 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
843 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
844 (VMOVUPSYmr addr:$dst, VR256:$src)>;
845 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
846 (VMOVUPDYmr addr:$dst, VR256:$src)>;
848 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
849 "movaps\t{$src, $dst|$dst, $src}",
850 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
851 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
852 "movapd\t{$src, $dst|$dst, $src}",
853 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
854 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
855 "movups\t{$src, $dst|$dst, $src}",
856 [(store (v4f32 VR128:$src), addr:$dst)]>;
857 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
858 "movupd\t{$src, $dst|$dst, $src}",
859 [(store (v2f64 VR128:$src), addr:$dst)]>;
862 let isCodeGenOnly = 1 in {
863 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
864 "movaps\t{$src, $dst|$dst, $src}", []>;
865 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
866 "movapd\t{$src, $dst|$dst, $src}", []>;
867 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
868 "movups\t{$src, $dst|$dst, $src}", []>;
869 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
870 "movupd\t{$src, $dst|$dst, $src}", []>;
873 let Predicates = [HasAVX] in {
874 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
875 (VMOVUPSmr addr:$dst, VR128:$src)>;
876 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
877 (VMOVUPDmr addr:$dst, VR128:$src)>;
880 let Predicates = [HasSSE1] in
881 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
882 (MOVUPSmr addr:$dst, VR128:$src)>;
883 let Predicates = [HasSSE2] in
884 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
885 (MOVUPDmr addr:$dst, VR128:$src)>;
887 // Use vmovaps/vmovups for AVX integer load/store.
888 let Predicates = [HasAVX] in {
889 // 128-bit load/store
890 def : Pat<(alignedloadv2i64 addr:$src),
891 (VMOVAPSrm addr:$src)>;
892 def : Pat<(loadv2i64 addr:$src),
893 (VMOVUPSrm addr:$src)>;
895 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
896 (VMOVAPSmr addr:$dst, VR128:$src)>;
897 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
898 (VMOVAPSmr addr:$dst, VR128:$src)>;
899 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
900 (VMOVAPSmr addr:$dst, VR128:$src)>;
901 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
902 (VMOVAPSmr addr:$dst, VR128:$src)>;
903 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
904 (VMOVUPSmr addr:$dst, VR128:$src)>;
905 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
906 (VMOVUPSmr addr:$dst, VR128:$src)>;
907 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
908 (VMOVUPSmr addr:$dst, VR128:$src)>;
909 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
910 (VMOVUPSmr addr:$dst, VR128:$src)>;
912 // 256-bit load/store
913 def : Pat<(alignedloadv4i64 addr:$src),
914 (VMOVAPSYrm addr:$src)>;
915 def : Pat<(loadv4i64 addr:$src),
916 (VMOVUPSYrm addr:$src)>;
917 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
918 (VMOVAPSYmr addr:$dst, VR256:$src)>;
919 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
920 (VMOVAPSYmr addr:$dst, VR256:$src)>;
921 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
922 (VMOVAPSYmr addr:$dst, VR256:$src)>;
923 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
924 (VMOVAPSYmr addr:$dst, VR256:$src)>;
925 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
926 (VMOVUPSYmr addr:$dst, VR256:$src)>;
927 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
928 (VMOVUPSYmr addr:$dst, VR256:$src)>;
929 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
930 (VMOVUPSYmr addr:$dst, VR256:$src)>;
931 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
932 (VMOVUPSYmr addr:$dst, VR256:$src)>;
935 // Use movaps / movups for SSE integer load / store (one byte shorter).
936 // The instructions selected below are then converted to MOVDQA/MOVDQU
937 // during the SSE domain pass.
938 let Predicates = [HasSSE1] in {
939 def : Pat<(alignedloadv2i64 addr:$src),
940 (MOVAPSrm addr:$src)>;
941 def : Pat<(loadv2i64 addr:$src),
942 (MOVUPSrm addr:$src)>;
944 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
945 (MOVAPSmr addr:$dst, VR128:$src)>;
946 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
947 (MOVAPSmr addr:$dst, VR128:$src)>;
948 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
949 (MOVAPSmr addr:$dst, VR128:$src)>;
950 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
951 (MOVAPSmr addr:$dst, VR128:$src)>;
952 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
953 (MOVUPSmr addr:$dst, VR128:$src)>;
954 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
955 (MOVUPSmr addr:$dst, VR128:$src)>;
956 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
957 (MOVUPSmr addr:$dst, VR128:$src)>;
958 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
959 (MOVUPSmr addr:$dst, VR128:$src)>;
962 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
963 // bits are disregarded. FIXME: Set encoding to pseudo!
964 let neverHasSideEffects = 1 in {
965 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
966 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
967 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
968 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
969 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
970 "movaps\t{$src, $dst|$dst, $src}", []>;
971 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
972 "movapd\t{$src, $dst|$dst, $src}", []>;
975 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
976 // bits are disregarded. FIXME: Set encoding to pseudo!
977 let canFoldAsLoad = 1, isReMaterializable = 1 in {
978 let isCodeGenOnly = 1 in {
979 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
980 "movaps\t{$src, $dst|$dst, $src}",
981 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
982 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
983 "movapd\t{$src, $dst|$dst, $src}",
984 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
986 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
987 "movaps\t{$src, $dst|$dst, $src}",
988 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
989 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
990 "movapd\t{$src, $dst|$dst, $src}",
991 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
994 //===----------------------------------------------------------------------===//
995 // SSE 1 & 2 - Move Low packed FP Instructions
996 //===----------------------------------------------------------------------===//
998 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
999 PatFrag mov_frag, string base_opc,
1001 def PSrm : PI<opc, MRMSrcMem,
1002 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1003 !strconcat(base_opc, "s", asm_opr),
1006 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1007 IIC_DEFAULT, SSEPackedSingle>, TB;
1009 def PDrm : PI<opc, MRMSrcMem,
1010 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1011 !strconcat(base_opc, "d", asm_opr),
1012 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
1013 (scalar_to_vector (loadf64 addr:$src2)))))],
1014 IIC_DEFAULT, SSEPackedDouble>, TB, OpSize;
1017 let AddedComplexity = 20 in {
1018 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1019 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1021 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1022 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1023 "\t{$src2, $dst|$dst, $src2}">;
1026 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1027 "movlps\t{$src, $dst|$dst, $src}",
1028 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1029 (iPTR 0))), addr:$dst)]>, VEX;
1030 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1031 "movlpd\t{$src, $dst|$dst, $src}",
1032 [(store (f64 (vector_extract (v2f64 VR128:$src),
1033 (iPTR 0))), addr:$dst)]>, VEX;
1034 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1035 "movlps\t{$src, $dst|$dst, $src}",
1036 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1037 (iPTR 0))), addr:$dst)]>;
1038 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1039 "movlpd\t{$src, $dst|$dst, $src}",
1040 [(store (f64 (vector_extract (v2f64 VR128:$src),
1041 (iPTR 0))), addr:$dst)]>;
1043 let Predicates = [HasAVX] in {
1044 let AddedComplexity = 20 in {
1045 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1046 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1047 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1048 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1049 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1050 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1051 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1052 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1053 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1054 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1057 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1058 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1059 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1060 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1061 VR128:$src2)), addr:$src1),
1062 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1064 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1065 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1066 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1067 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1068 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1070 // Shuffle with VMOVLPS
1071 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1072 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1073 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1074 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1075 def : Pat<(X86Movlps VR128:$src1,
1076 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1077 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1079 // Shuffle with VMOVLPD
1080 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1081 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1082 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1083 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1084 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1085 (scalar_to_vector (loadf64 addr:$src2)))),
1086 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1089 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1091 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1092 def : Pat<(store (v4i32 (X86Movlps
1093 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1094 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1095 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1097 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1098 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1100 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1103 let Predicates = [HasSSE1] in {
1104 let AddedComplexity = 20 in {
1105 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1106 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1107 (MOVLPSrm VR128:$src1, addr:$src2)>;
1108 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1109 (MOVLPSrm VR128:$src1, addr:$src2)>;
1112 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1113 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1114 (iPTR 0))), addr:$src1),
1115 (MOVLPSmr addr:$src1, VR128:$src2)>;
1116 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1117 (MOVLPSmr addr:$src1, VR128:$src2)>;
1118 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1119 VR128:$src2)), addr:$src1),
1120 (MOVLPSmr addr:$src1, VR128:$src2)>;
1122 // Shuffle with MOVLPS
1123 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1124 (MOVLPSrm VR128:$src1, addr:$src2)>;
1125 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1126 (MOVLPSrm VR128:$src1, addr:$src2)>;
1127 def : Pat<(X86Movlps VR128:$src1,
1128 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1129 (MOVLPSrm VR128:$src1, addr:$src2)>;
1130 def : Pat<(X86Movlps VR128:$src1,
1131 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1132 (MOVLPSrm VR128:$src1, addr:$src2)>;
1135 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1137 (MOVLPSmr addr:$src1, VR128:$src2)>;
1138 def : Pat<(store (v4i32 (X86Movlps
1139 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1141 (MOVLPSmr addr:$src1, VR128:$src2)>;
1144 let Predicates = [HasSSE2] in {
1145 let AddedComplexity = 20 in {
1146 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1147 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1148 (MOVLPDrm VR128:$src1, addr:$src2)>;
1149 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1150 (MOVLPDrm VR128:$src1, addr:$src2)>;
1153 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1154 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1155 (MOVLPDmr addr:$src1, VR128:$src2)>;
1156 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1157 (MOVLPDmr addr:$src1, VR128:$src2)>;
1159 // Shuffle with MOVLPD
1160 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1161 (MOVLPDrm VR128:$src1, addr:$src2)>;
1162 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1163 (MOVLPDrm VR128:$src1, addr:$src2)>;
1164 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1165 (scalar_to_vector (loadf64 addr:$src2)))),
1166 (MOVLPDrm VR128:$src1, addr:$src2)>;
1169 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1171 (MOVLPDmr addr:$src1, VR128:$src2)>;
1172 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1174 (MOVLPDmr addr:$src1, VR128:$src2)>;
1177 //===----------------------------------------------------------------------===//
1178 // SSE 1 & 2 - Move Hi packed FP Instructions
1179 //===----------------------------------------------------------------------===//
1181 let AddedComplexity = 20 in {
1182 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1183 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1185 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1186 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1187 "\t{$src2, $dst|$dst, $src2}">;
1190 // v2f64 extract element 1 is always custom lowered to unpack high to low
1191 // and extract element 0 so the non-store version isn't too horrible.
1192 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1193 "movhps\t{$src, $dst|$dst, $src}",
1194 [(store (f64 (vector_extract
1195 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1196 (undef)), (iPTR 0))), addr:$dst)]>,
1198 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1199 "movhpd\t{$src, $dst|$dst, $src}",
1200 [(store (f64 (vector_extract
1201 (v2f64 (unpckh VR128:$src, (undef))),
1202 (iPTR 0))), addr:$dst)]>,
1204 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1205 "movhps\t{$src, $dst|$dst, $src}",
1206 [(store (f64 (vector_extract
1207 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1208 (undef)), (iPTR 0))), addr:$dst)]>;
1209 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1210 "movhpd\t{$src, $dst|$dst, $src}",
1211 [(store (f64 (vector_extract
1212 (v2f64 (unpckh VR128:$src, (undef))),
1213 (iPTR 0))), addr:$dst)]>;
1215 let Predicates = [HasAVX] in {
1217 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1218 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1219 def : Pat<(X86Movlhps VR128:$src1,
1220 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1221 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1222 def : Pat<(X86Movlhps VR128:$src1,
1223 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1224 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1225 def : Pat<(X86Movlhps VR128:$src1,
1226 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1227 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1229 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1230 // is during lowering, where it's not possible to recognize the load fold
1231 // cause it has two uses through a bitcast. One use disappears at isel time
1232 // and the fold opportunity reappears.
1233 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1234 (scalar_to_vector (loadf64 addr:$src2)))),
1235 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1237 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1238 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1239 (scalar_to_vector (loadf64 addr:$src2)))),
1240 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1243 def : Pat<(store (f64 (vector_extract
1244 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1245 (bc_v2f64 (v4f32 VR128:$src))), (iPTR 0))), addr:$dst),
1246 (VMOVHPSmr addr:$dst, VR128:$src)>;
1247 def : Pat<(store (f64 (vector_extract
1248 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))), addr:$dst),
1249 (VMOVHPDmr addr:$dst, VR128:$src)>;
1252 let Predicates = [HasSSE1] in {
1254 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1255 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1256 def : Pat<(X86Movlhps VR128:$src1,
1257 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1258 (MOVHPSrm VR128:$src1, addr:$src2)>;
1259 def : Pat<(X86Movlhps VR128:$src1,
1260 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1261 (MOVHPSrm VR128:$src1, addr:$src2)>;
1262 def : Pat<(X86Movlhps VR128:$src1,
1263 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1264 (MOVHPSrm VR128:$src1, addr:$src2)>;
1267 def : Pat<(store (f64 (vector_extract
1268 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1269 (bc_v2f64 (v4f32 VR128:$src))), (iPTR 0))), addr:$dst),
1270 (MOVHPSmr addr:$dst, VR128:$src)>;
1273 let Predicates = [HasSSE2] in {
1274 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1275 // is during lowering, where it's not possible to recognize the load fold
1276 // cause it has two uses through a bitcast. One use disappears at isel time
1277 // and the fold opportunity reappears.
1278 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1279 (scalar_to_vector (loadf64 addr:$src2)))),
1280 (MOVHPDrm VR128:$src1, addr:$src2)>;
1282 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1283 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1284 (scalar_to_vector (loadf64 addr:$src2)))),
1285 (MOVHPDrm VR128:$src1, addr:$src2)>;
1288 def : Pat<(store (f64 (vector_extract
1289 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))),addr:$dst),
1290 (MOVHPDmr addr:$dst, VR128:$src)>;
1293 //===----------------------------------------------------------------------===//
1294 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1295 //===----------------------------------------------------------------------===//
1297 let AddedComplexity = 20 in {
1298 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1299 (ins VR128:$src1, VR128:$src2),
1300 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1302 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1304 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1305 (ins VR128:$src1, VR128:$src2),
1306 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1308 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1311 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1312 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1313 (ins VR128:$src1, VR128:$src2),
1314 "movlhps\t{$src2, $dst|$dst, $src2}",
1316 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1317 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1318 (ins VR128:$src1, VR128:$src2),
1319 "movhlps\t{$src2, $dst|$dst, $src2}",
1321 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1324 let Predicates = [HasAVX] in {
1326 let AddedComplexity = 20 in {
1327 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1328 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1329 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1331 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1332 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1333 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1334 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1335 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1336 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1339 let AddedComplexity = 20 in {
1340 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1341 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1342 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1344 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1345 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1346 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1347 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1348 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1351 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1352 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1353 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1354 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1357 let Predicates = [HasSSE1] in {
1359 let AddedComplexity = 20 in {
1360 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1361 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1362 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1364 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1365 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1366 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1367 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1368 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1369 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1372 let AddedComplexity = 20 in {
1373 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1374 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1375 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1377 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1378 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1379 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1380 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1381 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1384 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1385 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1386 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1387 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1390 //===----------------------------------------------------------------------===//
1391 // SSE 1 & 2 - Conversion Instructions
1392 //===----------------------------------------------------------------------===//
1394 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1395 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1397 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1398 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1399 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1400 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1403 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1404 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1405 string asm, Domain d> {
1406 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1407 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1409 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1410 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1414 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1415 X86MemOperand x86memop, string asm> {
1416 let neverHasSideEffects = 1 in {
1417 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1418 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1420 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1421 (ins DstRC:$src1, x86memop:$src),
1422 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1423 } // neverHasSideEffects = 1
1426 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1427 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1429 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1430 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1432 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1433 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1435 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1436 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1437 VEX, VEX_W, VEX_LIG;
1439 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1440 // register, but the same isn't true when only using memory operands,
1441 // provide other assembly "l" and "q" forms to address this explicitly
1442 // where appropriate to do so.
1443 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1445 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1446 VEX_4V, VEX_W, VEX_LIG;
1447 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1449 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1451 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1452 VEX_4V, VEX_W, VEX_LIG;
1454 let Predicates = [HasAVX], AddedComplexity = 1 in {
1455 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1456 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1457 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1458 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1459 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1460 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1461 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1462 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1464 def : Pat<(f32 (sint_to_fp GR32:$src)),
1465 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1466 def : Pat<(f32 (sint_to_fp GR64:$src)),
1467 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1468 def : Pat<(f64 (sint_to_fp GR32:$src)),
1469 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1470 def : Pat<(f64 (sint_to_fp GR64:$src)),
1471 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1474 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1475 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1476 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1477 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1478 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1479 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1480 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1481 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1482 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1483 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1484 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1485 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1486 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1487 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1488 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1489 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1491 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1492 // and/or XMM operand(s).
