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, $src2}",
1302 (v4f32 (X86Movlhps VR128:$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, $src2}",
1308 (v4f32 (X86Movhlps VR128:$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, $src2}",
1316 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))]>;
1317 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1318 (ins VR128:$src1, VR128:$src2),
1319 "movhlps\t{$src2, $dst|$dst, $src2}",
1321 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))]>;
1324 let Predicates = [HasAVX] in {
1326 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1327 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1328 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1329 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1332 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1333 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1336 let Predicates = [HasSSE1] in {
1338 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1339 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1340 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1341 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1344 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1345 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1348 //===----------------------------------------------------------------------===//
1349 // SSE 1 & 2 - Conversion Instructions
1350 //===----------------------------------------------------------------------===//
1352 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1353 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1355 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1356 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1357 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1358 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1361 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1362 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1363 string asm, Domain d> {
1364 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1365 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1367 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1368 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1372 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1373 X86MemOperand x86memop, string asm> {
1374 let neverHasSideEffects = 1 in {
1375 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1376 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1378 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1379 (ins DstRC:$src1, x86memop:$src),
1380 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1381 } // neverHasSideEffects = 1
1384 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1385 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1387 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1388 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1390 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1391 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1393 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1394 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1395 VEX, VEX_W, VEX_LIG;
1397 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1398 // register, but the same isn't true when only using memory operands,
1399 // provide other assembly "l" and "q" forms to address this explicitly
1400 // where appropriate to do so.
1401 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1403 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1404 VEX_4V, VEX_W, VEX_LIG;
1405 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1407 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1409 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1410 VEX_4V, VEX_W, VEX_LIG;
1412 let Predicates = [HasAVX], AddedComplexity = 1 in {
1413 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1414 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1415 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1416 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1417 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1418 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1419 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1420 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1422 def : Pat<(f32 (sint_to_fp GR32:$src)),
1423 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1424 def : Pat<(f32 (sint_to_fp GR64:$src)),
1425 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1426 def : Pat<(f64 (sint_to_fp GR32:$src)),
1427 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1428 def : Pat<(f64 (sint_to_fp GR64:$src)),
1429 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1432 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1433 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1434 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1435 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1436 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1437 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1438 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1439 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1440 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1441 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1442 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1443 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1444 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1445 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1446 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1447 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1449 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1450 // and/or XMM operand(s).
1452 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1453 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1455 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1456 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1457 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1458 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1459 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1460 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1463 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1464 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1465 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1466 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1468 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1469 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1470 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1471 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1472 (ins DstRC:$src1, x86memop:$src2),
1474 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1475 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1476 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1479 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1480 f128mem, load, "cvtsd2si">, XD, VEX, VEX_LIG;
1481 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1482 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1483 XD, VEX, VEX_W, VEX_LIG;
1485 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1486 f128mem, load, "cvtsd2si{l}">, XD;
1487 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1488 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1491 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1492 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1493 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1494 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1496 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1497 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1498 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1499 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1502 let Constraints = "$src1 = $dst" in {
1503 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1504 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1506 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1507 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1508 "cvtsi2ss{q}">, XS, REX_W;
1509 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1510 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1512 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1513 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1514 "cvtsi2sd">, XD, REX_W;
1519 // Aliases for intrinsics
1520 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1521 f32mem, load, "cvttss2si">, XS, VEX;
1522 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1523 int_x86_sse_cvttss2si64, f32mem, load,
1524 "cvttss2si">, XS, VEX, VEX_W;
1525 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1526 f128mem, load, "cvttsd2si">, XD, VEX;
1527 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1528 int_x86_sse2_cvttsd2si64, f128mem, load,
1529 "cvttsd2si">, XD, VEX, VEX_W;
1530 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1531 f32mem, load, "cvttss2si">, XS;
1532 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1533 int_x86_sse_cvttss2si64, f32mem, load,
1534 "cvttss2si{q}">, XS, REX_W;
1535 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1536 f128mem, load, "cvttsd2si">, XD;
1537 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1538 int_x86_sse2_cvttsd2si64, f128mem, load,
1539 "cvttsd2si{q}">, XD, REX_W;
1541 let Pattern = []<dag> in {
1542 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1543 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1545 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1546 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1548 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1549 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1550 SSEPackedSingle>, TB, VEX;
1551 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1552 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1553 SSEPackedSingle>, TB, VEX;
1556 let Pattern = []<dag> in {
1557 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1558 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1559 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1560 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1561 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1562 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1563 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1566 let Predicates = [HasAVX] in {
1567 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1568 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1569 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1570 (VCVTSS2SIrm addr:$src)>;
1571 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1572 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1573 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1574 (VCVTSS2SI64rm addr:$src)>;
1577 let Predicates = [HasSSE1] in {
1578 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1579 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1580 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1581 (CVTSS2SIrm addr:$src)>;
1582 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1583 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1584 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1585 (CVTSS2SI64rm addr:$src)>;
1590 // Convert scalar double to scalar single
1591 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1592 (ins FR64:$src1, FR64:$src2),
1593 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1596 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1597 (ins FR64:$src1, f64mem:$src2),
1598 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1599 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1601 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1604 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1605 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1606 [(set FR32:$dst, (fround FR64:$src))]>;
1607 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1608 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1609 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1610 Requires<[HasSSE2, OptForSize]>;
1612 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1613 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1615 let Constraints = "$src1 = $dst" in
1616 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1617 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1619 // Convert scalar single to scalar double
1620 // SSE2 instructions with XS prefix
1621 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1622 (ins FR32:$src1, FR32:$src2),
1623 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1624 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1626 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1627 (ins FR32:$src1, f32mem:$src2),
1628 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1629 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1631 let Predicates = [HasAVX] in {
1632 def : Pat<(f64 (fextend FR32:$src)),
1633 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1634 def : Pat<(fextend (loadf32 addr:$src)),
1635 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1636 def : Pat<(extloadf32 addr:$src),
1637 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1640 def : Pat<(extloadf32 addr:$src),
1641 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1642 Requires<[HasAVX, OptForSpeed]>;
1644 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1645 "cvtss2sd\t{$src, $dst|$dst, $src}",
1646 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1647 Requires<[HasSSE2]>;
1648 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1649 "cvtss2sd\t{$src, $dst|$dst, $src}",
1650 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1651 Requires<[HasSSE2, OptForSize]>;
1653 // extload f32 -> f64. This matches load+fextend because we have a hack in
1654 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1656 // Since these loads aren't folded into the fextend, we have to match it
1658 def : Pat<(fextend (loadf32 addr:$src)),
1659 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1660 def : Pat<(extloadf32 addr:$src),
1661 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1663 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1664 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1665 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1666 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1667 VR128:$src2))]>, XS, VEX_4V,
1669 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1670 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1671 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1672 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1673 (load addr:$src2)))]>, XS, VEX_4V,
1675 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1676 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1677 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1678 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1679 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1680 VR128:$src2))]>, XS,
1681 Requires<[HasSSE2]>;
1682 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1683 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1684 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1685 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1686 (load addr:$src2)))]>, XS,
1687 Requires<[HasSSE2]>;
1690 // Convert doubleword to packed single/double fp
1691 // SSE2 instructions without OpSize prefix
1692 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1693 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1694 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1695 TB, VEX, Requires<[HasAVX]>;
1696 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1697 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1698 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1699 (bitconvert (memopv2i64 addr:$src))))]>,
1700 TB, VEX, Requires<[HasAVX]>;
1701 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1702 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1703 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1704 TB, Requires<[HasSSE2]>;
1705 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1706 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1707 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1708 (bitconvert (memopv2i64 addr:$src))))]>,
1709 TB, Requires<[HasSSE2]>;
1711 // FIXME: why the non-intrinsic version is described as SSE3?
1712 // SSE2 instructions with XS prefix
1713 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1714 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1715 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1716 XS, VEX, Requires<[HasAVX]>;
1717 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1718 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1719 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1720 (bitconvert (memopv2i64 addr:$src))))]>,
1721 XS, VEX, Requires<[HasAVX]>;
1722 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1723 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1724 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1725 XS, Requires<[HasSSE2]>;
1726 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1727 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1728 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1729 (bitconvert (memopv2i64 addr:$src))))]>,
1730 XS, Requires<[HasSSE2]>;
1733 // Convert packed single/double fp to doubleword
1734 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1735 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1736 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1737 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1738 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1739 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1740 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1741 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1742 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1743 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1744 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1745 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1747 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1748 "cvtps2dq\t{$src, $dst|$dst, $src}",
1749 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1751 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1753 "cvtps2dq\t{$src, $dst|$dst, $src}",
1754 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1755 (memop addr:$src)))]>, VEX;
1756 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1757 "cvtps2dq\t{$src, $dst|$dst, $src}",
1758 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1759 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1760 "cvtps2dq\t{$src, $dst|$dst, $src}",
1761 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1762 (memop addr:$src)))]>;
1764 // SSE2 packed instructions with XD prefix
1765 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1766 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1767 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1768 XD, VEX, Requires<[HasAVX]>;
1769 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1770 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1771 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1772 (memop addr:$src)))]>,
1773 XD, VEX, Requires<[HasAVX]>;
1774 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1775 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1776 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1777 XD, Requires<[HasSSE2]>;
1778 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1779 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1780 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1781 (memop addr:$src)))]>,
1782 XD, Requires<[HasSSE2]>;
1785 // Convert with truncation packed single/double fp to doubleword
1786 // SSE2 packed instructions with XS prefix
1787 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1788 "cvttps2dq\t{$src, $dst|$dst, $src}",
1790 (int_x86_sse2_cvttps2dq VR128:$src))]>, VEX;
1791 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1792 "cvttps2dq\t{$src, $dst|$dst, $src}",
1793 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1794 (memop addr:$src)))]>, VEX;
1795 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1796 "cvttps2dq\t{$src, $dst|$dst, $src}",
1798 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))]>, VEX;
1799 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1800 "cvttps2dq\t{$src, $dst|$dst, $src}",
1801 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1802 (memopv8f32 addr:$src)))]>, VEX;
1804 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1805 "cvttps2dq\t{$src, $dst|$dst, $src}",
1807 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1808 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1809 "cvttps2dq\t{$src, $dst|$dst, $src}",
1811 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1813 let Predicates = [HasAVX] in {
1814 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1815 (Int_VCVTDQ2PSrr VR128:$src)>;
1816 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1817 (Int_VCVTDQ2PSrm addr:$src)>;
1819 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1820 (VCVTTPS2DQrr VR128:$src)>;
1821 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1822 (VCVTTPS2DQrm addr:$src)>;
1824 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1825 (VCVTDQ2PSYrr VR256:$src)>;
1826 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1827 (VCVTDQ2PSYrm addr:$src)>;
1829 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1830 (VCVTTPS2DQYrr VR256:$src)>;
1831 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1832 (VCVTTPS2DQYrm addr:$src)>;
1835 let Predicates = [HasSSE2] in {
1836 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1837 (Int_CVTDQ2PSrr VR128:$src)>;
1838 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1839 (Int_CVTDQ2PSrm addr:$src)>;
1841 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1842 (CVTTPS2DQrr VR128:$src)>;
1843 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1844 (CVTTPS2DQrm addr:$src)>;
1847 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1848 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1850 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1851 let isCodeGenOnly = 1 in
1852 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1853 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1854 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1855 (memop addr:$src)))]>, VEX;
1856 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1857 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1858 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1859 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1860 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1861 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1862 (memop addr:$src)))]>;
1864 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1865 // register, but the same isn't true when using memory operands instead.
1866 // Provide other assembly rr and rm forms to address this explicitly.
1867 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1868 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1871 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1872 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1873 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1874 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1877 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1878 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1879 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1880 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1882 // Convert packed single to packed double
1883 let Predicates = [HasAVX] in {
1884 // SSE2 instructions without OpSize prefix
1885 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1886 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1887 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1888 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1889 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1890 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1891 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1892 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1894 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1895 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1896 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1897 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1899 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1900 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1901 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1902 TB, VEX, Requires<[HasAVX]>;
1903 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1904 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1905 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1906 (load addr:$src)))]>,
1907 TB, VEX, Requires<[HasAVX]>;
1908 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1909 "cvtps2pd\t{$src, $dst|$dst, $src}",
1910 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1911 TB, Requires<[HasSSE2]>;
1912 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1913 "cvtps2pd\t{$src, $dst|$dst, $src}",
1914 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1915 (load addr:$src)))]>,
1916 TB, Requires<[HasSSE2]>;
1918 // Convert packed double to packed single
1919 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1920 // register, but the same isn't true when using memory operands instead.
1921 // Provide other assembly rr and rm forms to address this explicitly.
1922 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1923 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1924 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1925 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1928 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1929 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1930 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1931 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1934 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1935 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1936 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1937 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1938 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1939 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1940 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1941 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1944 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1945 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1946 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1947 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1949 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1950 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1951 (memop addr:$src)))]>;
1952 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1953 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1954 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1955 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1956 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1957 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1958 (memop addr:$src)))]>;
1960 // AVX 256-bit register conversion intrinsics
1961 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1962 // whenever possible to avoid declaring two versions of each one.
1963 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1964 (VCVTDQ2PSYrr VR256:$src)>;
1965 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
1966 (VCVTDQ2PSYrm addr:$src)>;
1968 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1969 (VCVTPD2PSYrr VR256:$src)>;
1970 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1971 (VCVTPD2PSYrm addr:$src)>;
1973 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1974 (VCVTPS2DQYrr VR256:$src)>;
1975 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1976 (VCVTPS2DQYrm addr:$src)>;
1978 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1979 (VCVTPS2PDYrr VR128:$src)>;
1980 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1981 (VCVTPS2PDYrm addr:$src)>;
1983 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1984 (VCVTTPD2DQYrr VR256:$src)>;
1985 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1986 (VCVTTPD2DQYrm addr:$src)>;
1988 // Match fround and fextend for 128/256-bit conversions
1989 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1990 (VCVTPD2PSYrr VR256:$src)>;
1991 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1992 (VCVTPD2PSYrm addr:$src)>;
1994 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1995 (VCVTPS2PDYrr VR128:$src)>;
1996 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1997 (VCVTPS2PDYrm addr:$src)>;
1999 //===----------------------------------------------------------------------===//
2000 // SSE 1 & 2 - Compare Instructions
2001 //===----------------------------------------------------------------------===//
2003 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2004 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2005 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2006 string asm, string asm_alt> {
2007 def rr : SIi8<0xC2, MRMSrcReg,
2008 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2009 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2010 def rm : SIi8<0xC2, MRMSrcMem,
2011 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2012 [(set RC:$dst, (OpNode (VT RC:$src1),
2013 (ld_frag addr:$src2), imm:$cc))]>;
2015 // Accept explicit immediate argument form instead of comparison code.
2016 let neverHasSideEffects = 1 in {
2017 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2018 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2020 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2021 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2025 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2026 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2027 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2028 XS, VEX_4V, VEX_LIG;
2029 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2030 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2031 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2032 XD, VEX_4V, VEX_LIG;
2034 let Constraints = "$src1 = $dst" in {
2035 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2036 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2037 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2039 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2040 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2041 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2045 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2046 Intrinsic Int, string asm> {
2047 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2048 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2049 [(set VR128:$dst, (Int VR128:$src1,
2050 VR128:$src, imm:$cc))]>;
2051 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2052 (ins VR128:$src1, x86memop:$src, SSECC:$cc), asm,
2053 [(set VR128:$dst, (Int VR128:$src1,
2054 (load addr:$src), imm:$cc))]>;
2057 // Aliases to match intrinsics which expect XMM operand(s).
2058 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2059 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2061 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2062 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2064 let Constraints = "$src1 = $dst" in {
2065 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2066 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2067 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2068 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2072 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2073 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2074 ValueType vt, X86MemOperand x86memop,
2075 PatFrag ld_frag, string OpcodeStr, Domain d> {
2076 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2077 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2078 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2080 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2081 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2082 [(set EFLAGS, (OpNode (vt RC:$src1),
2083 (ld_frag addr:$src2)))],
2087 let Defs = [EFLAGS] in {
2088 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2089 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2090 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2091 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2093 let Pattern = []<dag> in {
2094 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2095 "comiss", SSEPackedSingle>, TB, VEX,
2097 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2098 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2102 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2103 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2104 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2105 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2107 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2108 load, "comiss", SSEPackedSingle>, TB, VEX;
2109 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2110 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2111 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2112 "ucomiss", SSEPackedSingle>, TB;
2113 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2114 "ucomisd", SSEPackedDouble>, TB, OpSize;
2116 let Pattern = []<dag> in {
2117 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2118 "comiss", SSEPackedSingle>, TB;
2119 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2120 "comisd", SSEPackedDouble>, TB, OpSize;
2123 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2124 load, "ucomiss", SSEPackedSingle>, TB;
2125 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2126 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2128 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2129 "comiss", SSEPackedSingle>, TB;
2130 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2131 "comisd", SSEPackedDouble>, TB, OpSize;
2132 } // Defs = [EFLAGS]
2134 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2135 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2136 Intrinsic Int, string asm, string asm_alt,
2138 let isAsmParserOnly = 1 in {
2139 def rri : PIi8<0xC2, MRMSrcReg,
2140 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2141 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2143 def rmi : PIi8<0xC2, MRMSrcMem,
2144 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2145 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2149 // Accept explicit immediate argument form instead of comparison code.
