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, SDNode OpNode, ValueType vt, string asm> :
349 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
350 [(set VR128:$dst, (vt (OpNode VR128:$src1,
351 (scalar_to_vector RC:$src2))))]>;
353 // Loading from memory automatically zeroing upper bits.
354 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
355 PatFrag mem_pat, string OpcodeStr> :
356 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
357 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
358 [(set RC:$dst, (mem_pat addr:$src))]>;
361 def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
362 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
364 def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
365 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
368 // For the disassembler
369 let isCodeGenOnly = 1 in {
370 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
371 (ins VR128:$src1, FR32:$src2),
372 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
374 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
375 (ins VR128:$src1, FR64:$src2),
376 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
380 let canFoldAsLoad = 1, isReMaterializable = 1 in {
381 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
383 let AddedComplexity = 20 in
384 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
388 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
389 "movss\t{$src, $dst|$dst, $src}",
390 [(store FR32:$src, addr:$dst)]>, XS, VEX, VEX_LIG;
391 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
392 "movsd\t{$src, $dst|$dst, $src}",
393 [(store FR64:$src, addr:$dst)]>, XD, VEX, VEX_LIG;
396 let Constraints = "$src1 = $dst" in {
397 def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
398 "movss\t{$src2, $dst|$dst, $src2}">, XS;
399 def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
400 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
402 // For the disassembler
403 let isCodeGenOnly = 1 in {
404 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
405 (ins VR128:$src1, FR32:$src2),
406 "movss\t{$src2, $dst|$dst, $src2}", []>, XS;
407 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
408 (ins VR128:$src1, FR64:$src2),
409 "movsd\t{$src2, $dst|$dst, $src2}", []>, XD;
413 let canFoldAsLoad = 1, isReMaterializable = 1 in {
414 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
416 let AddedComplexity = 20 in
417 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
420 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
421 "movss\t{$src, $dst|$dst, $src}",
422 [(store FR32:$src, addr:$dst)]>;
423 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
424 "movsd\t{$src, $dst|$dst, $src}",
425 [(store FR64:$src, addr:$dst)]>;
428 let Predicates = [HasAVX] in {
429 let AddedComplexity = 15 in {
430 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
431 // MOVS{S,D} to the lower bits.
432 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
433 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
434 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
435 (VMOVSSrr (v4f32 (V_SET0)),
436 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
437 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
438 (VMOVSSrr (v4i32 (V_SET0)),
439 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
440 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
441 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
443 // Move low f32 and clear high bits.
444 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
445 (SUBREG_TO_REG (i32 0),
446 (VMOVSSrr (v4f32 (V_SET0)),
447 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
448 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
449 (SUBREG_TO_REG (i32 0),
450 (VMOVSSrr (v4i32 (V_SET0)),
451 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
454 let AddedComplexity = 20 in {
455 // MOVSSrm zeros the high parts of the register; represent this
456 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
457 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
458 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
459 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
460 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
461 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
462 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
464 // MOVSDrm zeros the high parts of the register; represent this
465 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
466 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
467 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
468 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
469 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
470 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
471 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
472 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
473 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
474 def : Pat<(v2f64 (X86vzload addr:$src)),
475 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
477 // Represent the same patterns above but in the form they appear for
479 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
480 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
481 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
482 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
483 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
484 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
485 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
486 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
487 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
489 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
490 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
491 (SUBREG_TO_REG (i32 0),
492 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
494 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
495 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
496 (SUBREG_TO_REG (i64 0),
497 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
499 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
500 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
501 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
503 // Move low f64 and clear high bits.
504 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
505 (SUBREG_TO_REG (i32 0),
506 (VMOVSDrr (v2f64 (V_SET0)),
507 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
509 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
510 (SUBREG_TO_REG (i32 0),
511 (VMOVSDrr (v2i64 (V_SET0)),
512 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_sd)), sub_xmm)>;
514 // Extract and store.
515 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
518 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
519 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
522 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
524 // Shuffle with VMOVSS
525 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
526 (VMOVSSrr (v4i32 VR128:$src1),
527 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
528 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
529 (VMOVSSrr (v4f32 VR128:$src1),
530 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
533 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
534 (SUBREG_TO_REG (i32 0),
535 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
536 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
537 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
538 (SUBREG_TO_REG (i32 0),
539 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
540 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
542 // Shuffle with VMOVSD
543 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
544 (VMOVSDrr (v2i64 VR128:$src1),
545 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
546 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
547 (VMOVSDrr (v2f64 VR128:$src1),
548 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
549 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
550 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
552 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
553 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
557 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
558 (SUBREG_TO_REG (i32 0),
559 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
560 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
561 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
562 (SUBREG_TO_REG (i32 0),
563 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
564 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
567 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
568 // is during lowering, where it's not possible to recognize the fold cause
569 // it has two uses through a bitcast. One use disappears at isel time and the
570 // fold opportunity reappears.
571 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
572 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
574 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
575 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
577 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
578 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
580 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
581 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
585 let Predicates = [HasSSE1] in {
586 let AddedComplexity = 15 in {
587 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
588 // MOVSS to the lower bits.
589 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
590 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
591 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
592 (MOVSSrr (v4f32 (V_SET0)),
593 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
594 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
595 (MOVSSrr (v4i32 (V_SET0)),
596 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
599 let AddedComplexity = 20 in {
600 // MOVSSrm zeros the high parts of the register; represent this
601 // with SUBREG_TO_REG.
602 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
603 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
604 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
605 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
606 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
607 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
610 // Extract and store.
611 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
614 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
616 // Shuffle with MOVSS
617 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
618 (MOVSSrr (v4i32 VR128:$src1),
619 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
620 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
621 (MOVSSrr (v4f32 VR128:$src1),
622 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
625 let Predicates = [HasSSE2] in {
626 let AddedComplexity = 15 in {
627 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
628 // MOVSD to the lower bits.
629 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
630 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
633 let AddedComplexity = 20 in {
634 // MOVSDrm zeros the high parts of the register; represent this
635 // with SUBREG_TO_REG.
636 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
637 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
638 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
639 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
640 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
641 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
642 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
643 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
644 def : Pat<(v2f64 (X86vzload addr:$src)),
645 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
648 // Extract and store.
649 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
652 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
654 // Shuffle with MOVSD
655 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
656 (MOVSDrr (v2i64 VR128:$src1),
657 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
658 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
659 (MOVSDrr (v2f64 VR128:$src1),
660 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
661 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
662 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
663 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
664 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
666 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
667 // is during lowering, where it's not possible to recognize the fold cause
668 // it has two uses through a bitcast. One use disappears at isel time and the
669 // fold opportunity reappears.
670 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
671 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
672 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
673 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
674 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
675 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
676 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
677 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
680 //===----------------------------------------------------------------------===//
681 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
682 //===----------------------------------------------------------------------===//
684 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
685 X86MemOperand x86memop, PatFrag ld_frag,
686 string asm, Domain d,
687 bit IsReMaterializable = 1> {
688 let neverHasSideEffects = 1 in
689 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
690 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], IIC_DEFAULT, d>;
691 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
692 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
693 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
694 [(set RC:$dst, (ld_frag addr:$src))], IIC_DEFAULT, d>;
697 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
698 "movaps", SSEPackedSingle>, TB, VEX;
699 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
700 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
701 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
702 "movups", SSEPackedSingle>, TB, VEX;
703 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
704 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
706 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
707 "movaps", SSEPackedSingle>, TB, VEX;
708 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
709 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
710 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
711 "movups", SSEPackedSingle>, TB, VEX;
712 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
713 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
714 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
715 "movaps", SSEPackedSingle>, TB;
716 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
717 "movapd", SSEPackedDouble>, TB, OpSize;
718 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
719 "movups", SSEPackedSingle>, TB;
720 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
721 "movupd", SSEPackedDouble, 0>, TB, OpSize;
723 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
724 "movaps\t{$src, $dst|$dst, $src}",
725 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
726 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
727 "movapd\t{$src, $dst|$dst, $src}",
728 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
729 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
730 "movups\t{$src, $dst|$dst, $src}",
731 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
732 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
733 "movupd\t{$src, $dst|$dst, $src}",
734 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
735 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
736 "movaps\t{$src, $dst|$dst, $src}",
737 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)]>, VEX;
738 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
739 "movapd\t{$src, $dst|$dst, $src}",
740 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)]>, VEX;
741 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
742 "movups\t{$src, $dst|$dst, $src}",
743 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
744 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
745 "movupd\t{$src, $dst|$dst, $src}",
746 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
749 let isCodeGenOnly = 1 in {
750 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
752 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
753 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
755 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
756 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
758 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
759 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
761 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
762 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
764 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
765 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
767 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
768 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
770 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
771 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
773 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
776 let Predicates = [HasAVX] in {
777 def : Pat<(v8i32 (X86vzmovl
778 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
779 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
780 def : Pat<(v4i64 (X86vzmovl
781 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
782 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
783 def : Pat<(v8f32 (X86vzmovl
784 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
785 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
786 def : Pat<(v4f64 (X86vzmovl
787 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
788 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
792 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
793 (VMOVUPSYmr addr:$dst, VR256:$src)>;
794 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
795 (VMOVUPDYmr addr:$dst, VR256:$src)>;
797 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
798 "movaps\t{$src, $dst|$dst, $src}",
799 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
800 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
801 "movapd\t{$src, $dst|$dst, $src}",
802 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
803 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
804 "movups\t{$src, $dst|$dst, $src}",
805 [(store (v4f32 VR128:$src), addr:$dst)]>;
806 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
807 "movupd\t{$src, $dst|$dst, $src}",
808 [(store (v2f64 VR128:$src), addr:$dst)]>;
811 let isCodeGenOnly = 1 in {
812 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
813 "movaps\t{$src, $dst|$dst, $src}", []>;
814 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
815 "movapd\t{$src, $dst|$dst, $src}", []>;
816 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
817 "movups\t{$src, $dst|$dst, $src}", []>;
818 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
819 "movupd\t{$src, $dst|$dst, $src}", []>;
822 let Predicates = [HasAVX] in {
823 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
824 (VMOVUPSmr addr:$dst, VR128:$src)>;
825 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
826 (VMOVUPDmr addr:$dst, VR128:$src)>;
829 let Predicates = [HasSSE1] in
830 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
831 (MOVUPSmr addr:$dst, VR128:$src)>;
832 let Predicates = [HasSSE2] in
833 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
834 (MOVUPDmr addr:$dst, VR128:$src)>;
836 // Use vmovaps/vmovups for AVX integer load/store.
837 let Predicates = [HasAVX] in {
838 // 128-bit load/store
839 def : Pat<(alignedloadv2i64 addr:$src),
840 (VMOVAPSrm addr:$src)>;
841 def : Pat<(loadv2i64 addr:$src),
842 (VMOVUPSrm addr:$src)>;
844 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
845 (VMOVAPSmr addr:$dst, VR128:$src)>;
846 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
847 (VMOVAPSmr addr:$dst, VR128:$src)>;
848 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
849 (VMOVAPSmr addr:$dst, VR128:$src)>;
850 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
851 (VMOVAPSmr addr:$dst, VR128:$src)>;
852 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
853 (VMOVUPSmr addr:$dst, VR128:$src)>;
854 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
855 (VMOVUPSmr addr:$dst, VR128:$src)>;
856 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
857 (VMOVUPSmr addr:$dst, VR128:$src)>;
858 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
859 (VMOVUPSmr addr:$dst, VR128:$src)>;
861 // 256-bit load/store
862 def : Pat<(alignedloadv4i64 addr:$src),
863 (VMOVAPSYrm addr:$src)>;
864 def : Pat<(loadv4i64 addr:$src),
865 (VMOVUPSYrm addr:$src)>;
866 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
867 (VMOVAPSYmr addr:$dst, VR256:$src)>;
868 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
869 (VMOVAPSYmr addr:$dst, VR256:$src)>;
870 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
871 (VMOVAPSYmr addr:$dst, VR256:$src)>;
872 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
873 (VMOVAPSYmr addr:$dst, VR256:$src)>;
874 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
875 (VMOVUPSYmr addr:$dst, VR256:$src)>;
876 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
877 (VMOVUPSYmr addr:$dst, VR256:$src)>;
878 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
879 (VMOVUPSYmr addr:$dst, VR256:$src)>;
880 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
881 (VMOVUPSYmr addr:$dst, VR256:$src)>;
884 // Use movaps / movups for SSE integer load / store (one byte shorter).
885 // The instructions selected below are then converted to MOVDQA/MOVDQU
886 // during the SSE domain pass.
887 let Predicates = [HasSSE1] in {
888 def : Pat<(alignedloadv2i64 addr:$src),
889 (MOVAPSrm addr:$src)>;
890 def : Pat<(loadv2i64 addr:$src),
891 (MOVUPSrm addr:$src)>;
893 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
894 (MOVAPSmr addr:$dst, VR128:$src)>;
895 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
896 (MOVAPSmr addr:$dst, VR128:$src)>;
897 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
898 (MOVAPSmr addr:$dst, VR128:$src)>;
899 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
900 (MOVAPSmr addr:$dst, VR128:$src)>;
901 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
902 (MOVUPSmr addr:$dst, VR128:$src)>;
903 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
904 (MOVUPSmr addr:$dst, VR128:$src)>;
905 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
906 (MOVUPSmr addr:$dst, VR128:$src)>;
907 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
908 (MOVUPSmr addr:$dst, VR128:$src)>;
911 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
912 // bits are disregarded. FIXME: Set encoding to pseudo!
913 let neverHasSideEffects = 1 in {
914 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
915 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
916 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
917 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
918 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
919 "movaps\t{$src, $dst|$dst, $src}", []>;
920 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
921 "movapd\t{$src, $dst|$dst, $src}", []>;
924 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
925 // bits are disregarded. FIXME: Set encoding to pseudo!
926 let canFoldAsLoad = 1, isReMaterializable = 1 in {
927 let isCodeGenOnly = 1 in {
928 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
929 "movaps\t{$src, $dst|$dst, $src}",
930 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
931 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
932 "movapd\t{$src, $dst|$dst, $src}",
933 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
935 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
936 "movaps\t{$src, $dst|$dst, $src}",
937 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
938 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
939 "movapd\t{$src, $dst|$dst, $src}",
940 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
943 //===----------------------------------------------------------------------===//
944 // SSE 1 & 2 - Move Low packed FP Instructions
945 //===----------------------------------------------------------------------===//
947 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
948 PatFrag mov_frag, string base_opc,
950 def PSrm : PI<opc, MRMSrcMem,
951 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
952 !strconcat(base_opc, "s", asm_opr),
955 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
956 IIC_DEFAULT, SSEPackedSingle>, TB;
958 def PDrm : PI<opc, MRMSrcMem,
959 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
960 !strconcat(base_opc, "d", asm_opr),
961 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
962 (scalar_to_vector (loadf64 addr:$src2)))))],
963 IIC_DEFAULT, SSEPackedDouble>, TB, OpSize;
966 let AddedComplexity = 20 in {
967 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
968 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
970 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
971 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
972 "\t{$src2, $dst|$dst, $src2}">;
975 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
976 "movlps\t{$src, $dst|$dst, $src}",
977 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
978 (iPTR 0))), addr:$dst)]>, VEX;
979 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
980 "movlpd\t{$src, $dst|$dst, $src}",
981 [(store (f64 (vector_extract (v2f64 VR128:$src),
982 (iPTR 0))), addr:$dst)]>, VEX;
983 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
984 "movlps\t{$src, $dst|$dst, $src}",
985 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
986 (iPTR 0))), addr:$dst)]>;
987 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
988 "movlpd\t{$src, $dst|$dst, $src}",
989 [(store (f64 (vector_extract (v2f64 VR128:$src),
990 (iPTR 0))), addr:$dst)]>;
992 let Predicates = [HasAVX] in {
993 let AddedComplexity = 20 in {
994 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
995 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
996 (VMOVLPSrm VR128:$src1, addr:$src2)>;
997 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
998 (VMOVLPSrm VR128:$src1, addr:$src2)>;
999 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1000 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1001 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1002 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1003 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1006 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1007 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1008 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1009 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1010 VR128:$src2)), addr:$src1),
1011 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1013 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1014 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1015 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1016 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1017 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1019 // Shuffle with VMOVLPS
1020 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1021 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1022 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1023 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1024 def : Pat<(X86Movlps VR128:$src1,
1025 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1026 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1028 // Shuffle with VMOVLPD
1029 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1030 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1031 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1032 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1033 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1034 (scalar_to_vector (loadf64 addr:$src2)))),
1035 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1038 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1040 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1041 def : Pat<(store (v4i32 (X86Movlps
1042 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1043 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1044 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1046 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1047 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1049 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1052 let Predicates = [HasSSE1] in {
1053 let AddedComplexity = 20 in {
1054 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1055 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1056 (MOVLPSrm VR128:$src1, addr:$src2)>;
1057 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1058 (MOVLPSrm VR128:$src1, addr:$src2)>;
1061 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1062 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1063 (iPTR 0))), addr:$src1),
1064 (MOVLPSmr addr:$src1, VR128:$src2)>;
1065 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1066 (MOVLPSmr addr:$src1, VR128:$src2)>;
1067 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1068 VR128:$src2)), addr:$src1),
1069 (MOVLPSmr addr:$src1, VR128:$src2)>;
1071 // Shuffle with MOVLPS
1072 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1073 (MOVLPSrm VR128:$src1, addr:$src2)>;
1074 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1075 (MOVLPSrm VR128:$src1, addr:$src2)>;
1076 def : Pat<(X86Movlps VR128:$src1,
1077 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1078 (MOVLPSrm VR128:$src1, addr:$src2)>;
1079 def : Pat<(X86Movlps VR128:$src1,
1080 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1081 (MOVLPSrm VR128:$src1, addr:$src2)>;
1084 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1086 (MOVLPSmr addr:$src1, VR128:$src2)>;
1087 def : Pat<(store (v4i32 (X86Movlps
1088 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1090 (MOVLPSmr addr:$src1, VR128:$src2)>;
1093 let Predicates = [HasSSE2] in {
1094 let AddedComplexity = 20 in {
1095 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1096 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1097 (MOVLPDrm VR128:$src1, addr:$src2)>;
1098 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1099 (MOVLPDrm VR128:$src1, addr:$src2)>;
1102 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1103 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1104 (MOVLPDmr addr:$src1, VR128:$src2)>;
1105 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1106 (MOVLPDmr addr:$src1, VR128:$src2)>;
1108 // Shuffle with MOVLPD
1109 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1110 (MOVLPDrm VR128:$src1, addr:$src2)>;
1111 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1112 (MOVLPDrm VR128:$src1, addr:$src2)>;
1113 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1114 (scalar_to_vector (loadf64 addr:$src2)))),
1115 (MOVLPDrm VR128:$src1, addr:$src2)>;
1118 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1120 (MOVLPDmr addr:$src1, VR128:$src2)>;
1121 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1123 (MOVLPDmr addr:$src1, VR128:$src2)>;
1126 //===----------------------------------------------------------------------===//
1127 // SSE 1 & 2 - Move Hi packed FP Instructions
1128 //===----------------------------------------------------------------------===//
1130 let AddedComplexity = 20 in {
1131 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1132 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1134 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1135 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1136 "\t{$src2, $dst|$dst, $src2}">;
1139 // v2f64 extract element 1 is always custom lowered to unpack high to low
1140 // and extract element 0 so the non-store version isn't too horrible.