1494 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1495 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1497 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1498 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1499 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1500 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1501 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1502 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1505 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1506 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1507 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1508 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1510 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1511 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1512 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1513 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1514 (ins DstRC:$src1, x86memop:$src2),
1516 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1517 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1518 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1521 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1522 f128mem, load, "cvtsd2si">, XD, VEX, VEX_LIG;
1523 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1524 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1525 XD, VEX, VEX_W, VEX_LIG;
1527 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1528 f128mem, load, "cvtsd2si{l}">, XD;
1529 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1530 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1533 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1534 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1535 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1536 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1538 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1539 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1540 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1541 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1544 let Constraints = "$src1 = $dst" in {
1545 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1546 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1548 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1549 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1550 "cvtsi2ss{q}">, XS, REX_W;
1551 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1552 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1554 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1555 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1556 "cvtsi2sd">, XD, REX_W;
1561 // Aliases for intrinsics
1562 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1563 f32mem, load, "cvttss2si">, XS, VEX;
1564 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1565 int_x86_sse_cvttss2si64, f32mem, load,
1566 "cvttss2si">, XS, VEX, VEX_W;
1567 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1568 f128mem, load, "cvttsd2si">, XD, VEX;
1569 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1570 int_x86_sse2_cvttsd2si64, f128mem, load,
1571 "cvttsd2si">, XD, VEX, VEX_W;
1572 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1573 f32mem, load, "cvttss2si">, XS;
1574 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1575 int_x86_sse_cvttss2si64, f32mem, load,
1576 "cvttss2si{q}">, XS, REX_W;
1577 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1578 f128mem, load, "cvttsd2si">, XD;
1579 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1580 int_x86_sse2_cvttsd2si64, f128mem, load,
1581 "cvttsd2si{q}">, XD, REX_W;
1583 let Pattern = []<dag> in {
1584 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1585 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1587 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1588 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1590 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1591 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1592 SSEPackedSingle>, TB, VEX;
1593 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1594 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1595 SSEPackedSingle>, TB, VEX;
1598 let Pattern = []<dag> in {
1599 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1600 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1601 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1602 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1603 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1604 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1605 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1608 let Predicates = [HasAVX] in {
1609 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1610 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1611 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1612 (VCVTSS2SIrm addr:$src)>;
1613 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1614 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1615 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1616 (VCVTSS2SI64rm addr:$src)>;
1619 let Predicates = [HasSSE1] in {
1620 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1621 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1622 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1623 (CVTSS2SIrm addr:$src)>;
1624 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1625 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1626 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1627 (CVTSS2SI64rm addr:$src)>;
1632 // Convert scalar double to scalar single
1633 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1634 (ins FR64:$src1, FR64:$src2),
1635 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1638 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1639 (ins FR64:$src1, f64mem:$src2),
1640 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1641 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1643 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1646 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1647 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1648 [(set FR32:$dst, (fround FR64:$src))]>;
1649 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1650 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1651 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1652 Requires<[HasSSE2, OptForSize]>;
1654 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1655 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1657 let Constraints = "$src1 = $dst" in
1658 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1659 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1661 // Convert scalar single to scalar double
1662 // SSE2 instructions with XS prefix
1663 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1664 (ins FR32:$src1, FR32:$src2),
1665 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1666 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1668 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1669 (ins FR32:$src1, f32mem:$src2),
1670 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1671 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1673 let Predicates = [HasAVX] in {
1674 def : Pat<(f64 (fextend FR32:$src)),
1675 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1676 def : Pat<(fextend (loadf32 addr:$src)),
1677 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1678 def : Pat<(extloadf32 addr:$src),
1679 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1682 def : Pat<(extloadf32 addr:$src),
1683 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1684 Requires<[HasAVX, OptForSpeed]>;
1686 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1687 "cvtss2sd\t{$src, $dst|$dst, $src}",
1688 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1689 Requires<[HasSSE2]>;
1690 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1691 "cvtss2sd\t{$src, $dst|$dst, $src}",
1692 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1693 Requires<[HasSSE2, OptForSize]>;
1695 // extload f32 -> f64. This matches load+fextend because we have a hack in
1696 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1698 // Since these loads aren't folded into the fextend, we have to match it
1700 def : Pat<(fextend (loadf32 addr:$src)),
1701 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1702 def : Pat<(extloadf32 addr:$src),
1703 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1705 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1706 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1707 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1708 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1709 VR128:$src2))]>, XS, VEX_4V,
1711 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1712 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1713 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1714 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1715 (load addr:$src2)))]>, XS, VEX_4V,
1717 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1718 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1719 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1720 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1721 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1722 VR128:$src2))]>, XS,
1723 Requires<[HasSSE2]>;
1724 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1725 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1726 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1727 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1728 (load addr:$src2)))]>, XS,
1729 Requires<[HasSSE2]>;
1732 // Convert doubleword to packed single/double fp
1733 // SSE2 instructions without OpSize prefix
1734 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1735 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1736 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1737 TB, VEX, Requires<[HasAVX]>;
1738 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1739 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1740 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1741 (bitconvert (memopv2i64 addr:$src))))]>,
1742 TB, VEX, Requires<[HasAVX]>;
1743 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1744 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1745 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1746 TB, Requires<[HasSSE2]>;
1747 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1748 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1749 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1750 (bitconvert (memopv2i64 addr:$src))))]>,
1751 TB, Requires<[HasSSE2]>;
1753 // FIXME: why the non-intrinsic version is described as SSE3?
1754 // SSE2 instructions with XS prefix
1755 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1756 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1757 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1758 XS, VEX, Requires<[HasAVX]>;
1759 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1760 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1761 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1762 (bitconvert (memopv2i64 addr:$src))))]>,
1763 XS, VEX, Requires<[HasAVX]>;
1764 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1765 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1766 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1767 XS, Requires<[HasSSE2]>;
1768 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1769 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1770 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1771 (bitconvert (memopv2i64 addr:$src))))]>,
1772 XS, Requires<[HasSSE2]>;
1775 // Convert packed single/double fp to doubleword
1776 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1777 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1778 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1779 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1780 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1781 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1782 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1783 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1784 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1785 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1786 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1787 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1789 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1790 "cvtps2dq\t{$src, $dst|$dst, $src}",
1791 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1793 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1795 "cvtps2dq\t{$src, $dst|$dst, $src}",
1796 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1797 (memop addr:$src)))]>, VEX;
1798 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1799 "cvtps2dq\t{$src, $dst|$dst, $src}",
1800 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1801 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1802 "cvtps2dq\t{$src, $dst|$dst, $src}",
1803 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1804 (memop addr:$src)))]>;
1806 // SSE2 packed instructions with XD prefix
1807 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1808 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1809 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1810 XD, VEX, Requires<[HasAVX]>;
1811 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1812 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1813 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1814 (memop addr:$src)))]>,
1815 XD, VEX, Requires<[HasAVX]>;
1816 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1817 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1818 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1819 XD, Requires<[HasSSE2]>;
1820 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1821 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1822 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1823 (memop addr:$src)))]>,
1824 XD, Requires<[HasSSE2]>;
1827 // Convert with truncation packed single/double fp to doubleword
1828 // SSE2 packed instructions with XS prefix
1829 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1830 "cvttps2dq\t{$src, $dst|$dst, $src}",
1832 (int_x86_sse2_cvttps2dq VR128:$src))]>, VEX;
1833 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1834 "cvttps2dq\t{$src, $dst|$dst, $src}",
1835 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1836 (memop addr:$src)))]>, VEX;
1837 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1838 "cvttps2dq\t{$src, $dst|$dst, $src}",
1840 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))]>, VEX;
1841 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1842 "cvttps2dq\t{$src, $dst|$dst, $src}",
1843 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1844 (memopv8f32 addr:$src)))]>, VEX;
1846 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1847 "cvttps2dq\t{$src, $dst|$dst, $src}",
1849 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1850 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1851 "cvttps2dq\t{$src, $dst|$dst, $src}",
1853 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1855 let Predicates = [HasAVX] in {
1856 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1857 (Int_VCVTDQ2PSrr VR128:$src)>;
1858 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1859 (Int_VCVTDQ2PSrm addr:$src)>;
1861 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1862 (VCVTTPS2DQrr VR128:$src)>;
1863 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1864 (VCVTTPS2DQrm addr:$src)>;
1866 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1867 (VCVTDQ2PSYrr VR256:$src)>;
1868 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1869 (VCVTDQ2PSYrm addr:$src)>;
1871 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1872 (VCVTTPS2DQYrr VR256:$src)>;
1873 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1874 (VCVTTPS2DQYrm addr:$src)>;
1877 let Predicates = [HasSSE2] in {
1878 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1879 (Int_CVTDQ2PSrr VR128:$src)>;
1880 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1881 (Int_CVTDQ2PSrm addr:$src)>;
1883 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1884 (CVTTPS2DQrr VR128:$src)>;
1885 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1886 (CVTTPS2DQrm addr:$src)>;
1889 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1890 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1892 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1893 let isCodeGenOnly = 1 in
1894 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1895 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1896 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1897 (memop addr:$src)))]>, VEX;
1898 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1899 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1900 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1901 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1902 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1903 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1904 (memop addr:$src)))]>;
1906 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1907 // register, but the same isn't true when using memory operands instead.
1908 // Provide other assembly rr and rm forms to address this explicitly.
1909 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1910 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1913 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1914 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1915 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1916 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1919 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1920 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1921 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1922 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1924 // Convert packed single to packed double
1925 let Predicates = [HasAVX] in {
1926 // SSE2 instructions without OpSize prefix
1927 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1928 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1929 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1930 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1931 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1932 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1933 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1934 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1936 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1937 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1938 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1939 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1941 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1942 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1943 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1944 TB, VEX, Requires<[HasAVX]>;
1945 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1946 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1947 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1948 (load addr:$src)))]>,
1949 TB, VEX, Requires<[HasAVX]>;
1950 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1951 "cvtps2pd\t{$src, $dst|$dst, $src}",
1952 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1953 TB, Requires<[HasSSE2]>;
1954 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1955 "cvtps2pd\t{$src, $dst|$dst, $src}",
1956 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1957 (load addr:$src)))]>,
1958 TB, Requires<[HasSSE2]>;
1960 // Convert packed double to packed single
1961 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1962 // register, but the same isn't true when using memory operands instead.
1963 // Provide other assembly rr and rm forms to address this explicitly.
1964 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1965 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1966 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1967 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1970 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1971 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1972 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1973 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1976 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1977 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1978 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1979 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1980 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1981 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1982 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1983 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1986 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1987 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1988 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1989 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1991 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1992 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1993 (memop addr:$src)))]>;
1994 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1995 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1996 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1997 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1998 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1999 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2000 (memop addr:$src)))]>;
2002 // AVX 256-bit register conversion intrinsics
2003 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2004 // whenever possible to avoid declaring two versions of each one.
2005 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2006 (VCVTDQ2PSYrr VR256:$src)>;
2007 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2008 (VCVTDQ2PSYrm addr:$src)>;
2010 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2011 (VCVTPD2PSYrr VR256:$src)>;
2012 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2013 (VCVTPD2PSYrm addr:$src)>;
2015 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2016 (VCVTPS2DQYrr VR256:$src)>;
2017 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2018 (VCVTPS2DQYrm addr:$src)>;
2020 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2021 (VCVTPS2PDYrr VR128:$src)>;
2022 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2023 (VCVTPS2PDYrm addr:$src)>;
2025 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2026 (VCVTTPD2DQYrr VR256:$src)>;
2027 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2028 (VCVTTPD2DQYrm addr:$src)>;
2030 // Match fround and fextend for 128/256-bit conversions
2031 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2032 (VCVTPD2PSYrr VR256:$src)>;
2033 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2034 (VCVTPD2PSYrm addr:$src)>;
2036 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2037 (VCVTPS2PDYrr VR128:$src)>;
2038 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2039 (VCVTPS2PDYrm addr:$src)>;
2041 //===----------------------------------------------------------------------===//
2042 // SSE 1 & 2 - Compare Instructions
2043 //===----------------------------------------------------------------------===//
2045 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2046 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2047 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2048 string asm, string asm_alt> {
2049 def rr : SIi8<0xC2, MRMSrcReg,
2050 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2051 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2052 def rm : SIi8<0xC2, MRMSrcMem,
2053 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2054 [(set RC:$dst, (OpNode (VT RC:$src1),
2055 (ld_frag addr:$src2), imm:$cc))]>;
2057 // Accept explicit immediate argument form instead of comparison code.
2058 let neverHasSideEffects = 1 in {
2059 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2060 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2062 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2063 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2067 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2068 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2069 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2070 XS, VEX_4V, VEX_LIG;
2071 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2072 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2073 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2074 XD, VEX_4V, VEX_LIG;
2076 let Constraints = "$src1 = $dst" in {
2077 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2078 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2079 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2081 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2082 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2083 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2087 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2088 Intrinsic Int, string asm> {
2089 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2090 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2091 [(set VR128:$dst, (Int VR128:$src1,
2092 VR128:$src, imm:$cc))]>;
2093 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2094 (ins VR128:$src1, x86memop:$src, SSECC:$cc), asm,
2095 [(set VR128:$dst, (Int VR128:$src1,
2096 (load addr:$src), imm:$cc))]>;
2099 // Aliases to match intrinsics which expect XMM operand(s).
2100 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2101 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2103 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2104 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2106 let Constraints = "$src1 = $dst" in {
2107 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2108 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2109 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2110 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2114 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2115 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2116 ValueType vt, X86MemOperand x86memop,
2117 PatFrag ld_frag, string OpcodeStr, Domain d> {
2118 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2119 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2120 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2122 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2123 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2124 [(set EFLAGS, (OpNode (vt RC:$src1),
2125 (ld_frag addr:$src2)))],
2129 let Defs = [EFLAGS] in {
2130 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2131 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2132 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2133 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2135 let Pattern = []<dag> in {
2136 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2137 "comiss", SSEPackedSingle>, TB, VEX,
2139 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2140 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2144 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2145 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2146 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2147 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2149 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2150 load, "comiss", SSEPackedSingle>, TB, VEX;
2151 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2152 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2153 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2154 "ucomiss", SSEPackedSingle>, TB;
2155 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2156 "ucomisd", SSEPackedDouble>, TB, OpSize;
2158 let Pattern = []<dag> in {
2159 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2160 "comiss", SSEPackedSingle>, TB;
2161 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2162 "comisd", SSEPackedDouble>, TB, OpSize;
2165 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2166 load, "ucomiss", SSEPackedSingle>, TB;
2167 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2168 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2170 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2171 "comiss", SSEPackedSingle>, TB;
2172 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2173 "comisd", SSEPackedDouble>, TB, OpSize;
2174 } // Defs = [EFLAGS]
2176 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2177 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2178 Intrinsic Int, string asm, string asm_alt,
2180 let isAsmParserOnly = 1 in {
2181 def rri : PIi8<0xC2, MRMSrcReg,
2182 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2183 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2185 def rmi : PIi8<0xC2, MRMSrcMem,
2186 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2187 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2191 // Accept explicit immediate argument form instead of comparison code.
2192 def rri_alt : PIi8<0xC2, MRMSrcReg,
2193 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2194 asm_alt, [], IIC_DEFAULT, d>;
2195 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2196 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2197 asm_alt, [], IIC_DEFAULT, d>;
2200 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2201 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2202 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2203 SSEPackedSingle>, TB, VEX_4V;
2204 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2205 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2206 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2207 SSEPackedDouble>, TB, OpSize, VEX_4V;
2208 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2209 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2210 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2211 SSEPackedSingle>, TB, VEX_4V;
2212 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2213 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2214 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2215 SSEPackedDouble>, TB, OpSize, VEX_4V;
2216 let Constraints = "$src1 = $dst" in {
2217 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2218 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2219 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2220 SSEPackedSingle>, TB;
2221 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2222 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2223 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2224 SSEPackedDouble>, TB, OpSize;
2227 let Predicates = [HasAVX] in {
2228 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2229 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2230 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2231 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2232 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2233 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2234 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2235 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2237 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2238 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2239 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2240 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2241 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2242 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2243 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2244 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2247 let Predicates = [HasSSE1] in {
2248 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2249 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2250 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2251 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2254 let Predicates = [HasSSE2] in {
2255 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2256 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2257 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2258 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2261 //===----------------------------------------------------------------------===//
2262 // SSE 1 & 2 - Shuffle Instructions
2263 //===----------------------------------------------------------------------===//
2265 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2266 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2267 ValueType vt, string asm, PatFrag mem_frag,
2268 Domain d, bit IsConvertibleToThreeAddress = 0> {
2269 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2270 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2271 [(set RC:$dst, (vt (shufp:$src3
2272 RC:$src1, (mem_frag addr:$src2))))],
2274 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2275 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2276 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2278 (vt (shufp:$src3 RC:$src1, RC:$src2)))],
2282 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2283 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2284 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2285 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2286 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2287 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2288 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2289 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2290 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2291 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2292 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2293 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2295 let Constraints = "$src1 = $dst" in {
2296 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2297 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2298 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2300 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2301 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2302 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2306 let Predicates = [HasAVX] in {
2307 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2308 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2309 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2310 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2311 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2312 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2313 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2314 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2315 def : Pat<(v4i32 (X86Shufp VR128:$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<(v2i64 (X86Shufp VR128:$src1,
2347 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2348 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2349 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2350 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2351 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2352 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2353 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2354 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2355 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2358 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2359 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2360 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2361 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2362 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2364 def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2365 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2366 def : Pat<(v8f32 (X86Shufp VR256:$src1,
2367 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2368 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2370 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2371 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2372 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2373 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2374 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2376 def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2377 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2378 def : Pat<(v4f64 (X86Shufp VR256:$src1,
2379 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2380 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2383 let Predicates = [HasSSE1] in {
2384 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2385 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2386 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2387 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2388 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2389 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2390 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2391 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2392 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2393 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2394 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2395 // fall back to this for SSE1)
2396 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2397 (SHUFPSrri VR128:$src2, VR128:$src1,
2398 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2399 // Special unary SHUFPSrri case.
2400 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2401 (SHUFPSrri VR128:$src1, VR128:$src1,
2402 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2405 let Predicates = [HasSSE2] in {
2406 // Special binary v4i32 shuffle cases with SHUFPS.
2407 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2408 (SHUFPSrri VR128:$src1, VR128:$src2,
2409 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2410 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2411 (bc_v4i32 (memopv2i64 addr:$src2)))),
2412 (SHUFPSrmi VR128:$src1, addr:$src2,
2413 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2414 // Special unary SHUFPDrri cases.
2415 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2416 (SHUFPDrri VR128:$src1, VR128:$src1,
2417 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2418 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2419 (SHUFPDrri VR128:$src1, VR128:$src1,
2420 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2421 // Special binary v2i64 shuffle cases using SHUFPDrri.