2150 def rri_alt : PIi8<0xC2, MRMSrcReg,
2151 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2152 asm_alt, [], IIC_DEFAULT, d>;
2153 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2154 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2155 asm_alt, [], IIC_DEFAULT, d>;
2158 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2159 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2160 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2161 SSEPackedSingle>, TB, VEX_4V;
2162 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2163 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2164 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2165 SSEPackedDouble>, TB, OpSize, VEX_4V;
2166 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2167 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2168 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2169 SSEPackedSingle>, TB, VEX_4V;
2170 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2171 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2172 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2173 SSEPackedDouble>, TB, OpSize, VEX_4V;
2174 let Constraints = "$src1 = $dst" in {
2175 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2176 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2177 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2178 SSEPackedSingle>, TB;
2179 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2180 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2181 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2182 SSEPackedDouble>, TB, OpSize;
2185 let Predicates = [HasAVX] in {
2186 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2187 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2188 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2189 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2190 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2191 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2192 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2193 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2195 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2196 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2197 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2198 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2199 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2200 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2201 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2202 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2205 let Predicates = [HasSSE1] in {
2206 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2207 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2208 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2209 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2212 let Predicates = [HasSSE2] in {
2213 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2214 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2215 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2216 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2219 //===----------------------------------------------------------------------===//
2220 // SSE 1 & 2 - Shuffle Instructions
2221 //===----------------------------------------------------------------------===//
2223 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2224 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2225 ValueType vt, string asm, PatFrag mem_frag,
2226 Domain d, bit IsConvertibleToThreeAddress = 0> {
2227 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2228 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2229 [(set RC:$dst, (vt (shufp:$src3
2230 RC:$src1, (mem_frag addr:$src2))))],
2232 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2233 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2234 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2236 (vt (shufp:$src3 RC:$src1, RC:$src2)))],
2240 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2241 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2242 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2243 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2244 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2245 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2246 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2247 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2248 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2249 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2250 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2251 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2253 let Constraints = "$src1 = $dst" in {
2254 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2255 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2256 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2258 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2259 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2260 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2264 let Predicates = [HasAVX] in {
2265 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2266 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2267 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2268 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2269 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2270 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2271 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2272 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2273 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2274 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2275 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2276 // fall back to this for SSE1)
2277 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2278 (VSHUFPSrri VR128:$src2, VR128:$src1,
2279 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2280 // Special unary SHUFPSrri case.
2281 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2282 (VSHUFPSrri VR128:$src1, VR128:$src1,
2283 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2284 // Special binary v4i32 shuffle cases with SHUFPS.
2285 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2286 (VSHUFPSrri VR128:$src1, VR128:$src2,
2287 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2288 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2289 (bc_v4i32 (memopv2i64 addr:$src2)))),
2290 (VSHUFPSrmi VR128:$src1, addr:$src2,
2291 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2292 // Special unary SHUFPDrri cases.
2293 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2294 (VSHUFPDrri VR128:$src1, VR128:$src1,
2295 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2296 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2297 (VSHUFPDrri VR128:$src1, VR128:$src1,
2298 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2299 // Special binary v2i64 shuffle cases using SHUFPDrri.
2300 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2301 (VSHUFPDrri VR128:$src1, VR128:$src2,
2302 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2304 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2305 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2306 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2307 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2308 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2309 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2310 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2311 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2312 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2313 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2316 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2317 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2318 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2319 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2320 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2322 def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2323 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2324 def : Pat<(v8f32 (X86Shufp VR256:$src1,
2325 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2326 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2328 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2329 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2330 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2331 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2332 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2334 def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2335 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2336 def : Pat<(v4f64 (X86Shufp VR256:$src1,
2337 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2338 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2341 let Predicates = [HasSSE1] in {
2342 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2343 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2344 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2345 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2346 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2347 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2348 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2349 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2350 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2351 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2352 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2353 // fall back to this for SSE1)
2354 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2355 (SHUFPSrri VR128:$src2, VR128:$src1,
2356 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2357 // Special unary SHUFPSrri case.
2358 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2359 (SHUFPSrri VR128:$src1, VR128:$src1,
2360 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2363 let Predicates = [HasSSE2] in {
2364 // Special binary v4i32 shuffle cases with SHUFPS.
2365 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2366 (SHUFPSrri VR128:$src1, VR128:$src2,
2367 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2368 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2369 (bc_v4i32 (memopv2i64 addr:$src2)))),
2370 (SHUFPSrmi VR128:$src1, addr:$src2,
2371 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2372 // Special unary SHUFPDrri cases.
2373 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2374 (SHUFPDrri VR128:$src1, VR128:$src1,
2375 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2376 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2377 (SHUFPDrri VR128:$src1, VR128:$src1,
2378 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2379 // Special binary v2i64 shuffle cases using SHUFPDrri.
2380 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2381 (SHUFPDrri VR128:$src1, VR128:$src2,
2382 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2383 // Generic SHUFPD patterns
2384 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2385 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2386 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2387 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2388 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2389 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2390 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2391 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2392 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2393 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2396 //===----------------------------------------------------------------------===//
2397 // SSE 1 & 2 - Unpack Instructions
2398 //===----------------------------------------------------------------------===//
2400 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2401 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2402 PatFrag mem_frag, RegisterClass RC,
2403 X86MemOperand x86memop, string asm,
2405 def rr : PI<opc, MRMSrcReg,
2406 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2408 (vt (OpNode RC:$src1, RC:$src2)))],
2410 def rm : PI<opc, MRMSrcMem,
2411 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2413 (vt (OpNode RC:$src1,
2414 (mem_frag addr:$src2))))],
2418 let AddedComplexity = 10 in {
2419 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2420 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2421 SSEPackedSingle>, TB, VEX_4V;
2422 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2423 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2424 SSEPackedDouble>, TB, OpSize, VEX_4V;
2425 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2426 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2427 SSEPackedSingle>, TB, VEX_4V;
2428 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2429 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2430 SSEPackedDouble>, TB, OpSize, VEX_4V;
2432 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2433 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2434 SSEPackedSingle>, TB, VEX_4V;
2435 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2436 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2437 SSEPackedDouble>, TB, OpSize, VEX_4V;
2438 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2439 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2440 SSEPackedSingle>, TB, VEX_4V;
2441 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2442 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2443 SSEPackedDouble>, TB, OpSize, VEX_4V;
2445 let Constraints = "$src1 = $dst" in {
2446 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2447 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2448 SSEPackedSingle>, TB;
2449 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2450 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2451 SSEPackedDouble>, TB, OpSize;
2452 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2453 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2454 SSEPackedSingle>, TB;
2455 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2456 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2457 SSEPackedDouble>, TB, OpSize;
2458 } // Constraints = "$src1 = $dst"
2459 } // AddedComplexity
2461 let Predicates = [HasAVX], AddedComplexity = 1 in {
2462 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2463 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2464 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2465 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2466 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2467 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2468 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2469 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2471 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2472 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2473 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2474 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2475 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2476 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2477 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2478 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2480 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2481 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2482 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2483 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2484 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2485 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2486 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2487 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2489 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2490 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2491 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2492 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2493 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2494 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2495 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2496 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2498 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2499 // problem is during lowering, where it's not possible to recognize the load
2500 // fold cause it has two uses through a bitcast. One use disappears at isel
2501 // time and the fold opportunity reappears.
2502 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2503 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2506 let Predicates = [HasSSE1] in {
2507 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2508 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2509 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2510 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2511 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2512 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2513 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2514 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2517 let Predicates = [HasSSE2] in {
2518 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2519 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2520 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2521 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2522 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2523 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2524 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2525 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2527 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2528 // problem is during lowering, where it's not possible to recognize the load
2529 // fold cause it has two uses through a bitcast. One use disappears at isel
2530 // time and the fold opportunity reappears.
2531 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2532 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2535 //===----------------------------------------------------------------------===//
2536 // SSE 1 & 2 - Extract Floating-Point Sign mask
2537 //===----------------------------------------------------------------------===//
2539 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2540 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2542 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2543 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2544 [(set GR32:$dst, (Int RC:$src))], IIC_DEFAULT, d>;
2545 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2546 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2547 IIC_DEFAULT, d>, REX_W;
2550 let Predicates = [HasAVX] in {
2551 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2552 "movmskps", SSEPackedSingle>, TB, VEX;
2553 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2554 "movmskpd", SSEPackedDouble>, TB,
2556 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2557 "movmskps", SSEPackedSingle>, TB, VEX;
2558 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2559 "movmskpd", SSEPackedDouble>, TB,
2562 def : Pat<(i32 (X86fgetsign FR32:$src)),
2563 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2565 def : Pat<(i64 (X86fgetsign FR32:$src)),
2566 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2568 def : Pat<(i32 (X86fgetsign FR64:$src)),
2569 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2571 def : Pat<(i64 (X86fgetsign FR64:$src)),
2572 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2576 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2577 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2578 SSEPackedSingle>, TB, VEX;
2579 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2580 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2581 SSEPackedDouble>, TB,
2583 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2584 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2585 SSEPackedSingle>, TB, VEX;
2586 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2587 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2588 SSEPackedDouble>, TB,
2592 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2593 SSEPackedSingle>, TB;
2594 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2595 SSEPackedDouble>, TB, OpSize;
2597 def : Pat<(i32 (X86fgetsign FR32:$src)),
2598 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2599 sub_ss))>, Requires<[HasSSE1]>;
2600 def : Pat<(i64 (X86fgetsign FR32:$src)),
2601 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2602 sub_ss))>, Requires<[HasSSE1]>;
2603 def : Pat<(i32 (X86fgetsign FR64:$src)),
2604 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2605 sub_sd))>, Requires<[HasSSE2]>;
2606 def : Pat<(i64 (X86fgetsign FR64:$src)),
2607 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2608 sub_sd))>, Requires<[HasSSE2]>;
2610 //===---------------------------------------------------------------------===//
2611 // SSE2 - Packed Integer Logical Instructions
2612 //===---------------------------------------------------------------------===//
2614 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2616 /// PDI_binop_rm - Simple SSE2 binary operator.
2617 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2618 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2619 X86MemOperand x86memop, bit IsCommutable = 0,
2621 let isCommutable = IsCommutable in
2622 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2623 (ins RC:$src1, RC:$src2),
2625 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2626 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2627 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2628 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2629 (ins RC:$src1, x86memop:$src2),
2631 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2632 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2633 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2634 (bitconvert (memop_frag addr:$src2)))))]>;
2636 } // ExeDomain = SSEPackedInt
2638 // These are ordered here for pattern ordering requirements with the fp versions
2640 let Predicates = [HasAVX] in {
2641 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2642 i128mem, 1, 0>, VEX_4V;
2643 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2644 i128mem, 1, 0>, VEX_4V;
2645 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2646 i128mem, 1, 0>, VEX_4V;
2647 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2648 i128mem, 0, 0>, VEX_4V;
2651 let Constraints = "$src1 = $dst" in {
2652 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2654 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2656 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2658 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2660 } // Constraints = "$src1 = $dst"
2662 let Predicates = [HasAVX2] in {
2663 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2664 i256mem, 1, 0>, VEX_4V;
2665 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2666 i256mem, 1, 0>, VEX_4V;
2667 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2668 i256mem, 1, 0>, VEX_4V;
2669 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2670 i256mem, 0, 0>, VEX_4V;
2673 //===----------------------------------------------------------------------===//
2674 // SSE 1 & 2 - Logical Instructions
2675 //===----------------------------------------------------------------------===//
2677 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2679 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2681 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2682 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2684 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2685 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2687 let Constraints = "$src1 = $dst" in {
2688 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2689 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2691 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2692 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2696 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2697 let mayLoad = 0 in {
2698 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2699 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2700 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2703 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2704 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2706 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2708 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2710 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2711 // are all promoted to v2i64, and the patterns are covered by the int
2712 // version. This is needed in SSE only, because v2i64 isn't supported on
2713 // SSE1, but only on SSE2.
2714 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2715 !strconcat(OpcodeStr, "ps"), f128mem, [],
2716 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2717 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2719 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2720 !strconcat(OpcodeStr, "pd"), f128mem,
2721 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2722 (bc_v2i64 (v2f64 VR128:$src2))))],
2723 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2724 (memopv2i64 addr:$src2)))], 0>,
2726 let Constraints = "$src1 = $dst" in {
2727 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2728 !strconcat(OpcodeStr, "ps"), f128mem,
2729 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2730 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2731 (memopv2i64 addr:$src2)))]>, TB;
2733 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2734 !strconcat(OpcodeStr, "pd"), f128mem,
2735 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2736 (bc_v2i64 (v2f64 VR128:$src2))))],
2737 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2738 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2742 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2744 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2746 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2747 !strconcat(OpcodeStr, "ps"), f256mem,
2748 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2749 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2750 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2752 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2753 !strconcat(OpcodeStr, "pd"), f256mem,
2754 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2755 (bc_v4i64 (v4f64 VR256:$src2))))],
2756 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2757 (memopv4i64 addr:$src2)))], 0>,
2761 // AVX 256-bit packed logical ops forms
2762 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2763 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2764 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2765 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2767 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2768 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2769 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2770 let isCommutable = 0 in
2771 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2773 //===----------------------------------------------------------------------===//
2774 // SSE 1 & 2 - Arithmetic Instructions
2775 //===----------------------------------------------------------------------===//
2777 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2780 /// In addition, we also have a special variant of the scalar form here to
2781 /// represent the associated intrinsic operation. This form is unlike the
2782 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2783 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2785 /// These three forms can each be reg+reg or reg+mem.
2788 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2790 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2792 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2793 OpNode, FR32, f32mem, Is2Addr>, XS;
2794 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2795 OpNode, FR64, f64mem, Is2Addr>, XD;
2798 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2800 let mayLoad = 0 in {
2801 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2802 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2803 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2804 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2808 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2810 let mayLoad = 0 in {
2811 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2812 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2813 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2814 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2818 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2820 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2821 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2822 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2823 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2826 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2828 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2829 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2830 SSEPackedSingle, Is2Addr>, TB;
2832 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2833 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2834 SSEPackedDouble, Is2Addr>, TB, OpSize;
2837 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2838 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2839 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2840 SSEPackedSingle, 0>, TB;
2842 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2843 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2844 SSEPackedDouble, 0>, TB, OpSize;
2847 // Binary Arithmetic instructions
2848 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2849 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2850 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2851 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2852 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2853 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2854 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2855 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2857 let isCommutable = 0 in {
2858 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2859 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2860 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2861 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2862 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2863 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2864 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2865 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2866 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2867 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2868 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2869 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2870 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2871 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2872 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2873 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2874 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2875 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2876 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2877 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2880 let Constraints = "$src1 = $dst" in {
2881 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2882 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2883 basic_sse12_fp_binop_s_int<0x58, "add">;
2884 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2885 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2886 basic_sse12_fp_binop_s_int<0x59, "mul">;
2888 let isCommutable = 0 in {
2889 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2890 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2891 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2892 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2893 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2894 basic_sse12_fp_binop_s_int<0x5E, "div">;
2895 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2896 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2897 basic_sse12_fp_binop_s_int<0x5F, "max">,
2898 basic_sse12_fp_binop_p_int<0x5F, "max">;
2899 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2900 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2901 basic_sse12_fp_binop_s_int<0x5D, "min">,
2902 basic_sse12_fp_binop_p_int<0x5D, "min">;
2907 /// In addition, we also have a special variant of the scalar form here to
2908 /// represent the associated intrinsic operation. This form is unlike the
2909 /// plain scalar form, in that it takes an entire vector (instead of a
2910 /// scalar) and leaves the top elements undefined.
2912 /// And, we have a special variant form for a full-vector intrinsic form.
