[X86] Fix test case for SoftPromoteHalf of STRICT_FP_EXTEND/STRICT_FP_ROUND.

[vtjnash]

Summary

Tests that should have been with
33b9f3abd78ffe31e2f468f64d36dbdf75b25d6e
when writing the tests that should have been with
0daf9b8e41327b1511b2bbc272184ff4fdb8de79.

[vtjnash]

I noticed the first 2 RUN statements in the file were accidentally identical, so I fixed one and reran the update_llc_test_checks.py to fix the tests

[craig.topper]

LGTM

llvm/test/CodeGen/X86/half-constrained.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i686-apple-darwin | FileCheck %s --check-prefix=X32-NOF16C
-; RUN: llc < %s -mtriple=i686-apple-darwin | FileCheck %s --check-prefix=X32-F16C
+; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=f16c | FileCheck %s --check-prefix=X32-F16C
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin | FileCheck %s --check-prefix=X64-NOF16C
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=f16c | FileCheck %s --check-prefix=X64-F16C
 
 @a = global half 0xH0000, align 2
 @b = global half 0xH0000, align 2
 @c = global half 0xH0000, align 2
 
 define float @half_to_float() strictfp {
 ; X32-NOF16C-LABEL: half_to_float:
 ; X32-NOF16C:       ## %bb.0:
 ; X32-NOF16C-NEXT:    subl $12, %esp
 ; X32-NOF16C-NEXT:    .cfi_def_cfa_offset 16
 ; X32-NOF16C-NEXT:    movzwl _a, %eax
 ; X32-NOF16C-NEXT:    movl %eax, (%esp)
 ; X32-NOF16C-NEXT:    calll ___extendhfsf2
 ; X32-NOF16C-NEXT:    addl $12, %esp
 ; X32-NOF16C-NEXT:    retl
 ;
 ; X32-F16C-LABEL: half_to_float:
 ; X32-F16C:       ## %bb.0:
-; X32-F16C-NEXT:    subl $12, %esp
-; X32-F16C-NEXT:    .cfi_def_cfa_offset 16
+; X32-F16C-NEXT:    pushl %eax
+; X32-F16C-NEXT:    .cfi_def_cfa_offset 8
 ; X32-F16C-NEXT:    movzwl _a, %eax
-; X32-F16C-NEXT:    movl %eax, (%esp)
-; X32-F16C-NEXT:    calll ___extendhfsf2
-; X32-F16C-NEXT:    addl $12, %esp
+; X32-F16C-NEXT:    vmovd %eax, %xmm0
+; X32-F16C-NEXT:    vcvtph2ps %xmm0, %xmm0
+; X32-F16C-NEXT:    vmovss %xmm0, (%esp)
+; X32-F16C-NEXT:    flds (%esp)
+; X32-F16C-NEXT:    wait
+; X32-F16C-NEXT:    popl %eax
 ; X32-F16C-NEXT:    retl
 ;
 ; X64-NOF16C-LABEL: half_to_float:
 ; X64-NOF16C:       ## %bb.0:
 ; X64-NOF16C-NEXT:    pushq %rax
 ; X64-NOF16C-NEXT:    .cfi_def_cfa_offset 16
 ; X64-NOF16C-NEXT:    movzwl _a(%rip), %edi
 ; X64-NOF16C-NEXT:    callq ___extendhfsf2

 ; X32-NOF16C-NEXT:    addl $12, %esp
 ; X32-NOF16C-NEXT:    retl
 ;
 ; X32-F16C-LABEL: half_to_double:
 ; X32-F16C:       ## %bb.0:
 ; X32-F16C-NEXT:    subl $12, %esp
 ; X32-F16C-NEXT:    .cfi_def_cfa_offset 16
 ; X32-F16C-NEXT:    movzwl _a, %eax
-; X32-F16C-NEXT:    movl %eax, (%esp)
-; X32-F16C-NEXT:    calll ___extendhfsf2
+; X32-F16C-NEXT:    vmovd %eax, %xmm0
+; X32-F16C-NEXT:    vcvtph2ps %xmm0, %xmm0
+; X32-F16C-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
+; X32-F16C-NEXT:    vmovsd %xmm0, (%esp)
+; X32-F16C-NEXT:    fldl (%esp)
+; X32-F16C-NEXT:    wait
 ; X32-F16C-NEXT:    addl $12, %esp
 ; X32-F16C-NEXT:    retl
 ;
 ; X64-NOF16C-LABEL: half_to_double:
 ; X64-NOF16C:       ## %bb.0:
 ; X64-NOF16C-NEXT:    pushq %rax
 ; X64-NOF16C-NEXT:    .cfi_def_cfa_offset 16
 ; X64-NOF16C-NEXT:    movzwl _a(%rip), %edi

