[llvm] [X86][AVX512] Use comx for compare (PR #113098)

[Phoebe Wang via llvm-commits]

================
@@ -0,0 +1,121 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; 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:
+; X64:       # %bb.0:
+; X64-NEXT:    vucomxss %xmm1, %xmm0
+; X64-NEXT:    sete %al
+; X64-NEXT:    retq
+;
+; X86-LABEL: oeq:
+; X86:       # %bb.0:
+; X86-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT:    vucomxss {{[0-9]+}}(%esp), %xmm0
+; X86-NEXT:    sete %al
+; X86-NEXT:    retl
+    %1 = fcmp oeq float %x, %y
+    ret i1 %1
+}
+
+define i1 @une(float %x, float %y) {
+; X64-LABEL: une:
+; X64:       # %bb.0:
+; X64-NEXT:    vucomxss %xmm1, %xmm0
+; X64-NEXT:    setne %al
+; X64-NEXT:    retq
+;
+; X86-LABEL: une:
+; 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:    retl
+    %1 = fcmp une float %x, %y
+    ret i1 %1
+}
+
+define i1 @ogt(float %x, float %y) {
+; X64-LABEL: ogt:
+; X64:       # %bb.0:
+; X64-NEXT:    vucomiss %xmm1, %xmm0
+; X64-NEXT:    seta %al
+; X64-NEXT:    retq
+;
+; X86-LABEL: ogt:
+; 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:    retl
+    %1 = fcmp ogt float %x, %y
+    ret i1 %1
+}
+
+define i1 @oeq_mem(ptr %xp, ptr %yp) {
+; X64-LABEL: oeq_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:       # %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:    vucomxss (%eax), %xmm0
+; X86-NEXT:    sete %al
+; X86-NEXT:    retl
+    %x = load float, ptr %xp
+    %y = load float, ptr %yp
+    %1 = fcmp oeq float %x, %y
+    ret i1 %1
+}
+
+define i1 @une_mem(ptr %xp, ptr %yp) {
+; X64-LABEL: une_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:       # %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:    vucomxss (%eax), %xmm0
+; X86-NEXT:    setne %al
+; X86-NEXT:    retl
+    %x = load float, ptr %xp
+    %y = load float, ptr %yp
+    %1 = fcmp une float %x, %y
+    ret i1 %1
+}
+
+
+define i1 @ogt_mem(ptr %xp, ptr %yp) {
----------------
phoebewang wrote:

ditto.

https://github.com/llvm/llvm-project/pull/113098