2422 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2423 (SHUFPDrri VR128:$src1, VR128:$src2,
2424 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2425 // Generic SHUFPD patterns
2426 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2427 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2428 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2429 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2430 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2431 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2432 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2433 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2434 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2435 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2438 //===----------------------------------------------------------------------===//
2439 // SSE 1 & 2 - Unpack Instructions
2440 //===----------------------------------------------------------------------===//
2442 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2443 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2444 PatFrag mem_frag, RegisterClass RC,
2445 X86MemOperand x86memop, string asm,
2447 def rr : PI<opc, MRMSrcReg,
2448 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2450 (vt (OpNode RC:$src1, RC:$src2)))],
2452 def rm : PI<opc, MRMSrcMem,
2453 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2455 (vt (OpNode RC:$src1,
2456 (mem_frag addr:$src2))))],
2460 let AddedComplexity = 10 in {
2461 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2462 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2463 SSEPackedSingle>, TB, VEX_4V;
2464 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2465 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2466 SSEPackedDouble>, TB, OpSize, VEX_4V;
2467 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2468 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2469 SSEPackedSingle>, TB, VEX_4V;
2470 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2471 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2472 SSEPackedDouble>, TB, OpSize, VEX_4V;
2474 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2475 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2476 SSEPackedSingle>, TB, VEX_4V;
2477 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2478 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2479 SSEPackedDouble>, TB, OpSize, VEX_4V;
2480 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2481 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2482 SSEPackedSingle>, TB, VEX_4V;
2483 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2484 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2485 SSEPackedDouble>, TB, OpSize, VEX_4V;
2487 let Constraints = "$src1 = $dst" in {
2488 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2489 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2490 SSEPackedSingle>, TB;
2491 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2492 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2493 SSEPackedDouble>, TB, OpSize;
2494 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2495 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2496 SSEPackedSingle>, TB;
2497 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2498 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2499 SSEPackedDouble>, TB, OpSize;
2500 } // Constraints = "$src1 = $dst"
2501 } // AddedComplexity
2503 let Predicates = [HasAVX], AddedComplexity = 1 in {
2504 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2505 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2506 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2507 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2508 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2509 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2510 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2511 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2513 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2514 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2515 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2516 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2517 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2518 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2519 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2520 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2522 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2523 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2524 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2525 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2526 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2527 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2528 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2529 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2531 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2532 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2533 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2534 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2535 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2536 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2537 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2538 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2540 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2541 // problem is during lowering, where it's not possible to recognize the load
2542 // fold cause it has two uses through a bitcast. One use disappears at isel
2543 // time and the fold opportunity reappears.
2544 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2545 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2548 let Predicates = [HasSSE1] in {
2549 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2550 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2551 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2552 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2553 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2554 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2555 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2556 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2559 let Predicates = [HasSSE2] in {
2560 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2561 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2562 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2563 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2564 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2565 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2566 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2567 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2569 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2570 // problem is during lowering, where it's not possible to recognize the load
2571 // fold cause it has two uses through a bitcast. One use disappears at isel
2572 // time and the fold opportunity reappears.
2573 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2574 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2577 //===----------------------------------------------------------------------===//
2578 // SSE 1 & 2 - Extract Floating-Point Sign mask
2579 //===----------------------------------------------------------------------===//
2581 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2582 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2584 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2585 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2586 [(set GR32:$dst, (Int RC:$src))], IIC_DEFAULT, d>;
2587 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2588 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2589 IIC_DEFAULT, d>, REX_W;
2592 let Predicates = [HasAVX] in {
2593 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2594 "movmskps", SSEPackedSingle>, TB, VEX;
2595 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2596 "movmskpd", SSEPackedDouble>, TB,
2598 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2599 "movmskps", SSEPackedSingle>, TB, VEX;
2600 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2601 "movmskpd", SSEPackedDouble>, TB,
2604 def : Pat<(i32 (X86fgetsign FR32:$src)),
2605 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2607 def : Pat<(i64 (X86fgetsign FR32:$src)),
2608 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2610 def : Pat<(i32 (X86fgetsign FR64:$src)),
2611 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2613 def : Pat<(i64 (X86fgetsign FR64:$src)),
2614 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2618 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2619 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2620 SSEPackedSingle>, TB, VEX;
2621 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2622 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2623 SSEPackedDouble>, TB,
2625 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2626 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2627 SSEPackedSingle>, TB, VEX;
2628 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2629 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2630 SSEPackedDouble>, TB,
2634 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2635 SSEPackedSingle>, TB;
2636 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2637 SSEPackedDouble>, TB, OpSize;
2639 def : Pat<(i32 (X86fgetsign FR32:$src)),
2640 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2641 sub_ss))>, Requires<[HasSSE1]>;
2642 def : Pat<(i64 (X86fgetsign FR32:$src)),
2643 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2644 sub_ss))>, Requires<[HasSSE1]>;
2645 def : Pat<(i32 (X86fgetsign FR64:$src)),
2646 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2647 sub_sd))>, Requires<[HasSSE2]>;
2648 def : Pat<(i64 (X86fgetsign FR64:$src)),
2649 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2650 sub_sd))>, Requires<[HasSSE2]>;
2652 //===---------------------------------------------------------------------===//
2653 // SSE2 - Packed Integer Logical Instructions
2654 //===---------------------------------------------------------------------===//
2656 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2658 /// PDI_binop_rm - Simple SSE2 binary operator.
2659 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2660 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2661 X86MemOperand x86memop, bit IsCommutable = 0,
2663 let isCommutable = IsCommutable in
2664 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2665 (ins RC:$src1, RC:$src2),
2667 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2668 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2669 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2670 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2671 (ins RC:$src1, x86memop:$src2),
2673 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2674 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2675 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2676 (bitconvert (memop_frag addr:$src2)))))]>;
2678 } // ExeDomain = SSEPackedInt
2680 // These are ordered here for pattern ordering requirements with the fp versions
2682 let Predicates = [HasAVX] in {
2683 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2684 i128mem, 1, 0>, VEX_4V;
2685 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2686 i128mem, 1, 0>, VEX_4V;
2687 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2688 i128mem, 1, 0>, VEX_4V;
2689 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2690 i128mem, 0, 0>, VEX_4V;
2693 let Constraints = "$src1 = $dst" in {
2694 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2696 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2698 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2700 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2702 } // Constraints = "$src1 = $dst"
2704 let Predicates = [HasAVX2] in {
2705 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2706 i256mem, 1, 0>, VEX_4V;
2707 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2708 i256mem, 1, 0>, VEX_4V;
2709 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2710 i256mem, 1, 0>, VEX_4V;
2711 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2712 i256mem, 0, 0>, VEX_4V;
2715 //===----------------------------------------------------------------------===//
2716 // SSE 1 & 2 - Logical Instructions
2717 //===----------------------------------------------------------------------===//
2719 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2721 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2723 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2724 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2726 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2727 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2729 let Constraints = "$src1 = $dst" in {
2730 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2731 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2733 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2734 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2738 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2739 let mayLoad = 0 in {
2740 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2741 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2742 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2745 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2746 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2748 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2750 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2752 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2753 // are all promoted to v2i64, and the patterns are covered by the int
2754 // version. This is needed in SSE only, because v2i64 isn't supported on
2755 // SSE1, but only on SSE2.
2756 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2757 !strconcat(OpcodeStr, "ps"), f128mem, [],
2758 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2759 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2761 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2762 !strconcat(OpcodeStr, "pd"), f128mem,
2763 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2764 (bc_v2i64 (v2f64 VR128:$src2))))],
2765 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2766 (memopv2i64 addr:$src2)))], 0>,
2768 let Constraints = "$src1 = $dst" in {
2769 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2770 !strconcat(OpcodeStr, "ps"), f128mem,
2771 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2772 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2773 (memopv2i64 addr:$src2)))]>, TB;
2775 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2776 !strconcat(OpcodeStr, "pd"), f128mem,
2777 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2778 (bc_v2i64 (v2f64 VR128:$src2))))],
2779 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2780 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2784 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2786 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2788 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2789 !strconcat(OpcodeStr, "ps"), f256mem,
2790 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2791 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2792 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2794 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2795 !strconcat(OpcodeStr, "pd"), f256mem,
2796 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2797 (bc_v4i64 (v4f64 VR256:$src2))))],
2798 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2799 (memopv4i64 addr:$src2)))], 0>,
2803 // AVX 256-bit packed logical ops forms
2804 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2805 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2806 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2807 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2809 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2810 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2811 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2812 let isCommutable = 0 in
2813 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2815 //===----------------------------------------------------------------------===//
2816 // SSE 1 & 2 - Arithmetic Instructions
2817 //===----------------------------------------------------------------------===//
2819 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2822 /// In addition, we also have a special variant of the scalar form here to
2823 /// represent the associated intrinsic operation. This form is unlike the
2824 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2825 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2827 /// These three forms can each be reg+reg or reg+mem.
2830 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2832 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2834 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2835 OpNode, FR32, f32mem, Is2Addr>, XS;
2836 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2837 OpNode, FR64, f64mem, Is2Addr>, XD;
2840 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2842 let mayLoad = 0 in {
2843 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2844 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2845 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2846 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2850 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2852 let mayLoad = 0 in {
2853 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2854 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2855 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2856 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2860 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2862 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2863 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2864 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2865 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2868 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2870 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2871 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2872 SSEPackedSingle, Is2Addr>, TB;
2874 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2875 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2876 SSEPackedDouble, Is2Addr>, TB, OpSize;
2879 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2880 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2881 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2882 SSEPackedSingle, 0>, TB;
2884 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2885 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2886 SSEPackedDouble, 0>, TB, OpSize;
2889 // Binary Arithmetic instructions
2890 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2891 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2892 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2893 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2894 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2895 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2896 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2897 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2899 let isCommutable = 0 in {
2900 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2901 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2902 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2903 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2904 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2905 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2906 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2907 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2908 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2909 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2910 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2911 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2912 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2913 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2914 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2915 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2916 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2917 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2918 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2919 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2922 let Constraints = "$src1 = $dst" in {
2923 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2924 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2925 basic_sse12_fp_binop_s_int<0x58, "add">;
2926 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2927 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2928 basic_sse12_fp_binop_s_int<0x59, "mul">;
2930 let isCommutable = 0 in {
2931 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2932 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2933 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2934 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2935 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2936 basic_sse12_fp_binop_s_int<0x5E, "div">;
2937 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2938 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2939 basic_sse12_fp_binop_s_int<0x5F, "max">,
2940 basic_sse12_fp_binop_p_int<0x5F, "max">;
2941 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2942 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2943 basic_sse12_fp_binop_s_int<0x5D, "min">,
2944 basic_sse12_fp_binop_p_int<0x5D, "min">;
2949 /// In addition, we also have a special variant of the scalar form here to
2950 /// represent the associated intrinsic operation. This form is unlike the
2951 /// plain scalar form, in that it takes an entire vector (instead of a
2952 /// scalar) and leaves the top elements undefined.
2954 /// And, we have a special variant form for a full-vector intrinsic form.
2956 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2957 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2958 SDNode OpNode, Intrinsic F32Int> {
2959 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2960 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2961 [(set FR32:$dst, (OpNode FR32:$src))]>;
2962 // For scalar unary operations, fold a load into the operation
2963 // only in OptForSize mode. It eliminates an instruction, but it also
2964 // eliminates a whole-register clobber (the load), so it introduces a
2965 // partial register update condition.
2966 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2967 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2968 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2969 Requires<[HasSSE1, OptForSize]>;
2970 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2971 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2972 [(set VR128:$dst, (F32Int VR128:$src))]>;
2973 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2974 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2975 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2978 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2979 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2980 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2981 !strconcat(OpcodeStr,
2982 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2984 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2985 !strconcat(OpcodeStr,
2986 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2987 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2988 (ins VR128:$src1, ssmem:$src2),
2989 !strconcat(OpcodeStr,
2990 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2993 /// sse1_fp_unop_p - SSE1 unops in packed form.
2994 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2995 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2996 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2997 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2998 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2999 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3000 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
3003 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3004 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3005 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3006 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3007 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
3008 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3009 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3010 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
3013 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3014 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3015 Intrinsic V4F32Int> {
3016 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3017 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3018 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
3019 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3020 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3021 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
3024 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3025 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3026 Intrinsic V4F32Int> {
3027 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3028 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3029 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
3030 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3031 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3032 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
3035 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3036 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3037 SDNode OpNode, Intrinsic F64Int> {
3038 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3039 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3040 [(set FR64:$dst, (OpNode FR64:$src))]>;
3041 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3042 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3043 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3044 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
3045 Requires<[HasSSE2, OptForSize]>;
3046 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3047 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3048 [(set VR128:$dst, (F64Int VR128:$src))]>;
3049 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3050 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3051 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
3054 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3055 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3056 let neverHasSideEffects = 1 in {
3057 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3058 !strconcat(OpcodeStr,
3059 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3061 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3062 !strconcat(OpcodeStr,
3063 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3065 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3066 (ins VR128:$src1, sdmem:$src2),
3067 !strconcat(OpcodeStr,
3068 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3071 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3072 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3074 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3075 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3076 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3077 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3078 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3079 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3082 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3083 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3084 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3085 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3086 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3087 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3088 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3089 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3092 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3093 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3094 Intrinsic V2F64Int> {
3095 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3096 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3097 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3098 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3099 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3100 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3103 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3104 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3105 Intrinsic V2F64Int> {
3106 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3107 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3108 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3109 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3110 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3111 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3114 let Predicates = [HasAVX] in {
3116 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3117 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3119 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3120 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3121 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3122 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3123 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3124 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3125 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3126 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3129 // Reciprocal approximations. Note that these typically require refinement
3130 // in order to obtain suitable precision.
3131 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3132 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3133 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3134 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3135 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3137 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3138 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3139 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3140 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3141 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3144 let AddedComplexity = 1 in {
3145 def : Pat<(f32 (fsqrt FR32:$src)),
3146 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3147 def : Pat<(f32 (fsqrt (load addr:$src))),
3148 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3149 Requires<[HasAVX, OptForSize]>;
3150 def : Pat<(f64 (fsqrt FR64:$src)),
3151 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3152 def : Pat<(f64 (fsqrt (load addr:$src))),
3153 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3154 Requires<[HasAVX, OptForSize]>;
3156 def : Pat<(f32 (X86frsqrt FR32:$src)),
3157 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3158 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3159 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3160 Requires<[HasAVX, OptForSize]>;
3162 def : Pat<(f32 (X86frcp FR32:$src)),
3163 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3164 def : Pat<(f32 (X86frcp (load addr:$src))),
3165 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3166 Requires<[HasAVX, OptForSize]>;
3169 let Predicates = [HasAVX], AddedComplexity = 1 in {
3170 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3171 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3172 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3173 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3175 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3176 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3178 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3179 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3180 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3181 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3183 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3184 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3186 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3187 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3188 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3189 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3191 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3192 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3194 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3195 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3196 (VRCPSSr (f32 (IMPLICIT_DEF)),
3197 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3199 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3200 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3204 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3205 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3206 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3207 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3208 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3209 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3211 // Reciprocal approximations. Note that these typically require refinement
3212 // in order to obtain suitable precision.
3213 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3214 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3215 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3216 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3217 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3218 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3220 // There is no f64 version of the reciprocal approximation instructions.
3222 //===----------------------------------------------------------------------===//
3223 // SSE 1 & 2 - Non-temporal stores
3224 //===----------------------------------------------------------------------===//
3226 let AddedComplexity = 400 in { // Prefer non-temporal versions
3227 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3228 (ins f128mem:$dst, VR128:$src),
3229 "movntps\t{$src, $dst|$dst, $src}",
3230 [(alignednontemporalstore (v4f32 VR128:$src),
3232 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3233 (ins f128mem:$dst, VR128:$src),
3234 "movntpd\t{$src, $dst|$dst, $src}",
3235 [(alignednontemporalstore (v2f64 VR128:$src),
3238 let ExeDomain = SSEPackedInt in
3239 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3240 (ins f128mem:$dst, VR128:$src),
3241 "movntdq\t{$src, $dst|$dst, $src}",
3242 [(alignednontemporalstore (v2i64 VR128:$src),
3245 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3246 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3248 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3249 (ins f256mem:$dst, VR256:$src),
3250 "movntps\t{$src, $dst|$dst, $src}",
3251 [(alignednontemporalstore (v8f32 VR256:$src),
3253 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3254 (ins f256mem:$dst, VR256:$src),
3255 "movntpd\t{$src, $dst|$dst, $src}",
3256 [(alignednontemporalstore (v4f64 VR256:$src),
3258 let ExeDomain = SSEPackedInt in
3259 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3260 (ins f256mem:$dst, VR256:$src),
3261 "movntdq\t{$src, $dst|$dst, $src}",
3262 [(alignednontemporalstore (v4i64 VR256:$src),
3266 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3267 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3268 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3269 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3270 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3271 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3273 let AddedComplexity = 400 in { // Prefer non-temporal versions
3274 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3275 "movntps\t{$src, $dst|$dst, $src}",
3276 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3277 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3278 "movntpd\t{$src, $dst|$dst, $src}",
3279 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3281 let ExeDomain = SSEPackedInt in
3282 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3283 "movntdq\t{$src, $dst|$dst, $src}",
3284 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)]>;
3286 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3287 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3289 // There is no AVX form for instructions below this point
3290 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3291 "movnti{l}\t{$src, $dst|$dst, $src}",
3292 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3293 TB, Requires<[HasSSE2]>;
3294 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3295 "movnti{q}\t{$src, $dst|$dst, $src}",
3296 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3297 TB, Requires<[HasSSE2]>;
3300 //===----------------------------------------------------------------------===//
3301 // SSE 1 & 2 - Prefetch and memory fence
3302 //===----------------------------------------------------------------------===//
3304 // Prefetch intrinsic.
3305 let Predicates = [HasSSE1] in {
3306 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3307 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3308 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3309 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3310 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3311 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3312 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3313 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3317 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3318 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3319 TB, Requires<[HasSSE2]>;
3321 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3322 // was introduced with SSE2, it's backward compatible.