1141 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1142 "movhps\t{$src, $dst|$dst, $src}",
1143 [(store (f64 (vector_extract
1144 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1145 (bc_v2f64 (v4f32 VR128:$src))),
1146 (iPTR 0))), addr:$dst)]>, VEX;
1147 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1148 "movhpd\t{$src, $dst|$dst, $src}",
1149 [(store (f64 (vector_extract
1150 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1151 (iPTR 0))), addr:$dst)]>, VEX;
1152 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1153 "movhps\t{$src, $dst|$dst, $src}",
1154 [(store (f64 (vector_extract
1155 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1156 (bc_v2f64 (v4f32 VR128:$src))),
1157 (iPTR 0))), addr:$dst)]>;
1158 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1159 "movhpd\t{$src, $dst|$dst, $src}",
1160 [(store (f64 (vector_extract
1161 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1162 (iPTR 0))), addr:$dst)]>;
1164 let Predicates = [HasAVX] in {
1166 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1167 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1168 def : Pat<(X86Movlhps VR128:$src1,
1169 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1170 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1171 def : Pat<(X86Movlhps VR128:$src1,
1172 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1173 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1174 def : Pat<(X86Movlhps VR128:$src1,
1175 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1176 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1178 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1179 // is during lowering, where it's not possible to recognize the load fold
1180 // cause it has two uses through a bitcast. One use disappears at isel time
1181 // and the fold opportunity reappears.
1182 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1183 (scalar_to_vector (loadf64 addr:$src2)))),
1184 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1186 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1187 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1188 (scalar_to_vector (loadf64 addr:$src2)))),
1189 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1192 let Predicates = [HasSSE1] in {
1194 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1195 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1196 def : Pat<(X86Movlhps VR128:$src1,
1197 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1198 (MOVHPSrm VR128:$src1, addr:$src2)>;
1199 def : Pat<(X86Movlhps VR128:$src1,
1200 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1201 (MOVHPSrm VR128:$src1, addr:$src2)>;
1202 def : Pat<(X86Movlhps VR128:$src1,
1203 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1204 (MOVHPSrm VR128:$src1, addr:$src2)>;
1207 let Predicates = [HasSSE2] in {
1208 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1209 // is during lowering, where it's not possible to recognize the load fold
1210 // cause it has two uses through a bitcast. One use disappears at isel time
1211 // and the fold opportunity reappears.
1212 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1213 (scalar_to_vector (loadf64 addr:$src2)))),
1214 (MOVHPDrm VR128:$src1, addr:$src2)>;
1216 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1217 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1218 (scalar_to_vector (loadf64 addr:$src2)))),
1219 (MOVHPDrm VR128:$src1, addr:$src2)>;
1222 //===----------------------------------------------------------------------===//
1223 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1224 //===----------------------------------------------------------------------===//
1226 let AddedComplexity = 20 in {
1227 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1228 (ins VR128:$src1, VR128:$src2),
1229 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1231 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))]>,
1233 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1234 (ins VR128:$src1, VR128:$src2),
1235 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1237 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))]>,
1240 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1241 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1242 (ins VR128:$src1, VR128:$src2),
1243 "movlhps\t{$src2, $dst|$dst, $src2}",
1245 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))]>;
1246 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1247 (ins VR128:$src1, VR128:$src2),
1248 "movhlps\t{$src2, $dst|$dst, $src2}",
1250 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))]>;
1253 let Predicates = [HasAVX] in {
1255 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1256 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1257 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1258 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1261 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1262 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1265 let Predicates = [HasSSE1] in {
1267 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1268 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1269 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1270 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1273 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1274 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1277 //===----------------------------------------------------------------------===//
1278 // SSE 1 & 2 - Conversion Instructions
1279 //===----------------------------------------------------------------------===//
1281 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1282 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1284 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1285 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1286 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1287 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1290 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1291 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1292 string asm, Domain d> {
1293 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1294 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1296 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1297 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1301 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1302 X86MemOperand x86memop, string asm> {
1303 let neverHasSideEffects = 1 in {
1304 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1305 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1307 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1308 (ins DstRC:$src1, x86memop:$src),
1309 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1310 } // neverHasSideEffects = 1
1313 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1314 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1316 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1317 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1319 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1320 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1322 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1323 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1324 VEX, VEX_W, VEX_LIG;
1326 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1327 // register, but the same isn't true when only using memory operands,
1328 // provide other assembly "l" and "q" forms to address this explicitly
1329 // where appropriate to do so.
1330 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1332 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1333 VEX_4V, VEX_W, VEX_LIG;
1334 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1336 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1338 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1339 VEX_4V, VEX_W, VEX_LIG;
1341 let Predicates = [HasAVX], AddedComplexity = 1 in {
1342 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1343 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1344 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1345 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1346 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1347 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1348 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1349 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1351 def : Pat<(f32 (sint_to_fp GR32:$src)),
1352 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1353 def : Pat<(f32 (sint_to_fp GR64:$src)),
1354 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1355 def : Pat<(f64 (sint_to_fp GR32:$src)),
1356 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1357 def : Pat<(f64 (sint_to_fp GR64:$src)),
1358 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1361 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1362 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1363 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1364 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1365 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1366 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1367 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1368 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1369 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1370 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1371 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1372 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1373 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1374 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1375 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1376 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1378 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1379 // and/or XMM operand(s).
1381 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1382 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1384 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1385 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1386 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1387 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1388 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1389 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1392 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1393 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1394 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1395 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1397 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1398 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1399 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1400 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1401 (ins DstRC:$src1, x86memop:$src2),
1403 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1404 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1405 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1408 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1409 f128mem, load, "cvtsd2si">, XD, VEX, VEX_LIG;
1410 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1411 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1412 XD, VEX, VEX_W, VEX_LIG;
1414 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1415 f128mem, load, "cvtsd2si{l}">, XD;
1416 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1417 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1420 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1421 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1422 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1423 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1425 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1426 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1427 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1428 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1431 let Constraints = "$src1 = $dst" in {
1432 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1433 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1435 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1436 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1437 "cvtsi2ss{q}">, XS, REX_W;
1438 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1439 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1441 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1442 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1443 "cvtsi2sd">, XD, REX_W;
1448 // Aliases for intrinsics
1449 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1450 f32mem, load, "cvttss2si">, XS, VEX;
1451 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1452 int_x86_sse_cvttss2si64, f32mem, load,
1453 "cvttss2si">, XS, VEX, VEX_W;
1454 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1455 f128mem, load, "cvttsd2si">, XD, VEX;
1456 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1457 int_x86_sse2_cvttsd2si64, f128mem, load,
1458 "cvttsd2si">, XD, VEX, VEX_W;
1459 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1460 f32mem, load, "cvttss2si">, XS;
1461 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1462 int_x86_sse_cvttss2si64, f32mem, load,
1463 "cvttss2si{q}">, XS, REX_W;
1464 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1465 f128mem, load, "cvttsd2si">, XD;
1466 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1467 int_x86_sse2_cvttsd2si64, f128mem, load,
1468 "cvttsd2si{q}">, XD, REX_W;
1470 let Pattern = []<dag> in {
1471 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1472 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1474 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1475 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1477 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1478 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1479 SSEPackedSingle>, TB, VEX;
1480 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1481 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1482 SSEPackedSingle>, TB, VEX;
1485 let Pattern = []<dag> in {
1486 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1487 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1488 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1489 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1490 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1491 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1492 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1495 let Predicates = [HasAVX] in {
1496 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1497 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1498 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1499 (VCVTSS2SIrm addr:$src)>;
1500 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1501 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1502 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1503 (VCVTSS2SI64rm addr:$src)>;
1506 let Predicates = [HasSSE1] in {
1507 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1508 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1509 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1510 (CVTSS2SIrm addr:$src)>;
1511 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1512 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1513 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1514 (CVTSS2SI64rm addr:$src)>;
1519 // Convert scalar double to scalar single
1520 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1521 (ins FR64:$src1, FR64:$src2),
1522 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1525 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1526 (ins FR64:$src1, f64mem:$src2),
1527 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1528 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1530 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1533 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1534 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1535 [(set FR32:$dst, (fround FR64:$src))]>;
1536 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1537 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1538 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1539 Requires<[HasSSE2, OptForSize]>;
1541 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1542 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1544 let Constraints = "$src1 = $dst" in
1545 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1546 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1548 // Convert scalar single to scalar double
1549 // SSE2 instructions with XS prefix
1550 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1551 (ins FR32:$src1, FR32:$src2),
1552 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1553 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1555 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1556 (ins FR32:$src1, f32mem:$src2),
1557 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1558 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1560 let Predicates = [HasAVX] in {
1561 def : Pat<(f64 (fextend FR32:$src)),
1562 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1563 def : Pat<(fextend (loadf32 addr:$src)),
1564 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1565 def : Pat<(extloadf32 addr:$src),
1566 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1569 def : Pat<(extloadf32 addr:$src),
1570 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1571 Requires<[HasAVX, OptForSpeed]>;
1573 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1574 "cvtss2sd\t{$src, $dst|$dst, $src}",
1575 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1576 Requires<[HasSSE2]>;
1577 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1578 "cvtss2sd\t{$src, $dst|$dst, $src}",
1579 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1580 Requires<[HasSSE2, OptForSize]>;
1582 // extload f32 -> f64. This matches load+fextend because we have a hack in
1583 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1585 // Since these loads aren't folded into the fextend, we have to match it
1587 def : Pat<(fextend (loadf32 addr:$src)),
1588 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1589 def : Pat<(extloadf32 addr:$src),
1590 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1592 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1593 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1594 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1595 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1596 VR128:$src2))]>, XS, VEX_4V,
1598 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1599 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1600 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1601 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1602 (load addr:$src2)))]>, XS, VEX_4V,
1604 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1605 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1606 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1607 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1608 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1609 VR128:$src2))]>, XS,
1610 Requires<[HasSSE2]>;
1611 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1612 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1613 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1614 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1615 (load addr:$src2)))]>, XS,
1616 Requires<[HasSSE2]>;
1619 // Convert doubleword to packed single/double fp
1620 // SSE2 instructions without OpSize prefix
1621 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1622 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1623 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1624 TB, VEX, Requires<[HasAVX]>;
1625 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1626 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1627 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1628 (bitconvert (memopv2i64 addr:$src))))]>,
1629 TB, VEX, Requires<[HasAVX]>;
1630 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1631 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1632 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1633 TB, Requires<[HasSSE2]>;
1634 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1635 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1636 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1637 (bitconvert (memopv2i64 addr:$src))))]>,
1638 TB, Requires<[HasSSE2]>;
1640 // FIXME: why the non-intrinsic version is described as SSE3?
1641 // SSE2 instructions with XS prefix
1642 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1643 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1644 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1645 XS, VEX, Requires<[HasAVX]>;
1646 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1647 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1648 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1649 (bitconvert (memopv2i64 addr:$src))))]>,
1650 XS, VEX, Requires<[HasAVX]>;
1651 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1652 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1653 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1654 XS, Requires<[HasSSE2]>;
1655 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1656 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1657 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1658 (bitconvert (memopv2i64 addr:$src))))]>,
1659 XS, Requires<[HasSSE2]>;
1662 // Convert packed single/double fp to doubleword
1663 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1664 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1665 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1666 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1667 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1668 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1669 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1670 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1671 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1672 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1673 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1674 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1676 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1677 "cvtps2dq\t{$src, $dst|$dst, $src}",
1678 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1680 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1682 "cvtps2dq\t{$src, $dst|$dst, $src}",
1683 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1684 (memop addr:$src)))]>, VEX;
1685 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1686 "cvtps2dq\t{$src, $dst|$dst, $src}",
1687 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1688 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1689 "cvtps2dq\t{$src, $dst|$dst, $src}",
1690 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1691 (memop addr:$src)))]>;
1693 // SSE2 packed instructions with XD prefix
1694 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1695 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1696 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1697 XD, VEX, Requires<[HasAVX]>;
1698 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1699 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1700 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1701 (memop addr:$src)))]>,
1702 XD, VEX, Requires<[HasAVX]>;
1703 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1704 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1705 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1706 XD, Requires<[HasSSE2]>;
1707 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1708 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1709 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1710 (memop addr:$src)))]>,
1711 XD, Requires<[HasSSE2]>;
1714 // Convert with truncation packed single/double fp to doubleword
1715 // SSE2 packed instructions with XS prefix
1716 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1717 "cvttps2dq\t{$src, $dst|$dst, $src}",
1719 (int_x86_sse2_cvttps2dq VR128:$src))]>, VEX;
1720 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1721 "cvttps2dq\t{$src, $dst|$dst, $src}",
1722 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1723 (memop addr:$src)))]>, VEX;
1724 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1725 "cvttps2dq\t{$src, $dst|$dst, $src}",
1727 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))]>, VEX;
1728 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1729 "cvttps2dq\t{$src, $dst|$dst, $src}",
1730 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1731 (memopv8f32 addr:$src)))]>, VEX;
1733 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1734 "cvttps2dq\t{$src, $dst|$dst, $src}",
1736 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1737 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1738 "cvttps2dq\t{$src, $dst|$dst, $src}",
1740 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1742 let Predicates = [HasAVX] in {
1743 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1744 (Int_VCVTDQ2PSrr VR128:$src)>;
1745 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1746 (Int_VCVTDQ2PSrm addr:$src)>;
1748 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1749 (VCVTTPS2DQrr VR128:$src)>;
1750 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1751 (VCVTTPS2DQrm addr:$src)>;
1753 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1754 (VCVTDQ2PSYrr VR256:$src)>;
1755 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1756 (VCVTDQ2PSYrm addr:$src)>;
1758 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1759 (VCVTTPS2DQYrr VR256:$src)>;
1760 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1761 (VCVTTPS2DQYrm addr:$src)>;
1764 let Predicates = [HasSSE2] in {
1765 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1766 (Int_CVTDQ2PSrr VR128:$src)>;
1767 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1768 (Int_CVTDQ2PSrm addr:$src)>;
1770 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1771 (CVTTPS2DQrr VR128:$src)>;
1772 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1773 (CVTTPS2DQrm addr:$src)>;
1776 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1777 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1779 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1780 let isCodeGenOnly = 1 in
1781 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1782 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1783 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1784 (memop addr:$src)))]>, VEX;
1785 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1786 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1787 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1788 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1789 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1790 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1791 (memop addr:$src)))]>;
1793 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1794 // register, but the same isn't true when using memory operands instead.
1795 // Provide other assembly rr and rm forms to address this explicitly.
1796 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1797 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1800 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1801 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1802 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1803 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1806 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1807 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1808 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1809 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1811 // Convert packed single to packed double
1812 let Predicates = [HasAVX] in {
1813 // SSE2 instructions without OpSize prefix
1814 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1815 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1816 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1817 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1818 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1819 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1820 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1821 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1823 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1824 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1825 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1826 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1828 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1829 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1830 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1831 TB, VEX, Requires<[HasAVX]>;
1832 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1833 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1834 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1835 (load addr:$src)))]>,
1836 TB, VEX, Requires<[HasAVX]>;
1837 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1838 "cvtps2pd\t{$src, $dst|$dst, $src}",
1839 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1840 TB, Requires<[HasSSE2]>;
1841 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1842 "cvtps2pd\t{$src, $dst|$dst, $src}",
1843 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1844 (load addr:$src)))]>,
1845 TB, Requires<[HasSSE2]>;
1847 // Convert packed double to packed single
1848 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1849 // register, but the same isn't true when using memory operands instead.
1850 // Provide other assembly rr and rm forms to address this explicitly.
1851 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1852 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1853 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1854 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1857 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1858 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1859 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1860 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1863 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1864 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1865 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1866 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1867 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1868 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1869 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1870 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1873 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1874 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1875 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1876 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1878 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1879 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1880 (memop addr:$src)))]>;
1881 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1882 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1883 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1884 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1885 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1886 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1887 (memop addr:$src)))]>;
1889 // AVX 256-bit register conversion intrinsics
1890 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1891 // whenever possible to avoid declaring two versions of each one.
1892 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1893 (VCVTDQ2PSYrr VR256:$src)>;
1894 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
1895 (VCVTDQ2PSYrm addr:$src)>;
1897 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1898 (VCVTPD2PSYrr VR256:$src)>;
1899 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1900 (VCVTPD2PSYrm addr:$src)>;
1902 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1903 (VCVTPS2DQYrr VR256:$src)>;
1904 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1905 (VCVTPS2DQYrm addr:$src)>;
1907 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1908 (VCVTPS2PDYrr VR128:$src)>;
1909 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1910 (VCVTPS2PDYrm addr:$src)>;
1912 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1913 (VCVTTPD2DQYrr VR256:$src)>;
1914 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1915 (VCVTTPD2DQYrm addr:$src)>;
1917 // Match fround and fextend for 128/256-bit conversions
1918 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1919 (VCVTPD2PSYrr VR256:$src)>;
1920 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1921 (VCVTPD2PSYrm addr:$src)>;
1923 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1924 (VCVTPS2PDYrr VR128:$src)>;
1925 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1926 (VCVTPS2PDYrm addr:$src)>;
1928 //===----------------------------------------------------------------------===//
1929 // SSE 1 & 2 - Compare Instructions
1930 //===----------------------------------------------------------------------===//
1932 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1933 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1934 SDNode OpNode, ValueType VT, PatFrag ld_frag,
1935 string asm, string asm_alt> {
1936 def rr : SIi8<0xC2, MRMSrcReg,
1937 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
1938 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
1939 def rm : SIi8<0xC2, MRMSrcMem,
1940 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
1941 [(set RC:$dst, (OpNode (VT RC:$src1),
1942 (ld_frag addr:$src2), imm:$cc))]>;
1944 // Accept explicit immediate argument form instead of comparison code.
1945 let neverHasSideEffects = 1 in {
1946 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
1947 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
1949 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
1950 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
1954 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
1955 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1956 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
1957 XS, VEX_4V, VEX_LIG;
1958 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
1959 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1960 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
1961 XD, VEX_4V, VEX_LIG;
1963 let Constraints = "$src1 = $dst" in {
1964 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
1965 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1966 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
1968 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
1969 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1970 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
1974 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1975 Intrinsic Int, string asm> {
1976 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1977 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1978 [(set VR128:$dst, (Int VR128:$src1,
1979 VR128:$src, imm:$cc))]>;
1980 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1981 (ins VR128:$src1, x86memop:$src, SSECC:$cc), asm,
1982 [(set VR128:$dst, (Int VR128:$src1,
1983 (load addr:$src), imm:$cc))]>;
1986 // Aliases to match intrinsics which expect XMM operand(s).
