[X86]Fix wrong asm match for VMASKMOVDQU

[XinWang10]

Summary

VMASKMOVDQU supports 32bit/64bit version in 64bitmode, previously we prefer to use VMASKMOVDQU64 in 64bitmode because the 32bit one need 0x67 prefix.
After D150436, asm match table changed a little, which makes in 64bit mode "vmaskmovdqu %xmm0, %xmm1" will match VMASKMOVDQU other than VMASKMOVDQU64, this patch correct the asm match order for this instruction.

[skan]

Although I know your point, but please explain why we'd like to match VMASKMOVDQU64 in the summary.

[skan]

LGTM

[craig.topper]

llvm/lib/Target/X86/X86InstrSSE.td

4091

Does the same bug exist for the SSE version?

[XinWang10]

llvm/lib/Target/X86/X86InstrSSE.td

4091

Exactly! Will give another patch to fix it.

llvm/lib/Target/X86/X86InstrSSE.td

 //===---------------------------------------------------------------------===//
 // SSE2 - Conditional Store
 //===---------------------------------------------------------------------===//
 
 let ExeDomain = SSEPackedInt, SchedRW = [SchedWriteVecMoveLS.XMM.MR] in {
 // As VEX does not have separate instruction contexts for address size
 // overrides, VMASKMOVDQU and VMASKMOVDQU64 would have a decode conflict.
 // Prefer VMASKMODDQU64.
-let Uses = [EDI], Predicates = [HasAVX], isAsmParserOnly = 1 in
-def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
-           (ins VR128:$src, VR128:$mask),
-           "maskmovdqu\t{$mask, $src|$src, $mask}",
-           [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>,
-           VEX, WIG;
 let Uses = [RDI], Predicates = [HasAVX,In64BitMode] in
 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
            (ins VR128:$src, VR128:$mask),
            "maskmovdqu\t{$mask, $src|$src, $mask}",
            [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>,
            VEX, WIG;
+let Uses = [EDI], Predicates = [HasAVX], isAsmParserOnly = 1 in
+def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
+           (ins VR128:$src, VR128:$mask),
+           "maskmovdqu\t{$mask, $src|$src, $mask}",
+           [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>,
+           VEX, WIG;
 
 let Uses = [EDI], Predicates = [UseSSE2] in
 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
            "maskmovdqu\t{$mask, $src|$src, $mask}",
            [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
 let Uses = [RDI], Predicates = [UseSSE2,In64BitMode] in
 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
            "maskmovdqu\t{$mask, $src|$src, $mask}",
            [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
 
 } // ExeDomain = SSEPackedInt
 
 //===---------------------------------------------------------------------===//
 // SSE2 - Move Doubleword/Quadword
 //===---------------------------------------------------------------------===//

llvm/test/MC/X86/x86_64-asm-match.s

 // CHECK: AsmMatcher: found 2 encodings with mnemonic 'crc32l'
 // CHECK: Trying to match opcode CRC32r32r32
 // CHECK:   Matching formal operand class MCK_GR32 against actual operand at index 1 (Memory: ModeSize=64,BaseReg=rbx,IndexReg=rcx,Scale=8,Disp=3735928559,SegReg=gs): Opcode result: multiple operand mismatches, ignoring this opcode
 // CHECK: Trying to match opcode CRC32r32m32
 // CHECK:   Matching formal operand class MCK_Mem32 against actual operand at index 1 (Memory: ModeSize=64,BaseReg=rbx,IndexReg=rcx,Scale=8,Disp=3735928559,SegReg=gs): match success using generic matcher
 // CHECK:   Matching formal operand class MCK_GR32 against actual operand at index 2 (Reg:ecx): match success using generic matcher
 // CHECK:   Matching formal operand class InvalidMatchClass against actual operand at index 3: actual operand index out of range
 // CHECK:   Opcode result: complete match, selecting this opcode
+// CHECK: AsmMatcher: found 2 encodings with mnemonic 'vmaskmovdqu'
+// CHECK: Trying to match opcode VMASKMOVDQU64
+// CHECK:   Matching formal operand class MCK_FR16 against actual operand at index 1 (Reg:xmm0): match success using generic matcher
+// CHECK:   Matching formal operand class MCK_FR16 against actual operand at index 2 (Reg:xmm1): match success using generic matcher
+// CHECK:   Matching formal operand class InvalidMatchClass against actual operand at index 3: actual operand index out of range
 // CHECK: AsmMatcher: found 4 encodings with mnemonic 'punpcklbw'
 // CHECK: Trying to match opcode MMX_PUNPCKLBWrr
 // CHECK:   Matching formal operand class MCK_VR64 against actual operand at index 1 (Reg:mm0): match success using generic matcher
 // CHECK:   Matching formal operand class MCK_VR64 against actual operand at index 2 (Memory: ModeSize=64,Size=32,BaseReg=rsp,Scale=1): Opcode result: multiple operand mismatches, ignoring this opcode
 // CHECK: Trying to match opcode MMX_PUNPCKLBWrm
 // CHECK:   Matching formal operand class MCK_VR64 against actual operand at index 1 (Reg:mm0): match success using generic matcher
 // CHECK:   Matching formal operand class MCK_Mem32 against actual operand at index 2 (Memory: ModeSize=64,Size=32,BaseReg=rsp,Scale=1): match success using generic matcher
 // CHECK:   Matching formal operand class InvalidMatchClass against actual operand at index 3: actual operand index out of range
 // CHECK:   Opcode result: complete match, selecting this opcode
 
 pshufb    CPI1_0(%rip), %xmm1
 sha1rnds4 $1, %xmm1, %xmm2
 pinsrw    $3, %ecx, %xmm5
 crc32l    %gs:0xdeadbeef(%rbx,%rcx,8),%ecx
+vmaskmovdqu     %xmm0, %xmm1
 
 .intel_syntax
 punpcklbw mm0, dword ptr [rsp]