add pattern to movberr

Author: XinWang10

llvm/lib/Target/X86/X86InstrMisc.td

@@ -1151,14 +1151,17 @@ let Predicates = [HasMOVBE, HasEGPR, In64BitMode] in {
                        EVEX, NoCD8, T_MAP4;
   }
   let SchedRW = [WriteALU] in {
-  def MOVBE16rr_EVEX : I<0x61, MRMDestReg, (outs), (ins GR16:$dst, GR16:$src),
-                         "movbe{w}\t{$src, $dst|$dst, $src}", []>,
+  def MOVBE16rr_EVEX : I<0x61, MRMDestReg, (outs GR16:$dst), (ins GR16:$src),
+                         "movbe{w}\t{$src, $dst|$dst, $src}", 
+                         [(set GR16:$dst, (bswap GR16:$src))]>,
                        EVEX, NoCD8, T_MAP4, PD;
-  def MOVBE32rr_EVEX : I<0x61, MRMDestReg, (outs), (ins GR32:$dst, GR32:$src),
-                         "movbe{l}\t{$src, $dst|$dst, $src}", []>,
+  def MOVBE32rr_EVEX : I<0x61, MRMDestReg, (outs GR32:$dst), (ins GR32:$src),
+                         "movbe{l}\t{$src, $dst|$dst, $src}", 
+                         [(set GR32:$dst, (bswap GR32:$src))]>,
                        EVEX, NoCD8, T_MAP4;
-  def MOVBE64rr_EVEX : RI<0x61, MRMDestReg, (outs), (ins GR64:$dst, GR64:$src),
-                          "movbe{q}\t{$src, $dst|$dst, $src}", []>,
+  def MOVBE64rr_EVEX : RI<0x61, MRMDestReg, (outs GR64:$dst), (ins GR64:$src),
+                          "movbe{q}\t{$src, $dst|$dst, $src}", 
+                          [(set GR64:$dst, (bswap GR64:$src))]>,
                        EVEX, NoCD8, T_MAP4;
 
   def MOVBE16rr_EVEX_REV : I<0x60, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),

llvm/test/CodeGen/X86/movbe.ll

@@ -162,3 +162,29 @@ define i64 @test6(ptr %x) nounwind {
   %bswap = call i64 @llvm.bswap.i64(i64 %load)
   ret i64 %bswap
 }
+
+define i64 @test7(i64 %x) nounwind {
+; CHECK-LABEL: test7:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    bswapq %rax
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    retq
+;
+; SLM-LABEL: test7:
+; SLM:       # %bb.0:
+; SLM-NEXT:    movq %rdi, %rax
+; SLM-NEXT:    bswapq %rax
+; SLM-NEXT:    retq
+;
+; EGPR-LABEL: test7:
+; EGPR:       # %bb.0:
+; EGPR-NEXT:    movq %rdi, %rax # encoding: [0x48,0x89,0xf8]
+; EGPR-NEXT:    bswapq %rax # encoding: [0x48,0x0f,0xc8]
+; EGPR-NEXT:    retq # encoding: [0xc3]
+  %bswap = call i64 @llvm.bswap.i64(i64 %x)
+  ret i64 %bswap
+}

llvm/test/TableGen/x86-fold-tables.inc

@@ -421,9 +421,9 @@ static const X86FoldTableEntry Table0[] = {
   {X86::MOV8rr_NOREX, X86::MOV8mr_NOREX, TB_FOLDED_STORE|TB_NO_REVERSE},
   {X86::MOVAPDrr, X86::MOVAPDmr, TB_FOLDED_STORE|TB_NO_REVERSE|TB_ALIGN_16},
   {X86::MOVAPSrr, X86::MOVAPSmr, TB_FOLDED_STORE|TB_NO_REVERSE|TB_ALIGN_16},
-  {X86::MOVBE16rr_EVEX, X86::MOVBE16mr_EVEX, TB_FOLDED_LOAD},
-  {X86::MOVBE32rr_EVEX, X86::MOVBE32mr_EVEX, TB_FOLDED_LOAD},
-  {X86::MOVBE64rr_EVEX, X86::MOVBE64mr_EVEX, TB_FOLDED_LOAD},
+  {X86::MOVBE16rr_EVEX, X86::MOVBE16mr_EVEX, TB_FOLDED_STORE},
+  {X86::MOVBE32rr_EVEX, X86::MOVBE32mr_EVEX, TB_FOLDED_STORE},
+  {X86::MOVBE64rr_EVEX, X86::MOVBE64mr_EVEX, TB_FOLDED_STORE},
   {X86::MOVDI2SSrr, X86::MOV32mr, TB_FOLDED_STORE|TB_NO_REVERSE},
   {X86::MOVDQArr, X86::MOVDQAmr, TB_FOLDED_STORE|TB_ALIGN_16},
   {X86::MOVDQUrr, X86::MOVDQUmr, TB_FOLDED_STORE},
@@ -775,6 +775,9 @@ static const X86FoldTableEntry Table1[] = {
   {X86::MOV8rr, X86::MOV8rm, 0},
   {X86::MOVAPDrr, X86::MOVAPDrm, TB_ALIGN_16},
   {X86::MOVAPSrr, X86::MOVAPSrm, TB_ALIGN_16},
+  {X86::MOVBE16rr_EVEX, X86::MOVBE16rm_EVEX, 0},
+  {X86::MOVBE32rr_EVEX, X86::MOVBE32rm_EVEX, 0},
+  {X86::MOVBE64rr_EVEX, X86::MOVBE64rm_EVEX, 0},
   {X86::MOVDDUPrr, X86::MOVDDUPrm, TB_NO_REVERSE},
   {X86::MOVDI2PDIrr, X86::MOVDI2PDIrm, 0},
   {X86::MOVDI2SSrr, X86::MOVSSrm_alt, 0},