1987 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1988 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1990 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1991 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1993 let Constraints = "$src1 = $dst" in {
1994 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1995 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1996 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1997 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2001 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2002 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2003 ValueType vt, X86MemOperand x86memop,
2004 PatFrag ld_frag, string OpcodeStr, Domain d> {
2005 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2006 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2007 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2009 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2010 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2011 [(set EFLAGS, (OpNode (vt RC:$src1),
2012 (ld_frag addr:$src2)))],
2016 let Defs = [EFLAGS] in {
2017 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2018 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2019 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2020 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2022 let Pattern = []<dag> in {
2023 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2024 "comiss", SSEPackedSingle>, TB, VEX,
2026 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2027 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2031 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2032 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2033 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2034 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2036 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2037 load, "comiss", SSEPackedSingle>, TB, VEX;
2038 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2039 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2040 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2041 "ucomiss", SSEPackedSingle>, TB;
2042 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2043 "ucomisd", SSEPackedDouble>, TB, OpSize;
2045 let Pattern = []<dag> in {
2046 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2047 "comiss", SSEPackedSingle>, TB;
2048 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2049 "comisd", SSEPackedDouble>, TB, OpSize;
2052 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2053 load, "ucomiss", SSEPackedSingle>, TB;
2054 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2055 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2057 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2058 "comiss", SSEPackedSingle>, TB;
2059 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2060 "comisd", SSEPackedDouble>, TB, OpSize;
2061 } // Defs = [EFLAGS]
2063 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2064 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2065 Intrinsic Int, string asm, string asm_alt,
2067 let isAsmParserOnly = 1 in {
2068 def rri : PIi8<0xC2, MRMSrcReg,
2069 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2070 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2072 def rmi : PIi8<0xC2, MRMSrcMem,
2073 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2074 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2078 // Accept explicit immediate argument form instead of comparison code.
2079 def rri_alt : PIi8<0xC2, MRMSrcReg,
2080 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2081 asm_alt, [], IIC_DEFAULT, d>;
2082 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2083 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2084 asm_alt, [], IIC_DEFAULT, d>;
2087 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2088 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2089 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2090 SSEPackedSingle>, TB, VEX_4V;
2091 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2092 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2093 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2094 SSEPackedDouble>, TB, OpSize, VEX_4V;
2095 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2096 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2097 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2098 SSEPackedSingle>, TB, VEX_4V;
2099 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2100 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2101 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2102 SSEPackedDouble>, TB, OpSize, VEX_4V;
2103 let Constraints = "$src1 = $dst" in {
2104 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2105 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2106 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2107 SSEPackedSingle>, TB;
2108 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2109 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2110 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2111 SSEPackedDouble>, TB, OpSize;
2114 let Predicates = [HasAVX] in {
2115 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2116 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2117 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2118 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2119 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2120 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2121 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2122 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2124 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2125 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2126 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2127 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2128 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2129 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2130 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2131 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2134 let Predicates = [HasSSE1] in {
2135 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2136 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2137 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2138 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2141 let Predicates = [HasSSE2] in {
2142 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2143 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2144 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2145 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2148 //===----------------------------------------------------------------------===//
2149 // SSE 1 & 2 - Shuffle Instructions
2150 //===----------------------------------------------------------------------===//
2152 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2153 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2154 ValueType vt, string asm, PatFrag mem_frag,
2155 Domain d, bit IsConvertibleToThreeAddress = 0> {
2156 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2157 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2158 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2159 (i8 imm:$src3))))], IIC_DEFAULT, d>;
2160 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2161 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2162 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2163 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2164 (i8 imm:$src3))))], IIC_DEFAULT, d>;
2167 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2168 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2169 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2170 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2171 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2172 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2173 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2174 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2175 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2176 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2177 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2178 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2180 let Constraints = "$src1 = $dst" in {
2181 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2182 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2183 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2185 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2186 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2187 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2191 let Predicates = [HasAVX] in {
2192 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2193 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2194 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2195 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2196 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2198 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2199 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2200 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2201 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2202 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2205 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2206 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2207 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2208 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2209 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2211 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2212 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2213 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2214 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2215 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2218 let Predicates = [HasSSE1] in {
2219 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2220 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2221 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2222 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2223 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2226 let Predicates = [HasSSE2] in {
2227 // Generic SHUFPD patterns
2228 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2229 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2230 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2231 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2232 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2235 //===----------------------------------------------------------------------===//
2236 // SSE 1 & 2 - Unpack Instructions
2237 //===----------------------------------------------------------------------===//
2239 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2240 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2241 PatFrag mem_frag, RegisterClass RC,
2242 X86MemOperand x86memop, string asm,
2244 def rr : PI<opc, MRMSrcReg,
2245 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2247 (vt (OpNode RC:$src1, RC:$src2)))],
2249 def rm : PI<opc, MRMSrcMem,
2250 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2252 (vt (OpNode RC:$src1,
2253 (mem_frag addr:$src2))))],
2257 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2258 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2259 SSEPackedSingle>, TB, VEX_4V;
2260 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2261 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2262 SSEPackedDouble>, TB, OpSize, VEX_4V;
2263 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2264 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2265 SSEPackedSingle>, TB, VEX_4V;
2266 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2267 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2268 SSEPackedDouble>, TB, OpSize, VEX_4V;
2270 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2271 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2272 SSEPackedSingle>, TB, VEX_4V;
2273 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2274 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2275 SSEPackedDouble>, TB, OpSize, VEX_4V;
2276 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2277 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2278 SSEPackedSingle>, TB, VEX_4V;
2279 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2280 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2281 SSEPackedDouble>, TB, OpSize, VEX_4V;
2283 let Constraints = "$src1 = $dst" in {
2284 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2285 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2286 SSEPackedSingle>, TB;
2287 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2288 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2289 SSEPackedDouble>, TB, OpSize;
2290 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2291 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2292 SSEPackedSingle>, TB;
2293 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2294 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2295 SSEPackedDouble>, TB, OpSize;
2296 } // Constraints = "$src1 = $dst"
2298 let Predicates = [HasAVX], AddedComplexity = 1 in {
2299 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2300 // problem is during lowering, where it's not possible to recognize the load
2301 // fold cause it has two uses through a bitcast. One use disappears at isel
2302 // time and the fold opportunity reappears.
2303 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2304 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2307 let Predicates = [HasSSE2] in {
2308 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2309 // problem is during lowering, where it's not possible to recognize the load
2310 // fold cause it has two uses through a bitcast. One use disappears at isel
2311 // time and the fold opportunity reappears.
2312 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2313 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2316 //===----------------------------------------------------------------------===//
2317 // SSE 1 & 2 - Extract Floating-Point Sign mask
2318 //===----------------------------------------------------------------------===//
2320 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2321 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2323 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2324 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2325 [(set GR32:$dst, (Int RC:$src))], IIC_DEFAULT, d>;
2326 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2327 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2328 IIC_DEFAULT, d>, REX_W;
2331 let Predicates = [HasAVX] in {
2332 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2333 "movmskps", SSEPackedSingle>, TB, VEX;
2334 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2335 "movmskpd", SSEPackedDouble>, TB,
2337 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2338 "movmskps", SSEPackedSingle>, TB, VEX;
2339 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2340 "movmskpd", SSEPackedDouble>, TB,
2343 def : Pat<(i32 (X86fgetsign FR32:$src)),
2344 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2346 def : Pat<(i64 (X86fgetsign FR32:$src)),
2347 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2349 def : Pat<(i32 (X86fgetsign FR64:$src)),
2350 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2352 def : Pat<(i64 (X86fgetsign FR64:$src)),
2353 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2357 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2358 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2359 SSEPackedSingle>, TB, VEX;
2360 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2361 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2362 SSEPackedDouble>, TB,
2364 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2365 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2366 SSEPackedSingle>, TB, VEX;
2367 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2368 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_DEFAULT,
2369 SSEPackedDouble>, TB,
2373 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2374 SSEPackedSingle>, TB;
2375 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2376 SSEPackedDouble>, TB, OpSize;
2378 def : Pat<(i32 (X86fgetsign FR32:$src)),
2379 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2380 sub_ss))>, Requires<[HasSSE1]>;
2381 def : Pat<(i64 (X86fgetsign FR32:$src)),
2382 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2383 sub_ss))>, Requires<[HasSSE1]>;
2384 def : Pat<(i32 (X86fgetsign FR64:$src)),
2385 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2386 sub_sd))>, Requires<[HasSSE2]>;
2387 def : Pat<(i64 (X86fgetsign FR64:$src)),
2388 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2389 sub_sd))>, Requires<[HasSSE2]>;
2391 //===---------------------------------------------------------------------===//
2392 // SSE2 - Packed Integer Logical Instructions
2393 //===---------------------------------------------------------------------===//
2395 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2397 /// PDI_binop_rm - Simple SSE2 binary operator.
2398 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2399 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2400 X86MemOperand x86memop, bit IsCommutable = 0,
2402 let isCommutable = IsCommutable in
2403 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2404 (ins RC:$src1, RC:$src2),
2406 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2407 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2408 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2409 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2410 (ins RC:$src1, x86memop:$src2),
2412 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2413 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2414 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2415 (bitconvert (memop_frag addr:$src2)))))]>;
2417 } // ExeDomain = SSEPackedInt
2419 // These are ordered here for pattern ordering requirements with the fp versions
2421 let Predicates = [HasAVX] in {
2422 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2423 i128mem, 1, 0>, VEX_4V;
2424 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2425 i128mem, 1, 0>, VEX_4V;
2426 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2427 i128mem, 1, 0>, VEX_4V;
2428 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2429 i128mem, 0, 0>, VEX_4V;
2432 let Constraints = "$src1 = $dst" in {
2433 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2435 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2437 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2439 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2441 } // Constraints = "$src1 = $dst"
2443 let Predicates = [HasAVX2] in {
2444 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2445 i256mem, 1, 0>, VEX_4V;
2446 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2447 i256mem, 1, 0>, VEX_4V;
2448 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2449 i256mem, 1, 0>, VEX_4V;
2450 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2451 i256mem, 0, 0>, VEX_4V;
2454 //===----------------------------------------------------------------------===//
2455 // SSE 1 & 2 - Logical Instructions
2456 //===----------------------------------------------------------------------===//
2458 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2460 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2462 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2463 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2465 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2466 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2468 let Constraints = "$src1 = $dst" in {
2469 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2470 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2472 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2473 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2477 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2478 let mayLoad = 0 in {
2479 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2480 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2481 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2484 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2485 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2487 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2489 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2491 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2492 // are all promoted to v2i64, and the patterns are covered by the int
2493 // version. This is needed in SSE only, because v2i64 isn't supported on
2494 // SSE1, but only on SSE2.
2495 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2496 !strconcat(OpcodeStr, "ps"), f128mem, [],
2497 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2498 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2500 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2501 !strconcat(OpcodeStr, "pd"), f128mem,
2502 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2503 (bc_v2i64 (v2f64 VR128:$src2))))],
2504 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2505 (memopv2i64 addr:$src2)))], 0>,
2507 let Constraints = "$src1 = $dst" in {
2508 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2509 !strconcat(OpcodeStr, "ps"), f128mem,
2510 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2511 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2512 (memopv2i64 addr:$src2)))]>, TB;
2514 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2515 !strconcat(OpcodeStr, "pd"), f128mem,
2516 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2517 (bc_v2i64 (v2f64 VR128:$src2))))],
2518 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2519 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2523 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2525 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2527 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2528 !strconcat(OpcodeStr, "ps"), f256mem,
2529 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2530 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2531 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2533 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2534 !strconcat(OpcodeStr, "pd"), f256mem,
2535 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2536 (bc_v4i64 (v4f64 VR256:$src2))))],
2537 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2538 (memopv4i64 addr:$src2)))], 0>,
2542 // AVX 256-bit packed logical ops forms
2543 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2544 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2545 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2546 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2548 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2549 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2550 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2551 let isCommutable = 0 in
2552 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2554 //===----------------------------------------------------------------------===//
2555 // SSE 1 & 2 - Arithmetic Instructions
2556 //===----------------------------------------------------------------------===//
2558 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2561 /// In addition, we also have a special variant of the scalar form here to
2562 /// represent the associated intrinsic operation. This form is unlike the
2563 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2564 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2566 /// These three forms can each be reg+reg or reg+mem.
2569 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2571 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2573 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2574 OpNode, FR32, f32mem, Is2Addr>, XS;
2575 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2576 OpNode, FR64, f64mem, Is2Addr>, XD;
2579 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2581 let mayLoad = 0 in {
2582 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2583 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2584 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2585 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2589 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2591 let mayLoad = 0 in {
2592 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2593 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2594 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2595 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2599 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2601 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2602 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2603 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2604 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2607 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2609 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2610 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2611 SSEPackedSingle, Is2Addr>, TB;
2613 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2614 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2615 SSEPackedDouble, Is2Addr>, TB, OpSize;
2618 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2619 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2620 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2621 SSEPackedSingle, 0>, TB;
2623 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2624 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2625 SSEPackedDouble, 0>, TB, OpSize;
2628 // Binary Arithmetic instructions
2629 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2630 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2631 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2632 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2633 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2634 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2635 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2636 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2638 let isCommutable = 0 in {
2639 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2640 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2641 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2642 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2643 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2644 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2645 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2646 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2647 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2648 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2649 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2650 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2651 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2652 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2653 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2654 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2655 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2656 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2657 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2658 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2661 let Constraints = "$src1 = $dst" in {
2662 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2663 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2664 basic_sse12_fp_binop_s_int<0x58, "add">;
2665 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2666 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2667 basic_sse12_fp_binop_s_int<0x59, "mul">;
2669 let isCommutable = 0 in {
2670 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2671 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2672 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2673 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2674 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2675 basic_sse12_fp_binop_s_int<0x5E, "div">;
2676 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2677 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2678 basic_sse12_fp_binop_s_int<0x5F, "max">,
2679 basic_sse12_fp_binop_p_int<0x5F, "max">;
2680 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2681 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2682 basic_sse12_fp_binop_s_int<0x5D, "min">,
2683 basic_sse12_fp_binop_p_int<0x5D, "min">;
2688 /// In addition, we also have a special variant of the scalar form here to
2689 /// represent the associated intrinsic operation. This form is unlike the
2690 /// plain scalar form, in that it takes an entire vector (instead of a
2691 /// scalar) and leaves the top elements undefined.
2693 /// And, we have a special variant form for a full-vector intrinsic form.
2695 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2696 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2697 SDNode OpNode, Intrinsic F32Int> {
2698 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2699 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2700 [(set FR32:$dst, (OpNode FR32:$src))]>;
2701 // For scalar unary operations, fold a load into the operation
2702 // only in OptForSize mode. It eliminates an instruction, but it also
2703 // eliminates a whole-register clobber (the load), so it introduces a
2704 // partial register update condition.
2705 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2706 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2707 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2708 Requires<[HasSSE1, OptForSize]>;
2709 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2710 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2711 [(set VR128:$dst, (F32Int VR128:$src))]>;
2712 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2713 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2714 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2717 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2718 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2719 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2720 !strconcat(OpcodeStr,
2721 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2723 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2724 !strconcat(OpcodeStr,
2725 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2726 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2727 (ins VR128:$src1, ssmem:$src2),
2728 !strconcat(OpcodeStr,
2729 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2732 /// sse1_fp_unop_p - SSE1 unops in packed form.
2733 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2734 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2735 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2736 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2737 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2738 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2739 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2742 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2743 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2744 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2745 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2746 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2747 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2748 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2749 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2752 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2753 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2754 Intrinsic V4F32Int> {
2755 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2756 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2757 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2758 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2759 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2760 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2763 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2764 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2765 Intrinsic V4F32Int> {
2766 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2767 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2768 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2769 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2770 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2771 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2774 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2775 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2776 SDNode OpNode, Intrinsic F64Int> {
2777 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2778 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2779 [(set FR64:$dst, (OpNode FR64:$src))]>;
2780 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2781 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2782 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2783 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2784 Requires<[HasSSE2, OptForSize]>;
2785 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2786 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2787 [(set VR128:$dst, (F64Int VR128:$src))]>;
2788 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2789 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2790 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2793 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2794 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2795 let neverHasSideEffects = 1 in {
2796 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2797 !strconcat(OpcodeStr,
2798 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2800 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2801 !strconcat(OpcodeStr,
2802 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2804 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2805 (ins VR128:$src1, sdmem:$src2),
2806 !strconcat(OpcodeStr,
2807 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2810 /// sse2_fp_unop_p - SSE2 unops in vector forms.
2811 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
2813 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2814 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2815 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
2816 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2817 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2818 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
2821 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
2822 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2823 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2824 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2825 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
2826 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2827 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2828 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
2831 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
2832 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2833 Intrinsic V2F64Int> {
2834 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2835 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2836 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
2837 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2838 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2839 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
2842 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
2843 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2844 Intrinsic V2F64Int> {
2845 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2846 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2847 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
2848 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2849 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2850 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
2853 let Predicates = [HasAVX] in {
2855 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
2856 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
2858 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
2859 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
2860 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2861 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2862 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
2863 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
2864 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
2865 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
2868 // Reciprocal approximations. Note that these typically require refinement
2869 // in order to obtain suitable precision.
2870 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
2871 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
2872 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
2873 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
2874 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
2876 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
2877 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
2878 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
2879 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
2880 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
2883 let AddedComplexity = 1 in {
2884 def : Pat<(f32 (fsqrt FR32:$src)),
2885 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2886 def : Pat<(f32 (fsqrt (load addr:$src))),
2887 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2888 Requires<[HasAVX, OptForSize]>;
2889 def : Pat<(f64 (fsqrt FR64:$src)),
2890 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
2891 def : Pat<(f64 (fsqrt (load addr:$src))),
2892 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
2893 Requires<[HasAVX, OptForSize]>;
2895 def : Pat<(f32 (X86frsqrt FR32:$src)),
2896 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2897 def : Pat<(f32 (X86frsqrt (load addr:$src))),
2898 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2899 Requires<[HasAVX, OptForSize]>;
2901 def : Pat<(f32 (X86frcp FR32:$src)),
2902 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2903 def : Pat<(f32 (X86frcp (load addr:$src))),
2904 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2905 Requires<[HasAVX, OptForSize]>;
2908 let Predicates = [HasAVX], AddedComplexity = 1 in {
2909 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
2910 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2911 (VSQRTSSr (f32 (IMPLICIT_DEF)),
2912 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2914 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
2915 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2917 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
2918 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
2919 (VSQRTSDr (f64 (IMPLICIT_DEF)),
2920 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
2922 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
2923 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
2925 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
2926 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2927 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
2928 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2930 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
2931 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2933 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
2934 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2935 (VRCPSSr (f32 (IMPLICIT_DEF)),
2936 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2938 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
2939 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2943 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
2944 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
2945 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
2946 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
2947 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
2948 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
2950 // Reciprocal approximations. Note that these typically require refinement
2951 // in order to obtain suitable precision.
2952 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
2953 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
2954 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
2955 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
2956 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
2957 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
2959 // There is no f64 version of the reciprocal approximation instructions.