2914 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2915 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2916 SDNode OpNode, Intrinsic F32Int> {
2917 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2918 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2919 [(set FR32:$dst, (OpNode FR32:$src))]>;
2920 // For scalar unary operations, fold a load into the operation
2921 // only in OptForSize mode. It eliminates an instruction, but it also
2922 // eliminates a whole-register clobber (the load), so it introduces a
2923 // partial register update condition.
2924 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2925 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2926 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2927 Requires<[HasSSE1, OptForSize]>;
2928 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2929 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2930 [(set VR128:$dst, (F32Int VR128:$src))]>;
2931 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2932 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2933 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2936 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2937 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2938 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2939 !strconcat(OpcodeStr,
2940 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2942 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2943 !strconcat(OpcodeStr,
2944 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2945 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2946 (ins VR128:$src1, ssmem:$src2),
2947 !strconcat(OpcodeStr,
2948 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2951 /// sse1_fp_unop_p - SSE1 unops in packed form.
2952 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2953 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2954 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2955 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2956 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2957 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2958 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2961 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2962 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2963 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2964 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2965 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2966 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2967 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2968 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2971 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2972 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2973 Intrinsic V4F32Int> {
2974 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2975 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2976 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2977 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2978 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2979 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2982 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2983 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2984 Intrinsic V4F32Int> {
2985 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2986 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2987 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2988 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2989 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2990 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2993 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2994 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2995 SDNode OpNode, Intrinsic F64Int> {
2996 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2997 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2998 [(set FR64:$dst, (OpNode FR64:$src))]>;
2999 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3000 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3001 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3002 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
3003 Requires<[HasSSE2, OptForSize]>;
3004 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3005 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3006 [(set VR128:$dst, (F64Int VR128:$src))]>;
3007 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3008 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3009 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
3012 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3013 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3014 let neverHasSideEffects = 1 in {
3015 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3016 !strconcat(OpcodeStr,
3017 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3019 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3020 !strconcat(OpcodeStr,
3021 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3023 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3024 (ins VR128:$src1, sdmem:$src2),
3025 !strconcat(OpcodeStr,
3026 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3029 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3030 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3032 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3033 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3034 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3035 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3036 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3037 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3040 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3041 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3042 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3043 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3044 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3045 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3046 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3047 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3050 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3051 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3052 Intrinsic V2F64Int> {
3053 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3054 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3055 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3056 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3057 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3058 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3061 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3062 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3063 Intrinsic V2F64Int> {
3064 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3065 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3066 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3067 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3068 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3069 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3072 let Predicates = [HasAVX] in {
3074 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3075 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3077 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3078 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3079 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3080 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3081 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3082 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3083 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3084 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3087 // Reciprocal approximations. Note that these typically require refinement
3088 // in order to obtain suitable precision.
3089 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3090 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3091 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3092 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3093 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3095 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3096 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3097 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3098 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3099 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3102 let AddedComplexity = 1 in {
3103 def : Pat<(f32 (fsqrt FR32:$src)),
3104 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3105 def : Pat<(f32 (fsqrt (load addr:$src))),
3106 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3107 Requires<[HasAVX, OptForSize]>;
3108 def : Pat<(f64 (fsqrt FR64:$src)),
3109 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3110 def : Pat<(f64 (fsqrt (load addr:$src))),
3111 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3112 Requires<[HasAVX, OptForSize]>;
3114 def : Pat<(f32 (X86frsqrt FR32:$src)),
3115 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3116 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3117 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3118 Requires<[HasAVX, OptForSize]>;
3120 def : Pat<(f32 (X86frcp FR32:$src)),
3121 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3122 def : Pat<(f32 (X86frcp (load addr:$src))),
3123 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3124 Requires<[HasAVX, OptForSize]>;
3127 let Predicates = [HasAVX], AddedComplexity = 1 in {
3128 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3129 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3130 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3131 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3133 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3134 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3136 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3137 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3138 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3139 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3141 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3142 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3144 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3145 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3146 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3147 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3149 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3150 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3152 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3153 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3154 (VRCPSSr (f32 (IMPLICIT_DEF)),
3155 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3157 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3158 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3162 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3163 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3164 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3165 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3166 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3167 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3169 // Reciprocal approximations. Note that these typically require refinement
3170 // in order to obtain suitable precision.
3171 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3172 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3173 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3174 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3175 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3176 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3178 // There is no f64 version of the reciprocal approximation instructions.
3180 //===----------------------------------------------------------------------===//
3181 // SSE 1 & 2 - Non-temporal stores
3182 //===----------------------------------------------------------------------===//
3184 let AddedComplexity = 400 in { // Prefer non-temporal versions
3185 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3186 (ins f128mem:$dst, VR128:$src),
3187 "movntps\t{$src, $dst|$dst, $src}",
3188 [(alignednontemporalstore (v4f32 VR128:$src),
3190 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3191 (ins f128mem:$dst, VR128:$src),
3192 "movntpd\t{$src, $dst|$dst, $src}",
3193 [(alignednontemporalstore (v2f64 VR128:$src),
3196 let ExeDomain = SSEPackedInt in
3197 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3198 (ins f128mem:$dst, VR128:$src),
3199 "movntdq\t{$src, $dst|$dst, $src}",
3200 [(alignednontemporalstore (v2i64 VR128:$src),
3203 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3204 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3206 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3207 (ins f256mem:$dst, VR256:$src),
3208 "movntps\t{$src, $dst|$dst, $src}",
3209 [(alignednontemporalstore (v8f32 VR256:$src),
3211 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3212 (ins f256mem:$dst, VR256:$src),
3213 "movntpd\t{$src, $dst|$dst, $src}",
3214 [(alignednontemporalstore (v4f64 VR256:$src),
3216 let ExeDomain = SSEPackedInt in
3217 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3218 (ins f256mem:$dst, VR256:$src),
3219 "movntdq\t{$src, $dst|$dst, $src}",
3220 [(alignednontemporalstore (v4i64 VR256:$src),
3224 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3225 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3226 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3227 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3228 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3229 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3231 let AddedComplexity = 400 in { // Prefer non-temporal versions
3232 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3233 "movntps\t{$src, $dst|$dst, $src}",
3234 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3235 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3236 "movntpd\t{$src, $dst|$dst, $src}",
3237 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3239 let ExeDomain = SSEPackedInt in
3240 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3241 "movntdq\t{$src, $dst|$dst, $src}",
3242 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)]>;
3244 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3245 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3247 // There is no AVX form for instructions below this point
3248 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3249 "movnti{l}\t{$src, $dst|$dst, $src}",
3250 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3251 TB, Requires<[HasSSE2]>;
3252 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3253 "movnti{q}\t{$src, $dst|$dst, $src}",
3254 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3255 TB, Requires<[HasSSE2]>;
3258 //===----------------------------------------------------------------------===//
3259 // SSE 1 & 2 - Prefetch and memory fence
3260 //===----------------------------------------------------------------------===//
3262 // Prefetch intrinsic.
3263 let Predicates = [HasSSE1] in {
3264 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3265 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3266 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3267 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3268 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3269 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3270 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3271 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3275 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3276 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3277 TB, Requires<[HasSSE2]>;
3279 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3280 // was introduced with SSE2, it's backward compatible.
3281 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3283 // Load, store, and memory fence
3284 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3285 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3286 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3287 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3288 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3289 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3291 def : Pat<(X86SFence), (SFENCE)>;
3292 def : Pat<(X86LFence), (LFENCE)>;
3293 def : Pat<(X86MFence), (MFENCE)>;
3295 //===----------------------------------------------------------------------===//
3296 // SSE 1 & 2 - Load/Store XCSR register
3297 //===----------------------------------------------------------------------===//
3299 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3300 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3301 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3302 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3304 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3305 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3306 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3307 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3309 //===---------------------------------------------------------------------===//
3310 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3311 //===---------------------------------------------------------------------===//
3313 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3315 let neverHasSideEffects = 1 in {
3316 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3317 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3318 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3319 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3321 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3322 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3323 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3324 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3327 let isCodeGenOnly = 1 in {
3328 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3329 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3330 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3331 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3332 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3333 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3334 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3335 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3338 let canFoldAsLoad = 1, mayLoad = 1 in {
3339 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3340 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3341 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3342 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3343 let Predicates = [HasAVX] in {
3344 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3345 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3346 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3347 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3351 let mayStore = 1 in {
3352 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3353 (ins i128mem:$dst, VR128:$src),
3354 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3355 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3356 (ins i256mem:$dst, VR256:$src),
3357 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3358 let Predicates = [HasAVX] in {
3359 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3360 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3361 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3362 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3366 let neverHasSideEffects = 1 in
3367 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3368 "movdqa\t{$src, $dst|$dst, $src}", []>;
3370 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3371 "movdqu\t{$src, $dst|$dst, $src}",
3372 []>, XS, Requires<[HasSSE2]>;
3375 let isCodeGenOnly = 1 in {
3376 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3377 "movdqa\t{$src, $dst|$dst, $src}", []>;
3379 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3380 "movdqu\t{$src, $dst|$dst, $src}",
3381 []>, XS, Requires<[HasSSE2]>;
3384 let canFoldAsLoad = 1, mayLoad = 1 in {
3385 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3386 "movdqa\t{$src, $dst|$dst, $src}",
3387 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3388 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3389 "movdqu\t{$src, $dst|$dst, $src}",
3390 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3391 XS, Requires<[HasSSE2]>;
3394 let mayStore = 1 in {
3395 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3396 "movdqa\t{$src, $dst|$dst, $src}",
3397 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3398 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3399 "movdqu\t{$src, $dst|$dst, $src}",
3400 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3401 XS, Requires<[HasSSE2]>;
3404 // Intrinsic forms of MOVDQU load and store
3405 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3406 "vmovdqu\t{$src, $dst|$dst, $src}",
3407 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3408 XS, VEX, Requires<[HasAVX]>;
3410 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3411 "movdqu\t{$src, $dst|$dst, $src}",
3412 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3413 XS, Requires<[HasSSE2]>;
3415 } // ExeDomain = SSEPackedInt
3417 let Predicates = [HasAVX] in {
3418 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3419 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3422 //===---------------------------------------------------------------------===//
3423 // SSE2 - Packed Integer Arithmetic Instructions
3424 //===---------------------------------------------------------------------===//
3426 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3428 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3429 RegisterClass RC, PatFrag memop_frag,
3430 X86MemOperand x86memop, bit IsCommutable = 0,
3432 let isCommutable = IsCommutable in
3433 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3434 (ins RC:$src1, RC:$src2),
3436 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3437 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3438 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3439 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3440 (ins RC:$src1, x86memop:$src2),
3442 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3443 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3444 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3447 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3448 string OpcodeStr, SDNode OpNode,
3449 SDNode OpNode2, RegisterClass RC,
3450 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3452 // src2 is always 128-bit
3453 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3454 (ins RC:$src1, VR128:$src2),
3456 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3457 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3458 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))]>;
3459 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3460 (ins RC:$src1, i128mem:$src2),
3462 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3463 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3464 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3465 (bc_frag (memopv2i64 addr:$src2)))))]>;
3466 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3467 (ins RC:$src1, i32i8imm:$src2),
3469 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3470 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3471 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))]>;
3474 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3475 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3476 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3477 PatFrag memop_frag, X86MemOperand x86memop,
3478 bit IsCommutable = 0, bit Is2Addr = 1> {
3479 let isCommutable = IsCommutable in
3480 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3481 (ins RC:$src1, RC:$src2),
3483 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3484 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3485 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3486 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3487 (ins RC:$src1, x86memop:$src2),
3489 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3490 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3491 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3492 (bitconvert (memop_frag addr:$src2)))))]>;
3494 } // ExeDomain = SSEPackedInt
3496 // 128-bit Integer Arithmetic
3498 let Predicates = [HasAVX] in {
3499 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3500 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3501 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3502 i128mem, 1, 0>, VEX_4V;
3503 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3504 i128mem, 1, 0>, VEX_4V;
3505 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3506 i128mem, 1, 0>, VEX_4V;
3507 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3508 i128mem, 1, 0>, VEX_4V;
3509 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3510 i128mem, 0, 0>, VEX_4V;
3511 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3512 i128mem, 0, 0>, VEX_4V;
3513 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3514 i128mem, 0, 0>, VEX_4V;
3515 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3516 i128mem, 0, 0>, VEX_4V;
3517 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3518 memopv2i64, i128mem, 1, 0>, VEX_4V;
3521 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3522 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3523 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3524 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3525 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3526 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3527 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3528 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3529 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3530 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3531 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3532 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3533 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3534 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3535 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3536 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3537 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3538 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3539 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3540 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3541 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3542 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3543 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3544 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3545 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3546 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3547 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3548 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3549 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3550 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3551 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3552 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3553 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3554 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3555 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3556 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3559 let Predicates = [HasAVX2] in {
3560 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3561 i256mem, 1, 0>, VEX_4V;
3562 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3563 i256mem, 1, 0>, VEX_4V;
3564 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3565 i256mem, 1, 0>, VEX_4V;
3566 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3567 i256mem, 1, 0>, VEX_4V;
3568 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3569 i256mem, 1, 0>, VEX_4V;
3570 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3571 i256mem, 0, 0>, VEX_4V;
3572 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3573 i256mem, 0, 0>, VEX_4V;
3574 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3575 i256mem, 0, 0>, VEX_4V;
3576 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3577 i256mem, 0, 0>, VEX_4V;
3578 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3579 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3582 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3583 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3584 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3585 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3586 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3587 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3588 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3589 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3590 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3591 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3592 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3593 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3594 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3595 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3596 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3597 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3598 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3599 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3600 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3601 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3602 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3603 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3604 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3605 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3606 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3607 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3608 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3609 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3610 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3611 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3612 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3613 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3614 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3615 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3616 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3617 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3620 let Constraints = "$src1 = $dst" in {
3621 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3623 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3625 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3627 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3629 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3631 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3633 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3635 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3637 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3639 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3640 memopv2i64, i128mem, 1>;
3643 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3644 VR128, memopv2i64, i128mem>;
3645 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3646 VR128, memopv2i64, i128mem>;
3647 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3648 VR128, memopv2i64, i128mem>;
3649 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3650 VR128, memopv2i64, i128mem>;
3651 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3652 VR128, memopv2i64, i128mem, 1>;
3653 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3654 VR128, memopv2i64, i128mem, 1>;
3655 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3656 VR128, memopv2i64, i128mem, 1>;
3657 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3658 VR128, memopv2i64, i128mem, 1>;
3659 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3660 VR128, memopv2i64, i128mem, 1>;
3661 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3662 VR128, memopv2i64, i128mem, 1>;
3663 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3664 VR128, memopv2i64, i128mem, 1>;
3665 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3666 VR128, memopv2i64, i128mem, 1>;
3667 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3668 VR128, memopv2i64, i128mem, 1>;
3669 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3670 VR128, memopv2i64, i128mem, 1>;
3671 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3672 VR128, memopv2i64, i128mem, 1>;
3673 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3674 VR128, memopv2i64, i128mem, 1>;
3675 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3676 VR128, memopv2i64, i128mem, 1>;
3677 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3678 VR128, memopv2i64, i128mem, 1>;
3680 } // Constraints = "$src1 = $dst"
3682 //===---------------------------------------------------------------------===//
3683 // SSE2 - Packed Integer Logical Instructions
3684 //===---------------------------------------------------------------------===//
3686 let Predicates = [HasAVX] in {
3687 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3688 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3689 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3690 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3691 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3692 VR128, v2i64, v2i64, bc_v2i64, 0>, VEX_4V;
3694 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3695 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3696 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3697 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3698 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3699 VR128, v2i64, v2i64, bc_v2i64, 0>, VEX_4V;
3701 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3702 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3703 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3704 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3706 let ExeDomain = SSEPackedInt in {
3707 // 128-bit logical shifts.
3708 def VPSLLDQri : PDIi8<0x73, MRM7r,
3709 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3710 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3712 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3714 def VPSRLDQri : PDIi8<0x73, MRM3r,
3715 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3716 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3718 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3720 // PSRADQri doesn't exist in SSE[1-3].
3722 } // Predicates = [HasAVX]
3724 let Predicates = [HasAVX2] in {
3725 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3726 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3727 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3728 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3729 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3730 VR256, v4i64, v2i64, bc_v2i64, 0>, VEX_4V;
3732 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3733 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3734 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3735 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3736 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3737 VR256, v4i64, v2i64, bc_v2i64, 0>, VEX_4V;
3739 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3740 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3741 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3742 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3744 let ExeDomain = SSEPackedInt in {
3745 // 256-bit logical shifts.
