[llvm] [X86] Support ATOMIC_LOAD_FP_BINOP_MI for other binops (PR #87524)

[Phoebe Wang via llvm-commits]

================
@@ -777,3 +777,2053 @@ bb:
   store atomic i64 %tmp9, ptr %tmp4 monotonic, align 8
   ret void
 }
+
+; ----- FSUB -----
+
+define dso_local void @fsub_32r(ptr %loc, float %val) nounwind {
+; X86-NOSSE-LABEL: fsub_32r:
+; X86-NOSSE:       # %bb.0:
+; X86-NOSSE-NEXT:    subl $8, %esp
+; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOSSE-NEXT:    movl (%eax), %ecx
+; X86-NOSSE-NEXT:    movl %ecx, (%esp)
+; X86-NOSSE-NEXT:    flds (%esp)
+; X86-NOSSE-NEXT:    fsubs {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NOSSE-NEXT:    movl %ecx, (%eax)
+; X86-NOSSE-NEXT:    addl $8, %esp
+; X86-NOSSE-NEXT:    retl
+;
+; X86-SSE1-LABEL: fsub_32r:
+; X86-SSE1:       # %bb.0:
+; X86-SSE1-NEXT:    subl $8, %esp
+; X86-SSE1-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE1-NEXT:    movl (%eax), %ecx
+; X86-SSE1-NEXT:    movl %ecx, (%esp)
+; X86-SSE1-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-SSE1-NEXT:    subss {{[0-9]+}}(%esp), %xmm0
+; X86-SSE1-NEXT:    movss %xmm0, {{[0-9]+}}(%esp)
+; X86-SSE1-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SSE1-NEXT:    movl %ecx, (%eax)
+; X86-SSE1-NEXT:    addl $8, %esp
+; X86-SSE1-NEXT:    retl
+;
+; X86-SSE2-LABEL: fsub_32r:
+; X86-SSE2:       # %bb.0:
+; X86-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE2-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-SSE2-NEXT:    subss {{[0-9]+}}(%esp), %xmm0
+; X86-SSE2-NEXT:    movss %xmm0, (%eax)
+; X86-SSE2-NEXT:    retl
+;
+; X86-AVX-LABEL: fsub_32r:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-AVX-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-AVX-NEXT:    vsubss {{[0-9]+}}(%esp), %xmm0, %xmm0
+; X86-AVX-NEXT:    vmovss %xmm0, (%eax)
+; X86-AVX-NEXT:    retl
+;
+; X64-SSE-LABEL: fsub_32r:
+; X64-SSE:       # %bb.0:
+; X64-SSE-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X64-SSE-NEXT:    subss %xmm0, %xmm1
+; X64-SSE-NEXT:    movss %xmm1, (%rdi)
+; X64-SSE-NEXT:    retq
+;
+; X64-AVX-LABEL: fsub_32r:
+; X64-AVX:       # %bb.0:
+; X64-AVX-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; X64-AVX-NEXT:    vsubss %xmm0, %xmm1, %xmm0
+; X64-AVX-NEXT:    vmovss %xmm0, (%rdi)
+; X64-AVX-NEXT:    retq
+  %1 = load atomic i32, ptr %loc seq_cst, align 4
+  %2 = bitcast i32 %1 to float
+  %sub = fsub float %2, %val
+  %3 = bitcast float %sub to i32
+  store atomic i32 %3, ptr %loc release, align 4
+  ret void
----------------
phoebewang wrote:

I think the code is duplicated 4 times, how about
```
; RUN: sed 's/OP/fadd/g' %s | ... | FileCheck --check-prefix=FADD...
... ...
  %1 = load atomic i32, ptr %loc seq_cst, align 4
  %2 = bitcast i32 %1 to float
  %sub = OP float %2, %val
  %3 = bitcast float %sub to i32
  store atomic i32 %3, ptr %loc release, align 4
  ret void
```

https://github.com/llvm/llvm-project/pull/87524