3323 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3325 // Load, store, and memory fence
3326 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3327 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3328 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3329 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3330 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3331 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3333 def : Pat<(X86SFence), (SFENCE)>;
3334 def : Pat<(X86LFence), (LFENCE)>;
3335 def : Pat<(X86MFence), (MFENCE)>;
3337 //===----------------------------------------------------------------------===//
3338 // SSE 1 & 2 - Load/Store XCSR register
3339 //===----------------------------------------------------------------------===//
3341 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3342 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3343 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3344 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3346 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3347 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3348 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3349 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3351 //===---------------------------------------------------------------------===//
3352 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3353 //===---------------------------------------------------------------------===//
3355 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3357 let neverHasSideEffects = 1 in {
3358 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3359 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3360 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3361 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3363 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3364 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3365 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3366 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3369 let isCodeGenOnly = 1 in {
3370 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3371 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3372 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3373 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3374 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3375 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3376 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3377 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3380 let canFoldAsLoad = 1, mayLoad = 1 in {
3381 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3382 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3383 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3384 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3385 let Predicates = [HasAVX] in {
3386 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3387 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3388 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3389 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3393 let mayStore = 1 in {
3394 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3395 (ins i128mem:$dst, VR128:$src),
3396 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3397 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3398 (ins i256mem:$dst, VR256:$src),
3399 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3400 let Predicates = [HasAVX] in {
3401 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3402 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3403 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3404 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3408 let neverHasSideEffects = 1 in
3409 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3410 "movdqa\t{$src, $dst|$dst, $src}", []>;
3412 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3413 "movdqu\t{$src, $dst|$dst, $src}",
3414 []>, XS, Requires<[HasSSE2]>;
3417 let isCodeGenOnly = 1 in {
3418 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3419 "movdqa\t{$src, $dst|$dst, $src}", []>;
3421 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3422 "movdqu\t{$src, $dst|$dst, $src}",
3423 []>, XS, Requires<[HasSSE2]>;
3426 let canFoldAsLoad = 1, mayLoad = 1 in {
3427 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3428 "movdqa\t{$src, $dst|$dst, $src}",
3429 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3430 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3431 "movdqu\t{$src, $dst|$dst, $src}",
3432 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3433 XS, Requires<[HasSSE2]>;
3436 let mayStore = 1 in {
3437 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3438 "movdqa\t{$src, $dst|$dst, $src}",
3439 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3440 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3441 "movdqu\t{$src, $dst|$dst, $src}",
3442 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3443 XS, Requires<[HasSSE2]>;
3446 // Intrinsic forms of MOVDQU load and store
3447 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3448 "vmovdqu\t{$src, $dst|$dst, $src}",
3449 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3450 XS, VEX, Requires<[HasAVX]>;
3452 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3453 "movdqu\t{$src, $dst|$dst, $src}",
3454 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3455 XS, Requires<[HasSSE2]>;
3457 } // ExeDomain = SSEPackedInt
3459 let Predicates = [HasAVX] in {
3460 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3461 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3464 //===---------------------------------------------------------------------===//
3465 // SSE2 - Packed Integer Arithmetic Instructions
3466 //===---------------------------------------------------------------------===//
3468 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3470 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3471 RegisterClass RC, PatFrag memop_frag,
3472 X86MemOperand x86memop, bit IsCommutable = 0,
3474 let isCommutable = IsCommutable in
3475 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3476 (ins RC:$src1, RC:$src2),
3478 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3479 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3480 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3481 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3482 (ins RC:$src1, x86memop:$src2),
3484 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3485 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3486 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3489 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3490 string OpcodeStr, SDNode OpNode,
3491 SDNode OpNode2, RegisterClass RC,
3492 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3494 // src2 is always 128-bit
3495 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3496 (ins RC:$src1, VR128:$src2),
3498 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3499 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3500 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))]>;
3501 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3502 (ins RC:$src1, i128mem:$src2),
3504 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3505 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3506 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3507 (bc_frag (memopv2i64 addr:$src2)))))]>;
3508 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3509 (ins RC:$src1, i32i8imm:$src2),
3511 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3512 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3513 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))]>;
3516 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3517 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3518 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3519 PatFrag memop_frag, X86MemOperand x86memop,
3520 bit IsCommutable = 0, bit Is2Addr = 1> {
3521 let isCommutable = IsCommutable in
3522 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3523 (ins RC:$src1, RC:$src2),
3525 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3526 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3527 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3528 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3529 (ins RC:$src1, x86memop:$src2),
3531 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3532 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3533 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3534 (bitconvert (memop_frag addr:$src2)))))]>;
3536 } // ExeDomain = SSEPackedInt
3538 // 128-bit Integer Arithmetic
3540 let Predicates = [HasAVX] in {
3541 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3542 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3543 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3544 i128mem, 1, 0>, VEX_4V;
3545 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3546 i128mem, 1, 0>, VEX_4V;
3547 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3548 i128mem, 1, 0>, VEX_4V;
3549 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3550 i128mem, 1, 0>, VEX_4V;
3551 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3552 i128mem, 0, 0>, VEX_4V;
3553 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3554 i128mem, 0, 0>, VEX_4V;
3555 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3556 i128mem, 0, 0>, VEX_4V;
3557 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3558 i128mem, 0, 0>, VEX_4V;
3559 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3560 memopv2i64, i128mem, 1, 0>, VEX_4V;
3563 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3564 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3565 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3566 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3567 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3568 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3569 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3570 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3571 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3572 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3573 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3574 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3575 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3576 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3577 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3578 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3579 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3580 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3581 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3582 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3583 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3584 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3585 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3586 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3587 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3588 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3589 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3590 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3591 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3592 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3593 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3594 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3595 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3596 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3597 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3598 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3601 let Predicates = [HasAVX2] in {
3602 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3603 i256mem, 1, 0>, VEX_4V;
3604 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3605 i256mem, 1, 0>, VEX_4V;
3606 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3607 i256mem, 1, 0>, VEX_4V;
3608 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3609 i256mem, 1, 0>, VEX_4V;
3610 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3611 i256mem, 1, 0>, VEX_4V;
3612 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3613 i256mem, 0, 0>, VEX_4V;
3614 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3615 i256mem, 0, 0>, VEX_4V;
3616 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3617 i256mem, 0, 0>, VEX_4V;
3618 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3619 i256mem, 0, 0>, VEX_4V;
3620 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3621 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3624 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3625 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3626 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3627 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3628 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3629 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3630 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3631 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3632 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3633 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3634 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3635 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3636 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3637 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3638 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3639 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3640 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3641 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3642 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3643 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3644 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3645 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3646 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3647 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3648 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3649 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3650 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3651 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3652 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3653 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3654 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3655 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3656 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3657 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3658 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3659 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3662 let Constraints = "$src1 = $dst" in {
3663 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3665 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3667 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3669 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3671 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3673 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3675 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3677 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3679 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3681 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3682 memopv2i64, i128mem, 1>;
3685 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3686 VR128, memopv2i64, i128mem>;
3687 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3688 VR128, memopv2i64, i128mem>;
3689 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3690 VR128, memopv2i64, i128mem>;
3691 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3692 VR128, memopv2i64, i128mem>;
3693 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3694 VR128, memopv2i64, i128mem, 1>;
3695 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3696 VR128, memopv2i64, i128mem, 1>;
3697 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3698 VR128, memopv2i64, i128mem, 1>;
3699 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3700 VR128, memopv2i64, i128mem, 1>;
3701 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3702 VR128, memopv2i64, i128mem, 1>;
3703 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3704 VR128, memopv2i64, i128mem, 1>;
3705 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3706 VR128, memopv2i64, i128mem, 1>;
3707 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3708 VR128, memopv2i64, i128mem, 1>;
3709 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3710 VR128, memopv2i64, i128mem, 1>;
3711 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3712 VR128, memopv2i64, i128mem, 1>;
3713 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3714 VR128, memopv2i64, i128mem, 1>;
3715 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3716 VR128, memopv2i64, i128mem, 1>;
3717 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3718 VR128, memopv2i64, i128mem, 1>;
3719 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3720 VR128, memopv2i64, i128mem, 1>;
3722 } // Constraints = "$src1 = $dst"
3724 //===---------------------------------------------------------------------===//
3725 // SSE2 - Packed Integer Logical Instructions
3726 //===---------------------------------------------------------------------===//
3728 let Predicates = [HasAVX] in {
3729 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3730 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3731 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3732 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3733 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3734 VR128, v2i64, v2i64, bc_v2i64, 0>, VEX_4V;
3736 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3737 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3738 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3739 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3740 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3741 VR128, v2i64, v2i64, bc_v2i64, 0>, VEX_4V;
3743 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3744 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3745 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3746 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3748 let ExeDomain = SSEPackedInt in {
3749 // 128-bit logical shifts.
3750 def VPSLLDQri : PDIi8<0x73, MRM7r,
3751 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3752 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3754 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3756 def VPSRLDQri : PDIi8<0x73, MRM3r,
3757 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3758 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3760 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3762 // PSRADQri doesn't exist in SSE[1-3].
3764 } // Predicates = [HasAVX]
3766 let Predicates = [HasAVX2] in {
3767 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3768 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3769 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3770 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3771 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3772 VR256, v4i64, v2i64, bc_v2i64, 0>, VEX_4V;
3774 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3775 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3776 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3777 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3778 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3779 VR256, v4i64, v2i64, bc_v2i64, 0>, VEX_4V;
3781 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3782 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3783 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3784 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3786 let ExeDomain = SSEPackedInt in {
3787 // 256-bit logical shifts.
3788 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3789 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3790 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3792 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3794 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3795 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3796 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3798 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3800 // PSRADQYri doesn't exist in SSE[1-3].
3802 } // Predicates = [HasAVX2]
3804 let Constraints = "$src1 = $dst" in {
3805 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3806 VR128, v8i16, v8i16, bc_v8i16>;
3807 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3808 VR128, v4i32, v4i32, bc_v4i32>;
3809 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3810 VR128, v2i64, v2i64, bc_v2i64>;
3812 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3813 VR128, v8i16, v8i16, bc_v8i16>;
3814 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3815 VR128, v4i32, v4i32, bc_v4i32>;
3816 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3817 VR128, v2i64, v2i64, bc_v2i64>;
3819 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3820 VR128, v8i16, v8i16, bc_v8i16>;
3821 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3822 VR128, v4i32, v4i32, bc_v4i32>;
3824 let ExeDomain = SSEPackedInt in {
3825 // 128-bit logical shifts.
3826 def PSLLDQri : PDIi8<0x73, MRM7r,
3827 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3828 "pslldq\t{$src2, $dst|$dst, $src2}",
3830 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3831 def PSRLDQri : PDIi8<0x73, MRM3r,
3832 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3833 "psrldq\t{$src2, $dst|$dst, $src2}",
3835 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3836 // PSRADQri doesn't exist in SSE[1-3].
3838 } // Constraints = "$src1 = $dst"
3840 let Predicates = [HasAVX] in {
3841 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3842 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3843 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3844 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3845 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3846 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3848 // Shift up / down and insert zero's.
3849 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3850 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3851 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3852 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3855 let Predicates = [HasAVX2] in {
3856 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3857 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3858 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3859 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3862 let Predicates = [HasSSE2] in {
3863 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3864 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3865 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3866 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3867 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3868 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3870 // Shift up / down and insert zero's.
3871 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3872 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3873 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3874 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3877 //===---------------------------------------------------------------------===//
3878 // SSE2 - Packed Integer Comparison Instructions
3879 //===---------------------------------------------------------------------===//
3881 let Predicates = [HasAVX] in {
3882 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
3883 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3884 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
3885 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3886 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
3887 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3888 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
3889 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3890 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
3891 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3892 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
3893 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3896 let Predicates = [HasAVX2] in {
3897 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
3898 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3899 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
3900 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3901 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
3902 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3903 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
3904 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3905 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
3906 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3907 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
3908 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3911 let Constraints = "$src1 = $dst" in {
3912 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
3913 VR128, memopv2i64, i128mem, 1>;
3914 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
3915 VR128, memopv2i64, i128mem, 1>;
3916 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
3917 VR128, memopv2i64, i128mem, 1>;
3918 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
3919 VR128, memopv2i64, i128mem>;
3920 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
3921 VR128, memopv2i64, i128mem>;
3922 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
3923 VR128, memopv2i64, i128mem>;
3924 } // Constraints = "$src1 = $dst"
3926 //===---------------------------------------------------------------------===//
3927 // SSE2 - Packed Integer Pack Instructions
3928 //===---------------------------------------------------------------------===//
3930 let Predicates = [HasAVX] in {
3931 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3932 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3933 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3934 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3935 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3936 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3939 let Predicates = [HasAVX2] in {
3940 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
3941 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3942 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
3943 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3944 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
3945 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3948 let Constraints = "$src1 = $dst" in {
3949 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
3950 VR128, memopv2i64, i128mem>;
3951 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
3952 VR128, memopv2i64, i128mem>;
3953 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
3954 VR128, memopv2i64, i128mem>;
3955 } // Constraints = "$src1 = $dst"
3957 //===---------------------------------------------------------------------===//
3958 // SSE2 - Packed Integer Shuffle Instructions
3959 //===---------------------------------------------------------------------===//
3961 let ExeDomain = SSEPackedInt in {
3962 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
3964 def ri : Ii8<0x70, MRMSrcReg,
3965 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3966 !strconcat(OpcodeStr,
3967 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3968 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
3970 def mi : Ii8<0x70, MRMSrcMem,
3971 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3972 !strconcat(OpcodeStr,
3973 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3974 [(set VR128:$dst, (vt (pshuf_frag:$src2
3975 (bc_frag (memopv2i64 addr:$src1)),
3979 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
3980 def Yri : Ii8<0x70, MRMSrcReg,
3981 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
3982 !strconcat(OpcodeStr,
3983 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3984 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
3985 def Ymi : Ii8<0x70, MRMSrcMem,
3986 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
3987 !strconcat(OpcodeStr,
3988 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3990 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
3991 (i8 imm:$src2))))]>;
3993 } // ExeDomain = SSEPackedInt
3995 let Predicates = [HasAVX] in {
3996 let AddedComplexity = 5 in
3997 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4000 // SSE2 with ImmT == Imm8 and XS prefix.
4001 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4004 // SSE2 with ImmT == Imm8 and XD prefix.
4005 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4008 let AddedComplexity = 5 in
4009 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4010 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4011 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4012 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4013 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4015 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4017 (VPSHUFDmi addr:$src1, imm:$imm)>;
4018 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4019 (VPSHUFDmi addr:$src1, imm:$imm)>;
4020 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4021 (VPSHUFDri VR128:$src1, imm:$imm)>;
4022 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4023 (VPSHUFDri VR128:$src1, imm:$imm)>;
4024 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4025 (VPSHUFHWri VR128:$src, imm:$imm)>;
4026 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4028 (VPSHUFHWmi addr:$src, imm:$imm)>;
4029 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4030 (VPSHUFLWri VR128:$src, imm:$imm)>;
4031 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4033 (VPSHUFLWmi addr:$src, imm:$imm)>;
4036 let Predicates = [HasAVX2] in {
4037 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
4038 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
4039 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
4042 let Predicates = [HasSSE2] in {
4043 let AddedComplexity = 5 in
4044 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4046 // SSE2 with ImmT == Imm8 and XS prefix.
4047 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4049 // SSE2 with ImmT == Imm8 and XD prefix.