3746 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3747 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3748 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3750 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3752 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3753 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3754 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3756 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3758 // PSRADQYri doesn't exist in SSE[1-3].
3760 } // Predicates = [HasAVX2]
3762 let Constraints = "$src1 = $dst" in {
3763 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3764 VR128, v8i16, v8i16, bc_v8i16>;
3765 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3766 VR128, v4i32, v4i32, bc_v4i32>;
3767 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3768 VR128, v2i64, v2i64, bc_v2i64>;
3770 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3771 VR128, v8i16, v8i16, bc_v8i16>;
3772 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3773 VR128, v4i32, v4i32, bc_v4i32>;
3774 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3775 VR128, v2i64, v2i64, bc_v2i64>;
3777 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3778 VR128, v8i16, v8i16, bc_v8i16>;
3779 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3780 VR128, v4i32, v4i32, bc_v4i32>;
3782 let ExeDomain = SSEPackedInt in {
3783 // 128-bit logical shifts.
3784 def PSLLDQri : PDIi8<0x73, MRM7r,
3785 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3786 "pslldq\t{$src2, $dst|$dst, $src2}",
3788 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3789 def PSRLDQri : PDIi8<0x73, MRM3r,
3790 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3791 "psrldq\t{$src2, $dst|$dst, $src2}",
3793 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3794 // PSRADQri doesn't exist in SSE[1-3].
3796 } // Constraints = "$src1 = $dst"
3798 let Predicates = [HasAVX] in {
3799 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3800 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3801 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3802 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3803 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3804 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3806 // Shift up / down and insert zero's.
3807 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3808 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3809 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3810 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3813 let Predicates = [HasAVX2] in {
3814 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3815 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3816 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3817 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3820 let Predicates = [HasSSE2] in {
3821 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3822 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3823 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3824 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3825 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3826 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3828 // Shift up / down and insert zero's.
3829 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3830 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3831 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3832 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3835 //===---------------------------------------------------------------------===//
3836 // SSE2 - Packed Integer Comparison Instructions
3837 //===---------------------------------------------------------------------===//
3839 let Predicates = [HasAVX] in {
3840 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
3841 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3842 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
3843 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3844 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
3845 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3846 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
3847 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3848 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
3849 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3850 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
3851 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3854 let Predicates = [HasAVX2] in {
3855 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
3856 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3857 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
3858 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3859 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
3860 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3861 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
3862 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3863 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
3864 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3865 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
3866 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3869 let Constraints = "$src1 = $dst" in {
3870 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
3871 VR128, memopv2i64, i128mem, 1>;
3872 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
3873 VR128, memopv2i64, i128mem, 1>;
3874 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
3875 VR128, memopv2i64, i128mem, 1>;
3876 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
3877 VR128, memopv2i64, i128mem>;
3878 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
3879 VR128, memopv2i64, i128mem>;
3880 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
3881 VR128, memopv2i64, i128mem>;
3882 } // Constraints = "$src1 = $dst"
3884 //===---------------------------------------------------------------------===//
3885 // SSE2 - Packed Integer Pack Instructions
3886 //===---------------------------------------------------------------------===//
3888 let Predicates = [HasAVX] in {
3889 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3890 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3891 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3892 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3893 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3894 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3897 let Predicates = [HasAVX2] in {
3898 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
3899 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3900 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
3901 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3902 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
3903 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3906 let Constraints = "$src1 = $dst" in {
3907 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
3908 VR128, memopv2i64, i128mem>;
3909 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
3910 VR128, memopv2i64, i128mem>;
3911 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
3912 VR128, memopv2i64, i128mem>;
3913 } // Constraints = "$src1 = $dst"
3915 //===---------------------------------------------------------------------===//
3916 // SSE2 - Packed Integer Shuffle Instructions
3917 //===---------------------------------------------------------------------===//
3919 let ExeDomain = SSEPackedInt in {
3920 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
3922 def ri : Ii8<0x70, MRMSrcReg,
3923 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3924 !strconcat(OpcodeStr,
3925 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3926 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
3928 def mi : Ii8<0x70, MRMSrcMem,
3929 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3930 !strconcat(OpcodeStr,
3931 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3932 [(set VR128:$dst, (vt (pshuf_frag:$src2
3933 (bc_frag (memopv2i64 addr:$src1)),
3937 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
3938 def Yri : Ii8<0x70, MRMSrcReg,
3939 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
3940 !strconcat(OpcodeStr,
3941 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3942 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
3943 def Ymi : Ii8<0x70, MRMSrcMem,
3944 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
3945 !strconcat(OpcodeStr,
3946 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3948 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
3949 (i8 imm:$src2))))]>;
3951 } // ExeDomain = SSEPackedInt
3953 let Predicates = [HasAVX] in {
3954 let AddedComplexity = 5 in
3955 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
3958 // SSE2 with ImmT == Imm8 and XS prefix.
3959 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
3962 // SSE2 with ImmT == Imm8 and XD prefix.
3963 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
3966 let AddedComplexity = 5 in
3967 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3968 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3969 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
3970 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3971 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3973 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3975 (VPSHUFDmi addr:$src1, imm:$imm)>;
3976 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3977 (VPSHUFDmi addr:$src1, imm:$imm)>;
3978 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3979 (VPSHUFDri VR128:$src1, imm:$imm)>;
3980 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3981 (VPSHUFDri VR128:$src1, imm:$imm)>;
3982 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3983 (VPSHUFHWri VR128:$src, imm:$imm)>;
3984 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3986 (VPSHUFHWmi addr:$src, imm:$imm)>;
3987 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3988 (VPSHUFLWri VR128:$src, imm:$imm)>;
3989 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3991 (VPSHUFLWmi addr:$src, imm:$imm)>;
3994 let Predicates = [HasAVX2] in {
3995 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
3996 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
3997 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
4000 let Predicates = [HasSSE2] in {
4001 let AddedComplexity = 5 in
4002 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4004 // SSE2 with ImmT == Imm8 and XS prefix.
4005 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4007 // SSE2 with ImmT == Imm8 and XD prefix.
4008 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4010 let AddedComplexity = 5 in
4011 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4012 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4013 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4014 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4015 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4017 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4019 (PSHUFDmi addr:$src1, imm:$imm)>;
4020 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4021 (PSHUFDmi addr:$src1, imm:$imm)>;
4022 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4023 (PSHUFDri VR128:$src1, imm:$imm)>;
4024 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4025 (PSHUFDri VR128:$src1, imm:$imm)>;
4026 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4027 (PSHUFHWri VR128:$src, imm:$imm)>;
4028 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4030 (PSHUFHWmi addr:$src, imm:$imm)>;
4031 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4032 (PSHUFLWri VR128:$src, imm:$imm)>;
4033 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4035 (PSHUFLWmi addr:$src, imm:$imm)>;
4038 //===---------------------------------------------------------------------===//
4039 // SSE2 - Packed Integer Unpack Instructions
4040 //===---------------------------------------------------------------------===//
4042 let ExeDomain = SSEPackedInt in {
4043 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4044 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4045 def rr : PDI<opc, MRMSrcReg,
4046 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4048 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4049 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4050 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4051 def rm : PDI<opc, MRMSrcMem,
4052 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4054 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4055 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4056 [(set VR128:$dst, (OpNode VR128:$src1,
4057 (bc_frag (memopv2i64
4061 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4062 SDNode OpNode, PatFrag bc_frag> {
4063 def Yrr : PDI<opc, MRMSrcReg,
4064 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4065 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4066 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4067 def Yrm : PDI<opc, MRMSrcMem,
4068 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4069 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4070 [(set VR256:$dst, (OpNode VR256:$src1,
4071 (bc_frag (memopv4i64 addr:$src2))))]>;
4074 let Predicates = [HasAVX] in {
4075 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4076 bc_v16i8, 0>, VEX_4V;
4077 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4078 bc_v8i16, 0>, VEX_4V;
4079 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4080 bc_v4i32, 0>, VEX_4V;
4081 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4082 bc_v2i64, 0>, VEX_4V;
4084 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4085 bc_v16i8, 0>, VEX_4V;
4086 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4087 bc_v8i16, 0>, VEX_4V;
4088 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4089 bc_v4i32, 0>, VEX_4V;
4090 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4091 bc_v2i64, 0>, VEX_4V;
4094 let Predicates = [HasAVX2] in {
4095 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4097 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4099 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4101 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4104 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4106 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4108 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4110 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4114 let Constraints = "$src1 = $dst" in {
4115 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4117 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4119 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4121 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4124 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4126 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4128 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4130 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4133 } // ExeDomain = SSEPackedInt
4135 // Patterns for using AVX1 instructions with integer vectors
4136 // Here to give AVX2 priority
4137 let Predicates = [HasAVX] in {
4138 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4139 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4140 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4141 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4142 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4143 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4144 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4145 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4147 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4148 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4149 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4150 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4151 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4152 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4153 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4154 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4157 //===---------------------------------------------------------------------===//
4158 // SSE2 - Packed Integer Extract and Insert
4159 //===---------------------------------------------------------------------===//
4161 let ExeDomain = SSEPackedInt in {
4162 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4163 def rri : Ii8<0xC4, MRMSrcReg,
4164 (outs VR128:$dst), (ins VR128:$src1,
4165 GR32:$src2, i32i8imm:$src3),
4167 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4168 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4170 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4171 def rmi : Ii8<0xC4, MRMSrcMem,
4172 (outs VR128:$dst), (ins VR128:$src1,
4173 i16mem:$src2, i32i8imm:$src3),
4175 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4176 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4178 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4183 let Predicates = [HasAVX] in
4184 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4185 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4186 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4187 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4188 imm:$src2))]>, TB, OpSize, VEX;
4189 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4190 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4191 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4192 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4196 let Predicates = [HasAVX] in {
4197 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4198 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4199 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4200 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4201 []>, TB, OpSize, VEX_4V;
4204 let Constraints = "$src1 = $dst" in
4205 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4207 } // ExeDomain = SSEPackedInt
4209 //===---------------------------------------------------------------------===//
4210 // SSE2 - Packed Mask Creation
4211 //===---------------------------------------------------------------------===//
4213 let ExeDomain = SSEPackedInt in {
4215 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4216 "pmovmskb\t{$src, $dst|$dst, $src}",
4217 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4218 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4219 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4221 let Predicates = [HasAVX2] in {
4222 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4223 "pmovmskb\t{$src, $dst|$dst, $src}",
4224 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4225 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4226 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4229 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4230 "pmovmskb\t{$src, $dst|$dst, $src}",
4231 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4233 } // ExeDomain = SSEPackedInt
4235 //===---------------------------------------------------------------------===//
4236 // SSE2 - Conditional Store
4237 //===---------------------------------------------------------------------===//
4239 let ExeDomain = SSEPackedInt in {
4242 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4243 (ins VR128:$src, VR128:$mask),
4244 "maskmovdqu\t{$mask, $src|$src, $mask}",
4245 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4247 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4248 (ins VR128:$src, VR128:$mask),
4249 "maskmovdqu\t{$mask, $src|$src, $mask}",
4250 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4253 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4254 "maskmovdqu\t{$mask, $src|$src, $mask}",
4255 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4257 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4258 "maskmovdqu\t{$mask, $src|$src, $mask}",
4259 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4261 } // ExeDomain = SSEPackedInt
4263 //===---------------------------------------------------------------------===//
4264 // SSE2 - Move Doubleword
4265 //===---------------------------------------------------------------------===//
4267 //===---------------------------------------------------------------------===//
4268 // Move Int Doubleword to Packed Double Int
4270 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4271 "movd\t{$src, $dst|$dst, $src}",
4273 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4274 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4275 "movd\t{$src, $dst|$dst, $src}",
4277 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4279 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4280 "mov{d|q}\t{$src, $dst|$dst, $src}",
4282 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4283 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4284 "mov{d|q}\t{$src, $dst|$dst, $src}",
4285 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4287 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4288 "movd\t{$src, $dst|$dst, $src}",
4290 (v4i32 (scalar_to_vector GR32:$src)))]>;
4291 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4292 "movd\t{$src, $dst|$dst, $src}",
4294 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4295 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4296 "mov{d|q}\t{$src, $dst|$dst, $src}",
4298 (v2i64 (scalar_to_vector GR64:$src)))]>;
4299 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4300 "mov{d|q}\t{$src, $dst|$dst, $src}",
4301 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4303 //===---------------------------------------------------------------------===//
4304 // Move Int Doubleword to Single Scalar
4306 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4307 "movd\t{$src, $dst|$dst, $src}",
4308 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4310 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4311 "movd\t{$src, $dst|$dst, $src}",
4312 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4314 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4315 "movd\t{$src, $dst|$dst, $src}",
4316 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4318 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4319 "movd\t{$src, $dst|$dst, $src}",
4320 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4322 //===---------------------------------------------------------------------===//
4323 // Move Packed Doubleword Int to Packed Double Int
4325 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4326 "movd\t{$src, $dst|$dst, $src}",
4327 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4329 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4330 (ins i32mem:$dst, VR128:$src),
4331 "movd\t{$src, $dst|$dst, $src}",
4332 [(store (i32 (vector_extract (v4i32 VR128:$src),
4333 (iPTR 0))), addr:$dst)]>, VEX;
4334 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4335 "movd\t{$src, $dst|$dst, $src}",
4336 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4338 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4339 "movd\t{$src, $dst|$dst, $src}",
4340 [(store (i32 (vector_extract (v4i32 VR128:$src),
4341 (iPTR 0))), addr:$dst)]>;
4343 //===---------------------------------------------------------------------===//
4344 // Move Packed Doubleword Int first element to Doubleword Int
4346 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4347 "mov{d|q}\t{$src, $dst|$dst, $src}",
4348 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4350 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4352 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4353 "mov{d|q}\t{$src, $dst|$dst, $src}",
4354 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4357 //===---------------------------------------------------------------------===//
4358 // Bitcast FR64 <-> GR64
4360 let Predicates = [HasAVX] in
4361 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4362 "vmovq\t{$src, $dst|$dst, $src}",
4363 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4365 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4366 "mov{d|q}\t{$src, $dst|$dst, $src}",
4367 [(set GR64:$dst, (bitconvert FR64:$src))]>, VEX;
4368 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4369 "movq\t{$src, $dst|$dst, $src}",
4370 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>,
4373 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4374 "movq\t{$src, $dst|$dst, $src}",
4375 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4376 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4377 "mov{d|q}\t{$src, $dst|$dst, $src}",
4378 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4379 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4380 "movq\t{$src, $dst|$dst, $src}",
4381 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4383 //===---------------------------------------------------------------------===//
4384 // Move Scalar Single to Double Int
4386 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4387 "movd\t{$src, $dst|$dst, $src}",
4388 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4389 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4390 "movd\t{$src, $dst|$dst, $src}",
4391 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4392 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4393 "movd\t{$src, $dst|$dst, $src}",
4394 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4395 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4396 "movd\t{$src, $dst|$dst, $src}",
4397 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4399 //===---------------------------------------------------------------------===//
4400 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4402 let AddedComplexity = 15 in {
4403 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4404 "movd\t{$src, $dst|$dst, $src}",
4405 [(set VR128:$dst, (v4i32 (X86vzmovl
4406 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4408 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4409 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4410 [(set VR128:$dst, (v2i64 (X86vzmovl
4411 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4414 let AddedComplexity = 15 in {
4415 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4416 "movd\t{$src, $dst|$dst, $src}",
4417 [(set VR128:$dst, (v4i32 (X86vzmovl
4418 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4419 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4420 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4421 [(set VR128:$dst, (v2i64 (X86vzmovl
4422 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4425 let AddedComplexity = 20 in {
4426 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4427 "movd\t{$src, $dst|$dst, $src}",
4429 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4430 (loadi32 addr:$src))))))]>,
4432 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4433 "movd\t{$src, $dst|$dst, $src}",
4435 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4436 (loadi32 addr:$src))))))]>;
4439 let Predicates = [HasAVX] in {
4440 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4441 let AddedComplexity = 20 in {
4442 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4443 (VMOVZDI2PDIrm addr:$src)>;
4444 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4445 (VMOVZDI2PDIrm addr:$src)>;
4447 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4448 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4449 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4450 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4451 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4452 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4453 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4456 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4457 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4458 (MOVZDI2PDIrm addr:$src)>;
4459 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4460 (MOVZDI2PDIrm addr:$src)>;
4463 // These are the correct encodings of the instructions so that we know how to
4464 // read correct assembly, even though we continue to emit the wrong ones for
4465 // compatibility with Darwin's buggy assembler.