2961 //===----------------------------------------------------------------------===//
2962 // SSE 1 & 2 - Non-temporal stores
2963 //===----------------------------------------------------------------------===//
2965 let AddedComplexity = 400 in { // Prefer non-temporal versions
2966 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
2967 (ins f128mem:$dst, VR128:$src),
2968 "movntps\t{$src, $dst|$dst, $src}",
2969 [(alignednontemporalstore (v4f32 VR128:$src),
2971 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
2972 (ins f128mem:$dst, VR128:$src),
2973 "movntpd\t{$src, $dst|$dst, $src}",
2974 [(alignednontemporalstore (v2f64 VR128:$src),
2977 let ExeDomain = SSEPackedInt in
2978 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
2979 (ins f128mem:$dst, VR128:$src),
2980 "movntdq\t{$src, $dst|$dst, $src}",
2981 [(alignednontemporalstore (v2i64 VR128:$src),
2984 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2985 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
2987 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
2988 (ins f256mem:$dst, VR256:$src),
2989 "movntps\t{$src, $dst|$dst, $src}",
2990 [(alignednontemporalstore (v8f32 VR256:$src),
2992 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2993 (ins f256mem:$dst, VR256:$src),
2994 "movntpd\t{$src, $dst|$dst, $src}",
2995 [(alignednontemporalstore (v4f64 VR256:$src),
2997 let ExeDomain = SSEPackedInt in
2998 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2999 (ins f256mem:$dst, VR256:$src),
3000 "movntdq\t{$src, $dst|$dst, $src}",
3001 [(alignednontemporalstore (v4i64 VR256:$src),
3005 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3006 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3007 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3008 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3009 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3010 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3012 let AddedComplexity = 400 in { // Prefer non-temporal versions
3013 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3014 "movntps\t{$src, $dst|$dst, $src}",
3015 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3016 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3017 "movntpd\t{$src, $dst|$dst, $src}",
3018 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3020 let ExeDomain = SSEPackedInt in
3021 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3022 "movntdq\t{$src, $dst|$dst, $src}",
3023 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)]>;
3025 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3026 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3028 // There is no AVX form for instructions below this point
3029 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3030 "movnti{l}\t{$src, $dst|$dst, $src}",
3031 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3032 TB, Requires<[HasSSE2]>;
3033 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3034 "movnti{q}\t{$src, $dst|$dst, $src}",
3035 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3036 TB, Requires<[HasSSE2]>;
3039 //===----------------------------------------------------------------------===//
3040 // SSE 1 & 2 - Prefetch and memory fence
3041 //===----------------------------------------------------------------------===//
3043 // Prefetch intrinsic.
3044 let Predicates = [HasSSE1] in {
3045 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3046 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3047 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3048 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3049 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3050 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3051 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3052 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3056 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3057 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3058 TB, Requires<[HasSSE2]>;
3060 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3061 // was introduced with SSE2, it's backward compatible.
3062 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3064 // Load, store, and memory fence
3065 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3066 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3067 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3068 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3069 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3070 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3072 def : Pat<(X86SFence), (SFENCE)>;
3073 def : Pat<(X86LFence), (LFENCE)>;
3074 def : Pat<(X86MFence), (MFENCE)>;
3076 //===----------------------------------------------------------------------===//
3077 // SSE 1 & 2 - Load/Store XCSR register
3078 //===----------------------------------------------------------------------===//
3080 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3081 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3082 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3083 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3085 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3086 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3087 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3088 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3090 //===---------------------------------------------------------------------===//
3091 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3092 //===---------------------------------------------------------------------===//
3094 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3096 let neverHasSideEffects = 1 in {
3097 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3098 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3099 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3100 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3102 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3103 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3104 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3105 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3108 let isCodeGenOnly = 1 in {
3109 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3110 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3111 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3112 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3113 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3114 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3115 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3116 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3119 let canFoldAsLoad = 1, mayLoad = 1 in {
3120 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3121 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3122 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3123 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3124 let Predicates = [HasAVX] in {
3125 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3126 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3127 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3128 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3132 let mayStore = 1 in {
3133 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3134 (ins i128mem:$dst, VR128:$src),
3135 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3136 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3137 (ins i256mem:$dst, VR256:$src),
3138 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3139 let Predicates = [HasAVX] in {
3140 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3141 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3142 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3143 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3147 let neverHasSideEffects = 1 in
3148 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3149 "movdqa\t{$src, $dst|$dst, $src}", []>;
3151 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3152 "movdqu\t{$src, $dst|$dst, $src}",
3153 []>, XS, Requires<[HasSSE2]>;
3156 let isCodeGenOnly = 1 in {
3157 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3158 "movdqa\t{$src, $dst|$dst, $src}", []>;
3160 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3161 "movdqu\t{$src, $dst|$dst, $src}",
3162 []>, XS, Requires<[HasSSE2]>;
3165 let canFoldAsLoad = 1, mayLoad = 1 in {
3166 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3167 "movdqa\t{$src, $dst|$dst, $src}",
3168 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3169 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3170 "movdqu\t{$src, $dst|$dst, $src}",
3171 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3172 XS, Requires<[HasSSE2]>;
3175 let mayStore = 1 in {
3176 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3177 "movdqa\t{$src, $dst|$dst, $src}",
3178 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3179 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3180 "movdqu\t{$src, $dst|$dst, $src}",
3181 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3182 XS, Requires<[HasSSE2]>;
3185 // Intrinsic forms of MOVDQU load and store
3186 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3187 "vmovdqu\t{$src, $dst|$dst, $src}",
3188 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3189 XS, VEX, Requires<[HasAVX]>;
3191 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3192 "movdqu\t{$src, $dst|$dst, $src}",
3193 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3194 XS, Requires<[HasSSE2]>;
3196 } // ExeDomain = SSEPackedInt
3198 let Predicates = [HasAVX] in {
3199 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3200 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3203 //===---------------------------------------------------------------------===//
3204 // SSE2 - Packed Integer Arithmetic Instructions
3205 //===---------------------------------------------------------------------===//
3207 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3209 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3210 RegisterClass RC, PatFrag memop_frag,
3211 X86MemOperand x86memop, bit IsCommutable = 0,
3213 let isCommutable = IsCommutable in
3214 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3215 (ins RC:$src1, RC:$src2),
3217 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3218 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3219 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3220 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3221 (ins RC:$src1, x86memop:$src2),
3223 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3224 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3225 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3228 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3229 string OpcodeStr, SDNode OpNode,
3230 SDNode OpNode2, RegisterClass RC,
3231 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3233 // src2 is always 128-bit
3234 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3235 (ins RC:$src1, VR128:$src2),
3237 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3238 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3239 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))]>;
3240 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3241 (ins RC:$src1, i128mem:$src2),
3243 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3244 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3245 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3246 (bc_frag (memopv2i64 addr:$src2)))))]>;
3247 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3248 (ins RC:$src1, i32i8imm:$src2),
3250 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3251 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3252 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))]>;
3255 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3256 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3257 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3258 PatFrag memop_frag, X86MemOperand x86memop,
3259 bit IsCommutable = 0, bit Is2Addr = 1> {
3260 let isCommutable = IsCommutable in
3261 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3262 (ins RC:$src1, RC:$src2),
3264 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3265 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3266 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3267 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3268 (ins RC:$src1, x86memop:$src2),
3270 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3271 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3272 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3273 (bitconvert (memop_frag addr:$src2)))))]>;
3275 } // ExeDomain = SSEPackedInt
3277 // 128-bit Integer Arithmetic
3279 let Predicates = [HasAVX] in {
3280 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3281 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3282 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3283 i128mem, 1, 0>, VEX_4V;
3284 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3285 i128mem, 1, 0>, VEX_4V;
3286 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3287 i128mem, 1, 0>, VEX_4V;
3288 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3289 i128mem, 1, 0>, VEX_4V;
3290 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3291 i128mem, 0, 0>, VEX_4V;
3292 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3293 i128mem, 0, 0>, VEX_4V;
3294 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3295 i128mem, 0, 0>, VEX_4V;
3296 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3297 i128mem, 0, 0>, VEX_4V;
3298 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3299 memopv2i64, i128mem, 1, 0>, VEX_4V;
3302 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3303 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3304 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3305 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3306 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3307 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3308 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3309 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3310 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3311 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3312 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3313 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3314 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3315 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3316 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3317 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3318 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3319 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3320 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3321 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3322 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3323 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3324 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3325 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3326 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3327 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3328 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3329 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3330 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3331 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3332 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3333 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3334 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3335 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3336 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3337 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3340 let Predicates = [HasAVX2] in {
3341 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3342 i256mem, 1, 0>, VEX_4V;
3343 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3344 i256mem, 1, 0>, VEX_4V;
3345 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3346 i256mem, 1, 0>, VEX_4V;
3347 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3348 i256mem, 1, 0>, VEX_4V;
3349 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3350 i256mem, 1, 0>, VEX_4V;
3351 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3352 i256mem, 0, 0>, VEX_4V;
3353 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3354 i256mem, 0, 0>, VEX_4V;
3355 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3356 i256mem, 0, 0>, VEX_4V;
3357 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3358 i256mem, 0, 0>, VEX_4V;
3359 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3360 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3363 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3364 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3365 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3366 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3367 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3368 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3369 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3370 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3371 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3372 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3373 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3374 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3375 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3376 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3377 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3378 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3379 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3380 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3381 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3382 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3383 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3384 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3385 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3386 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3387 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3388 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3389 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3390 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3391 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3392 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3393 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3394 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3395 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3396 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3397 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3398 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3401 let Constraints = "$src1 = $dst" in {
3402 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3404 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3406 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3408 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3410 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3412 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3414 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3416 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3418 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3420 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3421 memopv2i64, i128mem, 1>;
3424 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3425 VR128, memopv2i64, i128mem>;
3426 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3427 VR128, memopv2i64, i128mem>;
3428 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3429 VR128, memopv2i64, i128mem>;
3430 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3431 VR128, memopv2i64, i128mem>;
3432 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3433 VR128, memopv2i64, i128mem, 1>;
3434 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3435 VR128, memopv2i64, i128mem, 1>;
3436 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3437 VR128, memopv2i64, i128mem, 1>;
3438 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3439 VR128, memopv2i64, i128mem, 1>;
3440 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3441 VR128, memopv2i64, i128mem, 1>;
3442 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3443 VR128, memopv2i64, i128mem, 1>;
3444 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3445 VR128, memopv2i64, i128mem, 1>;
3446 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3447 VR128, memopv2i64, i128mem, 1>;
3448 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3449 VR128, memopv2i64, i128mem, 1>;
3450 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3451 VR128, memopv2i64, i128mem, 1>;
3452 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3453 VR128, memopv2i64, i128mem, 1>;
3454 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3455 VR128, memopv2i64, i128mem, 1>;
3456 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3457 VR128, memopv2i64, i128mem, 1>;
3458 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3459 VR128, memopv2i64, i128mem, 1>;
3461 } // Constraints = "$src1 = $dst"
3463 //===---------------------------------------------------------------------===//
3464 // SSE2 - Packed Integer Logical Instructions
3465 //===---------------------------------------------------------------------===//
3467 let Predicates = [HasAVX] in {
3468 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3469 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3470 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3471 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3472 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3473 VR128, v2i64, v2i64, bc_v2i64, 0>, VEX_4V;
3475 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3476 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3477 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3478 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3479 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3480 VR128, v2i64, v2i64, bc_v2i64, 0>, VEX_4V;
3482 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3483 VR128, v8i16, v8i16, bc_v8i16, 0>, VEX_4V;
3484 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3485 VR128, v4i32, v4i32, bc_v4i32, 0>, VEX_4V;
3487 let ExeDomain = SSEPackedInt in {
3488 // 128-bit logical shifts.
3489 def VPSLLDQri : PDIi8<0x73, MRM7r,
3490 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3491 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3493 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3495 def VPSRLDQri : PDIi8<0x73, MRM3r,
3496 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3497 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3499 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3501 // PSRADQri doesn't exist in SSE[1-3].
3503 } // Predicates = [HasAVX]
3505 let Predicates = [HasAVX2] in {
3506 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3507 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3508 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3509 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3510 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3511 VR256, v4i64, v2i64, bc_v2i64, 0>, VEX_4V;
3513 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3514 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3515 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3516 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3517 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3518 VR256, v4i64, v2i64, bc_v2i64, 0>, VEX_4V;
3520 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3521 VR256, v16i16, v8i16, bc_v8i16, 0>, VEX_4V;
3522 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3523 VR256, v8i32, v4i32, bc_v4i32, 0>, VEX_4V;
3525 let ExeDomain = SSEPackedInt in {
3526 // 256-bit logical shifts.
3527 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3528 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3529 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3531 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3533 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3534 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3535 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3537 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3539 // PSRADQYri doesn't exist in SSE[1-3].
3541 } // Predicates = [HasAVX2]
3543 let Constraints = "$src1 = $dst" in {
3544 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3545 VR128, v8i16, v8i16, bc_v8i16>;
3546 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3547 VR128, v4i32, v4i32, bc_v4i32>;
3548 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3549 VR128, v2i64, v2i64, bc_v2i64>;
3551 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3552 VR128, v8i16, v8i16, bc_v8i16>;
3553 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3554 VR128, v4i32, v4i32, bc_v4i32>;
3555 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3556 VR128, v2i64, v2i64, bc_v2i64>;
3558 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3559 VR128, v8i16, v8i16, bc_v8i16>;
3560 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3561 VR128, v4i32, v4i32, bc_v4i32>;
3563 let ExeDomain = SSEPackedInt in {
3564 // 128-bit logical shifts.
3565 def PSLLDQri : PDIi8<0x73, MRM7r,
3566 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3567 "pslldq\t{$src2, $dst|$dst, $src2}",
3569 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3570 def PSRLDQri : PDIi8<0x73, MRM3r,
3571 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3572 "psrldq\t{$src2, $dst|$dst, $src2}",
3574 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3575 // PSRADQri doesn't exist in SSE[1-3].
3577 } // Constraints = "$src1 = $dst"
3579 let Predicates = [HasAVX] in {
3580 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3581 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3582 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3583 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3584 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3585 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3587 // Shift up / down and insert zero's.
3588 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3589 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3590 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3591 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3594 let Predicates = [HasAVX2] in {
3595 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3596 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3597 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3598 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3601 let Predicates = [HasSSE2] in {
3602 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3603 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3604 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3605 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3606 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3607 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3609 // Shift up / down and insert zero's.
3610 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3611 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3612 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3613 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3616 //===---------------------------------------------------------------------===//
3617 // SSE2 - Packed Integer Comparison Instructions
3618 //===---------------------------------------------------------------------===//
3620 let Predicates = [HasAVX] in {
3621 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
3622 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3623 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
3624 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3625 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
3626 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3627 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
3628 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3629 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
3630 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3631 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
3632 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3635 let Predicates = [HasAVX2] in {
3636 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
3637 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3638 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
3639 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3640 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
3641 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3642 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
3643 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3644 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
3645 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3646 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
3647 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3650 let Constraints = "$src1 = $dst" in {
3651 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
3652 VR128, memopv2i64, i128mem, 1>;
3653 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
3654 VR128, memopv2i64, i128mem, 1>;
3655 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
3656 VR128, memopv2i64, i128mem, 1>;
3657 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
3658 VR128, memopv2i64, i128mem>;
3659 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
3660 VR128, memopv2i64, i128mem>;
3661 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
3662 VR128, memopv2i64, i128mem>;
3663 } // Constraints = "$src1 = $dst"
3665 //===---------------------------------------------------------------------===//
3666 // SSE2 - Packed Integer Pack Instructions
3667 //===---------------------------------------------------------------------===//
3669 let Predicates = [HasAVX] in {
3670 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3671 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3672 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3673 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3674 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3675 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3678 let Predicates = [HasAVX2] in {
3679 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
3680 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3681 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
3682 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3683 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
3684 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3687 let Constraints = "$src1 = $dst" in {
3688 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
3689 VR128, memopv2i64, i128mem>;
3690 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
3691 VR128, memopv2i64, i128mem>;
3692 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
3693 VR128, memopv2i64, i128mem>;
3694 } // Constraints = "$src1 = $dst"
3696 //===---------------------------------------------------------------------===//
3697 // SSE2 - Packed Integer Shuffle Instructions
3698 //===---------------------------------------------------------------------===//
3700 let ExeDomain = SSEPackedInt in {
3701 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
3702 def ri : Ii8<0x70, MRMSrcReg,
3703 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3704 !strconcat(OpcodeStr,
3705 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3706 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))]>;
3707 def mi : Ii8<0x70, MRMSrcMem,
3708 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3709 !strconcat(OpcodeStr,
3710 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3712 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
3713 (i8 imm:$src2))))]>;
3716 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
3717 def Yri : Ii8<0x70, MRMSrcReg,
3718 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
3719 !strconcat(OpcodeStr,
3720 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3721 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
3722 def Ymi : Ii8<0x70, MRMSrcMem,
3723 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
3724 !strconcat(OpcodeStr,
3725 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3727 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
3728 (i8 imm:$src2))))]>;
3730 } // ExeDomain = SSEPackedInt
3732 let Predicates = [HasAVX] in {
3733 let AddedComplexity = 5 in
3734 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
3736 // SSE2 with ImmT == Imm8 and XS prefix.
3737 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
3739 // SSE2 with ImmT == Imm8 and XD prefix.
3740 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
3742 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3743 (VPSHUFDmi addr:$src1, imm:$imm)>;
3744 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3745 (VPSHUFDri VR128:$src1, imm:$imm)>;
3748 let Predicates = [HasAVX2] in {
3749 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
3750 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
3751 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
3754 let Predicates = [HasSSE2] in {
3755 let AddedComplexity = 5 in
3756 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
3758 // SSE2 with ImmT == Imm8 and XS prefix.
3759 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
3761 // SSE2 with ImmT == Imm8 and XD prefix.