4050 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4052 let AddedComplexity = 5 in
4053 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4054 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4055 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4056 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4057 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4059 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4061 (PSHUFDmi addr:$src1, imm:$imm)>;
4062 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4063 (PSHUFDmi addr:$src1, imm:$imm)>;
4064 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4065 (PSHUFDri VR128:$src1, imm:$imm)>;
4066 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4067 (PSHUFDri VR128:$src1, imm:$imm)>;
4068 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4069 (PSHUFHWri VR128:$src, imm:$imm)>;
4070 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4072 (PSHUFHWmi addr:$src, imm:$imm)>;
4073 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4074 (PSHUFLWri VR128:$src, imm:$imm)>;
4075 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4077 (PSHUFLWmi addr:$src, imm:$imm)>;
4080 //===---------------------------------------------------------------------===//
4081 // SSE2 - Packed Integer Unpack Instructions
4082 //===---------------------------------------------------------------------===//
4084 let ExeDomain = SSEPackedInt in {
4085 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4086 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4087 def rr : PDI<opc, MRMSrcReg,
4088 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4090 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4091 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4092 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4093 def rm : PDI<opc, MRMSrcMem,
4094 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4096 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4097 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4098 [(set VR128:$dst, (OpNode VR128:$src1,
4099 (bc_frag (memopv2i64
4103 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4104 SDNode OpNode, PatFrag bc_frag> {
4105 def Yrr : PDI<opc, MRMSrcReg,
4106 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4107 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4108 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4109 def Yrm : PDI<opc, MRMSrcMem,
4110 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4111 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4112 [(set VR256:$dst, (OpNode VR256:$src1,
4113 (bc_frag (memopv4i64 addr:$src2))))]>;
4116 let Predicates = [HasAVX] in {
4117 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4118 bc_v16i8, 0>, VEX_4V;
4119 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4120 bc_v8i16, 0>, VEX_4V;
4121 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4122 bc_v4i32, 0>, VEX_4V;
4123 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4124 bc_v2i64, 0>, VEX_4V;
4126 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4127 bc_v16i8, 0>, VEX_4V;
4128 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4129 bc_v8i16, 0>, VEX_4V;
4130 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4131 bc_v4i32, 0>, VEX_4V;
4132 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4133 bc_v2i64, 0>, VEX_4V;
4136 let Predicates = [HasAVX2] in {
4137 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4139 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4141 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4143 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4146 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4148 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4150 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4152 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4156 let Constraints = "$src1 = $dst" in {
4157 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4159 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4161 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4163 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4166 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4168 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4170 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4172 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4175 } // ExeDomain = SSEPackedInt
4177 // Patterns for using AVX1 instructions with integer vectors
4178 // Here to give AVX2 priority
4179 let Predicates = [HasAVX] in {
4180 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4181 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4182 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4183 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4184 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4185 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4186 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4187 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4189 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4190 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4191 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4192 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4193 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4194 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4195 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4196 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4199 //===---------------------------------------------------------------------===//
4200 // SSE2 - Packed Integer Extract and Insert
4201 //===---------------------------------------------------------------------===//
4203 let ExeDomain = SSEPackedInt in {
4204 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4205 def rri : Ii8<0xC4, MRMSrcReg,
4206 (outs VR128:$dst), (ins VR128:$src1,
4207 GR32:$src2, i32i8imm:$src3),
4209 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4210 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4212 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4213 def rmi : Ii8<0xC4, MRMSrcMem,
4214 (outs VR128:$dst), (ins VR128:$src1,
4215 i16mem:$src2, i32i8imm:$src3),
4217 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4218 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4220 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4225 let Predicates = [HasAVX] in
4226 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4227 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4228 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4229 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4230 imm:$src2))]>, TB, OpSize, VEX;
4231 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4232 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4233 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4234 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4238 let Predicates = [HasAVX] in {
4239 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4240 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4241 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4242 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4243 []>, TB, OpSize, VEX_4V;
4246 let Constraints = "$src1 = $dst" in
4247 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4249 } // ExeDomain = SSEPackedInt
4251 //===---------------------------------------------------------------------===//
4252 // SSE2 - Packed Mask Creation
4253 //===---------------------------------------------------------------------===//
4255 let ExeDomain = SSEPackedInt in {
4257 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4258 "pmovmskb\t{$src, $dst|$dst, $src}",
4259 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4260 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4261 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4263 let Predicates = [HasAVX2] in {
4264 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4265 "pmovmskb\t{$src, $dst|$dst, $src}",
4266 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4267 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4268 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4271 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4272 "pmovmskb\t{$src, $dst|$dst, $src}",
4273 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4275 } // ExeDomain = SSEPackedInt
4277 //===---------------------------------------------------------------------===//
4278 // SSE2 - Conditional Store
4279 //===---------------------------------------------------------------------===//
4281 let ExeDomain = SSEPackedInt in {
4284 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4285 (ins VR128:$src, VR128:$mask),
4286 "maskmovdqu\t{$mask, $src|$src, $mask}",
4287 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4289 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4290 (ins VR128:$src, VR128:$mask),
4291 "maskmovdqu\t{$mask, $src|$src, $mask}",
4292 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4295 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4296 "maskmovdqu\t{$mask, $src|$src, $mask}",
4297 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4299 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4300 "maskmovdqu\t{$mask, $src|$src, $mask}",
4301 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4303 } // ExeDomain = SSEPackedInt
4305 //===---------------------------------------------------------------------===//
4306 // SSE2 - Move Doubleword
4307 //===---------------------------------------------------------------------===//
4309 //===---------------------------------------------------------------------===//
4310 // Move Int Doubleword to Packed Double Int
4312 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4313 "movd\t{$src, $dst|$dst, $src}",
4315 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4316 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4317 "movd\t{$src, $dst|$dst, $src}",
4319 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4321 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4322 "mov{d|q}\t{$src, $dst|$dst, $src}",
4324 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4325 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4326 "mov{d|q}\t{$src, $dst|$dst, $src}",
4327 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4329 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4330 "movd\t{$src, $dst|$dst, $src}",
4332 (v4i32 (scalar_to_vector GR32:$src)))]>;
4333 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4334 "movd\t{$src, $dst|$dst, $src}",
4336 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4337 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4338 "mov{d|q}\t{$src, $dst|$dst, $src}",
4340 (v2i64 (scalar_to_vector GR64:$src)))]>;
4341 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4342 "mov{d|q}\t{$src, $dst|$dst, $src}",
4343 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4345 //===---------------------------------------------------------------------===//
4346 // Move Int Doubleword to Single Scalar
4348 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4349 "movd\t{$src, $dst|$dst, $src}",
4350 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4352 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4353 "movd\t{$src, $dst|$dst, $src}",
4354 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4356 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4357 "movd\t{$src, $dst|$dst, $src}",
4358 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4360 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4361 "movd\t{$src, $dst|$dst, $src}",
4362 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4364 //===---------------------------------------------------------------------===//
4365 // Move Packed Doubleword Int to Packed Double Int
4367 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4368 "movd\t{$src, $dst|$dst, $src}",
4369 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4371 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4372 (ins i32mem:$dst, VR128:$src),
4373 "movd\t{$src, $dst|$dst, $src}",
4374 [(store (i32 (vector_extract (v4i32 VR128:$src),
4375 (iPTR 0))), addr:$dst)]>, VEX;
4376 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4377 "movd\t{$src, $dst|$dst, $src}",
4378 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4380 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4381 "movd\t{$src, $dst|$dst, $src}",
4382 [(store (i32 (vector_extract (v4i32 VR128:$src),
4383 (iPTR 0))), addr:$dst)]>;
4385 //===---------------------------------------------------------------------===//
4386 // Move Packed Doubleword Int first element to Doubleword Int
4388 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4389 "mov{d|q}\t{$src, $dst|$dst, $src}",
4390 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4392 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4394 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4395 "mov{d|q}\t{$src, $dst|$dst, $src}",
4396 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4399 //===---------------------------------------------------------------------===//
4400 // Bitcast FR64 <-> GR64
4402 let Predicates = [HasAVX] in
4403 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4404 "vmovq\t{$src, $dst|$dst, $src}",
4405 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4407 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4408 "mov{d|q}\t{$src, $dst|$dst, $src}",
4409 [(set GR64:$dst, (bitconvert FR64:$src))]>, VEX;
4410 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4411 "movq\t{$src, $dst|$dst, $src}",
4412 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>,
4415 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4416 "movq\t{$src, $dst|$dst, $src}",
4417 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4418 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4419 "mov{d|q}\t{$src, $dst|$dst, $src}",
4420 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4421 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4422 "movq\t{$src, $dst|$dst, $src}",
4423 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4425 //===---------------------------------------------------------------------===//
4426 // Move Scalar Single to Double Int
4428 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4429 "movd\t{$src, $dst|$dst, $src}",
4430 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4431 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4432 "movd\t{$src, $dst|$dst, $src}",
4433 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4434 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4435 "movd\t{$src, $dst|$dst, $src}",
4436 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4437 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4438 "movd\t{$src, $dst|$dst, $src}",
4439 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4441 //===---------------------------------------------------------------------===//
4442 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4444 let AddedComplexity = 15 in {
4445 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4446 "movd\t{$src, $dst|$dst, $src}",
4447 [(set VR128:$dst, (v4i32 (X86vzmovl
4448 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4450 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4451 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4452 [(set VR128:$dst, (v2i64 (X86vzmovl
4453 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4456 let AddedComplexity = 15 in {
4457 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4458 "movd\t{$src, $dst|$dst, $src}",
4459 [(set VR128:$dst, (v4i32 (X86vzmovl
4460 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4461 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4462 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4463 [(set VR128:$dst, (v2i64 (X86vzmovl
4464 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4467 let AddedComplexity = 20 in {
4468 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4469 "movd\t{$src, $dst|$dst, $src}",
4471 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4472 (loadi32 addr:$src))))))]>,
4474 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4475 "movd\t{$src, $dst|$dst, $src}",
4477 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4478 (loadi32 addr:$src))))))]>;
4481 let Predicates = [HasAVX] in {
4482 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4483 let AddedComplexity = 20 in {
4484 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4485 (VMOVZDI2PDIrm addr:$src)>;
4486 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4487 (VMOVZDI2PDIrm addr:$src)>;
4489 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4490 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4491 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4492 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4493 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4494 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4495 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4498 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4499 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4500 (MOVZDI2PDIrm addr:$src)>;
4501 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4502 (MOVZDI2PDIrm addr:$src)>;
4505 // These are the correct encodings of the instructions so that we know how to
4506 // read correct assembly, even though we continue to emit the wrong ones for
4507 // compatibility with Darwin's buggy assembler.
4508 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4509 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4510 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4511 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4512 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4513 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4514 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4515 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4516 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4517 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4518 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4519 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4521 //===---------------------------------------------------------------------===//
4522 // SSE2 - Move Quadword
4523 //===---------------------------------------------------------------------===//
4525 //===---------------------------------------------------------------------===//
4526 // Move Quadword Int to Packed Quadword Int
4528 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4529 "vmovq\t{$src, $dst|$dst, $src}",
4531 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4532 VEX, Requires<[HasAVX]>;
4533 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4534 "movq\t{$src, $dst|$dst, $src}",
4536 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4537 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4539 //===---------------------------------------------------------------------===//
4540 // Move Packed Quadword Int to Quadword Int
4542 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4543 "movq\t{$src, $dst|$dst, $src}",
4544 [(store (i64 (vector_extract (v2i64 VR128:$src),
4545 (iPTR 0))), addr:$dst)]>, VEX;
4546 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4547 "movq\t{$src, $dst|$dst, $src}",
4548 [(store (i64 (vector_extract (v2i64 VR128:$src),
4549 (iPTR 0))), addr:$dst)]>;
4551 //===---------------------------------------------------------------------===//
4552 // Store / copy lower 64-bits of a XMM register.
4554 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4555 "movq\t{$src, $dst|$dst, $src}",
4556 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4557 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4558 "movq\t{$src, $dst|$dst, $src}",
4559 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4561 let AddedComplexity = 20 in
4562 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4563 "vmovq\t{$src, $dst|$dst, $src}",
4565 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4566 (loadi64 addr:$src))))))]>,
4567 XS, VEX, Requires<[HasAVX]>;
4569 let AddedComplexity = 20 in
4570 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4571 "movq\t{$src, $dst|$dst, $src}",
4573 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4574 (loadi64 addr:$src))))))]>,
4575 XS, Requires<[HasSSE2]>;
4577 let Predicates = [HasAVX], AddedComplexity = 20 in {
4578 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4579 (VMOVZQI2PQIrm addr:$src)>;
4580 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4581 (VMOVZQI2PQIrm addr:$src)>;
4582 def : Pat<(v2i64 (X86vzload addr:$src)),
4583 (VMOVZQI2PQIrm addr:$src)>;
4586 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4587 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4588 (MOVZQI2PQIrm addr:$src)>;
4589 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4590 (MOVZQI2PQIrm addr:$src)>;
4591 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4594 let Predicates = [HasAVX] in {
4595 def : Pat<(v4i64 (X86vzload addr:$src)),
4596 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4599 //===---------------------------------------------------------------------===//
4600 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4601 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4603 let AddedComplexity = 15 in
4604 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4605 "vmovq\t{$src, $dst|$dst, $src}",
4606 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4607 XS, VEX, Requires<[HasAVX]>;
4608 let AddedComplexity = 15 in
4609 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4610 "movq\t{$src, $dst|$dst, $src}",
4611 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4612 XS, Requires<[HasSSE2]>;
4614 let AddedComplexity = 20 in
4615 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4616 "vmovq\t{$src, $dst|$dst, $src}",
4617 [(set VR128:$dst, (v2i64 (X86vzmovl
4618 (loadv2i64 addr:$src))))]>,
4619 XS, VEX, Requires<[HasAVX]>;
4620 let AddedComplexity = 20 in {
4621 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4622 "movq\t{$src, $dst|$dst, $src}",
4623 [(set VR128:$dst, (v2i64 (X86vzmovl
4624 (loadv2i64 addr:$src))))]>,
4625 XS, Requires<[HasSSE2]>;
4628 let AddedComplexity = 20 in {
4629 let Predicates = [HasAVX] in {
4630 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4631 (VMOVZPQILo2PQIrm addr:$src)>;
4632 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4633 (VMOVZPQILo2PQIrr VR128:$src)>;
4635 let Predicates = [HasSSE2] in {
4636 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4637 (MOVZPQILo2PQIrm addr:$src)>;
4638 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4639 (MOVZPQILo2PQIrr VR128:$src)>;
4643 // Instructions to match in the assembler
4644 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4645 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4646 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4647 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4648 // Recognize "movd" with GR64 destination, but encode as a "movq"
4649 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4650 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4652 // Instructions for the disassembler
4653 // xr = XMM register
4656 let Predicates = [HasAVX] in
4657 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4658 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4659 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4660 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4662 //===---------------------------------------------------------------------===//
4663 // SSE3 - Conversion Instructions
4664 //===---------------------------------------------------------------------===//
4666 // Convert Packed Double FP to Packed DW Integers
4667 let Predicates = [HasAVX] in {
4668 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4669 // register, but the same isn't true when using memory operands instead.
4670 // Provide other assembly rr and rm forms to address this explicitly.
4671 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4672 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4673 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4674 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4677 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4678 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4679 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4680 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4683 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4684 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4685 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4686 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4689 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4690 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4691 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4692 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4694 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4695 (VCVTTPD2DQYrr VR256:$src)>;
4696 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4697 (VCVTTPD2DQYrm addr:$src)>;
4699 // Convert Packed DW Integers to Packed Double FP
4700 let Predicates = [HasAVX] in {
4701 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4702 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4703 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4704 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4705 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4706 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4707 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4708 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4711 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4712 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4713 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4714 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4716 // AVX 256-bit register conversion intrinsics
4717 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4718 (VCVTDQ2PDYrr VR128:$src)>;
4719 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4720 (VCVTDQ2PDYrm addr:$src)>;
4722 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4723 (VCVTPD2DQYrr VR256:$src)>;
4724 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4725 (VCVTPD2DQYrm addr:$src)>;
4727 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4728 (VCVTDQ2PDYrr VR128:$src)>;
4729 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4730 (VCVTDQ2PDYrm addr:$src)>;
4732 //===---------------------------------------------------------------------===//
4733 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4734 //===---------------------------------------------------------------------===//
4735 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4736 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4737 X86MemOperand x86memop> {
4738 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4739 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4740 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4741 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4742 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4743 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4746 let Predicates = [HasAVX] in {
4747 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4748 v4f32, VR128, memopv4f32, f128mem>, VEX;
4749 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4750 v4f32, VR128, memopv4f32, f128mem>, VEX;
4751 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4752 v8f32, VR256, memopv8f32, f256mem>, VEX;
4753 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4754 v8f32, VR256, memopv8f32, f256mem>, VEX;
4756 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4757 memopv4f32, f128mem>;
4758 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4759 memopv4f32, f128mem>;
4761 let Predicates = [HasAVX] in {
4762 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4763 (VMOVSHDUPrr VR128:$src)>;
4764 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4765 (VMOVSHDUPrm addr:$src)>;
4766 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4767 (VMOVSLDUPrr VR128:$src)>;
4768 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4769 (VMOVSLDUPrm addr:$src)>;
4770 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4771 (VMOVSHDUPYrr VR256:$src)>;
4772 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4773 (VMOVSHDUPYrm addr:$src)>;
4774 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4775 (VMOVSLDUPYrr VR256:$src)>;
4776 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4777 (VMOVSLDUPYrm addr:$src)>;
4780 let Predicates = [HasSSE3] in {
4781 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4782 (MOVSHDUPrr VR128:$src)>;
4783 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4784 (MOVSHDUPrm addr:$src)>;
4785 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4786 (MOVSLDUPrr VR128:$src)>;
4787 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4788 (MOVSLDUPrm addr:$src)>;
4791 //===---------------------------------------------------------------------===//
4792 // SSE3 - Replicate Double FP - MOVDDUP
4793 //===---------------------------------------------------------------------===//
4795 multiclass sse3_replicate_dfp<string OpcodeStr> {
4796 let neverHasSideEffects = 1 in
4797 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4798 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4800 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4801 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4804 (scalar_to_vector (loadf64 addr:$src)))))]>;
4807 // FIXME: Merge with above classe when there're patterns for the ymm version
4808 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4809 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4810 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4811 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4812 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4813 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4816 (scalar_to_vector (loadf64 addr:$src)))))]>;
4819 let Predicates = [HasAVX] in {
4820 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4821 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4824 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4826 let Predicates = [HasAVX] in {
4827 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4828 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4829 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4830 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4831 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4832 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4833 def : Pat<(X86Movddup (bc_v2f64
4834 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4835 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4838 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4839 (VMOVDDUPYrm addr:$src)>;
4840 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4841 (VMOVDDUPYrm addr:$src)>;
4842 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4843 (VMOVDDUPYrm addr:$src)>;
4844 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4845 (VMOVDDUPYrr VR256:$src)>;
4848 let Predicates = [HasSSE3] in {
4849 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4850 (MOVDDUPrm addr:$src)>;
4851 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4852 (MOVDDUPrm addr:$src)>;
4853 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4854 (MOVDDUPrm addr:$src)>;
4855 def : Pat<(X86Movddup (bc_v2f64
4856 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4857 (MOVDDUPrm addr:$src)>;
4860 //===---------------------------------------------------------------------===//
4861 // SSE3 - Move Unaligned Integer
4862 //===---------------------------------------------------------------------===//
4864 let Predicates = [HasAVX] in {
4865 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4866 "vlddqu\t{$src, $dst|$dst, $src}",
4867 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4868 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4869 "vlddqu\t{$src, $dst|$dst, $src}",
4870 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4872 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4873 "lddqu\t{$src, $dst|$dst, $src}",
4874 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
4876 //===---------------------------------------------------------------------===//
4877 // SSE3 - Arithmetic
4878 //===---------------------------------------------------------------------===//
4880 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4881 X86MemOperand x86memop, bit Is2Addr = 1> {
4882 def rr : I<0xD0, MRMSrcReg,
4883 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4885 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4886 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4887 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
4888 def rm : I<0xD0, MRMSrcMem,
4889 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4891 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4892 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4893 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
4896 let Predicates = [HasAVX] in {
4897 let ExeDomain = SSEPackedSingle in {
4898 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4899 f128mem, 0>, TB, XD, VEX_4V;
4900 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4901 f256mem, 0>, TB, XD, VEX_4V;
4903 let ExeDomain = SSEPackedDouble in {
4904 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4905 f128mem, 0>, TB, OpSize, VEX_4V;
4906 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4907 f256mem, 0>, TB, OpSize, VEX_4V;
4910 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
4911 let ExeDomain = SSEPackedSingle in
4912 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4914 let ExeDomain = SSEPackedDouble in
4915 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4916 f128mem>, TB, OpSize;
4919 //===---------------------------------------------------------------------===//
4920 // SSE3 Instructions
4921 //===---------------------------------------------------------------------===//
4924 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4925 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4926 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4928 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4929 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4930 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
4932 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4934 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4935 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4936 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
4938 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4939 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4940 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4942 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4943 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4944 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
4946 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4948 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4949 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4950 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
4953 let Predicates = [HasAVX] in {
4954 let ExeDomain = SSEPackedSingle in {
4955 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4956 X86fhadd, 0>, VEX_4V;
4957 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4958 X86fhsub, 0>, VEX_4V;
4959 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4960 X86fhadd, 0>, VEX_4V;
4961 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4962 X86fhsub, 0>, VEX_4V;
4964 let ExeDomain = SSEPackedDouble in {
4965 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4966 X86fhadd, 0>, VEX_4V;
4967 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4968 X86fhsub, 0>, VEX_4V;
4969 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4970 X86fhadd, 0>, VEX_4V;
4971 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4972 X86fhsub, 0>, VEX_4V;
4976 let Constraints = "$src1 = $dst" in {
4977 let ExeDomain = SSEPackedSingle in {
4978 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
4979 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
4981 let ExeDomain = SSEPackedDouble in {
4982 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
4983 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
4987 //===---------------------------------------------------------------------===//
4988 // SSSE3 - Packed Absolute Instructions
4989 //===---------------------------------------------------------------------===//
4992 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4993 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4994 Intrinsic IntId128> {
4995 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4997 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4998 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5001 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5003 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5006 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5009 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5010 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5011 Intrinsic IntId256> {
5012 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5014 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5015 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5018 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5020 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5023 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5026 let Predicates = [HasAVX] in {
5027 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5028 int_x86_ssse3_pabs_b_128>, VEX;
5029 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5030 int_x86_ssse3_pabs_w_128>, VEX;
5031 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5032 int_x86_ssse3_pabs_d_128>, VEX;
5035 let Predicates = [HasAVX2] in {
5036 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5037 int_x86_avx2_pabs_b>, VEX;
5038 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5039 int_x86_avx2_pabs_w>, VEX;
5040 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5041 int_x86_avx2_pabs_d>, VEX;
5044 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5045 int_x86_ssse3_pabs_b_128>;
5046 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5047 int_x86_ssse3_pabs_w_128>;
5048 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5049 int_x86_ssse3_pabs_d_128>;
5051 //===---------------------------------------------------------------------===//
5052 // SSSE3 - Packed Binary Operator Instructions
5053 //===---------------------------------------------------------------------===//
5055 /// SS3I_binop_rm - Simple SSSE3 bin op
5056 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5057 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5058 X86MemOperand x86memop, bit Is2Addr = 1> {
5059 let isCommutable = 1 in
5060 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5061 (ins RC:$src1, RC:$src2),
5063 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5064 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5065 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
5067 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5068 (ins RC:$src1, x86memop:$src2),
5070 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5071 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5073 (OpVT (OpNode RC:$src1,
5074 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
5077 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5078 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5079 Intrinsic IntId128, bit Is2Addr = 1> {
5080 let isCommutable = 1 in
5081 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5082 (ins VR128:$src1, VR128:$src2),
5084 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5085 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5086 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5088 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5089 (ins VR128:$src1, i128mem:$src2),
5091 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5092 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5094 (IntId128 VR128:$src1,
5095 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5098 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5099 Intrinsic IntId256> {
5100 let isCommutable = 1 in
5101 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5102 (ins VR256:$src1, VR256:$src2),
5103 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5104 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5106 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5107 (ins VR256:$src1, i256mem:$src2),
5108 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5110 (IntId256 VR256:$src1,
5111 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5114 let ImmT = NoImm, Predicates = [HasAVX] in {
5115 let isCommutable = 0 in {
5116 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5117 memopv2i64, i128mem, 0>, VEX_4V;
5118 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5119 memopv2i64, i128mem, 0>, VEX_4V;
5120 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5121 memopv2i64, i128mem, 0>, VEX_4V;
5122 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5123 memopv2i64, i128mem, 0>, VEX_4V;
5124 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5125 memopv2i64, i128mem, 0>, VEX_4V;
5126 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5127 memopv2i64, i128mem, 0>, VEX_4V;
5128 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5129 memopv2i64, i128mem, 0>, VEX_4V;
5130 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5131 memopv2i64, i128mem, 0>, VEX_4V;
5132 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5133 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5134 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5135 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5136 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5137 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5139 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5140 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5143 let ImmT = NoImm, Predicates = [HasAVX2] in {
5144 let isCommutable = 0 in {
5145 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5146 memopv4i64, i256mem, 0>, VEX_4V;
5147 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5148 memopv4i64, i256mem, 0>, VEX_4V;
5149 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5150 memopv4i64, i256mem, 0>, VEX_4V;
5151 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5152 memopv4i64, i256mem, 0>, VEX_4V;
5153 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5154 memopv4i64, i256mem, 0>, VEX_4V;
5155 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5156 memopv4i64, i256mem, 0>, VEX_4V;
5157 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5158 memopv4i64, i256mem, 0>, VEX_4V;
5159 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5160 memopv4i64, i256mem, 0>, VEX_4V;
5161 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5162 int_x86_avx2_phadd_sw>, VEX_4V;
5163 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5164 int_x86_avx2_phsub_sw>, VEX_4V;
5165 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5166 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5168 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5169 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5172 // None of these have i8 immediate fields.