4466 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4467 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4468 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4469 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4470 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4471 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4472 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4473 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4474 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4475 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4476 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4477 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4479 //===---------------------------------------------------------------------===//
4480 // SSE2 - Move Quadword
4481 //===---------------------------------------------------------------------===//
4483 //===---------------------------------------------------------------------===//
4484 // Move Quadword Int to Packed Quadword Int
4486 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4487 "vmovq\t{$src, $dst|$dst, $src}",
4489 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4490 VEX, Requires<[HasAVX]>;
4491 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4492 "movq\t{$src, $dst|$dst, $src}",
4494 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4495 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4497 //===---------------------------------------------------------------------===//
4498 // Move Packed Quadword Int to Quadword Int
4500 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4501 "movq\t{$src, $dst|$dst, $src}",
4502 [(store (i64 (vector_extract (v2i64 VR128:$src),
4503 (iPTR 0))), addr:$dst)]>, VEX;
4504 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4505 "movq\t{$src, $dst|$dst, $src}",
4506 [(store (i64 (vector_extract (v2i64 VR128:$src),
4507 (iPTR 0))), addr:$dst)]>;
4509 //===---------------------------------------------------------------------===//
4510 // Store / copy lower 64-bits of a XMM register.
4512 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4513 "movq\t{$src, $dst|$dst, $src}",
4514 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4515 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4516 "movq\t{$src, $dst|$dst, $src}",
4517 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4519 let AddedComplexity = 20 in
4520 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4521 "vmovq\t{$src, $dst|$dst, $src}",
4523 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4524 (loadi64 addr:$src))))))]>,
4525 XS, VEX, Requires<[HasAVX]>;
4527 let AddedComplexity = 20 in
4528 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4529 "movq\t{$src, $dst|$dst, $src}",
4531 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4532 (loadi64 addr:$src))))))]>,
4533 XS, Requires<[HasSSE2]>;
4535 let Predicates = [HasAVX], AddedComplexity = 20 in {
4536 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4537 (VMOVZQI2PQIrm addr:$src)>;
4538 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4539 (VMOVZQI2PQIrm addr:$src)>;
4540 def : Pat<(v2i64 (X86vzload addr:$src)),
4541 (VMOVZQI2PQIrm addr:$src)>;
4544 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4545 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4546 (MOVZQI2PQIrm addr:$src)>;
4547 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4548 (MOVZQI2PQIrm addr:$src)>;
4549 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4552 let Predicates = [HasAVX] in {
4553 def : Pat<(v4i64 (X86vzload addr:$src)),
4554 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4557 //===---------------------------------------------------------------------===//
4558 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4559 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4561 let AddedComplexity = 15 in
4562 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4563 "vmovq\t{$src, $dst|$dst, $src}",
4564 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4565 XS, VEX, Requires<[HasAVX]>;
4566 let AddedComplexity = 15 in
4567 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4568 "movq\t{$src, $dst|$dst, $src}",
4569 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4570 XS, Requires<[HasSSE2]>;
4572 let AddedComplexity = 20 in
4573 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4574 "vmovq\t{$src, $dst|$dst, $src}",
4575 [(set VR128:$dst, (v2i64 (X86vzmovl
4576 (loadv2i64 addr:$src))))]>,
4577 XS, VEX, Requires<[HasAVX]>;
4578 let AddedComplexity = 20 in {
4579 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4580 "movq\t{$src, $dst|$dst, $src}",
4581 [(set VR128:$dst, (v2i64 (X86vzmovl
4582 (loadv2i64 addr:$src))))]>,
4583 XS, Requires<[HasSSE2]>;
4586 let AddedComplexity = 20 in {
4587 let Predicates = [HasAVX] in {
4588 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4589 (VMOVZPQILo2PQIrm addr:$src)>;
4590 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4591 (VMOVZPQILo2PQIrr VR128:$src)>;
4593 let Predicates = [HasSSE2] in {
4594 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4595 (MOVZPQILo2PQIrm addr:$src)>;
4596 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4597 (MOVZPQILo2PQIrr VR128:$src)>;
4601 // Instructions to match in the assembler
4602 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4603 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4604 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4605 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4606 // Recognize "movd" with GR64 destination, but encode as a "movq"
4607 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4608 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4610 // Instructions for the disassembler
4611 // xr = XMM register
4614 let Predicates = [HasAVX] in
4615 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4616 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4617 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4618 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4620 //===---------------------------------------------------------------------===//
4621 // SSE3 - Conversion Instructions
4622 //===---------------------------------------------------------------------===//
4624 // Convert Packed Double FP to Packed DW Integers
4625 let Predicates = [HasAVX] in {
4626 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4627 // register, but the same isn't true when using memory operands instead.
4628 // Provide other assembly rr and rm forms to address this explicitly.
4629 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4630 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4631 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4632 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4635 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4636 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4637 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4638 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4641 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4642 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4643 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4644 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4647 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4648 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4649 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4650 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4652 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4653 (VCVTTPD2DQYrr VR256:$src)>;
4654 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4655 (VCVTTPD2DQYrm addr:$src)>;
4657 // Convert Packed DW Integers to Packed Double FP
4658 let Predicates = [HasAVX] in {
4659 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4660 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4661 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4662 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4663 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4664 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4665 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4666 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4669 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4670 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4671 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4672 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4674 // AVX 256-bit register conversion intrinsics
4675 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4676 (VCVTDQ2PDYrr VR128:$src)>;
4677 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4678 (VCVTDQ2PDYrm addr:$src)>;
4680 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4681 (VCVTPD2DQYrr VR256:$src)>;
4682 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4683 (VCVTPD2DQYrm addr:$src)>;
4685 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4686 (VCVTDQ2PDYrr VR128:$src)>;
4687 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4688 (VCVTDQ2PDYrm addr:$src)>;
4690 //===---------------------------------------------------------------------===//
4691 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4692 //===---------------------------------------------------------------------===//
4693 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4694 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4695 X86MemOperand x86memop> {
4696 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4697 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4698 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4699 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4700 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4701 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4704 let Predicates = [HasAVX] in {
4705 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4706 v4f32, VR128, memopv4f32, f128mem>, VEX;
4707 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4708 v4f32, VR128, memopv4f32, f128mem>, VEX;
4709 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4710 v8f32, VR256, memopv8f32, f256mem>, VEX;
4711 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4712 v8f32, VR256, memopv8f32, f256mem>, VEX;
4714 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4715 memopv4f32, f128mem>;
4716 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4717 memopv4f32, f128mem>;
4719 let Predicates = [HasAVX] in {
4720 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4721 (VMOVSHDUPrr VR128:$src)>;
4722 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4723 (VMOVSHDUPrm addr:$src)>;
4724 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4725 (VMOVSLDUPrr VR128:$src)>;
4726 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4727 (VMOVSLDUPrm addr:$src)>;
4728 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4729 (VMOVSHDUPYrr VR256:$src)>;
4730 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4731 (VMOVSHDUPYrm addr:$src)>;
4732 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4733 (VMOVSLDUPYrr VR256:$src)>;
4734 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4735 (VMOVSLDUPYrm addr:$src)>;
4738 let Predicates = [HasSSE3] in {
4739 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4740 (MOVSHDUPrr VR128:$src)>;
4741 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4742 (MOVSHDUPrm addr:$src)>;
4743 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4744 (MOVSLDUPrr VR128:$src)>;
4745 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4746 (MOVSLDUPrm addr:$src)>;
4749 //===---------------------------------------------------------------------===//
4750 // SSE3 - Replicate Double FP - MOVDDUP
4751 //===---------------------------------------------------------------------===//
4753 multiclass sse3_replicate_dfp<string OpcodeStr> {
4754 let neverHasSideEffects = 1 in
4755 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4756 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4758 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4759 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4762 (scalar_to_vector (loadf64 addr:$src)))))]>;
4765 // FIXME: Merge with above classe when there're patterns for the ymm version
4766 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4767 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4768 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4769 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4770 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4771 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4774 (scalar_to_vector (loadf64 addr:$src)))))]>;
4777 let Predicates = [HasAVX] in {
4778 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4779 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4782 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4784 let Predicates = [HasAVX] in {
4785 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4786 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4787 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4788 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4789 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4790 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4791 def : Pat<(X86Movddup (bc_v2f64
4792 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4793 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4796 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4797 (VMOVDDUPYrm addr:$src)>;
4798 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4799 (VMOVDDUPYrm addr:$src)>;
4800 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4801 (VMOVDDUPYrm addr:$src)>;
4802 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4803 (VMOVDDUPYrr VR256:$src)>;
4806 let Predicates = [HasSSE3] in {
4807 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4808 (MOVDDUPrm addr:$src)>;
4809 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4810 (MOVDDUPrm addr:$src)>;
4811 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4812 (MOVDDUPrm addr:$src)>;
4813 def : Pat<(X86Movddup (bc_v2f64
4814 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4815 (MOVDDUPrm addr:$src)>;
4818 //===---------------------------------------------------------------------===//
4819 // SSE3 - Move Unaligned Integer
4820 //===---------------------------------------------------------------------===//
4822 let Predicates = [HasAVX] in {
4823 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4824 "vlddqu\t{$src, $dst|$dst, $src}",
4825 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4826 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4827 "vlddqu\t{$src, $dst|$dst, $src}",
4828 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4830 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4831 "lddqu\t{$src, $dst|$dst, $src}",
4832 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
4834 //===---------------------------------------------------------------------===//
4835 // SSE3 - Arithmetic
4836 //===---------------------------------------------------------------------===//
4838 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4839 X86MemOperand x86memop, bit Is2Addr = 1> {
4840 def rr : I<0xD0, MRMSrcReg,
4841 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4843 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4844 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4845 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
4846 def rm : I<0xD0, MRMSrcMem,
4847 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4849 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4850 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4851 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
4854 let Predicates = [HasAVX] in {
4855 let ExeDomain = SSEPackedSingle in {
4856 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4857 f128mem, 0>, TB, XD, VEX_4V;
4858 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4859 f256mem, 0>, TB, XD, VEX_4V;
4861 let ExeDomain = SSEPackedDouble in {
4862 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4863 f128mem, 0>, TB, OpSize, VEX_4V;
4864 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4865 f256mem, 0>, TB, OpSize, VEX_4V;
4868 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
4869 let ExeDomain = SSEPackedSingle in
4870 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4872 let ExeDomain = SSEPackedDouble in
4873 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4874 f128mem>, TB, OpSize;
4877 //===---------------------------------------------------------------------===//
4878 // SSE3 Instructions
4879 //===---------------------------------------------------------------------===//
4882 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4883 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4884 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4886 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4887 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4888 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
4890 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4892 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4893 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4894 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
4896 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4897 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4898 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4900 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4901 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4902 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
4904 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4906 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4907 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4908 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
4911 let Predicates = [HasAVX] in {
4912 let ExeDomain = SSEPackedSingle in {
4913 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4914 X86fhadd, 0>, VEX_4V;
4915 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4916 X86fhsub, 0>, VEX_4V;
4917 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4918 X86fhadd, 0>, VEX_4V;
4919 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4920 X86fhsub, 0>, VEX_4V;
4922 let ExeDomain = SSEPackedDouble in {
4923 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4924 X86fhadd, 0>, VEX_4V;
4925 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4926 X86fhsub, 0>, VEX_4V;
4927 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4928 X86fhadd, 0>, VEX_4V;
4929 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4930 X86fhsub, 0>, VEX_4V;
4934 let Constraints = "$src1 = $dst" in {
4935 let ExeDomain = SSEPackedSingle in {
4936 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
4937 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
4939 let ExeDomain = SSEPackedDouble in {
4940 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
4941 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
4945 //===---------------------------------------------------------------------===//
4946 // SSSE3 - Packed Absolute Instructions
4947 //===---------------------------------------------------------------------===//
4950 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4951 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4952 Intrinsic IntId128> {
4953 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4955 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4956 [(set VR128:$dst, (IntId128 VR128:$src))]>,
4959 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4961 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4964 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
4967 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4968 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
4969 Intrinsic IntId256> {
4970 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4972 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4973 [(set VR256:$dst, (IntId256 VR256:$src))]>,
4976 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4978 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4981 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
4984 let Predicates = [HasAVX] in {
4985 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
4986 int_x86_ssse3_pabs_b_128>, VEX;
4987 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
4988 int_x86_ssse3_pabs_w_128>, VEX;
4989 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
4990 int_x86_ssse3_pabs_d_128>, VEX;
4993 let Predicates = [HasAVX2] in {
4994 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
4995 int_x86_avx2_pabs_b>, VEX;
4996 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
4997 int_x86_avx2_pabs_w>, VEX;
4998 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
4999 int_x86_avx2_pabs_d>, VEX;
5002 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5003 int_x86_ssse3_pabs_b_128>;
5004 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5005 int_x86_ssse3_pabs_w_128>;
5006 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5007 int_x86_ssse3_pabs_d_128>;
5009 //===---------------------------------------------------------------------===//
5010 // SSSE3 - Packed Binary Operator Instructions
5011 //===---------------------------------------------------------------------===//
5013 /// SS3I_binop_rm - Simple SSSE3 bin op
5014 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5015 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5016 X86MemOperand x86memop, bit Is2Addr = 1> {
5017 let isCommutable = 1 in
5018 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5019 (ins RC:$src1, RC:$src2),
5021 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5022 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5023 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
5025 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5026 (ins RC:$src1, x86memop:$src2),
5028 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5029 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5031 (OpVT (OpNode RC:$src1,
5032 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
5035 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5036 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5037 Intrinsic IntId128, bit Is2Addr = 1> {
5038 let isCommutable = 1 in
5039 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5040 (ins VR128:$src1, VR128:$src2),
5042 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5043 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5044 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5046 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5047 (ins VR128:$src1, i128mem:$src2),
5049 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5050 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5052 (IntId128 VR128:$src1,
5053 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5056 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5057 Intrinsic IntId256> {
5058 let isCommutable = 1 in
5059 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5060 (ins VR256:$src1, VR256:$src2),
5061 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5062 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5064 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5065 (ins VR256:$src1, i256mem:$src2),
5066 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5068 (IntId256 VR256:$src1,
5069 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5072 let ImmT = NoImm, Predicates = [HasAVX] in {
5073 let isCommutable = 0 in {
5074 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5075 memopv2i64, i128mem, 0>, VEX_4V;
5076 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5077 memopv2i64, i128mem, 0>, VEX_4V;
5078 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5079 memopv2i64, i128mem, 0>, VEX_4V;
5080 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5081 memopv2i64, i128mem, 0>, VEX_4V;
5082 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5083 memopv2i64, i128mem, 0>, VEX_4V;
5084 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5085 memopv2i64, i128mem, 0>, VEX_4V;
5086 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5087 memopv2i64, i128mem, 0>, VEX_4V;
5088 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5089 memopv2i64, i128mem, 0>, VEX_4V;
5090 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5091 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5092 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5093 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5094 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5095 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5097 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5098 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5101 let ImmT = NoImm, Predicates = [HasAVX2] in {
5102 let isCommutable = 0 in {
5103 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5104 memopv4i64, i256mem, 0>, VEX_4V;
5105 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5106 memopv4i64, i256mem, 0>, VEX_4V;
5107 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5108 memopv4i64, i256mem, 0>, VEX_4V;
5109 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5110 memopv4i64, i256mem, 0>, VEX_4V;
5111 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5112 memopv4i64, i256mem, 0>, VEX_4V;
5113 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5114 memopv4i64, i256mem, 0>, VEX_4V;
5115 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5116 memopv4i64, i256mem, 0>, VEX_4V;
5117 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5118 memopv4i64, i256mem, 0>, VEX_4V;
5119 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5120 int_x86_avx2_phadd_sw>, VEX_4V;
5121 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5122 int_x86_avx2_phsub_sw>, VEX_4V;
5123 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5124 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5126 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5127 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5130 // None of these have i8 immediate fields.
