update review comments for sh,sd and format

mattarde · mattarde · commit 6d5f26cda85c · 2024-10-21T02:17:38.000-07:00
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -49521,8 +49521,8 @@ static SDValue combineCompareEqual(SDNode *N, SelectionDAG &DAG,
           // See X86ATTInstPrinter.cpp:printSSECC().
           unsigned x86cc = (cc0 == X86::COND_E) ? 0 : 4;
 
-          // VCOMXSS simplifies conditional code sequence into single setcc node.
-          // Earlier until COMI, it required upto 2 SETCC's to test CC.
+          // VCOMXSS simplifies conditional code sequence into single setcc
+          // node. Earlier until COMI, it required upto 2 SETCC's to test CC.
           if (Subtarget.hasAVX10_2()) {
             return getSETCC(
                 ((cc0 == X86::COND_E) ? X86::COND_E : X86::COND_NE),
diff --git a/llvm/lib/Target/X86/X86InstrAVX10.td b/llvm/lib/Target/X86/X86InstrAVX10.td
@@ -1583,6 +1583,12 @@ multiclass avx10_com_ef_int<bits<8> Opc, X86VectorVTInfo _, SDNode OpNode,
 
 let Defs = [EFLAGS], Uses = [MXCSR], Predicates = [HasAVX10_2] in {
   
+  defm VUCOMXSDZ  :  avx10_com_ef<0x2e, FR64X, f64, X86ucomi512,
+                                  "vucomxsd", f64mem, loadf64, SSEPackedSingle>,
+                                  TB, XS, VEX_LIG, REX_W, EVEX_CD8<64, CD8VT1>;
+  defm VUCOMXSHZ  :  avx10_com_ef<0x2e, FR16X, f16, X86ucomi512,
+                                  "vucomxsh", f16mem, loadf16, SSEPackedSingle>,
+                                  T_MAP5, XD, EVEX_CD8<16, CD8VT1>;
   defm VUCOMXSSZ  :  avx10_com_ef<0x2e, FR32X, f32, X86ucomi512,
                                   "vucomxss", f32mem, loadf32, SSEPackedSingle>,
                                   TB, XD, VEX_LIG, EVEX_CD8<32, CD8VT1>;
diff --git a/llvm/test/CodeGen/X86/avx10_2-cmp.ll b/llvm/test/CodeGen/X86/avx10_2-cmp.ll
@@ -2,66 +2,127 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx10.2-512 | FileCheck %s --check-prefix=X64
 ; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=+avx10.2-512 | FileCheck %s --check-prefix=X86
 
-define i1 @oeq(float %x, float %y) {
-; X64-LABEL: oeq:
+define i1 @hoeq(half %x, half %y) {
+; X64-LABEL: hoeq:
 ; X64:       # %bb.0:
-; X64-NEXT:    vucomxss %xmm1, %xmm0
+; X64-NEXT:    vucomxsh %xmm1, %xmm0
 ; X64-NEXT:    sete %al
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: oeq:
+; X86-LABEL: hoeq:
 ; X86:       # %bb.0:
-; X86-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-NEXT:    vucomxss {{[0-9]+}}(%esp), %xmm0
+; X86-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
+; X86-NEXT:    vucomxsh {{[0-9]+}}(%esp), %xmm0
 ; X86-NEXT:    sete %al
 ; X86-NEXT:    retl
-    %1 = fcmp oeq float %x, %y
+    %1 = fcmp oeq half %x, %y
     ret i1 %1
 }
 