5173 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5174 let isCommutable = 0 in {
5175 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5176 memopv2i64, i128mem>;
5177 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5178 memopv2i64, i128mem>;
5179 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5180 memopv2i64, i128mem>;
5181 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5182 memopv2i64, i128mem>;
5183 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5184 memopv2i64, i128mem>;
5185 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5186 memopv2i64, i128mem>;
5187 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5188 memopv2i64, i128mem>;
5189 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5190 memopv2i64, i128mem>;
5191 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5192 int_x86_ssse3_phadd_sw_128>;
5193 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5194 int_x86_ssse3_phsub_sw_128>;
5195 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5196 int_x86_ssse3_pmadd_ub_sw_128>;
5198 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5199 int_x86_ssse3_pmul_hr_sw_128>;
5202 //===---------------------------------------------------------------------===//
5203 // SSSE3 - Packed Align Instruction Patterns
5204 //===---------------------------------------------------------------------===//
5206 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5207 let neverHasSideEffects = 1 in {
5208 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5209 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5211 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5213 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5216 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5217 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5219 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5221 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5226 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5227 let neverHasSideEffects = 1 in {
5228 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5229 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5231 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5234 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5235 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5237 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5242 let Predicates = [HasAVX] in
5243 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5244 let Predicates = [HasAVX2] in
5245 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5246 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5247 defm PALIGN : ssse3_palign<"palignr">;
5249 let Predicates = [HasAVX2] in {
5250 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5251 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5252 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5253 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5254 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5255 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5256 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5257 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5260 let Predicates = [HasAVX] in {
5261 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5262 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5263 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5264 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5265 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5266 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5267 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5268 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5271 let Predicates = [HasSSSE3] in {
5272 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5273 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5274 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5275 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5276 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5277 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5278 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5279 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5282 //===---------------------------------------------------------------------===//
5283 // SSSE3 - Thread synchronization
5284 //===---------------------------------------------------------------------===//
5286 let usesCustomInserter = 1 in {
5287 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5288 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5289 Requires<[HasSSE3]>;
5290 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5291 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5292 Requires<[HasSSE3]>;
5295 let Uses = [EAX, ECX, EDX] in
5296 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5297 Requires<[HasSSE3]>;
5298 let Uses = [ECX, EAX] in
5299 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5300 Requires<[HasSSE3]>;
5302 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5303 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5305 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5306 Requires<[In32BitMode]>;
5307 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5308 Requires<[In64BitMode]>;
5310 //===----------------------------------------------------------------------===//
5311 // SSE4.1 - Packed Move with Sign/Zero Extend
5312 //===----------------------------------------------------------------------===//
5314 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5315 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5316 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5317 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5319 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5320 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5322 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5326 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5328 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5329 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5330 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5332 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5333 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5334 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5337 let Predicates = [HasAVX] in {
5338 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5340 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5342 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5344 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5346 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5348 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5352 let Predicates = [HasAVX2] in {
5353 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5354 int_x86_avx2_pmovsxbw>, VEX;
5355 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5356 int_x86_avx2_pmovsxwd>, VEX;
5357 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5358 int_x86_avx2_pmovsxdq>, VEX;
5359 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5360 int_x86_avx2_pmovzxbw>, VEX;
5361 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5362 int_x86_avx2_pmovzxwd>, VEX;
5363 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5364 int_x86_avx2_pmovzxdq>, VEX;
5367 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5368 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5369 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5370 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5371 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5372 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5374 let Predicates = [HasAVX] in {
5375 // Common patterns involving scalar load.
5376 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5377 (VPMOVSXBWrm addr:$src)>;
5378 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5379 (VPMOVSXBWrm addr:$src)>;
5381 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5382 (VPMOVSXWDrm addr:$src)>;
5383 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5384 (VPMOVSXWDrm addr:$src)>;
5386 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5387 (VPMOVSXDQrm addr:$src)>;
5388 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5389 (VPMOVSXDQrm addr:$src)>;
5391 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5392 (VPMOVZXBWrm addr:$src)>;
5393 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5394 (VPMOVZXBWrm addr:$src)>;
5396 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5397 (VPMOVZXWDrm addr:$src)>;
5398 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5399 (VPMOVZXWDrm addr:$src)>;
5401 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5402 (VPMOVZXDQrm addr:$src)>;
5403 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5404 (VPMOVZXDQrm addr:$src)>;
5407 let Predicates = [HasSSE41] in {
5408 // Common patterns involving scalar load.
5409 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5410 (PMOVSXBWrm addr:$src)>;
5411 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5412 (PMOVSXBWrm addr:$src)>;
5414 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5415 (PMOVSXWDrm addr:$src)>;
5416 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5417 (PMOVSXWDrm addr:$src)>;
5419 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5420 (PMOVSXDQrm addr:$src)>;
5421 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5422 (PMOVSXDQrm addr:$src)>;
5424 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5425 (PMOVZXBWrm addr:$src)>;
5426 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5427 (PMOVZXBWrm addr:$src)>;
5429 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5430 (PMOVZXWDrm addr:$src)>;
5431 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5432 (PMOVZXWDrm addr:$src)>;
5434 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5435 (PMOVZXDQrm addr:$src)>;
5436 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5437 (PMOVZXDQrm addr:$src)>;
5440 let Predicates = [HasAVX] in {
5441 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5442 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5445 let Predicates = [HasSSE41] in {
5446 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5447 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5451 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5452 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5453 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5454 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5456 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5457 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5459 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5463 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5465 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5466 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5467 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5469 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5470 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5472 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5476 let Predicates = [HasAVX] in {
5477 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5479 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5481 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5483 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5487 let Predicates = [HasAVX2] in {
5488 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5489 int_x86_avx2_pmovsxbd>, VEX;
5490 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5491 int_x86_avx2_pmovsxwq>, VEX;
5492 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5493 int_x86_avx2_pmovzxbd>, VEX;
5494 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5495 int_x86_avx2_pmovzxwq>, VEX;
5498 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5499 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5500 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5501 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5503 let Predicates = [HasAVX] in {
5504 // Common patterns involving scalar load
5505 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5506 (VPMOVSXBDrm addr:$src)>;
5507 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5508 (VPMOVSXWQrm addr:$src)>;
5510 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5511 (VPMOVZXBDrm addr:$src)>;
5512 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5513 (VPMOVZXWQrm addr:$src)>;
5516 let Predicates = [HasSSE41] in {
5517 // Common patterns involving scalar load
5518 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5519 (PMOVSXBDrm addr:$src)>;
5520 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5521 (PMOVSXWQrm addr:$src)>;
5523 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5524 (PMOVZXBDrm addr:$src)>;
5525 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5526 (PMOVZXWQrm addr:$src)>;
5529 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5530 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5531 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5532 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5534 // Expecting a i16 load any extended to i32 value.
5535 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5536 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5537 [(set VR128:$dst, (IntId (bitconvert
5538 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5542 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5544 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5545 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5546 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5548 // Expecting a i16 load any extended to i32 value.
5549 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5550 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5551 [(set VR256:$dst, (IntId (bitconvert
5552 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5556 let Predicates = [HasAVX] in {
5557 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5559 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5562 let Predicates = [HasAVX2] in {
5563 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5564 int_x86_avx2_pmovsxbq>, VEX;
5565 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5566 int_x86_avx2_pmovzxbq>, VEX;
5568 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5569 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5571 let Predicates = [HasAVX] in {
5572 // Common patterns involving scalar load
5573 def : Pat<(int_x86_sse41_pmovsxbq
5574 (bitconvert (v4i32 (X86vzmovl
5575 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5576 (VPMOVSXBQrm addr:$src)>;
5578 def : Pat<(int_x86_sse41_pmovzxbq
5579 (bitconvert (v4i32 (X86vzmovl
5580 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5581 (VPMOVZXBQrm addr:$src)>;
5584 let Predicates = [HasSSE41] in {
5585 // Common patterns involving scalar load
5586 def : Pat<(int_x86_sse41_pmovsxbq
5587 (bitconvert (v4i32 (X86vzmovl
5588 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5589 (PMOVSXBQrm addr:$src)>;
5591 def : Pat<(int_x86_sse41_pmovzxbq
5592 (bitconvert (v4i32 (X86vzmovl
5593 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5594 (PMOVZXBQrm addr:$src)>;
5597 //===----------------------------------------------------------------------===//
5598 // SSE4.1 - Extract Instructions
5599 //===----------------------------------------------------------------------===//
5601 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5602 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5603 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5604 (ins VR128:$src1, i32i8imm:$src2),
5605 !strconcat(OpcodeStr,
5606 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5607 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5609 let neverHasSideEffects = 1, mayStore = 1 in
5610 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5611 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5612 !strconcat(OpcodeStr,
5613 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5616 // There's an AssertZext in the way of writing the store pattern
5617 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5620 let Predicates = [HasAVX] in {
5621 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5622 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5623 (ins VR128:$src1, i32i8imm:$src2),
5624 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5627 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5630 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5631 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5632 let neverHasSideEffects = 1, mayStore = 1 in
5633 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5634 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5635 !strconcat(OpcodeStr,
5636 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5639 // There's an AssertZext in the way of writing the store pattern
5640 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5643 let Predicates = [HasAVX] in
5644 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5646 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5649 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5650 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5651 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5652 (ins VR128:$src1, i32i8imm:$src2),
5653 !strconcat(OpcodeStr,
5654 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5656 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5657 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5658 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5659 !strconcat(OpcodeStr,
5660 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5661 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5662 addr:$dst)]>, OpSize;
5665 let Predicates = [HasAVX] in
5666 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5668 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5670 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5671 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5672 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5673 (ins VR128:$src1, i32i8imm:$src2),
5674 !strconcat(OpcodeStr,
5675 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5677 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5678 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5679 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5680 !strconcat(OpcodeStr,
5681 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5682 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5683 addr:$dst)]>, OpSize, REX_W;
5686 let Predicates = [HasAVX] in
5687 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5689 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5691 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5693 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5694 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5695 (ins VR128:$src1, i32i8imm:$src2),
5696 !strconcat(OpcodeStr,
5697 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5699 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5701 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5702 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5703 !strconcat(OpcodeStr,
5704 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5705 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5706 addr:$dst)]>, OpSize;
5709 let ExeDomain = SSEPackedSingle in {
5710 let Predicates = [HasAVX] in {
5711 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5712 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5713 (ins VR128:$src1, i32i8imm:$src2),
5714 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5717 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5720 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5721 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5724 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5726 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5729 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5730 Requires<[HasSSE41]>;
5732 //===----------------------------------------------------------------------===//
5733 // SSE4.1 - Insert Instructions
5734 //===----------------------------------------------------------------------===//
5736 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5737 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5738 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5740 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5742 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5744 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5745 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5746 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5748 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5750 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5752 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5753 imm:$src3))]>, OpSize;
5756 let Predicates = [HasAVX] in
5757 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5758 let Constraints = "$src1 = $dst" in
5759 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5761 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5762 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5763 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5765 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5767 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5769 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5771 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5772 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5774 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5776 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5778 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5779 imm:$src3)))]>, OpSize;
5782 let Predicates = [HasAVX] in
5783 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5784 let Constraints = "$src1 = $dst" in
5785 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5787 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5788 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5789 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5791 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5793 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5795 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5797 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5798 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5800 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5802 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5804 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5805 imm:$src3)))]>, OpSize;
5808 let Predicates = [HasAVX] in
5809 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5810 let Constraints = "$src1 = $dst" in
5811 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5813 // insertps has a few different modes, there's the first two here below which
5814 // are optimized inserts that won't zero arbitrary elements in the destination
5815 // vector. The next one matches the intrinsic and could zero arbitrary elements
5816 // in the target vector.
5817 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5818 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5819 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5821 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5823 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5825 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5827 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5828 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5830 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5832 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5834 (X86insrtps VR128:$src1,
5835 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5836 imm:$src3))]>, OpSize;
5839 let ExeDomain = SSEPackedSingle in {
5840 let Predicates = [HasAVX] in
5841 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5842 let Constraints = "$src1 = $dst" in
5843 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5846 //===----------------------------------------------------------------------===//
5847 // SSE4.1 - Round Instructions
5848 //===----------------------------------------------------------------------===//
5850 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5851 X86MemOperand x86memop, RegisterClass RC,
5852 PatFrag mem_frag32, PatFrag mem_frag64,
5853 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5854 let ExeDomain = SSEPackedSingle in {
5855 // Intrinsic operation, reg.
5856 // Vector intrinsic operation, reg
5857 def PSr : SS4AIi8<opcps, MRMSrcReg,
5858 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5859 !strconcat(OpcodeStr,
5860 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5861 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5864 // Vector intrinsic operation, mem
5865 def PSm : SS4AIi8<opcps, MRMSrcMem,
5866 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5867 !strconcat(OpcodeStr,
5868 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5870 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
5872 } // ExeDomain = SSEPackedSingle
5874 let ExeDomain = SSEPackedDouble in {
5875 // Vector intrinsic operation, reg
5876 def PDr : SS4AIi8<opcpd, MRMSrcReg,
5877 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5878 !strconcat(OpcodeStr,
5879 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5880 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
5883 // Vector intrinsic operation, mem
5884 def PDm : SS4AIi8<opcpd, MRMSrcMem,
5885 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5886 !strconcat(OpcodeStr,
5887 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5889 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
5891 } // ExeDomain = SSEPackedDouble
5894 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
5897 Intrinsic F64Int, bit Is2Addr = 1> {
5898 let ExeDomain = GenericDomain in {
5900 def SSr : SS4AIi8<opcss, MRMSrcReg,
5901 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
5903 !strconcat(OpcodeStr,
5904 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5905 !strconcat(OpcodeStr,
5906 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5909 // Intrinsic operation, reg.