3762 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
3764 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3765 (PSHUFDmi addr:$src1, imm:$imm)>;
3766 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3767 (PSHUFDri VR128:$src1, imm:$imm)>;
3770 //===---------------------------------------------------------------------===//
3771 // SSE2 - Packed Integer Unpack Instructions
3772 //===---------------------------------------------------------------------===//
3774 let ExeDomain = SSEPackedInt in {
3775 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
3776 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
3777 def rr : PDI<opc, MRMSrcReg,
3778 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3780 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3781 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3782 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
3783 def rm : PDI<opc, MRMSrcMem,
3784 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3786 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3787 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3788 [(set VR128:$dst, (OpNode VR128:$src1,
3789 (bc_frag (memopv2i64
3793 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
3794 SDNode OpNode, PatFrag bc_frag> {
3795 def Yrr : PDI<opc, MRMSrcReg,
3796 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
3797 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3798 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
3799 def Yrm : PDI<opc, MRMSrcMem,
3800 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
3801 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3802 [(set VR256:$dst, (OpNode VR256:$src1,
3803 (bc_frag (memopv4i64 addr:$src2))))]>;
3806 let Predicates = [HasAVX] in {
3807 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
3808 bc_v16i8, 0>, VEX_4V;
3809 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
3810 bc_v8i16, 0>, VEX_4V;
3811 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
3812 bc_v4i32, 0>, VEX_4V;
3813 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
3814 bc_v2i64, 0>, VEX_4V;
3816 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
3817 bc_v16i8, 0>, VEX_4V;
3818 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
3819 bc_v8i16, 0>, VEX_4V;
3820 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
3821 bc_v4i32, 0>, VEX_4V;
3822 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
3823 bc_v2i64, 0>, VEX_4V;
3826 let Predicates = [HasAVX2] in {
3827 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
3829 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
3831 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
3833 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
3836 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
3838 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
3840 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
3842 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
3846 let Constraints = "$src1 = $dst" in {
3847 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
3849 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
3851 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
3853 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
3856 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
3858 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
3860 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
3862 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
3865 } // ExeDomain = SSEPackedInt
3867 // Patterns for using AVX1 instructions with integer vectors
3868 // Here to give AVX2 priority
3869 let Predicates = [HasAVX] in {
3870 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
3871 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
3872 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
3873 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
3874 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
3875 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
3876 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
3877 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
3879 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
3880 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
3881 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
3882 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
3883 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
3884 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
3885 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
3886 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
3889 //===---------------------------------------------------------------------===//
3890 // SSE2 - Packed Integer Extract and Insert
3891 //===---------------------------------------------------------------------===//
3893 let ExeDomain = SSEPackedInt in {
3894 multiclass sse2_pinsrw<bit Is2Addr = 1> {
3895 def rri : Ii8<0xC4, MRMSrcReg,
3896 (outs VR128:$dst), (ins VR128:$src1,
3897 GR32:$src2, i32i8imm:$src3),
3899 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3900 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3902 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
3903 def rmi : Ii8<0xC4, MRMSrcMem,
3904 (outs VR128:$dst), (ins VR128:$src1,
3905 i16mem:$src2, i32i8imm:$src3),
3907 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3908 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3910 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
3915 let Predicates = [HasAVX] in
3916 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
3917 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3918 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3919 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3920 imm:$src2))]>, TB, OpSize, VEX;
3921 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
3922 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3923 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3924 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3928 let Predicates = [HasAVX] in {
3929 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
3930 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
3931 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
3932 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
3933 []>, TB, OpSize, VEX_4V;
3936 let Constraints = "$src1 = $dst" in
3937 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
3939 } // ExeDomain = SSEPackedInt
3941 //===---------------------------------------------------------------------===//
3942 // SSE2 - Packed Mask Creation
3943 //===---------------------------------------------------------------------===//
3945 let ExeDomain = SSEPackedInt in {
3947 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3948 "pmovmskb\t{$src, $dst|$dst, $src}",
3949 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
3950 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
3951 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
3953 let Predicates = [HasAVX2] in {
3954 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
3955 "pmovmskb\t{$src, $dst|$dst, $src}",
3956 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
3957 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
3958 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
3961 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3962 "pmovmskb\t{$src, $dst|$dst, $src}",
3963 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
3965 } // ExeDomain = SSEPackedInt
3967 //===---------------------------------------------------------------------===//
3968 // SSE2 - Conditional Store
3969 //===---------------------------------------------------------------------===//
3971 let ExeDomain = SSEPackedInt in {
3974 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
3975 (ins VR128:$src, VR128:$mask),
3976 "maskmovdqu\t{$mask, $src|$src, $mask}",
3977 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
3979 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
3980 (ins VR128:$src, VR128:$mask),
3981 "maskmovdqu\t{$mask, $src|$src, $mask}",
3982 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
3985 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
3986 "maskmovdqu\t{$mask, $src|$src, $mask}",
3987 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
3989 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
3990 "maskmovdqu\t{$mask, $src|$src, $mask}",
3991 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
3993 } // ExeDomain = SSEPackedInt
3995 //===---------------------------------------------------------------------===//
3996 // SSE2 - Move Doubleword
3997 //===---------------------------------------------------------------------===//
3999 //===---------------------------------------------------------------------===//
4000 // Move Int Doubleword to Packed Double Int
4002 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4003 "movd\t{$src, $dst|$dst, $src}",
4005 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4006 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4007 "movd\t{$src, $dst|$dst, $src}",
4009 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4011 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4012 "mov{d|q}\t{$src, $dst|$dst, $src}",
4014 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4015 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4016 "mov{d|q}\t{$src, $dst|$dst, $src}",
4017 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4019 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4020 "movd\t{$src, $dst|$dst, $src}",
4022 (v4i32 (scalar_to_vector GR32:$src)))]>;
4023 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4024 "movd\t{$src, $dst|$dst, $src}",
4026 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4027 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4028 "mov{d|q}\t{$src, $dst|$dst, $src}",
4030 (v2i64 (scalar_to_vector GR64:$src)))]>;
4031 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4032 "mov{d|q}\t{$src, $dst|$dst, $src}",
4033 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4035 //===---------------------------------------------------------------------===//
4036 // Move Int Doubleword to Single Scalar
4038 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4039 "movd\t{$src, $dst|$dst, $src}",
4040 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4042 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4043 "movd\t{$src, $dst|$dst, $src}",
4044 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4046 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4047 "movd\t{$src, $dst|$dst, $src}",
4048 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4050 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4051 "movd\t{$src, $dst|$dst, $src}",
4052 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4054 //===---------------------------------------------------------------------===//
4055 // Move Packed Doubleword Int to Packed Double Int
4057 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4058 "movd\t{$src, $dst|$dst, $src}",
4059 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4061 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4062 (ins i32mem:$dst, VR128:$src),
4063 "movd\t{$src, $dst|$dst, $src}",
4064 [(store (i32 (vector_extract (v4i32 VR128:$src),
4065 (iPTR 0))), addr:$dst)]>, VEX;
4066 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4067 "movd\t{$src, $dst|$dst, $src}",
4068 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4070 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4071 "movd\t{$src, $dst|$dst, $src}",
4072 [(store (i32 (vector_extract (v4i32 VR128:$src),
4073 (iPTR 0))), addr:$dst)]>;
4075 //===---------------------------------------------------------------------===//
4076 // Move Packed Doubleword Int first element to Doubleword Int
4078 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4079 "mov{d|q}\t{$src, $dst|$dst, $src}",
4080 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4082 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4084 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4085 "mov{d|q}\t{$src, $dst|$dst, $src}",
4086 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4089 //===---------------------------------------------------------------------===//
4090 // Bitcast FR64 <-> GR64
4092 let Predicates = [HasAVX] in
4093 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4094 "vmovq\t{$src, $dst|$dst, $src}",
4095 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4097 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4098 "mov{d|q}\t{$src, $dst|$dst, $src}",
4099 [(set GR64:$dst, (bitconvert FR64:$src))]>, VEX;
4100 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4101 "movq\t{$src, $dst|$dst, $src}",
4102 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>,
4105 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4106 "movq\t{$src, $dst|$dst, $src}",
4107 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4108 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4109 "mov{d|q}\t{$src, $dst|$dst, $src}",
4110 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4111 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4112 "movq\t{$src, $dst|$dst, $src}",
4113 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4115 //===---------------------------------------------------------------------===//
4116 // Move Scalar Single to Double Int
4118 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4119 "movd\t{$src, $dst|$dst, $src}",
4120 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4121 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4122 "movd\t{$src, $dst|$dst, $src}",
4123 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4124 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4125 "movd\t{$src, $dst|$dst, $src}",
4126 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4127 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4128 "movd\t{$src, $dst|$dst, $src}",
4129 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4131 //===---------------------------------------------------------------------===//
4132 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4134 let AddedComplexity = 15 in {
4135 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4136 "movd\t{$src, $dst|$dst, $src}",
4137 [(set VR128:$dst, (v4i32 (X86vzmovl
4138 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4140 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4141 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4142 [(set VR128:$dst, (v2i64 (X86vzmovl
4143 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4146 let AddedComplexity = 15 in {
4147 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4148 "movd\t{$src, $dst|$dst, $src}",
4149 [(set VR128:$dst, (v4i32 (X86vzmovl
4150 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4151 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4152 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4153 [(set VR128:$dst, (v2i64 (X86vzmovl
4154 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4157 let AddedComplexity = 20 in {
4158 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4159 "movd\t{$src, $dst|$dst, $src}",
4161 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4162 (loadi32 addr:$src))))))]>,
4164 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4165 "movd\t{$src, $dst|$dst, $src}",
4167 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4168 (loadi32 addr:$src))))))]>;
4171 let Predicates = [HasAVX] in {
4172 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4173 let AddedComplexity = 20 in {
4174 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4175 (VMOVZDI2PDIrm addr:$src)>;
4176 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4177 (VMOVZDI2PDIrm addr:$src)>;
4179 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4180 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4181 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4182 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4183 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4184 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4185 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4188 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4189 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4190 (MOVZDI2PDIrm addr:$src)>;
4191 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4192 (MOVZDI2PDIrm addr:$src)>;
4195 // These are the correct encodings of the instructions so that we know how to
4196 // read correct assembly, even though we continue to emit the wrong ones for
4197 // compatibility with Darwin's buggy assembler.
4198 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4199 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4200 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4201 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4202 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4203 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4204 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4205 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4206 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4207 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4208 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4209 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4211 //===---------------------------------------------------------------------===//
4212 // SSE2 - Move Quadword
4213 //===---------------------------------------------------------------------===//
4215 //===---------------------------------------------------------------------===//
4216 // Move Quadword Int to Packed Quadword Int
4218 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4219 "vmovq\t{$src, $dst|$dst, $src}",
4221 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4222 VEX, Requires<[HasAVX]>;
4223 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4224 "movq\t{$src, $dst|$dst, $src}",
4226 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4227 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4229 //===---------------------------------------------------------------------===//
4230 // Move Packed Quadword Int to Quadword Int
4232 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4233 "movq\t{$src, $dst|$dst, $src}",
4234 [(store (i64 (vector_extract (v2i64 VR128:$src),
4235 (iPTR 0))), addr:$dst)]>, VEX;
4236 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4237 "movq\t{$src, $dst|$dst, $src}",
4238 [(store (i64 (vector_extract (v2i64 VR128:$src),
4239 (iPTR 0))), addr:$dst)]>;
4241 //===---------------------------------------------------------------------===//
4242 // Store / copy lower 64-bits of a XMM register.
4244 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4245 "movq\t{$src, $dst|$dst, $src}",
4246 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4247 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4248 "movq\t{$src, $dst|$dst, $src}",
4249 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4251 let AddedComplexity = 20 in
4252 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4253 "vmovq\t{$src, $dst|$dst, $src}",
4255 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4256 (loadi64 addr:$src))))))]>,
4257 XS, VEX, Requires<[HasAVX]>;
4259 let AddedComplexity = 20 in
4260 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4261 "movq\t{$src, $dst|$dst, $src}",
4263 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4264 (loadi64 addr:$src))))))]>,
4265 XS, Requires<[HasSSE2]>;
4267 let Predicates = [HasAVX], AddedComplexity = 20 in {
4268 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4269 (VMOVZQI2PQIrm addr:$src)>;
4270 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4271 (VMOVZQI2PQIrm addr:$src)>;
4272 def : Pat<(v2i64 (X86vzload addr:$src)),
4273 (VMOVZQI2PQIrm addr:$src)>;
4276 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4277 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4278 (MOVZQI2PQIrm addr:$src)>;
4279 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4280 (MOVZQI2PQIrm addr:$src)>;
4281 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4284 let Predicates = [HasAVX] in {
4285 def : Pat<(v4i64 (X86vzload addr:$src)),
4286 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4289 //===---------------------------------------------------------------------===//
4290 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4291 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4293 let AddedComplexity = 15 in
4294 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4295 "vmovq\t{$src, $dst|$dst, $src}",
4296 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4297 XS, VEX, Requires<[HasAVX]>;
4298 let AddedComplexity = 15 in
4299 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4300 "movq\t{$src, $dst|$dst, $src}",
4301 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4302 XS, Requires<[HasSSE2]>;
4304 let AddedComplexity = 20 in
4305 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4306 "vmovq\t{$src, $dst|$dst, $src}",
4307 [(set VR128:$dst, (v2i64 (X86vzmovl
4308 (loadv2i64 addr:$src))))]>,
4309 XS, VEX, Requires<[HasAVX]>;
4310 let AddedComplexity = 20 in {
4311 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4312 "movq\t{$src, $dst|$dst, $src}",
4313 [(set VR128:$dst, (v2i64 (X86vzmovl
4314 (loadv2i64 addr:$src))))]>,
4315 XS, Requires<[HasSSE2]>;
4318 let AddedComplexity = 20 in {
4319 let Predicates = [HasAVX] in {
4320 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4321 (VMOVZPQILo2PQIrm addr:$src)>;
4322 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4323 (VMOVZPQILo2PQIrr VR128:$src)>;
4325 let Predicates = [HasSSE2] in {
4326 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4327 (MOVZPQILo2PQIrm addr:$src)>;
4328 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4329 (MOVZPQILo2PQIrr VR128:$src)>;
4333 // Instructions to match in the assembler
4334 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4335 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4336 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4337 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4338 // Recognize "movd" with GR64 destination, but encode as a "movq"
4339 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4340 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4342 // Instructions for the disassembler
4343 // xr = XMM register
4346 let Predicates = [HasAVX] in
4347 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4348 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4349 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4350 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4352 //===---------------------------------------------------------------------===//
4353 // SSE3 - Conversion Instructions
4354 //===---------------------------------------------------------------------===//
4356 // Convert Packed Double FP to Packed DW Integers
4357 let Predicates = [HasAVX] in {
4358 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4359 // register, but the same isn't true when using memory operands instead.
4360 // Provide other assembly rr and rm forms to address this explicitly.