-define i1 @une(float %x, float %y) {
-; X64-LABEL: une:
+define i1 @hune(half %x, half %y) {
+; X64-LABEL: hune:
 ; X64:       # %bb.0:
-; X64-NEXT:    vucomxss %xmm1, %xmm0
+; X64-NEXT:    vucomxsh %xmm1, %xmm0
 ; X64-NEXT:    setne %al
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: une:
+; X86-LABEL: hune:
+; X86:       # %bb.0:
+; X86-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
+; X86-NEXT:    vucomxsh {{[0-9]+}}(%esp), %xmm0
+; X86-NEXT:    setne %al
+; X86-NEXT:    retl
+    %1 = fcmp une half %x, %y
+    ret i1 %1
+}
+
+define i1 @hoeq_mem(ptr %xp, ptr %yp) {
+; X64-LABEL: hoeq_mem:
+; X64:       # %bb.0:
+; X64-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
+; X64-NEXT:    vucomxsh (%rsi), %xmm0
+; X64-NEXT:    sete %al
+; X64-NEXT:    retq
+;
+; X86-LABEL: hoeq_mem:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
+; X86-NEXT:    vucomxsh (%eax), %xmm0
+; X86-NEXT:    sete %al
+; X86-NEXT:    retl
+    %x = load half, ptr %xp
+    %y = load half, ptr %yp
+    %1 = fcmp oeq half %x, %y
+    ret i1 %1
+}
+
+define i1 @hune_mem(ptr %xp, ptr %yp) {
+; X64-LABEL: hune_mem:
+; X64:       # %bb.0:
+; X64-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
+; X64-NEXT:    vucomxsh (%rsi), %xmm0
+; X64-NEXT:    setne %al
+; X64-NEXT:    retq
+;
+; X86-LABEL: hune_mem:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
+; X86-NEXT:    vucomxsh (%eax), %xmm0
+; X86-NEXT:    setne %al
+; X86-NEXT:    retl
+    %x = load half, ptr %xp
+    %y = load half, ptr %yp
+    %1 = fcmp une half %x, %y
+    ret i1 %1
+}
+
+define i1 @foeq(float %x, float %y) {
+; X64-LABEL: foeq:
+; X64:       # %bb.0:
+; X64-NEXT:    vucomxss %xmm1, %xmm0
+; X64-NEXT:    sete %al
+; X64-NEXT:    retq
+;
+; X86-LABEL: foeq:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; X86-NEXT:    vucomxss {{[0-9]+}}(%esp), %xmm0
-; X86-NEXT:    setne %al
+; X86-NEXT:    sete %al
 ; X86-NEXT:    retl
-    %1 = fcmp une float %x, %y
+    %1 = fcmp oeq float %x, %y
     ret i1 %1
 }
 
-define i1 @ogt(float %x, float %y) {
-; X64-LABEL: ogt:
+define i1 @fune(float %x, float %y) {
+; X64-LABEL: fune:
 ; X64:       # %bb.0:
-; X64-NEXT:    vucomiss %xmm1, %xmm0
-; X64-NEXT:    seta %al
+; X64-NEXT:    vucomxss %xmm1, %xmm0
+; X64-NEXT:    setne %al
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: ogt:
+; X86-LABEL: fune:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-NEXT:    vucomiss {{[0-9]+}}(%esp), %xmm0
-; X86-NEXT:    seta %al
+; X86-NEXT:    vucomxss {{[0-9]+}}(%esp), %xmm0
+; X86-NEXT:    setne %al
 ; X86-NEXT:    retl
-    %1 = fcmp ogt float %x, %y
+    %1 = fcmp une float %x, %y
     ret i1 %1
 }
 
-define i1 @oeq_mem(ptr %xp, ptr %yp) {
-; X64-LABEL: oeq_mem:
+define i1 @foeq_mem(ptr %xp, ptr %yp) {
+; X64-LABEL: foeq_mem:
 ; X64:       # %bb.0:
 ; X64-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; X64-NEXT:    vucomxss (%rsi), %xmm0
 ; X64-NEXT:    sete %al
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: oeq_mem:
+; X86-LABEL: foeq_mem:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
@@ -75,15 +136,15 @@ define i1 @oeq_mem(ptr %xp, ptr %yp) {
     ret i1 %1
 }
 
-define i1 @une_mem(ptr %xp, ptr %yp) {
-; X64-LABEL: une_mem:
+define i1 @fune_mem(ptr %xp, ptr %yp) {
+; X64-LABEL: fune_mem:
 ; X64:       # %bb.0:
 ; X64-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; X64-NEXT:    vucomxss (%rsi), %xmm0
 ; X64-NEXT:    setne %al
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: une_mem:
+; X86-LABEL: fune_mem:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
@@ -97,25 +158,80 @@ define i1 @une_mem(ptr %xp, ptr %yp) {
     ret i1 %1
 }
 