5131 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5132 let isCommutable = 0 in {
5133 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5134 memopv2i64, i128mem>;
5135 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5136 memopv2i64, i128mem>;
5137 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5138 memopv2i64, i128mem>;
5139 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5140 memopv2i64, i128mem>;
5141 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5142 memopv2i64, i128mem>;
5143 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5144 memopv2i64, i128mem>;
5145 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5146 memopv2i64, i128mem>;
5147 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5148 memopv2i64, i128mem>;
5149 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5150 int_x86_ssse3_phadd_sw_128>;
5151 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5152 int_x86_ssse3_phsub_sw_128>;
5153 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5154 int_x86_ssse3_pmadd_ub_sw_128>;
5156 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5157 int_x86_ssse3_pmul_hr_sw_128>;
5160 //===---------------------------------------------------------------------===//
5161 // SSSE3 - Packed Align Instruction Patterns
5162 //===---------------------------------------------------------------------===//
5164 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5165 let neverHasSideEffects = 1 in {
5166 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5167 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5169 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5171 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5174 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5175 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5177 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5179 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5184 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5185 let neverHasSideEffects = 1 in {
5186 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5187 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5189 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5192 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5193 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5195 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5200 let Predicates = [HasAVX] in
5201 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5202 let Predicates = [HasAVX2] in
5203 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5204 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5205 defm PALIGN : ssse3_palign<"palignr">;
5207 let Predicates = [HasAVX2] in {
5208 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5209 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5210 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5211 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5212 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5213 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5214 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5215 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5218 let Predicates = [HasAVX] in {
5219 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5220 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5221 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5222 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5223 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5224 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5225 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5226 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5229 let Predicates = [HasSSSE3] in {
5230 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5231 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5232 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5233 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5234 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5235 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5236 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5237 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5240 //===---------------------------------------------------------------------===//
5241 // SSSE3 - Thread synchronization
5242 //===---------------------------------------------------------------------===//
5244 let usesCustomInserter = 1 in {
5245 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5246 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5247 Requires<[HasSSE3]>;
5248 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5249 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5250 Requires<[HasSSE3]>;
5253 let Uses = [EAX, ECX, EDX] in
5254 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5255 Requires<[HasSSE3]>;
5256 let Uses = [ECX, EAX] in
5257 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5258 Requires<[HasSSE3]>;
5260 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5261 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5263 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5264 Requires<[In32BitMode]>;
5265 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5266 Requires<[In64BitMode]>;
5268 //===----------------------------------------------------------------------===//
5269 // SSE4.1 - Packed Move with Sign/Zero Extend
5270 //===----------------------------------------------------------------------===//
5272 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5273 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5274 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5275 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5277 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5278 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5280 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5284 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5286 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5287 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5288 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5290 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5291 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5292 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5295 let Predicates = [HasAVX] in {
5296 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5298 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5300 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5302 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5304 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5306 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5310 let Predicates = [HasAVX2] in {
5311 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5312 int_x86_avx2_pmovsxbw>, VEX;
5313 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5314 int_x86_avx2_pmovsxwd>, VEX;
5315 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5316 int_x86_avx2_pmovsxdq>, VEX;
5317 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5318 int_x86_avx2_pmovzxbw>, VEX;
5319 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5320 int_x86_avx2_pmovzxwd>, VEX;
5321 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5322 int_x86_avx2_pmovzxdq>, VEX;
5325 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5326 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5327 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5328 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5329 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5330 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5332 let Predicates = [HasAVX] in {
5333 // Common patterns involving scalar load.
5334 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5335 (VPMOVSXBWrm addr:$src)>;
5336 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5337 (VPMOVSXBWrm addr:$src)>;
5339 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5340 (VPMOVSXWDrm addr:$src)>;
5341 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5342 (VPMOVSXWDrm addr:$src)>;
5344 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5345 (VPMOVSXDQrm addr:$src)>;
5346 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5347 (VPMOVSXDQrm addr:$src)>;
5349 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5350 (VPMOVZXBWrm addr:$src)>;
5351 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5352 (VPMOVZXBWrm addr:$src)>;
5354 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5355 (VPMOVZXWDrm addr:$src)>;
5356 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5357 (VPMOVZXWDrm addr:$src)>;
5359 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5360 (VPMOVZXDQrm addr:$src)>;
5361 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5362 (VPMOVZXDQrm addr:$src)>;
5365 let Predicates = [HasSSE41] in {
5366 // Common patterns involving scalar load.
5367 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5368 (PMOVSXBWrm addr:$src)>;
5369 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5370 (PMOVSXBWrm addr:$src)>;
5372 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5373 (PMOVSXWDrm addr:$src)>;
5374 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5375 (PMOVSXWDrm addr:$src)>;
5377 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5378 (PMOVSXDQrm addr:$src)>;
5379 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5380 (PMOVSXDQrm addr:$src)>;
5382 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5383 (PMOVZXBWrm addr:$src)>;
5384 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5385 (PMOVZXBWrm addr:$src)>;
5387 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5388 (PMOVZXWDrm addr:$src)>;
5389 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5390 (PMOVZXWDrm addr:$src)>;
5392 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5393 (PMOVZXDQrm addr:$src)>;
5394 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5395 (PMOVZXDQrm addr:$src)>;
5398 let Predicates = [HasAVX] in {
5399 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5400 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5403 let Predicates = [HasSSE41] in {
5404 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5405 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5409 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5410 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5411 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5412 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5414 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5415 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5417 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5421 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5423 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5424 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5425 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5427 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5428 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5430 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5434 let Predicates = [HasAVX] in {
5435 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5437 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5439 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5441 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5445 let Predicates = [HasAVX2] in {
5446 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5447 int_x86_avx2_pmovsxbd>, VEX;
5448 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5449 int_x86_avx2_pmovsxwq>, VEX;
5450 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5451 int_x86_avx2_pmovzxbd>, VEX;
5452 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5453 int_x86_avx2_pmovzxwq>, VEX;
5456 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5457 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5458 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5459 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5461 let Predicates = [HasAVX] in {
5462 // Common patterns involving scalar load
5463 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5464 (VPMOVSXBDrm addr:$src)>;
5465 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5466 (VPMOVSXWQrm addr:$src)>;
5468 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5469 (VPMOVZXBDrm addr:$src)>;
5470 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5471 (VPMOVZXWQrm addr:$src)>;
5474 let Predicates = [HasSSE41] in {
5475 // Common patterns involving scalar load
5476 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5477 (PMOVSXBDrm addr:$src)>;
5478 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5479 (PMOVSXWQrm addr:$src)>;
5481 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5482 (PMOVZXBDrm addr:$src)>;
5483 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5484 (PMOVZXWQrm addr:$src)>;
5487 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5488 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5489 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5490 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5492 // Expecting a i16 load any extended to i32 value.
5493 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5494 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5495 [(set VR128:$dst, (IntId (bitconvert
5496 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5500 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5502 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5503 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5504 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5506 // Expecting a i16 load any extended to i32 value.
5507 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5508 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5509 [(set VR256:$dst, (IntId (bitconvert
5510 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5514 let Predicates = [HasAVX] in {
5515 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5517 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5520 let Predicates = [HasAVX2] in {
5521 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5522 int_x86_avx2_pmovsxbq>, VEX;
5523 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5524 int_x86_avx2_pmovzxbq>, VEX;
5526 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5527 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5529 let Predicates = [HasAVX] in {
5530 // Common patterns involving scalar load
5531 def : Pat<(int_x86_sse41_pmovsxbq
5532 (bitconvert (v4i32 (X86vzmovl
5533 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5534 (VPMOVSXBQrm addr:$src)>;
5536 def : Pat<(int_x86_sse41_pmovzxbq
5537 (bitconvert (v4i32 (X86vzmovl
5538 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5539 (VPMOVZXBQrm addr:$src)>;
5542 let Predicates = [HasSSE41] in {
5543 // Common patterns involving scalar load
5544 def : Pat<(int_x86_sse41_pmovsxbq
5545 (bitconvert (v4i32 (X86vzmovl
5546 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5547 (PMOVSXBQrm addr:$src)>;
5549 def : Pat<(int_x86_sse41_pmovzxbq
5550 (bitconvert (v4i32 (X86vzmovl
5551 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5552 (PMOVZXBQrm addr:$src)>;
5555 //===----------------------------------------------------------------------===//
5556 // SSE4.1 - Extract Instructions
5557 //===----------------------------------------------------------------------===//
5559 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5560 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5561 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5562 (ins VR128:$src1, i32i8imm:$src2),
5563 !strconcat(OpcodeStr,
5564 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5565 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5567 let neverHasSideEffects = 1, mayStore = 1 in
5568 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5569 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5570 !strconcat(OpcodeStr,
5571 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5574 // There's an AssertZext in the way of writing the store pattern
5575 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5578 let Predicates = [HasAVX] in {
5579 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5580 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5581 (ins VR128:$src1, i32i8imm:$src2),
5582 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5585 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5588 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5589 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5590 let neverHasSideEffects = 1, mayStore = 1 in
5591 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5592 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5593 !strconcat(OpcodeStr,
5594 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5597 // There's an AssertZext in the way of writing the store pattern
5598 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5601 let Predicates = [HasAVX] in
5602 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5604 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5607 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5608 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5609 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5610 (ins VR128:$src1, i32i8imm:$src2),
5611 !strconcat(OpcodeStr,
5612 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5614 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5615 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5616 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5617 !strconcat(OpcodeStr,
5618 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5619 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5620 addr:$dst)]>, OpSize;
5623 let Predicates = [HasAVX] in
5624 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5626 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5628 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5629 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5630 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5631 (ins VR128:$src1, i32i8imm:$src2),
5632 !strconcat(OpcodeStr,
5633 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5635 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5636 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5637 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5638 !strconcat(OpcodeStr,
5639 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5640 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5641 addr:$dst)]>, OpSize, REX_W;
5644 let Predicates = [HasAVX] in
5645 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5647 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5649 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5651 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5652 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5653 (ins VR128:$src1, i32i8imm:$src2),
5654 !strconcat(OpcodeStr,
5655 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5657 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5659 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5660 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5661 !strconcat(OpcodeStr,
5662 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5663 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5664 addr:$dst)]>, OpSize;
5667 let ExeDomain = SSEPackedSingle in {
5668 let Predicates = [HasAVX] in {
5669 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5670 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5671 (ins VR128:$src1, i32i8imm:$src2),
5672 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5675 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5678 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5679 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5682 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5684 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5687 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5688 Requires<[HasSSE41]>;
5690 //===----------------------------------------------------------------------===//
5691 // SSE4.1 - Insert Instructions
5692 //===----------------------------------------------------------------------===//
5694 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5695 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5696 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5698 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5700 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5702 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5703 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5704 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5706 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5708 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5710 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5711 imm:$src3))]>, OpSize;
5714 let Predicates = [HasAVX] in
5715 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5716 let Constraints = "$src1 = $dst" in
5717 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5719 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5720 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5721 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5723 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5725 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5727 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5729 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5730 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5732 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5734 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5736 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5737 imm:$src3)))]>, OpSize;
5740 let Predicates = [HasAVX] in
5741 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5742 let Constraints = "$src1 = $dst" in
5743 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5745 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5746 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5747 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5749 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5751 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5753 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5755 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5756 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5758 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5760 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5762 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5763 imm:$src3)))]>, OpSize;
5766 let Predicates = [HasAVX] in
5767 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5768 let Constraints = "$src1 = $dst" in
5769 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5771 // insertps has a few different modes, there's the first two here below which
5772 // are optimized inserts that won't zero arbitrary elements in the destination
5773 // vector. The next one matches the intrinsic and could zero arbitrary elements
5774 // in the target vector.
5775 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5776 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5777 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5779 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5781 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5783 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5785 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5786 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5788 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5790 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5792 (X86insrtps VR128:$src1,
5793 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5794 imm:$src3))]>, OpSize;
5797 let ExeDomain = SSEPackedSingle in {
5798 let Predicates = [HasAVX] in
5799 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5800 let Constraints = "$src1 = $dst" in
5801 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5804 //===----------------------------------------------------------------------===//
5805 // SSE4.1 - Round Instructions
5806 //===----------------------------------------------------------------------===//
5808 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5809 X86MemOperand x86memop, RegisterClass RC,
5810 PatFrag mem_frag32, PatFrag mem_frag64,
5811 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5812 let ExeDomain = SSEPackedSingle in {
5813 // Intrinsic operation, reg.
5814 // Vector intrinsic operation, reg
5815 def PSr : SS4AIi8<opcps, MRMSrcReg,
5816 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5817 !strconcat(OpcodeStr,
5818 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5819 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5822 // Vector intrinsic operation, mem
5823 def PSm : SS4AIi8<opcps, MRMSrcMem,
5824 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5825 !strconcat(OpcodeStr,
5826 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5828 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
5830 } // ExeDomain = SSEPackedSingle
5832 let ExeDomain = SSEPackedDouble in {
5833 // Vector intrinsic operation, reg
5834 def PDr : SS4AIi8<opcpd, MRMSrcReg,
5835 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5836 !strconcat(OpcodeStr,
5837 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5838 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
5841 // Vector intrinsic operation, mem
5842 def PDm : SS4AIi8<opcpd, MRMSrcMem,
5843 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5844 !strconcat(OpcodeStr,
5845 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5847 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
5849 } // ExeDomain = SSEPackedDouble
5852 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
5855 Intrinsic F64Int, bit Is2Addr = 1> {
5856 let ExeDomain = GenericDomain in {
5858 def SSr : SS4AIi8<opcss, MRMSrcReg,
5859 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
5861 !strconcat(OpcodeStr,
5862 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5863 !strconcat(OpcodeStr,
5864 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5867 // Intrinsic operation, reg.
5868 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
5869 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5871 !strconcat(OpcodeStr,
5872 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5873 !strconcat(OpcodeStr,
5874 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5875 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5878 // Intrinsic operation, mem.
5879 def SSm : SS4AIi8<opcss, MRMSrcMem,
5880 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5882 !strconcat(OpcodeStr,
5883 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5884 !strconcat(OpcodeStr,
5885 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5887 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
5891 def SDr : SS4AIi8<opcsd, MRMSrcReg,
5892 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
5894 !strconcat(OpcodeStr,
5895 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5896 !strconcat(OpcodeStr,
5897 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5900 // Intrinsic operation, reg.
5901 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
5902 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5904 !strconcat(OpcodeStr,
5905 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5906 !strconcat(OpcodeStr,
5907 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5908 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5911 // Intrinsic operation, mem.
5912 def SDm : SS4AIi8<opcsd, MRMSrcMem,
5913 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5915 !strconcat(OpcodeStr,
5916 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5917 !strconcat(OpcodeStr,
5918 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5920 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
5922 } // ExeDomain = GenericDomain
5925 // FP round - roundss, roundps, roundsd, roundpd
5926 let Predicates = [HasAVX] in {
5928 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
5929 memopv4f32, memopv2f64,
5930 int_x86_sse41_round_ps,
5931 int_x86_sse41_round_pd>, VEX;
5932 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
5933 memopv8f32, memopv4f64,
5934 int_x86_avx_round_ps_256,
5935 int_x86_avx_round_pd_256>, VEX;
5936 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
5937 int_x86_sse41_round_ss,
5938 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
5940 def : Pat<(ffloor FR32:$src),
5941 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
5942 def : Pat<(f64 (ffloor FR64:$src)),
5943 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
5944 def : Pat<(f32 (fnearbyint FR32:$src)),
5945 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
5946 def : Pat<(f64 (fnearbyint FR64:$src)),
5947 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
5948 def : Pat<(f32 (fceil FR32:$src)),
5949 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
5950 def : Pat<(f64 (fceil FR64:$src)),
5951 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
5952 def : Pat<(f32 (frint FR32:$src)),
5953 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
5954 def : Pat<(f64 (frint FR64:$src)),
5955 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
5956 def : Pat<(f32 (ftrunc FR32:$src)),
5957 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
5958 def : Pat<(f64 (ftrunc FR64:$src)),
5959 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
5962 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
5963 memopv4f32, memopv2f64,
5964 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
5965 let Constraints = "$src1 = $dst" in
5966 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
5967 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
5969 def : Pat<(ffloor FR32:$src),
5970 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
5971 def : Pat<(f64 (ffloor FR64:$src)),
5972 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
5973 def : Pat<(f32 (fnearbyint FR32:$src)),
5974 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
5975 def : Pat<(f64 (fnearbyint FR64:$src)),
5976 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
5977 def : Pat<(f32 (fceil FR32:$src)),
5978 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
5979 def : Pat<(f64 (fceil FR64:$src)),
5980 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
5981 def : Pat<(f32 (frint FR32:$src)),
5982 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
5983 def : Pat<(f64 (frint FR64:$src)),
5984 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
5985 def : Pat<(f32 (ftrunc FR32:$src)),
5986 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
5987 def : Pat<(f64 (ftrunc FR64:$src)),
5988 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
5990 //===----------------------------------------------------------------------===//
5991 // SSE4.1 - Packed Bit Test
5992 //===----------------------------------------------------------------------===//
5994 // ptest instruction we'll lower to this in X86ISelLowering primarily from
5995 // the intel intrinsic that corresponds to this.