4361 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4362 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4363 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4364 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4367 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4368 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4369 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4370 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4373 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4374 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4375 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4376 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4379 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4380 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4381 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4382 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4384 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4385 (VCVTTPD2DQYrr VR256:$src)>;
4386 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4387 (VCVTTPD2DQYrm addr:$src)>;
4389 // Convert Packed DW Integers to Packed Double FP
4390 let Predicates = [HasAVX] in {
4391 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4392 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4393 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4394 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4395 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4396 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4397 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4398 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4401 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4402 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4403 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4404 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4406 // AVX 256-bit register conversion intrinsics
4407 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4408 (VCVTDQ2PDYrr VR128:$src)>;
4409 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4410 (VCVTDQ2PDYrm addr:$src)>;
4412 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4413 (VCVTPD2DQYrr VR256:$src)>;
4414 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4415 (VCVTPD2DQYrm addr:$src)>;
4417 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4418 (VCVTDQ2PDYrr VR128:$src)>;
4419 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4420 (VCVTDQ2PDYrm addr:$src)>;
4422 //===---------------------------------------------------------------------===//
4423 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4424 //===---------------------------------------------------------------------===//
4425 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4426 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4427 X86MemOperand x86memop> {
4428 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4429 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4430 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4431 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4432 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4433 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4436 let Predicates = [HasAVX] in {
4437 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4438 v4f32, VR128, memopv4f32, f128mem>, VEX;
4439 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4440 v4f32, VR128, memopv4f32, f128mem>, VEX;
4441 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4442 v8f32, VR256, memopv8f32, f256mem>, VEX;
4443 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4444 v8f32, VR256, memopv8f32, f256mem>, VEX;
4446 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4447 memopv4f32, f128mem>;
4448 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4449 memopv4f32, f128mem>;
4451 let Predicates = [HasAVX] in {
4452 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4453 (VMOVSHDUPrr VR128:$src)>;
4454 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4455 (VMOVSHDUPrm addr:$src)>;
4456 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4457 (VMOVSLDUPrr VR128:$src)>;
4458 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4459 (VMOVSLDUPrm addr:$src)>;
4460 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4461 (VMOVSHDUPYrr VR256:$src)>;
4462 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4463 (VMOVSHDUPYrm addr:$src)>;
4464 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4465 (VMOVSLDUPYrr VR256:$src)>;
4466 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4467 (VMOVSLDUPYrm addr:$src)>;
4470 let Predicates = [HasSSE3] in {
4471 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4472 (MOVSHDUPrr VR128:$src)>;
4473 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4474 (MOVSHDUPrm addr:$src)>;
4475 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4476 (MOVSLDUPrr VR128:$src)>;
4477 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4478 (MOVSLDUPrm addr:$src)>;
4481 //===---------------------------------------------------------------------===//
4482 // SSE3 - Replicate Double FP - MOVDDUP
4483 //===---------------------------------------------------------------------===//
4485 multiclass sse3_replicate_dfp<string OpcodeStr> {
4486 let neverHasSideEffects = 1 in
4487 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4488 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4490 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4491 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4494 (scalar_to_vector (loadf64 addr:$src)))))]>;
4497 // FIXME: Merge with above classe when there're patterns for the ymm version
4498 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4499 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4500 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4501 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4502 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4503 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4506 (scalar_to_vector (loadf64 addr:$src)))))]>;
4509 let Predicates = [HasAVX] in {
4510 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4511 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4514 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4516 let Predicates = [HasAVX] in {
4517 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4518 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4519 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4520 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4521 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4522 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4523 def : Pat<(X86Movddup (bc_v2f64
4524 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4525 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4528 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4529 (VMOVDDUPYrm addr:$src)>;
4530 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4531 (VMOVDDUPYrm addr:$src)>;
4532 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4533 (VMOVDDUPYrm addr:$src)>;
4534 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4535 (VMOVDDUPYrr VR256:$src)>;
4538 let Predicates = [HasSSE3] in {
4539 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4540 (MOVDDUPrm addr:$src)>;
4541 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4542 (MOVDDUPrm addr:$src)>;
4543 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4544 (MOVDDUPrm addr:$src)>;
4545 def : Pat<(X86Movddup (bc_v2f64
4546 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4547 (MOVDDUPrm addr:$src)>;
4550 //===---------------------------------------------------------------------===//
4551 // SSE3 - Move Unaligned Integer
4552 //===---------------------------------------------------------------------===//
4554 let Predicates = [HasAVX] in {
4555 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4556 "vlddqu\t{$src, $dst|$dst, $src}",
4557 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4558 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4559 "vlddqu\t{$src, $dst|$dst, $src}",
4560 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4562 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4563 "lddqu\t{$src, $dst|$dst, $src}",
4564 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
4566 //===---------------------------------------------------------------------===//
4567 // SSE3 - Arithmetic
4568 //===---------------------------------------------------------------------===//
4570 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4571 X86MemOperand x86memop, bit Is2Addr = 1> {
4572 def rr : I<0xD0, MRMSrcReg,
4573 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4575 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4576 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4577 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
4578 def rm : I<0xD0, MRMSrcMem,
4579 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4581 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4582 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4583 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
4586 let Predicates = [HasAVX] in {
4587 let ExeDomain = SSEPackedSingle in {
4588 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4589 f128mem, 0>, TB, XD, VEX_4V;
4590 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4591 f256mem, 0>, TB, XD, VEX_4V;
4593 let ExeDomain = SSEPackedDouble in {
4594 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4595 f128mem, 0>, TB, OpSize, VEX_4V;
4596 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4597 f256mem, 0>, TB, OpSize, VEX_4V;
4600 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
4601 let ExeDomain = SSEPackedSingle in
4602 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4604 let ExeDomain = SSEPackedDouble in
4605 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4606 f128mem>, TB, OpSize;
4609 //===---------------------------------------------------------------------===//
4610 // SSE3 Instructions
4611 //===---------------------------------------------------------------------===//
4614 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4615 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4616 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4618 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4619 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4620 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
4622 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4624 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4625 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4626 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
4628 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4629 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4630 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4632 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4633 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4634 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
4636 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4638 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4639 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4640 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
4643 let Predicates = [HasAVX] in {
4644 let ExeDomain = SSEPackedSingle in {
4645 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4646 X86fhadd, 0>, VEX_4V;
4647 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4648 X86fhsub, 0>, VEX_4V;
4649 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4650 X86fhadd, 0>, VEX_4V;
4651 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4652 X86fhsub, 0>, VEX_4V;
4654 let ExeDomain = SSEPackedDouble in {
4655 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4656 X86fhadd, 0>, VEX_4V;
4657 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4658 X86fhsub, 0>, VEX_4V;
4659 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4660 X86fhadd, 0>, VEX_4V;
4661 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4662 X86fhsub, 0>, VEX_4V;
4666 let Constraints = "$src1 = $dst" in {
4667 let ExeDomain = SSEPackedSingle in {
4668 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
4669 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
4671 let ExeDomain = SSEPackedDouble in {
4672 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
4673 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
4677 //===---------------------------------------------------------------------===//
4678 // SSSE3 - Packed Absolute Instructions
4679 //===---------------------------------------------------------------------===//
4682 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4683 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4684 Intrinsic IntId128> {
4685 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4687 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4688 [(set VR128:$dst, (IntId128 VR128:$src))]>,
4691 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4693 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4696 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
4699 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4700 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
4701 Intrinsic IntId256> {
4702 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4704 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4705 [(set VR256:$dst, (IntId256 VR256:$src))]>,
4708 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4710 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4713 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
4716 let Predicates = [HasAVX] in {
4717 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
4718 int_x86_ssse3_pabs_b_128>, VEX;
4719 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
4720 int_x86_ssse3_pabs_w_128>, VEX;
4721 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
4722 int_x86_ssse3_pabs_d_128>, VEX;
4725 let Predicates = [HasAVX2] in {
4726 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
4727 int_x86_avx2_pabs_b>, VEX;
4728 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
4729 int_x86_avx2_pabs_w>, VEX;
4730 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
4731 int_x86_avx2_pabs_d>, VEX;
4734 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
4735 int_x86_ssse3_pabs_b_128>;
4736 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
4737 int_x86_ssse3_pabs_w_128>;
4738 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
4739 int_x86_ssse3_pabs_d_128>;
4741 //===---------------------------------------------------------------------===//
4742 // SSSE3 - Packed Binary Operator Instructions
4743 //===---------------------------------------------------------------------===//
4745 /// SS3I_binop_rm - Simple SSSE3 bin op
4746 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4747 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
4748 X86MemOperand x86memop, bit Is2Addr = 1> {
4749 let isCommutable = 1 in
4750 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
4751 (ins RC:$src1, RC:$src2),
4753 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4754 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4755 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
4757 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
4758 (ins RC:$src1, x86memop:$src2),
4760 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4761 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4763 (OpVT (OpNode RC:$src1,
4764 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
4767 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
4768 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
4769 Intrinsic IntId128, bit Is2Addr = 1> {
4770 let isCommutable = 1 in
4771 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4772 (ins VR128:$src1, VR128:$src2),
4774 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4775 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4776 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4778 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4779 (ins VR128:$src1, i128mem:$src2),
4781 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4782 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4784 (IntId128 VR128:$src1,
4785 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
4788 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
4789 Intrinsic IntId256> {
4790 let isCommutable = 1 in
4791 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4792 (ins VR256:$src1, VR256:$src2),
4793 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4794 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
4796 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4797 (ins VR256:$src1, i256mem:$src2),
4798 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4800 (IntId256 VR256:$src1,
4801 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
4804 let ImmT = NoImm, Predicates = [HasAVX] in {
4805 let isCommutable = 0 in {
4806 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
4807 memopv2i64, i128mem, 0>, VEX_4V;
4808 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
4809 memopv2i64, i128mem, 0>, VEX_4V;
4810 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
4811 memopv2i64, i128mem, 0>, VEX_4V;
4812 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
4813 memopv2i64, i128mem, 0>, VEX_4V;
4814 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
4815 memopv2i64, i128mem, 0>, VEX_4V;
4816 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
4817 memopv2i64, i128mem, 0>, VEX_4V;
4818 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
4819 memopv2i64, i128mem, 0>, VEX_4V;
4820 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
4821 memopv2i64, i128mem, 0>, VEX_4V;
4822 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
4823 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
4824 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
4825 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
4826 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
4827 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
4829 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
4830 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
4833 let ImmT = NoImm, Predicates = [HasAVX2] in {
4834 let isCommutable = 0 in {
4835 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
4836 memopv4i64, i256mem, 0>, VEX_4V;
4837 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
4838 memopv4i64, i256mem, 0>, VEX_4V;
4839 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
4840 memopv4i64, i256mem, 0>, VEX_4V;
4841 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
4842 memopv4i64, i256mem, 0>, VEX_4V;
4843 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
4844 memopv4i64, i256mem, 0>, VEX_4V;
4845 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
4846 memopv4i64, i256mem, 0>, VEX_4V;
4847 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
4848 memopv4i64, i256mem, 0>, VEX_4V;
4849 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
4850 memopv4i64, i256mem, 0>, VEX_4V;
4851 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
4852 int_x86_avx2_phadd_sw>, VEX_4V;
4853 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
4854 int_x86_avx2_phsub_sw>, VEX_4V;
4855 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
4856 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
4858 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
4859 int_x86_avx2_pmul_hr_sw>, VEX_4V;
4862 // None of these have i8 immediate fields.
4863 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
4864 let isCommutable = 0 in {
4865 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
4866 memopv2i64, i128mem>;
4867 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
4868 memopv2i64, i128mem>;
4869 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
4870 memopv2i64, i128mem>;
4871 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
4872 memopv2i64, i128mem>;
4873 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
4874 memopv2i64, i128mem>;
4875 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
4876 memopv2i64, i128mem>;
4877 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
4878 memopv2i64, i128mem>;
4879 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
4880 memopv2i64, i128mem>;
4881 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
4882 int_x86_ssse3_phadd_sw_128>;
4883 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
4884 int_x86_ssse3_phsub_sw_128>;
4885 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
4886 int_x86_ssse3_pmadd_ub_sw_128>;
4888 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
4889 int_x86_ssse3_pmul_hr_sw_128>;
4892 //===---------------------------------------------------------------------===//
4893 // SSSE3 - Packed Align Instruction Patterns
4894 //===---------------------------------------------------------------------===//
4896 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
4897 let neverHasSideEffects = 1 in {
4898 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
4899 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4901 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4903 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4906 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
4907 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4909 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4911 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4916 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
4917 let neverHasSideEffects = 1 in {
4918 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
4919 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
4921 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4924 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
4925 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
4927 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4932 let Predicates = [HasAVX] in
4933 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
4934 let Predicates = [HasAVX2] in
4935 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
4936 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
4937 defm PALIGN : ssse3_palign<"palignr">;
4939 let Predicates = [HasAVX2] in {
4940 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
4941 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
4942 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
4943 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
4944 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
4945 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
4946 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
4947 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
4950 let Predicates = [HasAVX] in {
4951 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4952 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4953 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4954 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4955 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4956 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4957 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4958 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4961 let Predicates = [HasSSSE3] in {
4962 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4963 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4964 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4965 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4966 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4967 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4968 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4969 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4972 //===---------------------------------------------------------------------===//
4973 // SSSE3 - Thread synchronization
4974 //===---------------------------------------------------------------------===//
4976 let usesCustomInserter = 1 in {
4977 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
4978 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
4979 Requires<[HasSSE3]>;
4980 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
4981 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
4982 Requires<[HasSSE3]>;
4985 let Uses = [EAX, ECX, EDX] in
4986 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
4987 Requires<[HasSSE3]>;
4988 let Uses = [ECX, EAX] in
4989 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
4990 Requires<[HasSSE3]>;
4992 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
4993 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
4995 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
4996 Requires<[In32BitMode]>;
4997 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
4998 Requires<[In64BitMode]>;
5000 //===----------------------------------------------------------------------===//
5001 // SSE4.1 - Packed Move with Sign/Zero Extend
5002 //===----------------------------------------------------------------------===//
5004 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5005 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5006 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5007 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5009 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5010 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5012 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5016 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5018 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5019 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5020 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5022 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5023 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5024 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5027 let Predicates = [HasAVX] in {
5028 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5030 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5032 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5034 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5036 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5038 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5042 let Predicates = [HasAVX2] in {
5043 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5044 int_x86_avx2_pmovsxbw>, VEX;
5045 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5046 int_x86_avx2_pmovsxwd>, VEX;
5047 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5048 int_x86_avx2_pmovsxdq>, VEX;
5049 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5050 int_x86_avx2_pmovzxbw>, VEX;
5051 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5052 int_x86_avx2_pmovzxwd>, VEX;
5053 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5054 int_x86_avx2_pmovzxdq>, VEX;
5057 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5058 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5059 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5060 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5061 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5062 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5064 let Predicates = [HasAVX] in {
5065 // Common patterns involving scalar load.
5066 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5067 (VPMOVSXBWrm addr:$src)>;
5068 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5069 (VPMOVSXBWrm addr:$src)>;
5071 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5072 (VPMOVSXWDrm addr:$src)>;
5073 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5074 (VPMOVSXWDrm addr:$src)>;
5076 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5077 (VPMOVSXDQrm addr:$src)>;
5078 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5079 (VPMOVSXDQrm addr:$src)>;
5081 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5082 (VPMOVZXBWrm addr:$src)>;
5083 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5084 (VPMOVZXBWrm addr:$src)>;
5086 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5087 (VPMOVZXWDrm addr:$src)>;
5088 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5089 (VPMOVZXWDrm addr:$src)>;
5091 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5092 (VPMOVZXDQrm addr:$src)>;
5093 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5094 (VPMOVZXDQrm addr:$src)>;
5097 let Predicates = [HasSSE41] in {
5098 // Common patterns involving scalar load.
5099 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5100 (PMOVSXBWrm addr:$src)>;
5101 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5102 (PMOVSXBWrm addr:$src)>;
5104 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5105 (PMOVSXWDrm addr:$src)>;
5106 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5107 (PMOVSXWDrm addr:$src)>;
5109 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5110 (PMOVSXDQrm addr:$src)>;
5111 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5112 (PMOVSXDQrm addr:$src)>;
5114 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5115 (PMOVZXBWrm addr:$src)>;
5116 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5117 (PMOVZXBWrm addr:$src)>;
5119 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5120 (PMOVZXWDrm addr:$src)>;
5121 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5122 (PMOVZXWDrm addr:$src)>;
5124 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5125 (PMOVZXDQrm addr:$src)>;
5126 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5127 (PMOVZXDQrm addr:$src)>;
5130 let Predicates = [HasAVX] in {
5131 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5132 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5135 let Predicates = [HasSSE41] in {
5136 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5137 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5141 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5142 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5143 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5144 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5146 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5147 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5149 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5153 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5155 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5156 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5157 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5159 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5160 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5162 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5166 let Predicates = [HasAVX] in {
5167 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5169 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5171 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5173 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5177 let Predicates = [HasAVX2] in {
5178 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5179 int_x86_avx2_pmovsxbd>, VEX;
5180 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5181 int_x86_avx2_pmovsxwq>, VEX;
5182 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5183 int_x86_avx2_pmovzxbd>, VEX;
5184 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5185 int_x86_avx2_pmovzxwq>, VEX;
5188 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5189 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5190 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5191 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5193 let Predicates = [HasAVX] in {
5194 // Common patterns involving scalar load
5195 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5196 (VPMOVSXBDrm addr:$src)>;
5197 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5198 (VPMOVSXWQrm addr:$src)>;
5200 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5201 (VPMOVZXBDrm addr:$src)>;
5202 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5203 (VPMOVZXWQrm addr:$src)>;
5206 let Predicates = [HasSSE41] in {
5207 // Common patterns involving scalar load
5208 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5209 (PMOVSXBDrm addr:$src)>;
5210 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5211 (PMOVSXWQrm addr:$src)>;
5213 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5214 (PMOVZXBDrm addr:$src)>;
5215 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5216 (PMOVZXWQrm addr:$src)>;
5219 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5220 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5221 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5222 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5224 // Expecting a i16 load any extended to i32 value.
5225 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5226 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5227 [(set VR128:$dst, (IntId (bitconvert
5228 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5232 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5234 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5235 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5236 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5238 // Expecting a i16 load any extended to i32 value.
5239 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5240 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5241 [(set VR256:$dst, (IntId (bitconvert
5242 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5246 let Predicates = [HasAVX] in {
5247 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5249 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5252 let Predicates = [HasAVX2] in {
5253 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5254 int_x86_avx2_pmovsxbq>, VEX;
5255 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5256 int_x86_avx2_pmovzxbq>, VEX;
5258 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5259 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5261 let Predicates = [HasAVX] in {
5262 // Common patterns involving scalar load
5263 def : Pat<(int_x86_sse41_pmovsxbq
5264 (bitconvert (v4i32 (X86vzmovl
5265 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5266 (VPMOVSXBQrm addr:$src)>;
5268 def : Pat<(int_x86_sse41_pmovzxbq
5269 (bitconvert (v4i32 (X86vzmovl
5270 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5271 (VPMOVZXBQrm addr:$src)>;
5274 let Predicates = [HasSSE41] in {
5275 // Common patterns involving scalar load
5276 def : Pat<(int_x86_sse41_pmovsxbq
5277 (bitconvert (v4i32 (X86vzmovl
5278 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5279 (PMOVSXBQrm addr:$src)>;
5281 def : Pat<(int_x86_sse41_pmovzxbq
5282 (bitconvert (v4i32 (X86vzmovl
5283 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5284 (PMOVZXBQrm addr:$src)>;
5287 //===----------------------------------------------------------------------===//
5288 // SSE4.1 - Extract Instructions
5289 //===----------------------------------------------------------------------===//
5291 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5292 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5293 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5294 (ins VR128:$src1, i32i8imm:$src2),
5295 !strconcat(OpcodeStr,
5296 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5297 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5299 let neverHasSideEffects = 1, mayStore = 1 in
5300 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5301 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5302 !strconcat(OpcodeStr,
5303 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5306 // There's an AssertZext in the way of writing the store pattern
5307 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5310 let Predicates = [HasAVX] in {
5311 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5312 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5313 (ins VR128:$src1, i32i8imm:$src2),
5314 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5317 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5320 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5321 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5322 let neverHasSideEffects = 1, mayStore = 1 in
5323 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5324 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5325 !strconcat(OpcodeStr,
5326 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5329 // There's an AssertZext in the way of writing the store pattern
5330 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5333 let Predicates = [HasAVX] in
5334 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5336 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5339 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5340 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5341 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5342 (ins VR128:$src1, i32i8imm:$src2),
5343 !strconcat(OpcodeStr,
5344 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5346 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5347 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5348 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5349 !strconcat(OpcodeStr,
5350 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5351 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5352 addr:$dst)]>, OpSize;
5355 let Predicates = [HasAVX] in
5356 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5358 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5360 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5361 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5362 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5363 (ins VR128:$src1, i32i8imm:$src2),
5364 !strconcat(OpcodeStr,
5365 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5367 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5368 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5369 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5370 !strconcat(OpcodeStr,
5371 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5372 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5373 addr:$dst)]>, OpSize, REX_W;
5376 let Predicates = [HasAVX] in
5377 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5379 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5381 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5383 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5384 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5385 (ins VR128:$src1, i32i8imm:$src2),
5386 !strconcat(OpcodeStr,
5387 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5389 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5391 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5392 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5393 !strconcat(OpcodeStr,
5394 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5395 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5396 addr:$dst)]>, OpSize;
5399 let ExeDomain = SSEPackedSingle in {
5400 let Predicates = [HasAVX] in {
5401 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5402 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5403 (ins VR128:$src1, i32i8imm:$src2),
5404 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5407 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5410 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5411 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5414 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5416 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5419 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5420 Requires<[HasSSE41]>;
5422 //===----------------------------------------------------------------------===//
5423 // SSE4.1 - Insert Instructions
5424 //===----------------------------------------------------------------------===//
5426 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5427 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5428 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5430 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5432 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5434 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5435 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5436 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5438 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5440 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5442 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5443 imm:$src3))]>, OpSize;
5446 let Predicates = [HasAVX] in
5447 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5448 let Constraints = "$src1 = $dst" in
5449 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5451 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5452 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5453 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5455 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5457 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5459 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5461 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5462 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5464 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5466 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5468 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5469 imm:$src3)))]>, OpSize;
5472 let Predicates = [HasAVX] in
5473 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5474 let Constraints = "$src1 = $dst" in
5475 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5477 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5478 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5479 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5481 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5483 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5485 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5487 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5488 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5490 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5492 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5494 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5495 imm:$src3)))]>, OpSize;
5498 let Predicates = [HasAVX] in
5499 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5500 let Constraints = "$src1 = $dst" in
5501 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5503 // insertps has a few different modes, there's the first two here below which
5504 // are optimized inserts that won't zero arbitrary elements in the destination
5505 // vector. The next one matches the intrinsic and could zero arbitrary elements
5506 // in the target vector.
5507 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5508 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5509 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5511 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5513 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5515 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5517 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5518 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5520 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5522 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5524 (X86insrtps VR128:$src1,
5525 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5526 imm:$src3))]>, OpSize;
5529 let ExeDomain = SSEPackedSingle in {
5530 let Predicates = [HasAVX] in
5531 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5532 let Constraints = "$src1 = $dst" in
5533 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5536 //===----------------------------------------------------------------------===//
5537 // SSE4.1 - Round Instructions
5538 //===----------------------------------------------------------------------===//
5540 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5541 X86MemOperand x86memop, RegisterClass RC,
5542 PatFrag mem_frag32, PatFrag mem_frag64,
5543 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5544 let ExeDomain = SSEPackedSingle in {
5545 // Intrinsic operation, reg.