5910 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
5911 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5913 !strconcat(OpcodeStr,
5914 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5915 !strconcat(OpcodeStr,
5916 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5917 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5920 // Intrinsic operation, mem.
5921 def SSm : SS4AIi8<opcss, MRMSrcMem,
5922 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5924 !strconcat(OpcodeStr,
5925 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5926 !strconcat(OpcodeStr,
5927 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5929 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
5933 def SDr : SS4AIi8<opcsd, MRMSrcReg,
5934 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
5936 !strconcat(OpcodeStr,
5937 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5938 !strconcat(OpcodeStr,
5939 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5942 // Intrinsic operation, reg.
5943 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
5944 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5946 !strconcat(OpcodeStr,
5947 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5948 !strconcat(OpcodeStr,
5949 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5950 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5953 // Intrinsic operation, mem.
5954 def SDm : SS4AIi8<opcsd, MRMSrcMem,
5955 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5957 !strconcat(OpcodeStr,
5958 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5959 !strconcat(OpcodeStr,
5960 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5962 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
5964 } // ExeDomain = GenericDomain
5967 // FP round - roundss, roundps, roundsd, roundpd
5968 let Predicates = [HasAVX] in {
5970 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
5971 memopv4f32, memopv2f64,
5972 int_x86_sse41_round_ps,
5973 int_x86_sse41_round_pd>, VEX;
5974 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
5975 memopv8f32, memopv4f64,
5976 int_x86_avx_round_ps_256,
5977 int_x86_avx_round_pd_256>, VEX;
5978 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
5979 int_x86_sse41_round_ss,
5980 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
5982 def : Pat<(ffloor FR32:$src),
5983 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
5984 def : Pat<(f64 (ffloor FR64:$src)),
5985 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
5986 def : Pat<(f32 (fnearbyint FR32:$src)),
5987 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
5988 def : Pat<(f64 (fnearbyint FR64:$src)),
5989 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
5990 def : Pat<(f32 (fceil FR32:$src)),
5991 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
5992 def : Pat<(f64 (fceil FR64:$src)),
5993 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
5994 def : Pat<(f32 (frint FR32:$src)),
5995 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
5996 def : Pat<(f64 (frint FR64:$src)),
5997 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
5998 def : Pat<(f32 (ftrunc FR32:$src)),
5999 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6000 def : Pat<(f64 (ftrunc FR64:$src)),
6001 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6004 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6005 memopv4f32, memopv2f64,
6006 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6007 let Constraints = "$src1 = $dst" in
6008 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6009 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6011 def : Pat<(ffloor FR32:$src),
6012 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6013 def : Pat<(f64 (ffloor FR64:$src)),
6014 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6015 def : Pat<(f32 (fnearbyint FR32:$src)),
6016 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6017 def : Pat<(f64 (fnearbyint FR64:$src)),
6018 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6019 def : Pat<(f32 (fceil FR32:$src)),
6020 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6021 def : Pat<(f64 (fceil FR64:$src)),
6022 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6023 def : Pat<(f32 (frint FR32:$src)),
6024 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6025 def : Pat<(f64 (frint FR64:$src)),
6026 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6027 def : Pat<(f32 (ftrunc FR32:$src)),
6028 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6029 def : Pat<(f64 (ftrunc FR64:$src)),
6030 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6032 //===----------------------------------------------------------------------===//
6033 // SSE4.1 - Packed Bit Test
6034 //===----------------------------------------------------------------------===//
6036 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6037 // the intel intrinsic that corresponds to this.
6038 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6039 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6040 "vptest\t{$src2, $src1|$src1, $src2}",
6041 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6043 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6044 "vptest\t{$src2, $src1|$src1, $src2}",
6045 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6048 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6049 "vptest\t{$src2, $src1|$src1, $src2}",
6050 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6052 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6053 "vptest\t{$src2, $src1|$src1, $src2}",
6054 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6058 let Defs = [EFLAGS] in {
6059 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6060 "ptest\t{$src2, $src1|$src1, $src2}",
6061 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6063 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6064 "ptest\t{$src2, $src1|$src1, $src2}",
6065 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6069 // The bit test instructions below are AVX only
6070 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6071 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6072 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6073 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6074 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6075 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6076 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6077 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6081 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6082 let ExeDomain = SSEPackedSingle in {
6083 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6084 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6086 let ExeDomain = SSEPackedDouble in {
6087 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6088 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6092 //===----------------------------------------------------------------------===//
6093 // SSE4.1 - Misc Instructions
6094 //===----------------------------------------------------------------------===//
6096 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6097 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6098 "popcnt{w}\t{$src, $dst|$dst, $src}",
6099 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6101 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6102 "popcnt{w}\t{$src, $dst|$dst, $src}",
6103 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6104 (implicit EFLAGS)]>, OpSize, XS;
6106 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6107 "popcnt{l}\t{$src, $dst|$dst, $src}",
6108 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6110 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6111 "popcnt{l}\t{$src, $dst|$dst, $src}",
6112 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6113 (implicit EFLAGS)]>, XS;
6115 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6116 "popcnt{q}\t{$src, $dst|$dst, $src}",
6117 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6119 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6120 "popcnt{q}\t{$src, $dst|$dst, $src}",
6121 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6122 (implicit EFLAGS)]>, XS;
6127 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6128 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6129 Intrinsic IntId128> {
6130 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6132 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6133 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6134 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6136 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6139 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6142 let Predicates = [HasAVX] in
6143 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6144 int_x86_sse41_phminposuw>, VEX;
6145 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6146 int_x86_sse41_phminposuw>;
6148 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6149 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6150 Intrinsic IntId128, bit Is2Addr = 1> {
6151 let isCommutable = 1 in
6152 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6153 (ins VR128:$src1, VR128:$src2),
6155 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6156 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6157 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6158 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6159 (ins VR128:$src1, i128mem:$src2),
6161 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6162 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6164 (IntId128 VR128:$src1,
6165 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6168 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6169 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6170 Intrinsic IntId256> {
6171 let isCommutable = 1 in
6172 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6173 (ins VR256:$src1, VR256:$src2),
6174 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6175 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6176 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6177 (ins VR256:$src1, i256mem:$src2),
6178 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6180 (IntId256 VR256:$src1,
6181 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6184 let Predicates = [HasAVX] in {
6185 let isCommutable = 0 in
6186 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6188 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6190 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6192 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6194 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6196 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6198 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6200 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6202 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6204 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6208 let Predicates = [HasAVX2] in {
6209 let isCommutable = 0 in
6210 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6211 int_x86_avx2_packusdw>, VEX_4V;
6212 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6213 int_x86_avx2_pmins_b>, VEX_4V;
6214 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6215 int_x86_avx2_pmins_d>, VEX_4V;
6216 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6217 int_x86_avx2_pminu_d>, VEX_4V;
6218 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6219 int_x86_avx2_pminu_w>, VEX_4V;
6220 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6221 int_x86_avx2_pmaxs_b>, VEX_4V;
6222 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6223 int_x86_avx2_pmaxs_d>, VEX_4V;
6224 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6225 int_x86_avx2_pmaxu_d>, VEX_4V;
6226 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6227 int_x86_avx2_pmaxu_w>, VEX_4V;
6228 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6229 int_x86_avx2_pmul_dq>, VEX_4V;
6232 let Constraints = "$src1 = $dst" in {
6233 let isCommutable = 0 in
6234 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6235 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6236 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6237 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6238 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6239 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6240 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6241 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6242 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6243 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6246 /// SS48I_binop_rm - Simple SSE41 binary operator.
6247 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6248 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6249 X86MemOperand x86memop, bit Is2Addr = 1> {
6250 let isCommutable = 1 in
6251 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6252 (ins RC:$src1, RC:$src2),
6254 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6255 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6256 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6257 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6258 (ins RC:$src1, x86memop:$src2),
6260 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6261 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6263 (OpVT (OpNode RC:$src1,
6264 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6267 let Predicates = [HasAVX] in {
6268 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6269 memopv2i64, i128mem, 0>, VEX_4V;
6270 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6271 memopv2i64, i128mem, 0>, VEX_4V;
6273 let Predicates = [HasAVX2] in {
6274 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6275 memopv4i64, i256mem, 0>, VEX_4V;
6276 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6277 memopv4i64, i256mem, 0>, VEX_4V;
6280 let Constraints = "$src1 = $dst" in {
6281 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6282 memopv2i64, i128mem>;
6283 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6284 memopv2i64, i128mem>;
6287 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6288 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6289 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6290 X86MemOperand x86memop, bit Is2Addr = 1> {
6291 let isCommutable = 1 in
6292 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6293 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6295 !strconcat(OpcodeStr,
6296 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6297 !strconcat(OpcodeStr,
6298 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6299 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6301 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6302 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6304 !strconcat(OpcodeStr,
6305 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6306 !strconcat(OpcodeStr,
6307 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6310 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6314 let Predicates = [HasAVX] in {
6315 let isCommutable = 0 in {
6316 let ExeDomain = SSEPackedSingle in {
6317 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6318 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6319 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6320 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6322 let ExeDomain = SSEPackedDouble in {
6323 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6324 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6325 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6326 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6328 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6329 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6330 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6331 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6333 let ExeDomain = SSEPackedSingle in
6334 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6335 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6336 let ExeDomain = SSEPackedDouble in
6337 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6338 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6339 let ExeDomain = SSEPackedSingle in
6340 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6341 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6344 let Predicates = [HasAVX2] in {
6345 let isCommutable = 0 in {
6346 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6347 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6348 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6349 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6353 let Constraints = "$src1 = $dst" in {
6354 let isCommutable = 0 in {
6355 let ExeDomain = SSEPackedSingle in
6356 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6357 VR128, memopv4f32, i128mem>;
6358 let ExeDomain = SSEPackedDouble in
6359 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6360 VR128, memopv2f64, i128mem>;
6361 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6362 VR128, memopv2i64, i128mem>;
6363 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6364 VR128, memopv2i64, i128mem>;
6366 let ExeDomain = SSEPackedSingle in
6367 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6368 VR128, memopv4f32, i128mem>;
6369 let ExeDomain = SSEPackedDouble in
6370 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6371 VR128, memopv2f64, i128mem>;
6374 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6375 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6376 RegisterClass RC, X86MemOperand x86memop,
6377 PatFrag mem_frag, Intrinsic IntId> {
6378 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6379 (ins RC:$src1, RC:$src2, RC:$src3),
6380 !strconcat(OpcodeStr,
6381 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6382 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6383 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6385 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6386 (ins RC:$src1, x86memop:$src2, RC:$src3),
6387 !strconcat(OpcodeStr,
6388 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6390 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6392 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6395 let Predicates = [HasAVX] in {
6396 let ExeDomain = SSEPackedDouble in {
6397 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6398 memopv2f64, int_x86_sse41_blendvpd>;
6399 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6400 memopv4f64, int_x86_avx_blendv_pd_256>;
6401 } // ExeDomain = SSEPackedDouble
6402 let ExeDomain = SSEPackedSingle in {
6403 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6404 memopv4f32, int_x86_sse41_blendvps>;
6405 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6406 memopv8f32, int_x86_avx_blendv_ps_256>;
6407 } // ExeDomain = SSEPackedSingle
6408 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6409 memopv2i64, int_x86_sse41_pblendvb>;
6412 let Predicates = [HasAVX2] in {
6413 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6414 memopv4i64, int_x86_avx2_pblendvb>;
6417 let Predicates = [HasAVX] in {
6418 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6419 (v16i8 VR128:$src2))),
6420 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6421 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6422 (v4i32 VR128:$src2))),
6423 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6424 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6425 (v4f32 VR128:$src2))),
6426 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6427 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6428 (v2i64 VR128:$src2))),
6429 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6430 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6431 (v2f64 VR128:$src2))),
6432 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6433 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6434 (v8i32 VR256:$src2))),
6435 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6436 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6437 (v8f32 VR256:$src2))),
6438 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6439 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6440 (v4i64 VR256:$src2))),
6441 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6442 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6443 (v4f64 VR256:$src2))),
6444 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6447 let Predicates = [HasAVX2] in {
6448 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6449 (v32i8 VR256:$src2))),
6450 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6453 /// SS41I_ternary_int - SSE 4.1 ternary operator
6454 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6455 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6457 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6458 (ins VR128:$src1, VR128:$src2),
6459 !strconcat(OpcodeStr,
6460 "\t{$src2, $dst|$dst, $src2}"),
6461 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6464 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6465 (ins VR128:$src1, i128mem:$src2),
6466 !strconcat(OpcodeStr,
6467 "\t{$src2, $dst|$dst, $src2}"),
6470 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6474 let ExeDomain = SSEPackedDouble in
6475 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6476 int_x86_sse41_blendvpd>;
6477 let ExeDomain = SSEPackedSingle in
6478 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6479 int_x86_sse41_blendvps>;
6480 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6481 int_x86_sse41_pblendvb>;
6483 let Predicates = [HasSSE41] in {
6484 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6485 (v16i8 VR128:$src2))),
6486 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6487 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6488 (v4i32 VR128:$src2))),
6489 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6490 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6491 (v4f32 VR128:$src2))),
6492 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6493 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6494 (v2i64 VR128:$src2))),
6495 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6496 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6497 (v2f64 VR128:$src2))),
6498 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6501 let Predicates = [HasAVX] in
6502 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6503 "vmovntdqa\t{$src, $dst|$dst, $src}",
6504 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6506 let Predicates = [HasAVX2] in
6507 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6508 "vmovntdqa\t{$src, $dst|$dst, $src}",
6509 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6511 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6512 "movntdqa\t{$src, $dst|$dst, $src}",
6513 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6516 //===----------------------------------------------------------------------===//
6517 // SSE4.2 - Compare Instructions
6518 //===----------------------------------------------------------------------===//
6520 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6521 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6522 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6523 X86MemOperand x86memop, bit Is2Addr = 1> {
6524 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6525 (ins RC:$src1, RC:$src2),
6527 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6528 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6529 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6531 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6532 (ins RC:$src1, x86memop:$src2),
6534 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6535 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6537 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6540 let Predicates = [HasAVX] in
6541 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6542 memopv2i64, i128mem, 0>, VEX_4V;
6544 let Predicates = [HasAVX2] in
6545 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6546 memopv4i64, i256mem, 0>, VEX_4V;
6548 let Constraints = "$src1 = $dst" in
6549 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6550 memopv2i64, i128mem>;
6552 //===----------------------------------------------------------------------===//
6553 // SSE4.2 - String/text Processing Instructions
6554 //===----------------------------------------------------------------------===//
6556 // Packed Compare Implicit Length Strings, Return Mask
6557 multiclass pseudo_pcmpistrm<string asm> {
6558 def REG : PseudoI<(outs VR128:$dst),
6559 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6560 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6562 def MEM : PseudoI<(outs VR128:$dst),
6563 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6564 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6565 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6568 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6569 let AddedComplexity = 1 in
6570 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6571 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6574 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6575 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6576 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6577 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6579 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6580 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6581 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6584 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6585 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6586 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6587 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6589 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6590 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6591 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6594 // Packed Compare Explicit Length Strings, Return Mask
6595 multiclass pseudo_pcmpestrm<string asm> {
6596 def REG : PseudoI<(outs VR128:$dst),
6597 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6598 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6599 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6600 def MEM : PseudoI<(outs VR128:$dst),
6601 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6602 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6603 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6606 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6607 let AddedComplexity = 1 in
6608 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6609 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6612 let Predicates = [HasAVX],
6613 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6614 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6615 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6616 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6618 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6619 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6620 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6623 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6624 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6625 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6626 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6628 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6629 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6630 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6633 // Packed Compare Implicit Length Strings, Return Index
6634 let Defs = [ECX, EFLAGS] in {
6635 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6636 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6637 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6638 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6639 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6640 (implicit EFLAGS)]>, OpSize;
6641 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6642 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6643 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6644 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6645 (implicit EFLAGS)]>, OpSize;
6649 let Predicates = [HasAVX] in {
6650 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6652 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6654 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6656 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6658 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6660 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6664 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6665 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6666 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6667 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6668 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6669 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6671 // Packed Compare Explicit Length Strings, Return Index
6672 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6673 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6674 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6675 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6676 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6677 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6678 (implicit EFLAGS)]>, OpSize;
6679 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6680 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6681 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6683 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6684 (implicit EFLAGS)]>, OpSize;
6688 let Predicates = [HasAVX] in {
6689 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6691 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6693 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6695 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6697 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6699 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6703 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6704 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6705 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6706 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6707 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6708 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6710 //===----------------------------------------------------------------------===//
6711 // SSE4.2 - CRC Instructions
6712 //===----------------------------------------------------------------------===//
6714 // No CRC instructions have AVX equivalents
6716 // crc intrinsic instruction
6717 // This set of instructions are only rm, the only difference is the size
6719 let Constraints = "$src1 = $dst" in {
6720 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6721 (ins GR32:$src1, i8mem:$src2),
6722 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6724 (int_x86_sse42_crc32_32_8 GR32:$src1,
6725 (load addr:$src2)))]>;
6726 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6727 (ins GR32:$src1, GR8:$src2),
6728 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6730 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6731 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6732 (ins GR32:$src1, i16mem:$src2),
6733 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6735 (int_x86_sse42_crc32_32_16 GR32:$src1,
6736 (load addr:$src2)))]>,
6738 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6739 (ins GR32:$src1, GR16:$src2),
6740 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6742 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6744 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6745 (ins GR32:$src1, i32mem:$src2),
6746 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6748 (int_x86_sse42_crc32_32_32 GR32:$src1,
6749 (load addr:$src2)))]>;
6750 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6751 (ins GR32:$src1, GR32:$src2),
6752 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6754 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6755 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6756 (ins GR64:$src1, i8mem:$src2),
6757 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6759 (int_x86_sse42_crc32_64_8 GR64:$src1,
6760 (load addr:$src2)))]>,
6762 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6763 (ins GR64:$src1, GR8:$src2),
6764 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6766 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6768 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6769 (ins GR64:$src1, i64mem:$src2),
6770 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6772 (int_x86_sse42_crc32_64_64 GR64:$src1,
6773 (load addr:$src2)))]>,