5996 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5997 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5998 "vptest\t{$src2, $src1|$src1, $src2}",
5999 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6001 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6002 "vptest\t{$src2, $src1|$src1, $src2}",
6003 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6006 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6007 "vptest\t{$src2, $src1|$src1, $src2}",
6008 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6010 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6011 "vptest\t{$src2, $src1|$src1, $src2}",
6012 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6016 let Defs = [EFLAGS] in {
6017 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6018 "ptest\t{$src2, $src1|$src1, $src2}",
6019 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6021 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6022 "ptest\t{$src2, $src1|$src1, $src2}",
6023 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6027 // The bit test instructions below are AVX only
6028 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6029 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6030 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6031 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6032 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6033 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6034 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6035 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6039 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6040 let ExeDomain = SSEPackedSingle in {
6041 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6042 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6044 let ExeDomain = SSEPackedDouble in {
6045 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6046 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6050 //===----------------------------------------------------------------------===//
6051 // SSE4.1 - Misc Instructions
6052 //===----------------------------------------------------------------------===//
6054 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6055 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6056 "popcnt{w}\t{$src, $dst|$dst, $src}",
6057 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6059 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6060 "popcnt{w}\t{$src, $dst|$dst, $src}",
6061 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6062 (implicit EFLAGS)]>, OpSize, XS;
6064 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6065 "popcnt{l}\t{$src, $dst|$dst, $src}",
6066 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6068 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6069 "popcnt{l}\t{$src, $dst|$dst, $src}",
6070 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6071 (implicit EFLAGS)]>, XS;
6073 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6074 "popcnt{q}\t{$src, $dst|$dst, $src}",
6075 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6077 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6078 "popcnt{q}\t{$src, $dst|$dst, $src}",
6079 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6080 (implicit EFLAGS)]>, XS;
6085 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6086 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6087 Intrinsic IntId128> {
6088 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6090 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6091 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6092 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6094 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6097 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6100 let Predicates = [HasAVX] in
6101 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6102 int_x86_sse41_phminposuw>, VEX;
6103 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6104 int_x86_sse41_phminposuw>;
6106 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6107 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6108 Intrinsic IntId128, bit Is2Addr = 1> {
6109 let isCommutable = 1 in
6110 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6111 (ins VR128:$src1, VR128:$src2),
6113 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6114 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6115 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6116 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6117 (ins VR128:$src1, i128mem:$src2),
6119 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6120 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6122 (IntId128 VR128:$src1,
6123 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6126 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6127 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6128 Intrinsic IntId256> {
6129 let isCommutable = 1 in
6130 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6131 (ins VR256:$src1, VR256:$src2),
6132 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6133 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6134 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6135 (ins VR256:$src1, i256mem:$src2),
6136 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6138 (IntId256 VR256:$src1,
6139 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6142 let Predicates = [HasAVX] in {
6143 let isCommutable = 0 in
6144 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6146 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6148 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6150 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6152 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6154 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6156 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6158 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6160 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6162 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6166 let Predicates = [HasAVX2] in {
6167 let isCommutable = 0 in
6168 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6169 int_x86_avx2_packusdw>, VEX_4V;
6170 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6171 int_x86_avx2_pmins_b>, VEX_4V;
6172 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6173 int_x86_avx2_pmins_d>, VEX_4V;
6174 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6175 int_x86_avx2_pminu_d>, VEX_4V;
6176 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6177 int_x86_avx2_pminu_w>, VEX_4V;
6178 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6179 int_x86_avx2_pmaxs_b>, VEX_4V;
6180 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6181 int_x86_avx2_pmaxs_d>, VEX_4V;
6182 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6183 int_x86_avx2_pmaxu_d>, VEX_4V;
6184 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6185 int_x86_avx2_pmaxu_w>, VEX_4V;
6186 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6187 int_x86_avx2_pmul_dq>, VEX_4V;
6190 let Constraints = "$src1 = $dst" in {
6191 let isCommutable = 0 in
6192 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6193 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6194 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6195 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6196 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6197 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6198 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6199 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6200 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6201 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6204 /// SS48I_binop_rm - Simple SSE41 binary operator.
6205 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6206 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6207 X86MemOperand x86memop, bit Is2Addr = 1> {
6208 let isCommutable = 1 in
6209 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6210 (ins RC:$src1, RC:$src2),
6212 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6213 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6214 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6215 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6216 (ins RC:$src1, x86memop:$src2),
6218 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6219 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6221 (OpVT (OpNode RC:$src1,
6222 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6225 let Predicates = [HasAVX] in {
6226 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6227 memopv2i64, i128mem, 0>, VEX_4V;
6228 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6229 memopv2i64, i128mem, 0>, VEX_4V;
6231 let Predicates = [HasAVX2] in {
6232 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6233 memopv4i64, i256mem, 0>, VEX_4V;
6234 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6235 memopv4i64, i256mem, 0>, VEX_4V;
6238 let Constraints = "$src1 = $dst" in {
6239 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6240 memopv2i64, i128mem>;
6241 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6242 memopv2i64, i128mem>;
6245 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6246 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6247 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6248 X86MemOperand x86memop, bit Is2Addr = 1> {
6249 let isCommutable = 1 in
6250 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6251 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6253 !strconcat(OpcodeStr,
6254 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6255 !strconcat(OpcodeStr,
6256 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6257 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6259 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6260 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6262 !strconcat(OpcodeStr,
6263 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6264 !strconcat(OpcodeStr,
6265 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6268 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6272 let Predicates = [HasAVX] in {
6273 let isCommutable = 0 in {
6274 let ExeDomain = SSEPackedSingle in {
6275 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6276 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6277 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6278 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6280 let ExeDomain = SSEPackedDouble in {
6281 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6282 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6283 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6284 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6286 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6287 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6288 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6289 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6291 let ExeDomain = SSEPackedSingle in
6292 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6293 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6294 let ExeDomain = SSEPackedDouble in
6295 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6296 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6297 let ExeDomain = SSEPackedSingle in
6298 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6299 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6302 let Predicates = [HasAVX2] in {
6303 let isCommutable = 0 in {
6304 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6305 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6306 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6307 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6311 let Constraints = "$src1 = $dst" in {
6312 let isCommutable = 0 in {
6313 let ExeDomain = SSEPackedSingle in
6314 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6315 VR128, memopv4f32, i128mem>;
6316 let ExeDomain = SSEPackedDouble in
6317 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6318 VR128, memopv2f64, i128mem>;
6319 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6320 VR128, memopv2i64, i128mem>;
6321 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6322 VR128, memopv2i64, i128mem>;
6324 let ExeDomain = SSEPackedSingle in
6325 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6326 VR128, memopv4f32, i128mem>;
6327 let ExeDomain = SSEPackedDouble in
6328 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6329 VR128, memopv2f64, i128mem>;
6332 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6333 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6334 RegisterClass RC, X86MemOperand x86memop,
6335 PatFrag mem_frag, Intrinsic IntId> {
6336 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6337 (ins RC:$src1, RC:$src2, RC:$src3),
6338 !strconcat(OpcodeStr,
6339 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6340 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6341 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6343 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6344 (ins RC:$src1, x86memop:$src2, RC:$src3),
6345 !strconcat(OpcodeStr,
6346 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6348 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6350 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6353 let Predicates = [HasAVX] in {
6354 let ExeDomain = SSEPackedDouble in {
6355 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6356 memopv2f64, int_x86_sse41_blendvpd>;
6357 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6358 memopv4f64, int_x86_avx_blendv_pd_256>;
6359 } // ExeDomain = SSEPackedDouble
6360 let ExeDomain = SSEPackedSingle in {
6361 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6362 memopv4f32, int_x86_sse41_blendvps>;
6363 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6364 memopv8f32, int_x86_avx_blendv_ps_256>;
6365 } // ExeDomain = SSEPackedSingle
6366 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6367 memopv2i64, int_x86_sse41_pblendvb>;
6370 let Predicates = [HasAVX2] in {
6371 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6372 memopv4i64, int_x86_avx2_pblendvb>;
6375 let Predicates = [HasAVX] in {
6376 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6377 (v16i8 VR128:$src2))),
6378 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6379 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6380 (v4i32 VR128:$src2))),
6381 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6382 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6383 (v4f32 VR128:$src2))),
6384 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6385 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6386 (v2i64 VR128:$src2))),
6387 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6388 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6389 (v2f64 VR128:$src2))),
6390 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6391 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6392 (v8i32 VR256:$src2))),
6393 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6394 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6395 (v8f32 VR256:$src2))),
6396 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6397 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6398 (v4i64 VR256:$src2))),
6399 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6400 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6401 (v4f64 VR256:$src2))),
6402 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6405 let Predicates = [HasAVX2] in {
6406 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6407 (v32i8 VR256:$src2))),
6408 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6411 /// SS41I_ternary_int - SSE 4.1 ternary operator
6412 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6413 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6415 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6416 (ins VR128:$src1, VR128:$src2),
6417 !strconcat(OpcodeStr,
6418 "\t{$src2, $dst|$dst, $src2}"),
6419 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6422 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6423 (ins VR128:$src1, i128mem:$src2),
6424 !strconcat(OpcodeStr,
6425 "\t{$src2, $dst|$dst, $src2}"),
6428 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6432 let ExeDomain = SSEPackedDouble in
6433 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6434 int_x86_sse41_blendvpd>;
6435 let ExeDomain = SSEPackedSingle in
6436 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6437 int_x86_sse41_blendvps>;
6438 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6439 int_x86_sse41_pblendvb>;
6441 let Predicates = [HasSSE41] in {
6442 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6443 (v16i8 VR128:$src2))),
6444 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6445 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6446 (v4i32 VR128:$src2))),
6447 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6448 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6449 (v4f32 VR128:$src2))),
6450 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6451 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6452 (v2i64 VR128:$src2))),
6453 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6454 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6455 (v2f64 VR128:$src2))),
6456 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6459 let Predicates = [HasAVX] in
6460 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6461 "vmovntdqa\t{$src, $dst|$dst, $src}",
6462 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6464 let Predicates = [HasAVX2] in
6465 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6466 "vmovntdqa\t{$src, $dst|$dst, $src}",
6467 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6469 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6470 "movntdqa\t{$src, $dst|$dst, $src}",
6471 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6474 //===----------------------------------------------------------------------===//
6475 // SSE4.2 - Compare Instructions
6476 //===----------------------------------------------------------------------===//
6478 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6479 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6480 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6481 X86MemOperand x86memop, bit Is2Addr = 1> {
6482 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6483 (ins RC:$src1, RC:$src2),
6485 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6486 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6487 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6489 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6490 (ins RC:$src1, x86memop:$src2),
6492 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6493 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6495 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6498 let Predicates = [HasAVX] in
6499 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6500 memopv2i64, i128mem, 0>, VEX_4V;
6502 let Predicates = [HasAVX2] in
6503 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6504 memopv4i64, i256mem, 0>, VEX_4V;
6506 let Constraints = "$src1 = $dst" in
6507 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6508 memopv2i64, i128mem>;
6510 //===----------------------------------------------------------------------===//
6511 // SSE4.2 - String/text Processing Instructions
6512 //===----------------------------------------------------------------------===//
6514 // Packed Compare Implicit Length Strings, Return Mask
6515 multiclass pseudo_pcmpistrm<string asm> {
6516 def REG : PseudoI<(outs VR128:$dst),
6517 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6518 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6520 def MEM : PseudoI<(outs VR128:$dst),
6521 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6522 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6523 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6526 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6527 let AddedComplexity = 1 in
6528 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6529 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6532 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6533 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6534 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6535 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6537 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6538 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6539 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6542 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6543 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6544 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6545 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6547 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6548 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6549 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6552 // Packed Compare Explicit Length Strings, Return Mask
6553 multiclass pseudo_pcmpestrm<string asm> {
6554 def REG : PseudoI<(outs VR128:$dst),
6555 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6556 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6557 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6558 def MEM : PseudoI<(outs VR128:$dst),
6559 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6560 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6561 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6564 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6565 let AddedComplexity = 1 in
6566 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6567 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6570 let Predicates = [HasAVX],
6571 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6572 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6573 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6574 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6576 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6577 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6578 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6581 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6582 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6583 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6584 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6586 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6587 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6588 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6591 // Packed Compare Implicit Length Strings, Return Index
6592 let Defs = [ECX, EFLAGS] in {
6593 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6594 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6595 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6596 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6597 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6598 (implicit EFLAGS)]>, OpSize;
6599 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6600 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6601 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6602 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6603 (implicit EFLAGS)]>, OpSize;
6607 let Predicates = [HasAVX] in {
6608 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6610 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6612 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6614 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6616 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6618 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6622 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6623 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6624 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6625 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6626 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6627 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6629 // Packed Compare Explicit Length Strings, Return Index
6630 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6631 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6632 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6633 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6634 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6635 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6636 (implicit EFLAGS)]>, OpSize;
6637 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6638 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6639 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6641 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6642 (implicit EFLAGS)]>, OpSize;
6646 let Predicates = [HasAVX] in {
6647 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6649 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6651 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6653 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6655 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6657 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6661 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6662 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6663 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6664 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6665 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6666 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6668 //===----------------------------------------------------------------------===//
6669 // SSE4.2 - CRC Instructions
6670 //===----------------------------------------------------------------------===//
6672 // No CRC instructions have AVX equivalents
6674 // crc intrinsic instruction
6675 // This set of instructions are only rm, the only difference is the size
6677 let Constraints = "$src1 = $dst" in {
6678 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6679 (ins GR32:$src1, i8mem:$src2),
6680 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6682 (int_x86_sse42_crc32_32_8 GR32:$src1,
6683 (load addr:$src2)))]>;
6684 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6685 (ins GR32:$src1, GR8:$src2),
6686 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6688 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6689 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6690 (ins GR32:$src1, i16mem:$src2),
6691 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6693 (int_x86_sse42_crc32_32_16 GR32:$src1,
6694 (load addr:$src2)))]>,
6696 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6697 (ins GR32:$src1, GR16:$src2),
6698 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6700 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6702 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6703 (ins GR32:$src1, i32mem:$src2),
6704 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6706 (int_x86_sse42_crc32_32_32 GR32:$src1,
6707 (load addr:$src2)))]>;
6708 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6709 (ins GR32:$src1, GR32:$src2),
6710 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6712 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6713 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6714 (ins GR64:$src1, i8mem:$src2),
6715 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6717 (int_x86_sse42_crc32_64_8 GR64:$src1,
6718 (load addr:$src2)))]>,
6720 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6721 (ins GR64:$src1, GR8:$src2),
6722 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6724 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6726 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6727 (ins GR64:$src1, i64mem:$src2),
6728 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6730 (int_x86_sse42_crc32_64_64 GR64:$src1,
6731 (load addr:$src2)))]>,
6733 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6734 (ins GR64:$src1, GR64:$src2),
6735 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6737 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6741 //===----------------------------------------------------------------------===//
6742 // AES-NI Instructions
6743 //===----------------------------------------------------------------------===//
6745 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6746 Intrinsic IntId128, bit Is2Addr = 1> {
6747 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6748 (ins VR128:$src1, VR128:$src2),
6750 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6751 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6752 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6754 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6755 (ins VR128:$src1, i128mem:$src2),
6757 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6758 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6760 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6763 // Perform One Round of an AES Encryption/Decryption Flow
6764 let Predicates = [HasAVX, HasAES] in {
6765 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
6766 int_x86_aesni_aesenc, 0>, VEX_4V;
6767 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
6768 int_x86_aesni_aesenclast, 0>, VEX_4V;
6769 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
6770 int_x86_aesni_aesdec, 0>, VEX_4V;
6771 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
6772 int_x86_aesni_aesdeclast, 0>, VEX_4V;
6775 let Constraints = "$src1 = $dst" in {
6776 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
6777 int_x86_aesni_aesenc>;
6778 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
6779 int_x86_aesni_aesenclast>;
6780 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
6781 int_x86_aesni_aesdec>;
6782 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
6783 int_x86_aesni_aesdeclast>;
6786 // Perform the AES InvMixColumn Transformation
6787 let Predicates = [HasAVX, HasAES] in {
6788 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6790 "vaesimc\t{$src1, $dst|$dst, $src1}",
6792 (int_x86_aesni_aesimc VR128:$src1))]>,
6794 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6795 (ins i128mem:$src1),
6796 "vaesimc\t{$src1, $dst|$dst, $src1}",
6797 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
6800 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6802 "aesimc\t{$src1, $dst|$dst, $src1}",