+define i1 @doeq(double %x, double %y) {
+; X64-LABEL: doeq:
+; X64:       # %bb.0:
+; X64-NEXT:    vucomxsd %xmm1, %xmm0
+; X64-NEXT:    sete %al
+; X64-NEXT:    retq
+;
+; X86-LABEL: doeq:
+; X86:       # %bb.0:
+; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X86-NEXT:    vucomxsd {{[0-9]+}}(%esp), %xmm0
+; X86-NEXT:    sete %al
+; X86-NEXT:    retl
+    %1 = fcmp oeq double %x, %y
+    ret i1 %1
+}
 
-define i1 @ogt_mem(ptr %xp, ptr %yp) {
-; X64-LABEL: ogt_mem:
+define i1 @dune(double %x, double %y) {
+; X64-LABEL: dune:
 ; X64:       # %bb.0:
-; X64-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X64-NEXT:    vucomiss (%rsi), %xmm0
-; X64-NEXT:    seta %al
+; X64-NEXT:    vucomxsd %xmm1, %xmm0
+; X64-NEXT:    setne %al
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: ogt_mem:
+; X86-LABEL: dune:
+; X86:       # %bb.0:
+; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X86-NEXT:    vucomxsd {{[0-9]+}}(%esp), %xmm0
+; X86-NEXT:    setne %al
+; X86-NEXT:    retl
+    %1 = fcmp une double %x, %y
+    ret i1 %1
+}
+
+define i1 @doeq_mem(ptr %xp, ptr %yp) {
+; X64-LABEL: doeq_mem:
+; X64:       # %bb.0:
+; X64-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    vucomxsd (%rsi), %xmm0
+; X64-NEXT:    sete %al
+; X64-NEXT:    retq
+;
+; X86-LABEL: doeq_mem:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-NEXT:    vucomiss (%eax), %xmm0
-; X86-NEXT:    seta %al
+; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X86-NEXT:    vucomxsd (%eax), %xmm0
+; X86-NEXT:    sete %al
 ; X86-NEXT:    retl
-    %x = load float, ptr %xp
-    %y = load float, ptr %yp
-    %1 = fcmp ogt float %x, %y
+    %x = load double, ptr %xp
+    %y = load double, ptr %yp
+    %1 = fcmp oeq double %x, %y
+    ret i1 %1
+}
+
+define i1 @dune_mem(ptr %xp, ptr %yp) {
+; X64-LABEL: dune_mem:
+; X64:       # %bb.0:
+; X64-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    vucomxsd (%rsi), %xmm0
+; X64-NEXT:    setne %al
+; X64-NEXT:    retq
+;
+; X86-LABEL: dune_mem:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X86-NEXT:    vucomxsd (%eax), %xmm0
+; X86-NEXT:    setne %al
+; X86-NEXT:    retl
+    %x = load double, ptr %xp
+    %y = load double, ptr %yp
+    %1 = fcmp une double %x, %y
     ret i1 %1
 }
diff --git a/llvm/test/TableGen/x86-fold-tables.inc b/llvm/test/TableGen/x86-fold-tables.inc
@@ -1959,7 +1959,9 @@ static const X86FoldTableEntry Table1[] = {
   {X86::VUCOMISSZrr_Int, X86::VUCOMISSZrm_Int, TB_NO_REVERSE},
   {X86::VUCOMISSrr, X86::VUCOMISSrm, 0},
   {X86::VUCOMISSrr_Int, X86::VUCOMISSrm_Int, TB_NO_REVERSE},
+  {X86::VUCOMXSDZrr, X86::VUCOMXSDZrm, 0},
   {X86::VUCOMXSDZrr_Int, X86::VUCOMXSDZrm_Int, TB_NO_REVERSE},
+  {X86::VUCOMXSHZrr, X86::VUCOMXSHZrm, 0},
   {X86::VUCOMXSHZrr_Int, X86::VUCOMXSHZrm_Int, TB_NO_REVERSE},
   {X86::VUCOMXSSZrr, X86::VUCOMXSSZrm, 0},
   {X86::VUCOMXSSZrr_Int, X86::VUCOMXSSZrm_Int, TB_NO_REVERSE},