6775 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6776 (ins GR64:$src1, GR64:$src2),
6777 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6779 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6783 //===----------------------------------------------------------------------===//
6784 // AES-NI Instructions
6785 //===----------------------------------------------------------------------===//
6787 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6788 Intrinsic IntId128, bit Is2Addr = 1> {
6789 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6790 (ins VR128:$src1, VR128:$src2),
6792 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6793 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6794 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6796 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6797 (ins VR128:$src1, i128mem:$src2),
6799 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6800 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6802 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6805 // Perform One Round of an AES Encryption/Decryption Flow
6806 let Predicates = [HasAVX, HasAES] in {
6807 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
6808 int_x86_aesni_aesenc, 0>, VEX_4V;
6809 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
6810 int_x86_aesni_aesenclast, 0>, VEX_4V;
6811 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
6812 int_x86_aesni_aesdec, 0>, VEX_4V;
6813 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
6814 int_x86_aesni_aesdeclast, 0>, VEX_4V;
6817 let Constraints = "$src1 = $dst" in {
6818 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
6819 int_x86_aesni_aesenc>;
6820 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
6821 int_x86_aesni_aesenclast>;
6822 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
6823 int_x86_aesni_aesdec>;
6824 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
6825 int_x86_aesni_aesdeclast>;
6828 // Perform the AES InvMixColumn Transformation
6829 let Predicates = [HasAVX, HasAES] in {
6830 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6832 "vaesimc\t{$src1, $dst|$dst, $src1}",
6834 (int_x86_aesni_aesimc VR128:$src1))]>,
6836 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6837 (ins i128mem:$src1),
6838 "vaesimc\t{$src1, $dst|$dst, $src1}",
6839 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
6842 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6844 "aesimc\t{$src1, $dst|$dst, $src1}",
6846 (int_x86_aesni_aesimc VR128:$src1))]>,
6848 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6849 (ins i128mem:$src1),
6850 "aesimc\t{$src1, $dst|$dst, $src1}",
6851 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
6854 // AES Round Key Generation Assist
6855 let Predicates = [HasAVX, HasAES] in {
6856 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6857 (ins VR128:$src1, i8imm:$src2),
6858 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6860 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6862 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6863 (ins i128mem:$src1, i8imm:$src2),
6864 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6866 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
6869 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6870 (ins VR128:$src1, i8imm:$src2),
6871 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6873 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6875 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6876 (ins i128mem:$src1, i8imm:$src2),
6877 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6879 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
6882 //===----------------------------------------------------------------------===//
6883 // CLMUL Instructions
6884 //===----------------------------------------------------------------------===//
6886 // Carry-less Multiplication instructions
6887 let neverHasSideEffects = 1 in {
6888 // AVX carry-less Multiplication instructions
6889 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6890 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6891 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6895 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6896 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6897 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6900 let Constraints = "$src1 = $dst" in {
6901 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6902 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6903 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6907 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6908 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6909 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6911 } // Constraints = "$src1 = $dst"
6912 } // neverHasSideEffects = 1
6915 multiclass pclmul_alias<string asm, int immop> {
6916 def : InstAlias<!strconcat("pclmul", asm,
6917 "dq {$src, $dst|$dst, $src}"),
6918 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
6920 def : InstAlias<!strconcat("pclmul", asm,
6921 "dq {$src, $dst|$dst, $src}"),
6922 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
6924 def : InstAlias<!strconcat("vpclmul", asm,
6925 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6926 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
6928 def : InstAlias<!strconcat("vpclmul", asm,
6929 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6930 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
6932 defm : pclmul_alias<"hqhq", 0x11>;
6933 defm : pclmul_alias<"hqlq", 0x01>;
6934 defm : pclmul_alias<"lqhq", 0x10>;
6935 defm : pclmul_alias<"lqlq", 0x00>;
6937 //===----------------------------------------------------------------------===//
6939 //===----------------------------------------------------------------------===//
6941 //===----------------------------------------------------------------------===//
6942 // VBROADCAST - Load from memory and broadcast to all elements of the
6943 // destination operand
6945 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
6946 X86MemOperand x86memop, Intrinsic Int> :
6947 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
6948 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6949 [(set RC:$dst, (Int addr:$src))]>, VEX;
6951 // AVX2 adds register forms
6952 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
6954 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
6955 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6956 [(set RC:$dst, (Int VR128:$src))]>, VEX;
6958 let ExeDomain = SSEPackedSingle in {
6959 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
6960 int_x86_avx_vbroadcast_ss>;
6961 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
6962 int_x86_avx_vbroadcast_ss_256>;
6964 let ExeDomain = SSEPackedDouble in
6965 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
6966 int_x86_avx_vbroadcast_sd_256>;
6967 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
6968 int_x86_avx_vbroadcastf128_pd_256>;
6970 let ExeDomain = SSEPackedSingle in {
6971 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
6972 int_x86_avx2_vbroadcast_ss_ps>;
6973 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
6974 int_x86_avx2_vbroadcast_ss_ps_256>;
6976 let ExeDomain = SSEPackedDouble in
6977 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
6978 int_x86_avx2_vbroadcast_sd_pd_256>;
6980 let Predicates = [HasAVX2] in
6981 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
6982 int_x86_avx2_vbroadcasti128>;
6984 let Predicates = [HasAVX] in
6985 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
6986 (VBROADCASTF128 addr:$src)>;
6989 //===----------------------------------------------------------------------===//
6990 // VINSERTF128 - Insert packed floating-point values
6992 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
6993 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
6994 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
6995 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6998 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
6999 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7000 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7004 let Predicates = [HasAVX] in {
7005 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7006 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7007 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7008 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7009 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7010 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7013 //===----------------------------------------------------------------------===//
7014 // VEXTRACTF128 - Extract packed floating-point values
7016 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7017 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7018 (ins VR256:$src1, i8imm:$src2),
7019 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7022 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7023 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7024 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7028 let Predicates = [HasAVX] in {
7029 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7030 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7031 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7032 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7033 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7034 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7037 //===----------------------------------------------------------------------===//
7038 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7040 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7041 Intrinsic IntLd, Intrinsic IntLd256,
7042 Intrinsic IntSt, Intrinsic IntSt256> {
7043 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7044 (ins VR128:$src1, f128mem:$src2),
7045 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7046 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7048 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7049 (ins VR256:$src1, f256mem:$src2),
7050 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7051 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7053 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7054 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7055 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7056 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7057 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7058 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7059 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7060 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7063 let ExeDomain = SSEPackedSingle in
7064 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7065 int_x86_avx_maskload_ps,
7066 int_x86_avx_maskload_ps_256,
7067 int_x86_avx_maskstore_ps,
7068 int_x86_avx_maskstore_ps_256>;
7069 let ExeDomain = SSEPackedDouble in
7070 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7071 int_x86_avx_maskload_pd,
7072 int_x86_avx_maskload_pd_256,
7073 int_x86_avx_maskstore_pd,
7074 int_x86_avx_maskstore_pd_256>;
7076 //===----------------------------------------------------------------------===//
7077 // VPERMIL - Permute Single and Double Floating-Point Values
7079 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7080 RegisterClass RC, X86MemOperand x86memop_f,
7081 X86MemOperand x86memop_i, PatFrag i_frag,
7082 Intrinsic IntVar, ValueType vt> {
7083 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7084 (ins RC:$src1, RC:$src2),
7085 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7086 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7087 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7088 (ins RC:$src1, x86memop_i:$src2),
7089 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7090 [(set RC:$dst, (IntVar RC:$src1,
7091 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7093 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7094 (ins RC:$src1, i8imm:$src2),
7095 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7096 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7097 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7098 (ins x86memop_f:$src1, i8imm:$src2),
7099 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7101 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7104 let ExeDomain = SSEPackedSingle in {
7105 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7106 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7107 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7108 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
7110 let ExeDomain = SSEPackedDouble in {
7111 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7112 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7113 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7114 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
7117 let Predicates = [HasAVX] in {
7118 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7119 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7120 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7121 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7122 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7124 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7125 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7126 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7128 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7129 (VPERMILPDri VR128:$src1, imm:$imm)>;
7130 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7131 (VPERMILPDmi addr:$src1, imm:$imm)>;
7134 //===----------------------------------------------------------------------===//
7135 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7137 let ExeDomain = SSEPackedSingle in {
7138 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7139 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7140 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7141 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7142 (i8 imm:$src3))))]>, VEX_4V;
7143 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7144 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7145 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7146 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7147 (i8 imm:$src3)))]>, VEX_4V;
7150 let Predicates = [HasAVX] in {
7151 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7152 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7153 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7154 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7155 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7156 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7157 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7158 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7159 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7160 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7162 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7163 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7164 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7165 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7166 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7167 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7168 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7169 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7170 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7171 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7172 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7173 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7174 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7175 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7176 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7177 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7178 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7179 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7182 //===----------------------------------------------------------------------===//
7183 // VZERO - Zero YMM registers
7185 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7186 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7187 // Zero All YMM registers
7188 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7189 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7191 // Zero Upper bits of YMM registers
7192 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7193 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7196 //===----------------------------------------------------------------------===//
7197 // Half precision conversion instructions
7198 //===----------------------------------------------------------------------===//
7199 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7200 let Predicates = [HasAVX, HasF16C] in {
7201 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7202 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7203 [(set RC:$dst, (Int VR128:$src))]>,
7205 let neverHasSideEffects = 1, mayLoad = 1 in
7206 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7207 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7211 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7212 let Predicates = [HasAVX, HasF16C] in {
7213 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7214 (ins RC:$src1, i32i8imm:$src2),
7215 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7216 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7218 let neverHasSideEffects = 1, mayLoad = 1 in
7219 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7220 (ins RC:$src1, i32i8imm:$src2),
7221 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7226 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7227 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7228 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7229 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7231 //===----------------------------------------------------------------------===//
7232 // AVX2 Instructions
7233 //===----------------------------------------------------------------------===//
7235 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7236 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7237 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7238 X86MemOperand x86memop> {
7239 let isCommutable = 1 in
7240 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7241 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7242 !strconcat(OpcodeStr,
7243 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7244 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7246 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7247 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7248 !strconcat(OpcodeStr,
7249 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7252 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7256 let isCommutable = 0 in {
7257 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7258 VR128, memopv2i64, i128mem>;
7259 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7260 VR256, memopv4i64, i256mem>;
7263 //===----------------------------------------------------------------------===//
7264 // VPBROADCAST - Load from memory and broadcast to all elements of the
7265 // destination operand
7267 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7268 X86MemOperand x86memop, PatFrag ld_frag,
7269 Intrinsic Int128, Intrinsic Int256> {
7270 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7271 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7272 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7273 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7274 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7276 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7277 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7278 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7279 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7280 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7281 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7283 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7286 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7287 int_x86_avx2_pbroadcastb_128,
7288 int_x86_avx2_pbroadcastb_256>;
7289 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7290 int_x86_avx2_pbroadcastw_128,
7291 int_x86_avx2_pbroadcastw_256>;
7292 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7293 int_x86_avx2_pbroadcastd_128,
7294 int_x86_avx2_pbroadcastd_256>;
7295 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7296 int_x86_avx2_pbroadcastq_128,
7297 int_x86_avx2_pbroadcastq_256>;
7299 let Predicates = [HasAVX2] in {
7300 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7301 (VPBROADCASTBrm addr:$src)>;
7302 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7303 (VPBROADCASTBYrm addr:$src)>;
7304 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7305 (VPBROADCASTWrm addr:$src)>;
7306 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7307 (VPBROADCASTWYrm addr:$src)>;
7308 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7309 (VPBROADCASTDrm addr:$src)>;
7310 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7311 (VPBROADCASTDYrm addr:$src)>;
7312 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7313 (VPBROADCASTQrm addr:$src)>;
7314 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7315 (VPBROADCASTQYrm addr:$src)>;
7318 // AVX1 broadcast patterns
7319 let Predicates = [HasAVX] in {
7320 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7321 (VBROADCASTSSYrm addr:$src)>;
7322 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7323 (VBROADCASTSDrm addr:$src)>;
7324 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7325 (VBROADCASTSSYrm addr:$src)>;
7326 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7327 (VBROADCASTSDrm addr:$src)>;
7329 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7330 (VBROADCASTSSrm addr:$src)>;
7331 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7332 (VBROADCASTSSrm addr:$src)>;
7335 //===----------------------------------------------------------------------===//
7336 // VPERM - Permute instructions
7339 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7341 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7342 (ins VR256:$src1, VR256:$src2),
7343 !strconcat(OpcodeStr,
7344 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7345 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7346 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7347 (ins VR256:$src1, i256mem:$src2),
7348 !strconcat(OpcodeStr,
7349 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7350 [(set VR256:$dst, (Int VR256:$src1,
7351 (bitconvert (mem_frag addr:$src2))))]>,
7355 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7356 let ExeDomain = SSEPackedSingle in
7357 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7359 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7361 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7362 (ins VR256:$src1, i8imm:$src2),
7363 !strconcat(OpcodeStr,
7364 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7365 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7366 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7367 (ins i256mem:$src1, i8imm:$src2),
7368 !strconcat(OpcodeStr,
7369 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7370 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7374 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7376 let ExeDomain = SSEPackedDouble in
7377 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7380 //===----------------------------------------------------------------------===//
7381 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7383 let AddedComplexity = 1 in {
7384 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7385 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7386 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7387 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7388 (i8 imm:$src3))))]>, VEX_4V;
7389 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7390 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7391 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7392 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7393 (i8 imm:$src3)))]>, VEX_4V;
7396 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7397 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7398 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7399 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7400 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7401 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7402 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7404 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7406 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7407 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7408 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7409 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7410 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7412 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7416 //===----------------------------------------------------------------------===//
7417 // VINSERTI128 - Insert packed integer values
7419 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7420 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7421 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7423 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7425 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7426 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7427 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7429 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7430 imm:$src3))]>, VEX_4V;
7432 let Predicates = [HasAVX2] in {
7433 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7435 (VINSERTI128rr VR256:$src1, VR128:$src2,
7436 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7437 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7439 (VINSERTI128rr VR256:$src1, VR128:$src2,
7440 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7441 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7443 (VINSERTI128rr VR256:$src1, VR128:$src2,
7444 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7445 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7447 (VINSERTI128rr VR256:$src1, VR128:$src2,
7448 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7452 let Predicates = [HasAVX] in {
7453 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7455 (VINSERTF128rr VR256:$src1, VR128:$src2,
7456 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7457 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7459 (VINSERTF128rr VR256:$src1, VR128:$src2,
7460 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7461 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7463 (VINSERTF128rr VR256:$src1, VR128:$src2,
7464 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7465 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7467 (VINSERTF128rr VR256:$src1, VR128:$src2,
7468 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7469 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7471 (VINSERTF128rr VR256:$src1, VR128:$src2,
7472 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7473 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7475 (VINSERTF128rr VR256:$src1, VR128:$src2,
7476 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7479 //===----------------------------------------------------------------------===//
7480 // VEXTRACTI128 - Extract packed integer values
7482 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7483 (ins VR256:$src1, i8imm:$src2),
7484 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7486 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7488 let neverHasSideEffects = 1, mayStore = 1 in
7489 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7490 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7491 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7493 let Predicates = [HasAVX2] in {
7494 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7495 (v2i64 (VEXTRACTI128rr
7496 (v4i64 VR256:$src1),
7497 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7498 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7499 (v4i32 (VEXTRACTI128rr
7500 (v8i32 VR256:$src1),
7501 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7502 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7503 (v8i16 (VEXTRACTI128rr
7504 (v16i16 VR256:$src1),
7505 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7506 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7507 (v16i8 (VEXTRACTI128rr
7508 (v32i8 VR256:$src1),
7509 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7513 let Predicates = [HasAVX] in {
7514 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7515 (v4f32 (VEXTRACTF128rr
7516 (v8f32 VR256:$src1),
7517 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7518 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7519 (v2f64 (VEXTRACTF128rr
7520 (v4f64 VR256:$src1),
7521 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7522 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7523 (v2i64 (VEXTRACTF128rr
7524 (v4i64 VR256:$src1),
7525 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7526 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7527 (v4i32 (VEXTRACTF128rr
7528 (v8i32 VR256:$src1),
7529 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7530 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7531 (v8i16 (VEXTRACTF128rr
7532 (v16i16 VR256:$src1),
7533 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7534 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7535 (v16i8 (VEXTRACTF128rr
7536 (v32i8 VR256:$src1),
7537 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7540 //===----------------------------------------------------------------------===//
7541 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7543 multiclass avx2_pmovmask<string OpcodeStr,
7544 Intrinsic IntLd128, Intrinsic IntLd256,
7545 Intrinsic IntSt128, Intrinsic IntSt256> {
7546 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7547 (ins VR128:$src1, i128mem:$src2),
7548 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7549 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7550 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7551 (ins VR256:$src1, i256mem:$src2),
7552 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7553 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7554 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7555 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7556 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7557 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7558 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7559 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7560 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7561 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7564 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7565 int_x86_avx2_maskload_d,
7566 int_x86_avx2_maskload_d_256,
7567 int_x86_avx2_maskstore_d,
7568 int_x86_avx2_maskstore_d_256>;
7569 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7570 int_x86_avx2_maskload_q,
7571 int_x86_avx2_maskload_q_256,
7572 int_x86_avx2_maskstore_q,
7573 int_x86_avx2_maskstore_q_256>, VEX_W;
7576 //===----------------------------------------------------------------------===//
7577 // Variable Bit Shifts
7579 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7580 ValueType vt128, ValueType vt256> {
7581 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7582 (ins VR128:$src1, VR128:$src2),
7583 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7585 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7587 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7588 (ins VR128:$src1, i128mem:$src2),
7589 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7591 (vt128 (OpNode VR128:$src1,
7592 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7594 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7595 (ins VR256:$src1, VR256:$src2),
7596 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7598 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7600 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7601 (ins VR256:$src1, i256mem:$src2),
7602 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7604 (vt256 (OpNode VR256:$src1,
7605 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7609 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7610 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7611 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7612 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7613 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;