6804 (int_x86_aesni_aesimc VR128:$src1))]>,
6806 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6807 (ins i128mem:$src1),
6808 "aesimc\t{$src1, $dst|$dst, $src1}",
6809 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
6812 // AES Round Key Generation Assist
6813 let Predicates = [HasAVX, HasAES] in {
6814 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6815 (ins VR128:$src1, i8imm:$src2),
6816 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6818 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6820 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6821 (ins i128mem:$src1, i8imm:$src2),
6822 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6824 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
6827 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6828 (ins VR128:$src1, i8imm:$src2),
6829 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6831 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6833 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6834 (ins i128mem:$src1, i8imm:$src2),
6835 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6837 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
6840 //===----------------------------------------------------------------------===//
6841 // CLMUL Instructions
6842 //===----------------------------------------------------------------------===//
6844 // Carry-less Multiplication instructions
6845 let neverHasSideEffects = 1 in {
6846 // AVX carry-less Multiplication instructions
6847 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6848 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6849 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6853 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6854 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6855 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6858 let Constraints = "$src1 = $dst" in {
6859 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6860 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6861 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6865 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6866 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6867 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6869 } // Constraints = "$src1 = $dst"
6870 } // neverHasSideEffects = 1
6873 multiclass pclmul_alias<string asm, int immop> {
6874 def : InstAlias<!strconcat("pclmul", asm,
6875 "dq {$src, $dst|$dst, $src}"),
6876 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
6878 def : InstAlias<!strconcat("pclmul", asm,
6879 "dq {$src, $dst|$dst, $src}"),
6880 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
6882 def : InstAlias<!strconcat("vpclmul", asm,
6883 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6884 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
6886 def : InstAlias<!strconcat("vpclmul", asm,
6887 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6888 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
6890 defm : pclmul_alias<"hqhq", 0x11>;
6891 defm : pclmul_alias<"hqlq", 0x01>;
6892 defm : pclmul_alias<"lqhq", 0x10>;
6893 defm : pclmul_alias<"lqlq", 0x00>;
6895 //===----------------------------------------------------------------------===//
6897 //===----------------------------------------------------------------------===//
6899 //===----------------------------------------------------------------------===//
6900 // VBROADCAST - Load from memory and broadcast to all elements of the
6901 // destination operand
6903 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
6904 X86MemOperand x86memop, Intrinsic Int> :
6905 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
6906 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6907 [(set RC:$dst, (Int addr:$src))]>, VEX;
6909 // AVX2 adds register forms
6910 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
6912 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
6913 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6914 [(set RC:$dst, (Int VR128:$src))]>, VEX;
6916 let ExeDomain = SSEPackedSingle in {
6917 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
6918 int_x86_avx_vbroadcast_ss>;
6919 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
6920 int_x86_avx_vbroadcast_ss_256>;
6922 let ExeDomain = SSEPackedDouble in
6923 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
6924 int_x86_avx_vbroadcast_sd_256>;
6925 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
6926 int_x86_avx_vbroadcastf128_pd_256>;
6928 let ExeDomain = SSEPackedSingle in {
6929 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
6930 int_x86_avx2_vbroadcast_ss_ps>;
6931 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
6932 int_x86_avx2_vbroadcast_ss_ps_256>;
6934 let ExeDomain = SSEPackedDouble in
6935 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
6936 int_x86_avx2_vbroadcast_sd_pd_256>;
6938 let Predicates = [HasAVX2] in
6939 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
6940 int_x86_avx2_vbroadcasti128>;
6942 let Predicates = [HasAVX] in
6943 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
6944 (VBROADCASTF128 addr:$src)>;
6947 //===----------------------------------------------------------------------===//
6948 // VINSERTF128 - Insert packed floating-point values
6950 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
6951 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
6952 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
6953 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6956 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
6957 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
6958 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6962 let Predicates = [HasAVX] in {
6963 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
6964 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6965 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
6966 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6967 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
6968 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6971 //===----------------------------------------------------------------------===//
6972 // VEXTRACTF128 - Extract packed floating-point values
6974 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
6975 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
6976 (ins VR256:$src1, i8imm:$src2),
6977 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6980 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
6981 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
6982 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6986 let Predicates = [HasAVX] in {
6987 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
6988 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6989 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
6990 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6991 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
6992 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6995 //===----------------------------------------------------------------------===//
6996 // VMASKMOV - Conditional SIMD Packed Loads and Stores
6998 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
6999 Intrinsic IntLd, Intrinsic IntLd256,
7000 Intrinsic IntSt, Intrinsic IntSt256> {
7001 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7002 (ins VR128:$src1, f128mem:$src2),
7003 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7004 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7006 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7007 (ins VR256:$src1, f256mem:$src2),
7008 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7009 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7011 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7012 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7013 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7014 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7015 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7016 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7017 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7018 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7021 let ExeDomain = SSEPackedSingle in
7022 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7023 int_x86_avx_maskload_ps,
7024 int_x86_avx_maskload_ps_256,
7025 int_x86_avx_maskstore_ps,
7026 int_x86_avx_maskstore_ps_256>;
7027 let ExeDomain = SSEPackedDouble in
7028 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7029 int_x86_avx_maskload_pd,
7030 int_x86_avx_maskload_pd_256,
7031 int_x86_avx_maskstore_pd,
7032 int_x86_avx_maskstore_pd_256>;
7034 //===----------------------------------------------------------------------===//
7035 // VPERMIL - Permute Single and Double Floating-Point Values
7037 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7038 RegisterClass RC, X86MemOperand x86memop_f,
7039 X86MemOperand x86memop_i, PatFrag i_frag,
7040 Intrinsic IntVar, ValueType vt> {
7041 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7042 (ins RC:$src1, RC:$src2),
7043 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7044 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7045 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7046 (ins RC:$src1, x86memop_i:$src2),
7047 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7048 [(set RC:$dst, (IntVar RC:$src1,
7049 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7051 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7052 (ins RC:$src1, i8imm:$src2),
7053 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7054 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7055 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7056 (ins x86memop_f:$src1, i8imm:$src2),
7057 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7059 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7062 let ExeDomain = SSEPackedSingle in {
7063 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7064 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7065 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7066 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
7068 let ExeDomain = SSEPackedDouble in {
7069 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7070 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7071 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7072 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
7075 let Predicates = [HasAVX] in {
7076 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7077 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7078 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7079 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7080 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7082 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7083 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7084 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7086 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7087 (VPERMILPDri VR128:$src1, imm:$imm)>;
7088 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7089 (VPERMILPDmi addr:$src1, imm:$imm)>;
7092 //===----------------------------------------------------------------------===//
7093 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7095 let ExeDomain = SSEPackedSingle in {
7096 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7097 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7098 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7099 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7100 (i8 imm:$src3))))]>, VEX_4V;
7101 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7102 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7103 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7104 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7105 (i8 imm:$src3)))]>, VEX_4V;
7108 let Predicates = [HasAVX] in {
7109 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7110 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7111 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7112 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7113 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7114 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7115 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7116 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7117 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7118 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7120 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7121 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7122 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7123 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7124 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7125 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7126 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7127 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7128 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7129 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7130 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7131 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7132 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7133 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7134 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7135 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7136 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7137 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7140 //===----------------------------------------------------------------------===//
7141 // VZERO - Zero YMM registers
7143 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7144 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7145 // Zero All YMM registers
7146 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7147 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7149 // Zero Upper bits of YMM registers
7150 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7151 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7154 //===----------------------------------------------------------------------===//
7155 // Half precision conversion instructions
7156 //===----------------------------------------------------------------------===//
7157 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7158 let Predicates = [HasAVX, HasF16C] in {
7159 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7160 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7161 [(set RC:$dst, (Int VR128:$src))]>,
7163 let neverHasSideEffects = 1, mayLoad = 1 in
7164 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7165 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7169 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7170 let Predicates = [HasAVX, HasF16C] in {
7171 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7172 (ins RC:$src1, i32i8imm:$src2),
7173 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7174 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7176 let neverHasSideEffects = 1, mayLoad = 1 in
7177 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7178 (ins RC:$src1, i32i8imm:$src2),
7179 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7184 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7185 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7186 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7187 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7189 //===----------------------------------------------------------------------===//
7190 // AVX2 Instructions
7191 //===----------------------------------------------------------------------===//
7193 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7194 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7195 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7196 X86MemOperand x86memop> {
7197 let isCommutable = 1 in
7198 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7199 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7200 !strconcat(OpcodeStr,
7201 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7202 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7204 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7205 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7206 !strconcat(OpcodeStr,
7207 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7210 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7214 let isCommutable = 0 in {
7215 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7216 VR128, memopv2i64, i128mem>;
7217 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7218 VR256, memopv4i64, i256mem>;
7221 //===----------------------------------------------------------------------===//
7222 // VPBROADCAST - Load from memory and broadcast to all elements of the
7223 // destination operand
7225 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7226 X86MemOperand x86memop, PatFrag ld_frag,
7227 Intrinsic Int128, Intrinsic Int256> {
7228 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7229 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7230 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7231 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7232 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7234 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7235 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7236 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7237 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7238 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7239 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7241 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7244 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7245 int_x86_avx2_pbroadcastb_128,
7246 int_x86_avx2_pbroadcastb_256>;
7247 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7248 int_x86_avx2_pbroadcastw_128,
7249 int_x86_avx2_pbroadcastw_256>;
7250 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7251 int_x86_avx2_pbroadcastd_128,
7252 int_x86_avx2_pbroadcastd_256>;
7253 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7254 int_x86_avx2_pbroadcastq_128,
7255 int_x86_avx2_pbroadcastq_256>;
7257 let Predicates = [HasAVX2] in {
7258 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7259 (VPBROADCASTBrm addr:$src)>;
7260 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7261 (VPBROADCASTBYrm addr:$src)>;
7262 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7263 (VPBROADCASTWrm addr:$src)>;
7264 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7265 (VPBROADCASTWYrm addr:$src)>;
7266 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7267 (VPBROADCASTDrm addr:$src)>;
7268 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7269 (VPBROADCASTDYrm addr:$src)>;
7270 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7271 (VPBROADCASTQrm addr:$src)>;
7272 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7273 (VPBROADCASTQYrm addr:$src)>;
7276 // AVX1 broadcast patterns
7277 let Predicates = [HasAVX] in {
7278 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7279 (VBROADCASTSSYrm addr:$src)>;
7280 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7281 (VBROADCASTSDrm addr:$src)>;
7282 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7283 (VBROADCASTSSYrm addr:$src)>;
7284 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7285 (VBROADCASTSDrm addr:$src)>;
7287 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7288 (VBROADCASTSSrm addr:$src)>;
7289 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7290 (VBROADCASTSSrm addr:$src)>;
7293 //===----------------------------------------------------------------------===//
7294 // VPERM - Permute instructions
7297 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7299 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7300 (ins VR256:$src1, VR256:$src2),
7301 !strconcat(OpcodeStr,
7302 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7303 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7304 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7305 (ins VR256:$src1, i256mem:$src2),
7306 !strconcat(OpcodeStr,
7307 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7308 [(set VR256:$dst, (Int VR256:$src1,
7309 (bitconvert (mem_frag addr:$src2))))]>,
7313 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7314 let ExeDomain = SSEPackedSingle in
7315 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7317 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7319 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7320 (ins VR256:$src1, i8imm:$src2),
7321 !strconcat(OpcodeStr,
7322 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7323 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7324 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7325 (ins i256mem:$src1, i8imm:$src2),
7326 !strconcat(OpcodeStr,
7327 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7328 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7332 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7334 let ExeDomain = SSEPackedDouble in
7335 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7338 //===----------------------------------------------------------------------===//
7339 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7341 let AddedComplexity = 1 in {
7342 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7343 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7344 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7345 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7346 (i8 imm:$src3))))]>, VEX_4V;
7347 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7348 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7349 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7350 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7351 (i8 imm:$src3)))]>, VEX_4V;
7354 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7355 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7356 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7357 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7358 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7359 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7360 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7362 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7364 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7365 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7366 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7367 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7368 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7370 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7374 //===----------------------------------------------------------------------===//
7375 // VINSERTI128 - Insert packed integer values
7377 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7378 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7379 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7381 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7383 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7384 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7385 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7387 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7388 imm:$src3))]>, VEX_4V;
7390 let Predicates = [HasAVX2] in {
7391 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7393 (VINSERTI128rr VR256:$src1, VR128:$src2,
7394 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7395 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7397 (VINSERTI128rr VR256:$src1, VR128:$src2,
7398 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7399 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7401 (VINSERTI128rr VR256:$src1, VR128:$src2,
7402 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7403 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7405 (VINSERTI128rr VR256:$src1, VR128:$src2,
7406 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7410 let Predicates = [HasAVX] in {
7411 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7413 (VINSERTF128rr VR256:$src1, VR128:$src2,
7414 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7415 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7417 (VINSERTF128rr VR256:$src1, VR128:$src2,
7418 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7419 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7421 (VINSERTF128rr VR256:$src1, VR128:$src2,
7422 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7423 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7425 (VINSERTF128rr VR256:$src1, VR128:$src2,
7426 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7427 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7429 (VINSERTF128rr VR256:$src1, VR128:$src2,
7430 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7431 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7433 (VINSERTF128rr VR256:$src1, VR128:$src2,
7434 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7437 //===----------------------------------------------------------------------===//
7438 // VEXTRACTI128 - Extract packed integer values
7440 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7441 (ins VR256:$src1, i8imm:$src2),
7442 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7444 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7446 let neverHasSideEffects = 1, mayStore = 1 in
7447 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7448 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7449 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7451 let Predicates = [HasAVX2] in {
7452 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7453 (v2i64 (VEXTRACTI128rr
7454 (v4i64 VR256:$src1),
7455 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7456 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7457 (v4i32 (VEXTRACTI128rr
7458 (v8i32 VR256:$src1),
7459 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7460 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7461 (v8i16 (VEXTRACTI128rr
7462 (v16i16 VR256:$src1),
7463 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7464 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7465 (v16i8 (VEXTRACTI128rr
7466 (v32i8 VR256:$src1),
7467 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7471 let Predicates = [HasAVX] in {
7472 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7473 (v4f32 (VEXTRACTF128rr
7474 (v8f32 VR256:$src1),
7475 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7476 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7477 (v2f64 (VEXTRACTF128rr
7478 (v4f64 VR256:$src1),
7479 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7480 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7481 (v2i64 (VEXTRACTF128rr
7482 (v4i64 VR256:$src1),
7483 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7484 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7485 (v4i32 (VEXTRACTF128rr
7486 (v8i32 VR256:$src1),
7487 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7488 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7489 (v8i16 (VEXTRACTF128rr
7490 (v16i16 VR256:$src1),
7491 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7492 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7493 (v16i8 (VEXTRACTF128rr
7494 (v32i8 VR256:$src1),
7495 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7498 //===----------------------------------------------------------------------===//
7499 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7501 multiclass avx2_pmovmask<string OpcodeStr,
7502 Intrinsic IntLd128, Intrinsic IntLd256,
7503 Intrinsic IntSt128, Intrinsic IntSt256> {
7504 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7505 (ins VR128:$src1, i128mem:$src2),
7506 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7507 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7508 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7509 (ins VR256:$src1, i256mem:$src2),
7510 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7511 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7512 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7513 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7514 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7515 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7516 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7517 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7518 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7519 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7522 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7523 int_x86_avx2_maskload_d,
7524 int_x86_avx2_maskload_d_256,
7525 int_x86_avx2_maskstore_d,
7526 int_x86_avx2_maskstore_d_256>;
7527 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7528 int_x86_avx2_maskload_q,
7529 int_x86_avx2_maskload_q_256,
7530 int_x86_avx2_maskstore_q,
7531 int_x86_avx2_maskstore_q_256>, VEX_W;
7534 //===----------------------------------------------------------------------===//
7535 // Variable Bit Shifts
7537 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7538 ValueType vt128, ValueType vt256> {
7539 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7540 (ins VR128:$src1, VR128:$src2),
7541 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7543 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7545 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7546 (ins VR128:$src1, i128mem:$src2),
7547 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7549 (vt128 (OpNode VR128:$src1,
7550 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7552 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7553 (ins VR256:$src1, VR256:$src2),
7554 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7556 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7558 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7559 (ins VR256:$src1, i256mem:$src2),
7560 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7562 (vt256 (OpNode VR256:$src1,
7563 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7567 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7568 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7569 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7570 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7571 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;