5546 // Vector intrinsic operation, reg
5547 def PSr : SS4AIi8<opcps, MRMSrcReg,
5548 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5549 !strconcat(OpcodeStr,
5550 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5551 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5554 // Vector intrinsic operation, mem
5555 def PSm : SS4AIi8<opcps, MRMSrcMem,
5556 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5557 !strconcat(OpcodeStr,
5558 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5560 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
5562 } // ExeDomain = SSEPackedSingle
5564 let ExeDomain = SSEPackedDouble in {
5565 // Vector intrinsic operation, reg
5566 def PDr : SS4AIi8<opcpd, MRMSrcReg,
5567 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5568 !strconcat(OpcodeStr,
5569 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5570 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
5573 // Vector intrinsic operation, mem
5574 def PDm : SS4AIi8<opcpd, MRMSrcMem,
5575 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5576 !strconcat(OpcodeStr,
5577 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5579 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
5581 } // ExeDomain = SSEPackedDouble
5584 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
5587 Intrinsic F64Int, bit Is2Addr = 1> {
5588 let ExeDomain = GenericDomain in {
5590 def SSr : SS4AIi8<opcss, MRMSrcReg,
5591 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
5593 !strconcat(OpcodeStr,
5594 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5595 !strconcat(OpcodeStr,
5596 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5599 // Intrinsic operation, reg.
5600 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
5601 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5603 !strconcat(OpcodeStr,
5604 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5605 !strconcat(OpcodeStr,
5606 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5607 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5610 // Intrinsic operation, mem.
5611 def SSm : SS4AIi8<opcss, MRMSrcMem,
5612 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5614 !strconcat(OpcodeStr,
5615 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5616 !strconcat(OpcodeStr,
5617 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5619 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
5623 def SDr : SS4AIi8<opcsd, MRMSrcReg,
5624 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
5626 !strconcat(OpcodeStr,
5627 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5628 !strconcat(OpcodeStr,
5629 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5632 // Intrinsic operation, reg.
5633 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
5634 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5636 !strconcat(OpcodeStr,
5637 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5638 !strconcat(OpcodeStr,
5639 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5640 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5643 // Intrinsic operation, mem.
5644 def SDm : SS4AIi8<opcsd, MRMSrcMem,
5645 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5647 !strconcat(OpcodeStr,
5648 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5649 !strconcat(OpcodeStr,
5650 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5652 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
5654 } // ExeDomain = GenericDomain
5657 // FP round - roundss, roundps, roundsd, roundpd
5658 let Predicates = [HasAVX] in {
5660 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
5661 memopv4f32, memopv2f64,
5662 int_x86_sse41_round_ps,
5663 int_x86_sse41_round_pd>, VEX;
5664 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
5665 memopv8f32, memopv4f64,
5666 int_x86_avx_round_ps_256,
5667 int_x86_avx_round_pd_256>, VEX;
5668 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
5669 int_x86_sse41_round_ss,
5670 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
5672 def : Pat<(ffloor FR32:$src),
5673 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
5674 def : Pat<(f64 (ffloor FR64:$src)),
5675 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
5676 def : Pat<(f32 (fnearbyint FR32:$src)),
5677 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
5678 def : Pat<(f64 (fnearbyint FR64:$src)),
5679 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
5680 def : Pat<(f32 (fceil FR32:$src)),
5681 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
5682 def : Pat<(f64 (fceil FR64:$src)),
5683 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
5684 def : Pat<(f32 (frint FR32:$src)),
5685 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
5686 def : Pat<(f64 (frint FR64:$src)),
5687 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
5688 def : Pat<(f32 (ftrunc FR32:$src)),
5689 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
5690 def : Pat<(f64 (ftrunc FR64:$src)),
5691 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
5694 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
5695 memopv4f32, memopv2f64,
5696 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
5697 let Constraints = "$src1 = $dst" in
5698 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
5699 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
5701 def : Pat<(ffloor FR32:$src),
5702 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
5703 def : Pat<(f64 (ffloor FR64:$src)),
5704 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
5705 def : Pat<(f32 (fnearbyint FR32:$src)),
5706 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
5707 def : Pat<(f64 (fnearbyint FR64:$src)),
5708 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
5709 def : Pat<(f32 (fceil FR32:$src)),
5710 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
5711 def : Pat<(f64 (fceil FR64:$src)),
5712 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
5713 def : Pat<(f32 (frint FR32:$src)),
5714 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
5715 def : Pat<(f64 (frint FR64:$src)),
5716 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
5717 def : Pat<(f32 (ftrunc FR32:$src)),
5718 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
5719 def : Pat<(f64 (ftrunc FR64:$src)),
5720 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
5722 //===----------------------------------------------------------------------===//
5723 // SSE4.1 - Packed Bit Test
5724 //===----------------------------------------------------------------------===//
5726 // ptest instruction we'll lower to this in X86ISelLowering primarily from
5727 // the intel intrinsic that corresponds to this.
5728 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5729 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5730 "vptest\t{$src2, $src1|$src1, $src2}",
5731 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5733 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5734 "vptest\t{$src2, $src1|$src1, $src2}",
5735 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5738 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
5739 "vptest\t{$src2, $src1|$src1, $src2}",
5740 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
5742 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
5743 "vptest\t{$src2, $src1|$src1, $src2}",
5744 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
5748 let Defs = [EFLAGS] in {
5749 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5750 "ptest\t{$src2, $src1|$src1, $src2}",
5751 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5753 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5754 "ptest\t{$src2, $src1|$src1, $src2}",
5755 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5759 // The bit test instructions below are AVX only
5760 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
5761 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
5762 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
5763 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5764 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
5765 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
5766 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5767 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
5771 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5772 let ExeDomain = SSEPackedSingle in {
5773 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
5774 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
5776 let ExeDomain = SSEPackedDouble in {
5777 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
5778 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
5782 //===----------------------------------------------------------------------===//
5783 // SSE4.1 - Misc Instructions
5784 //===----------------------------------------------------------------------===//
5786 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
5787 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
5788 "popcnt{w}\t{$src, $dst|$dst, $src}",
5789 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
5791 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
5792 "popcnt{w}\t{$src, $dst|$dst, $src}",
5793 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
5794 (implicit EFLAGS)]>, OpSize, XS;
5796 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
5797 "popcnt{l}\t{$src, $dst|$dst, $src}",
5798 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
5800 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
5801 "popcnt{l}\t{$src, $dst|$dst, $src}",
5802 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
5803 (implicit EFLAGS)]>, XS;
5805 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
5806 "popcnt{q}\t{$src, $dst|$dst, $src}",
5807 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
5809 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
5810 "popcnt{q}\t{$src, $dst|$dst, $src}",
5811 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
5812 (implicit EFLAGS)]>, XS;
5817 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
5818 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
5819 Intrinsic IntId128> {
5820 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5822 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5823 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
5824 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5826 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5829 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5832 let Predicates = [HasAVX] in
5833 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
5834 int_x86_sse41_phminposuw>, VEX;
5835 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
5836 int_x86_sse41_phminposuw>;
5838 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
5839 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
5840 Intrinsic IntId128, bit Is2Addr = 1> {
5841 let isCommutable = 1 in
5842 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5843 (ins VR128:$src1, VR128:$src2),
5845 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5846 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5847 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
5848 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5849 (ins VR128:$src1, i128mem:$src2),
5851 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5852 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5854 (IntId128 VR128:$src1,
5855 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5858 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
5859 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5860 Intrinsic IntId256> {
5861 let isCommutable = 1 in
5862 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
5863 (ins VR256:$src1, VR256:$src2),
5864 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5865 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
5866 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
5867 (ins VR256:$src1, i256mem:$src2),
5868 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5870 (IntId256 VR256:$src1,
5871 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5874 let Predicates = [HasAVX] in {
5875 let isCommutable = 0 in
5876 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
5878 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
5880 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
5882 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
5884 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
5886 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
5888 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
5890 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
5892 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
5894 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
5898 let Predicates = [HasAVX2] in {
5899 let isCommutable = 0 in
5900 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
5901 int_x86_avx2_packusdw>, VEX_4V;
5902 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
5903 int_x86_avx2_pmins_b>, VEX_4V;
5904 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
5905 int_x86_avx2_pmins_d>, VEX_4V;
5906 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
5907 int_x86_avx2_pminu_d>, VEX_4V;
5908 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
5909 int_x86_avx2_pminu_w>, VEX_4V;
5910 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
5911 int_x86_avx2_pmaxs_b>, VEX_4V;
5912 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
5913 int_x86_avx2_pmaxs_d>, VEX_4V;
5914 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
5915 int_x86_avx2_pmaxu_d>, VEX_4V;
5916 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
5917 int_x86_avx2_pmaxu_w>, VEX_4V;
5918 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
5919 int_x86_avx2_pmul_dq>, VEX_4V;
5922 let Constraints = "$src1 = $dst" in {
5923 let isCommutable = 0 in
5924 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
5925 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
5926 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
5927 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
5928 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
5929 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
5930 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
5931 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
5932 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
5933 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
5936 /// SS48I_binop_rm - Simple SSE41 binary operator.
5937 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5938 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5939 X86MemOperand x86memop, bit Is2Addr = 1> {
5940 let isCommutable = 1 in
5941 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
5942 (ins RC:$src1, RC:$src2),
5944 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5945 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5946 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
5947 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
5948 (ins RC:$src1, x86memop:$src2),
5950 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5951 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5953 (OpVT (OpNode RC:$src1,
5954 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
5957 let Predicates = [HasAVX] in {
5958 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
5959 memopv2i64, i128mem, 0>, VEX_4V;
5960 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
5961 memopv2i64, i128mem, 0>, VEX_4V;
5963 let Predicates = [HasAVX2] in {
5964 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
5965 memopv4i64, i256mem, 0>, VEX_4V;
5966 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
5967 memopv4i64, i256mem, 0>, VEX_4V;
5970 let Constraints = "$src1 = $dst" in {
5971 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
5972 memopv2i64, i128mem>;
5973 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
5974 memopv2i64, i128mem>;
5977 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
5978 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
5979 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
5980 X86MemOperand x86memop, bit Is2Addr = 1> {
5981 let isCommutable = 1 in
5982 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
5983 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
5985 !strconcat(OpcodeStr,
5986 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5987 !strconcat(OpcodeStr,
5988 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5989 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
5991 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
5992 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
5994 !strconcat(OpcodeStr,
5995 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5996 !strconcat(OpcodeStr,
5997 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6000 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6004 let Predicates = [HasAVX] in {
6005 let isCommutable = 0 in {
6006 let ExeDomain = SSEPackedSingle in {
6007 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6008 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6009 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6010 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6012 let ExeDomain = SSEPackedDouble in {
6013 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6014 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6015 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6016 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6018 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6019 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6020 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6021 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6023 let ExeDomain = SSEPackedSingle in
6024 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6025 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6026 let ExeDomain = SSEPackedDouble in
6027 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6028 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6029 let ExeDomain = SSEPackedSingle in
6030 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6031 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6034 let Predicates = [HasAVX2] in {
6035 let isCommutable = 0 in {
6036 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6037 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6038 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6039 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6043 let Constraints = "$src1 = $dst" in {
6044 let isCommutable = 0 in {
6045 let ExeDomain = SSEPackedSingle in
6046 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6047 VR128, memopv4f32, i128mem>;
6048 let ExeDomain = SSEPackedDouble in
6049 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6050 VR128, memopv2f64, i128mem>;
6051 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6052 VR128, memopv2i64, i128mem>;
6053 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6054 VR128, memopv2i64, i128mem>;
6056 let ExeDomain = SSEPackedSingle in
6057 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6058 VR128, memopv4f32, i128mem>;
6059 let ExeDomain = SSEPackedDouble in
6060 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6061 VR128, memopv2f64, i128mem>;
6064 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6065 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6066 RegisterClass RC, X86MemOperand x86memop,
6067 PatFrag mem_frag, Intrinsic IntId> {
6068 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6069 (ins RC:$src1, RC:$src2, RC:$src3),
6070 !strconcat(OpcodeStr,
6071 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6072 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6073 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6075 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6076 (ins RC:$src1, x86memop:$src2, RC:$src3),
6077 !strconcat(OpcodeStr,
6078 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6080 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6082 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6085 let Predicates = [HasAVX] in {
6086 let ExeDomain = SSEPackedDouble in {
6087 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6088 memopv2f64, int_x86_sse41_blendvpd>;
6089 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6090 memopv4f64, int_x86_avx_blendv_pd_256>;
6091 } // ExeDomain = SSEPackedDouble
6092 let ExeDomain = SSEPackedSingle in {
6093 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6094 memopv4f32, int_x86_sse41_blendvps>;
6095 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6096 memopv8f32, int_x86_avx_blendv_ps_256>;
6097 } // ExeDomain = SSEPackedSingle
6098 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6099 memopv2i64, int_x86_sse41_pblendvb>;
6102 let Predicates = [HasAVX2] in {
6103 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6104 memopv4i64, int_x86_avx2_pblendvb>;
6107 let Predicates = [HasAVX] in {
6108 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6109 (v16i8 VR128:$src2))),
6110 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6111 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6112 (v4i32 VR128:$src2))),
6113 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6114 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6115 (v4f32 VR128:$src2))),
6116 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6117 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6118 (v2i64 VR128:$src2))),
6119 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6120 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6121 (v2f64 VR128:$src2))),
6122 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6123 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6124 (v8i32 VR256:$src2))),
6125 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6126 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6127 (v8f32 VR256:$src2))),
6128 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6129 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6130 (v4i64 VR256:$src2))),
6131 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6132 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6133 (v4f64 VR256:$src2))),
6134 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6137 let Predicates = [HasAVX2] in {
6138 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6139 (v32i8 VR256:$src2))),
6140 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6143 /// SS41I_ternary_int - SSE 4.1 ternary operator
6144 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6145 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6147 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6148 (ins VR128:$src1, VR128:$src2),
6149 !strconcat(OpcodeStr,
6150 "\t{$src2, $dst|$dst, $src2}"),
6151 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6154 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6155 (ins VR128:$src1, i128mem:$src2),
6156 !strconcat(OpcodeStr,
6157 "\t{$src2, $dst|$dst, $src2}"),
6160 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6164 let ExeDomain = SSEPackedDouble in
6165 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6166 int_x86_sse41_blendvpd>;
6167 let ExeDomain = SSEPackedSingle in
6168 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6169 int_x86_sse41_blendvps>;
6170 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6171 int_x86_sse41_pblendvb>;
6173 let Predicates = [HasSSE41] in {
6174 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6175 (v16i8 VR128:$src2))),
6176 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6177 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6178 (v4i32 VR128:$src2))),
6179 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6180 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6181 (v4f32 VR128:$src2))),
6182 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6183 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6184 (v2i64 VR128:$src2))),
6185 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6186 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6187 (v2f64 VR128:$src2))),
6188 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6191 let Predicates = [HasAVX] in
6192 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6193 "vmovntdqa\t{$src, $dst|$dst, $src}",
6194 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6196 let Predicates = [HasAVX2] in
6197 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6198 "vmovntdqa\t{$src, $dst|$dst, $src}",
6199 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6201 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6202 "movntdqa\t{$src, $dst|$dst, $src}",
6203 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6206 //===----------------------------------------------------------------------===//
6207 // SSE4.2 - Compare Instructions
6208 //===----------------------------------------------------------------------===//
6210 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6211 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6212 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6213 X86MemOperand x86memop, bit Is2Addr = 1> {
6214 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6215 (ins RC:$src1, RC:$src2),
6217 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6218 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6219 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6221 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6222 (ins RC:$src1, x86memop:$src2),
6224 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6225 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6227 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6230 let Predicates = [HasAVX] in
6231 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6232 memopv2i64, i128mem, 0>, VEX_4V;
6234 let Predicates = [HasAVX2] in
6235 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6236 memopv4i64, i256mem, 0>, VEX_4V;
6238 let Constraints = "$src1 = $dst" in
6239 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6240 memopv2i64, i128mem>;
6242 //===----------------------------------------------------------------------===//
6243 // SSE4.2 - String/text Processing Instructions
6244 //===----------------------------------------------------------------------===//
6246 // Packed Compare Implicit Length Strings, Return Mask
6247 multiclass pseudo_pcmpistrm<string asm> {
6248 def REG : PseudoI<(outs VR128:$dst),
6249 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6250 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6252 def MEM : PseudoI<(outs VR128:$dst),
6253 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6254 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6255 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6258 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6259 let AddedComplexity = 1 in
6260 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6261 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6264 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6265 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6266 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6267 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6269 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6270 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6271 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6274 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6275 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6276 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6277 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6279 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6280 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6281 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6284 // Packed Compare Explicit Length Strings, Return Mask
6285 multiclass pseudo_pcmpestrm<string asm> {
6286 def REG : PseudoI<(outs VR128:$dst),
6287 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6288 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6289 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6290 def MEM : PseudoI<(outs VR128:$dst),
6291 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6292 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6293 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6296 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6297 let AddedComplexity = 1 in
6298 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6299 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6302 let Predicates = [HasAVX],
6303 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6304 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6305 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6306 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6308 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6309 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6310 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6313 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6314 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6315 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6316 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6318 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6319 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6320 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6323 // Packed Compare Implicit Length Strings, Return Index
6324 let Defs = [ECX, EFLAGS] in {
6325 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6326 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6327 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6328 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6329 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6330 (implicit EFLAGS)]>, OpSize;
6331 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6332 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6333 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6334 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6335 (implicit EFLAGS)]>, OpSize;
6339 let Predicates = [HasAVX] in {
6340 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6342 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6344 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6346 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6348 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6350 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6354 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6355 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6356 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6357 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6358 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6359 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6361 // Packed Compare Explicit Length Strings, Return Index
6362 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6363 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6364 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6365 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6366 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6367 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6368 (implicit EFLAGS)]>, OpSize;
6369 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6370 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6371 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6373 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6374 (implicit EFLAGS)]>, OpSize;
6378 let Predicates = [HasAVX] in {
6379 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6381 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6383 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6385 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6387 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6389 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6393 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6394 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6395 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6396 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6397 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6398 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6400 //===----------------------------------------------------------------------===//
6401 // SSE4.2 - CRC Instructions
6402 //===----------------------------------------------------------------------===//
6404 // No CRC instructions have AVX equivalents
6406 // crc intrinsic instruction
6407 // This set of instructions are only rm, the only difference is the size
6409 let Constraints = "$src1 = $dst" in {
6410 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6411 (ins GR32:$src1, i8mem:$src2),
6412 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6414 (int_x86_sse42_crc32_32_8 GR32:$src1,
6415 (load addr:$src2)))]>;
6416 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6417 (ins GR32:$src1, GR8:$src2),
6418 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6420 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6421 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6422 (ins GR32:$src1, i16mem:$src2),
6423 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6425 (int_x86_sse42_crc32_32_16 GR32:$src1,
6426 (load addr:$src2)))]>,
6428 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6429 (ins GR32:$src1, GR16:$src2),
6430 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6432 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6434 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6435 (ins GR32:$src1, i32mem:$src2),
6436 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6438 (int_x86_sse42_crc32_32_32 GR32:$src1,
6439 (load addr:$src2)))]>;
6440 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6441 (ins GR32:$src1, GR32:$src2),
6442 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6444 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6445 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6446 (ins GR64:$src1, i8mem:$src2),
6447 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6449 (int_x86_sse42_crc32_64_8 GR64:$src1,
6450 (load addr:$src2)))]>,
6452 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6453 (ins GR64:$src1, GR8:$src2),
6454 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6456 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6458 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6459 (ins GR64:$src1, i64mem:$src2),
6460 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6462 (int_x86_sse42_crc32_64_64 GR64:$src1,
6463 (load addr:$src2)))]>,
6465 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6466 (ins GR64:$src1, GR64:$src2),
6467 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6469 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6473 //===----------------------------------------------------------------------===//
6474 // AES-NI Instructions
6475 //===----------------------------------------------------------------------===//
6477 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6478 Intrinsic IntId128, bit Is2Addr = 1> {
6479 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6480 (ins VR128:$src1, VR128:$src2),
6482 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6483 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6484 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6486 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6487 (ins VR128:$src1, i128mem:$src2),
6489 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6490 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6492 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6495 // Perform One Round of an AES Encryption/Decryption Flow
6496 let Predicates = [HasAVX, HasAES] in {
6497 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
6498 int_x86_aesni_aesenc, 0>, VEX_4V;
6499 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
6500 int_x86_aesni_aesenclast, 0>, VEX_4V;
6501 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
6502 int_x86_aesni_aesdec, 0>, VEX_4V;
6503 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
6504 int_x86_aesni_aesdeclast, 0>, VEX_4V;
6507 let Constraints = "$src1 = $dst" in {
6508 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
6509 int_x86_aesni_aesenc>;
6510 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
6511 int_x86_aesni_aesenclast>;
6512 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
6513 int_x86_aesni_aesdec>;
6514 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
6515 int_x86_aesni_aesdeclast>;
6518 // Perform the AES InvMixColumn Transformation
6519 let Predicates = [HasAVX, HasAES] in {
6520 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6522 "vaesimc\t{$src1, $dst|$dst, $src1}",
6524 (int_x86_aesni_aesimc VR128:$src1))]>,
6526 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6527 (ins i128mem:$src1),
6528 "vaesimc\t{$src1, $dst|$dst, $src1}",
6529 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
6532 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6534 "aesimc\t{$src1, $dst|$dst, $src1}",
6536 (int_x86_aesni_aesimc VR128:$src1))]>,
6538 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6539 (ins i128mem:$src1),
6540 "aesimc\t{$src1, $dst|$dst, $src1}",
6541 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
6544 // AES Round Key Generation Assist
6545 let Predicates = [HasAVX, HasAES] in {
6546 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6547 (ins VR128:$src1, i8imm:$src2),
6548 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6550 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6552 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6553 (ins i128mem:$src1, i8imm:$src2),
6554 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6556 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
6559 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
6560 (ins VR128:$src1, i8imm:$src2),
6561 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6563 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
6565 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
6566 (ins i128mem:$src1, i8imm:$src2),
6567 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6569 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
6572 //===----------------------------------------------------------------------===//
6573 // CLMUL Instructions
6574 //===----------------------------------------------------------------------===//
6576 // Carry-less Multiplication instructions
6577 let neverHasSideEffects = 1 in {
6578 // AVX carry-less Multiplication instructions
6579 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6580 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6581 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6585 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6586 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6587 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6590 let Constraints = "$src1 = $dst" in {
6591 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6592 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6593 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6597 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6598 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6599 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
6601 } // Constraints = "$src1 = $dst"
6602 } // neverHasSideEffects = 1
6605 multiclass pclmul_alias<string asm, int immop> {
6606 def : InstAlias<!strconcat("pclmul", asm,
6607 "dq {$src, $dst|$dst, $src}"),
6608 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
6610 def : InstAlias<!strconcat("pclmul", asm,
6611 "dq {$src, $dst|$dst, $src}"),
6612 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
6614 def : InstAlias<!strconcat("vpclmul", asm,
6615 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6616 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
6618 def : InstAlias<!strconcat("vpclmul", asm,
6619 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6620 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
6622 defm : pclmul_alias<"hqhq", 0x11>;
6623 defm : pclmul_alias<"hqlq", 0x01>;
6624 defm : pclmul_alias<"lqhq", 0x10>;
6625 defm : pclmul_alias<"lqlq", 0x00>;
6627 //===----------------------------------------------------------------------===//
6629 //===----------------------------------------------------------------------===//
6631 //===----------------------------------------------------------------------===//
6632 // VBROADCAST - Load from memory and broadcast to all elements of the
6633 // destination operand
6635 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
6636 X86MemOperand x86memop, Intrinsic Int> :
6637 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
6638 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6639 [(set RC:$dst, (Int addr:$src))]>, VEX;
6641 // AVX2 adds register forms
6642 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
6644 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
6645 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6646 [(set RC:$dst, (Int VR128:$src))]>, VEX;
6648 let ExeDomain = SSEPackedSingle in {
6649 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
6650 int_x86_avx_vbroadcast_ss>;
6651 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
6652 int_x86_avx_vbroadcast_ss_256>;
6654 let ExeDomain = SSEPackedDouble in
6655 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
6656 int_x86_avx_vbroadcast_sd_256>;
6657 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
6658 int_x86_avx_vbroadcastf128_pd_256>;
6660 let ExeDomain = SSEPackedSingle in {
6661 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
6662 int_x86_avx2_vbroadcast_ss_ps>;
6663 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
6664 int_x86_avx2_vbroadcast_ss_ps_256>;
6666 let ExeDomain = SSEPackedDouble in
6667 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
6668 int_x86_avx2_vbroadcast_sd_pd_256>;
6670 let Predicates = [HasAVX2] in
6671 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
6672 int_x86_avx2_vbroadcasti128>;
6674 let Predicates = [HasAVX] in
6675 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
6676 (VBROADCASTF128 addr:$src)>;
6679 //===----------------------------------------------------------------------===//
6680 // VINSERTF128 - Insert packed floating-point values
6682 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
6683 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
6684 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
6685 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6688 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
6689 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
6690 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6694 let Predicates = [HasAVX] in {
6695 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
6696 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6697 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
6698 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6699 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
6700 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6703 //===----------------------------------------------------------------------===//
6704 // VEXTRACTF128 - Extract packed floating-point values
6706 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
6707 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
6708 (ins VR256:$src1, i8imm:$src2),
6709 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6712 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
6713 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
6714 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6718 let Predicates = [HasAVX] in {
6719 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
6720 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6721 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
6722 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6723 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
6724 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6727 //===----------------------------------------------------------------------===//
6728 // VMASKMOV - Conditional SIMD Packed Loads and Stores
6730 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
6731 Intrinsic IntLd, Intrinsic IntLd256,
6732 Intrinsic IntSt, Intrinsic IntSt256> {
6733 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
6734 (ins VR128:$src1, f128mem:$src2),
6735 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6736 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
6738 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
6739 (ins VR256:$src1, f256mem:$src2),
6740 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6741 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
6743 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
6744 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
6745 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6746 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
6747 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
6748 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
6749 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6750 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
6753 let ExeDomain = SSEPackedSingle in
6754 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
6755 int_x86_avx_maskload_ps,
6756 int_x86_avx_maskload_ps_256,
6757 int_x86_avx_maskstore_ps,
6758 int_x86_avx_maskstore_ps_256>;
6759 let ExeDomain = SSEPackedDouble in
6760 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
6761 int_x86_avx_maskload_pd,
6762 int_x86_avx_maskload_pd_256,
6763 int_x86_avx_maskstore_pd,
6764 int_x86_avx_maskstore_pd_256>;
6766 //===----------------------------------------------------------------------===//
6767 // VPERMIL - Permute Single and Double Floating-Point Values
6769 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
6770 RegisterClass RC, X86MemOperand x86memop_f,
6771 X86MemOperand x86memop_i, PatFrag i_frag,
6772 Intrinsic IntVar, ValueType vt> {
6773 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
6774 (ins RC:$src1, RC:$src2),
6775 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6776 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
6777 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
6778 (ins RC:$src1, x86memop_i:$src2),
6779 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6780 [(set RC:$dst, (IntVar RC:$src1,
6781 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
6783 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
6784 (ins RC:$src1, i8imm:$src2),
6785 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6786 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
6787 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
6788 (ins x86memop_f:$src1, i8imm:$src2),
6789 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6791 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
6794 let ExeDomain = SSEPackedSingle in {
6795 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
6796 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
6797 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
6798 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
6800 let ExeDomain = SSEPackedDouble in {
6801 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
6802 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
6803 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
6804 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
6807 let Predicates = [HasAVX] in {
6808 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
6809 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6810 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
6811 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6812 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
6814 (VPERMILPSYmi addr:$src1, imm:$imm)>;
6815 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
6816 (VPERMILPDYmi addr:$src1, imm:$imm)>;
6818 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
6819 (VPERMILPDri VR128:$src1, imm:$imm)>;
6820 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
6821 (VPERMILPDmi addr:$src1, imm:$imm)>;
6824 //===----------------------------------------------------------------------===//
6825 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
6827 let ExeDomain = SSEPackedSingle in {
6828 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
6829 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
6830 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6831 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
6832 (i8 imm:$src3))))]>, VEX_4V;
6833 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
6834 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
6835 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6836 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
6837 (i8 imm:$src3)))]>, VEX_4V;
6840 let Predicates = [HasAVX] in {
6841 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6842 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6843 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6844 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6845 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6846 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6847 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6848 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6849 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6850 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6852 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
6853 (memopv8f32 addr:$src2), (i8 imm:$imm))),
6854 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6855 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
6856 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
6857 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6858 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
6859 (memopv4i64 addr:$src2), (i8 imm:$imm))),
6860 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6861 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
6862 (memopv4f64 addr:$src2), (i8 imm:$imm))),
6863 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6864 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
6865 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
6866 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6867 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
6868 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
6869 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
6872 //===----------------------------------------------------------------------===//
6873 // VZERO - Zero YMM registers
6875 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
6876 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
6877 // Zero All YMM registers
6878 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
6879 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
6881 // Zero Upper bits of YMM registers
6882 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
6883 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
6886 //===----------------------------------------------------------------------===//
6887 // Half precision conversion instructions
6888 //===----------------------------------------------------------------------===//
6889 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
6890 let Predicates = [HasAVX, HasF16C] in {
6891 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
6892 "vcvtph2ps\t{$src, $dst|$dst, $src}",
6893 [(set RC:$dst, (Int VR128:$src))]>,
6895 let neverHasSideEffects = 1, mayLoad = 1 in
6896 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
6897 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
6901 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
6902 let Predicates = [HasAVX, HasF16C] in {
6903 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
6904 (ins RC:$src1, i32i8imm:$src2),
6905 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6906 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
6908 let neverHasSideEffects = 1, mayLoad = 1 in
6909 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
6910 (ins RC:$src1, i32i8imm:$src2),
6911 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
6916 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
6917 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
6918 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
6919 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
6921 //===----------------------------------------------------------------------===//
6922 // AVX2 Instructions
6923 //===----------------------------------------------------------------------===//
6925 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
6926 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
6927 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6928 X86MemOperand x86memop> {
6929 let isCommutable = 1 in
6930 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
6931 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6932 !strconcat(OpcodeStr,
6933 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6934 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6936 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
6937 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6938 !strconcat(OpcodeStr,
6939 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6942 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6946 let isCommutable = 0 in {
6947 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
6948 VR128, memopv2i64, i128mem>;
6949 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
6950 VR256, memopv4i64, i256mem>;
6953 //===----------------------------------------------------------------------===//
6954 // VPBROADCAST - Load from memory and broadcast to all elements of the
6955 // destination operand
6957 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
6958 X86MemOperand x86memop, PatFrag ld_frag,
6959 Intrinsic Int128, Intrinsic Int256> {
6960 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
6961 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6962 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
6963 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
6964 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6966 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
6967 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
6968 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6969 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
6970 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
6971 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6973 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
6976 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
6977 int_x86_avx2_pbroadcastb_128,
6978 int_x86_avx2_pbroadcastb_256>;
6979 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
6980 int_x86_avx2_pbroadcastw_128,
6981 int_x86_avx2_pbroadcastw_256>;
6982 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
6983 int_x86_avx2_pbroadcastd_128,
6984 int_x86_avx2_pbroadcastd_256>;
6985 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
6986 int_x86_avx2_pbroadcastq_128,
6987 int_x86_avx2_pbroadcastq_256>;
6989 let Predicates = [HasAVX2] in {
6990 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
6991 (VPBROADCASTBrm addr:$src)>;
6992 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
6993 (VPBROADCASTBYrm addr:$src)>;
6994 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
6995 (VPBROADCASTWrm addr:$src)>;
6996 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
6997 (VPBROADCASTWYrm addr:$src)>;
6998 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
6999 (VPBROADCASTDrm addr:$src)>;
7000 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7001 (VPBROADCASTDYrm addr:$src)>;
7002 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7003 (VPBROADCASTQrm addr:$src)>;
7004 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7005 (VPBROADCASTQYrm addr:$src)>;
7008 // AVX1 broadcast patterns
7009 let Predicates = [HasAVX] in {
7010 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7011 (VBROADCASTSSYrm addr:$src)>;
7012 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7013 (VBROADCASTSDrm addr:$src)>;
7014 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7015 (VBROADCASTSSYrm addr:$src)>;
7016 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7017 (VBROADCASTSDrm addr:$src)>;
7019 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7020 (VBROADCASTSSrm addr:$src)>;
7021 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7022 (VBROADCASTSSrm addr:$src)>;
7025 //===----------------------------------------------------------------------===//
7026 // VPERM - Permute instructions
7029 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7031 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7032 (ins VR256:$src1, VR256:$src2),
7033 !strconcat(OpcodeStr,
7034 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7035 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7036 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7037 (ins VR256:$src1, i256mem:$src2),
7038 !strconcat(OpcodeStr,
7039 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7040 [(set VR256:$dst, (Int VR256:$src1,
7041 (bitconvert (mem_frag addr:$src2))))]>,
7045 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7046 let ExeDomain = SSEPackedSingle in
7047 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7049 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7051 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7052 (ins VR256:$src1, i8imm:$src2),
7053 !strconcat(OpcodeStr,
7054 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7055 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7056 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7057 (ins i256mem:$src1, i8imm:$src2),
7058 !strconcat(OpcodeStr,
7059 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7060 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7064 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7066 let ExeDomain = SSEPackedDouble in
7067 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7070 //===----------------------------------------------------------------------===//
7071 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7073 let AddedComplexity = 1 in {
7074 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7075 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7076 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7077 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7078 (i8 imm:$src3))))]>, VEX_4V;
7079 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7080 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7081 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7082 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7083 (i8 imm:$src3)))]>, VEX_4V;
7086 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7087 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7088 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7089 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7090 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7091 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7092 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7094 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7096 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7097 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7098 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7099 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7100 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7102 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7106 //===----------------------------------------------------------------------===//
7107 // VINSERTI128 - Insert packed integer values
7109 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7110 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7111 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7113 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7115 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7116 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7117 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7119 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7120 imm:$src3))]>, VEX_4V;
7122 let Predicates = [HasAVX2] in {
7123 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7125 (VINSERTI128rr VR256:$src1, VR128:$src2,
7126 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7127 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7129 (VINSERTI128rr VR256:$src1, VR128:$src2,
7130 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7131 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7133 (VINSERTI128rr VR256:$src1, VR128:$src2,
7134 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7135 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7137 (VINSERTI128rr VR256:$src1, VR128:$src2,
7138 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7142 let Predicates = [HasAVX] in {
7143 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7145 (VINSERTF128rr VR256:$src1, VR128:$src2,
7146 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7147 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7149 (VINSERTF128rr VR256:$src1, VR128:$src2,
7150 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7151 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7153 (VINSERTF128rr VR256:$src1, VR128:$src2,
7154 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7155 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7157 (VINSERTF128rr VR256:$src1, VR128:$src2,
7158 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7159 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7161 (VINSERTF128rr VR256:$src1, VR128:$src2,
7162 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7163 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7165 (VINSERTF128rr VR256:$src1, VR128:$src2,
7166 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7169 //===----------------------------------------------------------------------===//
7170 // VEXTRACTI128 - Extract packed integer values
7172 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7173 (ins VR256:$src1, i8imm:$src2),
7174 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7176 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7178 let neverHasSideEffects = 1, mayStore = 1 in
7179 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7180 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7181 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7183 let Predicates = [HasAVX2] in {
7184 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7185 (v2i64 (VEXTRACTI128rr
7186 (v4i64 VR256:$src1),
7187 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7188 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7189 (v4i32 (VEXTRACTI128rr
7190 (v8i32 VR256:$src1),
7191 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7192 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7193 (v8i16 (VEXTRACTI128rr
7194 (v16i16 VR256:$src1),
7195 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7196 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7197 (v16i8 (VEXTRACTI128rr
7198 (v32i8 VR256:$src1),
7199 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7203 let Predicates = [HasAVX] in {
7204 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7205 (v4f32 (VEXTRACTF128rr
7206 (v8f32 VR256:$src1),
7207 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7208 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7209 (v2f64 (VEXTRACTF128rr
7210 (v4f64 VR256:$src1),
7211 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7212 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7213 (v2i64 (VEXTRACTF128rr
7214 (v4i64 VR256:$src1),
7215 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7216 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7217 (v4i32 (VEXTRACTF128rr
7218 (v8i32 VR256:$src1),
7219 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7220 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7221 (v8i16 (VEXTRACTF128rr
7222 (v16i16 VR256:$src1),
7223 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7224 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7225 (v16i8 (VEXTRACTF128rr
7226 (v32i8 VR256:$src1),
7227 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7230 //===----------------------------------------------------------------------===//
7231 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7233 multiclass avx2_pmovmask<string OpcodeStr,
7234 Intrinsic IntLd128, Intrinsic IntLd256,
7235 Intrinsic IntSt128, Intrinsic IntSt256> {
7236 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7237 (ins VR128:$src1, i128mem:$src2),
7238 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7239 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7240 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7241 (ins VR256:$src1, i256mem:$src2),
7242 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7243 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7244 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7245 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7246 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7247 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7248 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7249 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7250 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7251 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7254 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7255 int_x86_avx2_maskload_d,
7256 int_x86_avx2_maskload_d_256,
7257 int_x86_avx2_maskstore_d,
7258 int_x86_avx2_maskstore_d_256>;
7259 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7260 int_x86_avx2_maskload_q,
7261 int_x86_avx2_maskload_q_256,
7262 int_x86_avx2_maskstore_q,
7263 int_x86_avx2_maskstore_q_256>, VEX_W;
7266 //===----------------------------------------------------------------------===//
7267 // Variable Bit Shifts
7269 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7270 ValueType vt128, ValueType vt256> {
7271 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7272 (ins VR128:$src1, VR128:$src2),
7273 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7275 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7277 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7278 (ins VR128:$src1, i128mem:$src2),
7279 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7281 (vt128 (OpNode VR128:$src1,
7282 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7284 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7285 (ins VR256:$src1, VR256:$src2),
7286 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7288 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7290 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7291 (ins VR256:$src1, i256mem:$src2),
7292 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7294 (vt256 (OpNode VR256:$src1,
7295 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7299 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7300 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7301 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7302 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7303 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;