 ; X32-NOF16C-NEXT:    movzwl _a, %eax
 ; X32-NOF16C-NEXT:    movl %eax, (%esp)
 ; X32-NOF16C-NEXT:    calll ___extendhfsf2
 ; X32-NOF16C-NEXT:    addl $12, %esp
 ; X32-NOF16C-NEXT:    retl
 ;
 ; X32-F16C-LABEL: half_to_fp80:
 ; X32-F16C:       ## %bb.0:
-; X32-F16C-NEXT:    subl $12, %esp
-; X32-F16C-NEXT:    .cfi_def_cfa_offset 16
+; X32-F16C-NEXT:    pushl %eax
+; X32-F16C-NEXT:    .cfi_def_cfa_offset 8
 ; X32-F16C-NEXT:    movzwl _a, %eax
-; X32-F16C-NEXT:    movl %eax, (%esp)
-; X32-F16C-NEXT:    calll ___extendhfsf2
-; X32-F16C-NEXT:    addl $12, %esp
+; X32-F16C-NEXT:    vmovd %eax, %xmm0
+; X32-F16C-NEXT:    vcvtph2ps %xmm0, %xmm0
+; X32-F16C-NEXT:    vmovss %xmm0, (%esp)
+; X32-F16C-NEXT:    flds (%esp)
+; X32-F16C-NEXT:    wait
+; X32-F16C-NEXT:    popl %eax
 ; X32-F16C-NEXT:    retl
 ;
 ; X64-NOF16C-LABEL: half_to_fp80:
 ; X64-NOF16C:       ## %bb.0:
 ; X64-NOF16C-NEXT:    pushq %rax
 ; X64-NOF16C-NEXT:    .cfi_def_cfa_offset 16
 ; X64-NOF16C-NEXT:    movzwl _a(%rip), %edi
 ; X64-NOF16C-NEXT:    callq ___extendhfsf2

 ; X32-NOF16C-NEXT:    wait
 ; X32-NOF16C-NEXT:    calll ___truncsfhf2
 ; X32-NOF16C-NEXT:    movw %ax, _a
 ; X32-NOF16C-NEXT:    addl $12, %esp
 ; X32-NOF16C-NEXT:    retl
 ;
 ; X32-F16C-LABEL: float_to_half:
 ; X32-F16C:       ## %bb.0:
-; X32-F16C-NEXT:    subl $12, %esp
-; X32-F16C-NEXT:    .cfi_def_cfa_offset 16
-; X32-F16C-NEXT:    flds {{[0-9]+}}(%esp)
-; X32-F16C-NEXT:    fstps (%esp)
-; X32-F16C-NEXT:    wait
-; X32-F16C-NEXT:    calll ___truncsfhf2
-; X32-F16C-NEXT:    movw %ax, _a
-; X32-F16C-NEXT:    addl $12, %esp
+; X32-F16C-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X32-F16C-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
+; X32-F16C-NEXT:    vpextrw $0, %xmm0, _a
 ; X32-F16C-NEXT:    retl
 ;
 ; X64-NOF16C-LABEL: float_to_half:
 ; X64-NOF16C:       ## %bb.0:
 ; X64-NOF16C-NEXT:    pushq %rax
 ; X64-NOF16C-NEXT:    .cfi_def_cfa_offset 16
 ; X64-NOF16C-NEXT:    callq ___truncsfhf2
 ; X64-NOF16C-NEXT:    movw %ax, _a(%rip)

 ; X32-NOF16C-NEXT:    movw %ax, _a
 ; X32-NOF16C-NEXT:    addl $12, %esp
 ; X32-NOF16C-NEXT:    retl
 ;
 ; X32-F16C-LABEL: double_to_half:
 ; X32-F16C:       ## %bb.0:
 ; X32-F16C-NEXT:    subl $12, %esp
 ; X32-F16C-NEXT:    .cfi_def_cfa_offset 16
-; X32-F16C-NEXT:    fldl {{[0-9]+}}(%esp)
-; X32-F16C-NEXT:    fstpl (%esp)
-; X32-F16C-NEXT:    wait
+; X32-F16C-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X32-F16C-NEXT:    vmovsd %xmm0, (%esp)
 ; X32-F16C-NEXT:    calll ___truncdfhf2
 ; X32-F16C-NEXT:    movw %ax, _a
 ; X32-F16C-NEXT:    addl $12, %esp
 ; X32-F16C-NEXT:    retl
 ;
 ; X64-NOF16C-LABEL: double_to_half:
 ; X64-NOF16C:       ## %bb.0:
 ; X64-NOF16C-NEXT:    pushq %rax

 ; X32-NOF16C-NEXT:    wait
 ; X32-NOF16C-NEXT:    calll ___truncsfhf2
 ; X32-NOF16C-NEXT:    movw %ax, _c
 ; X32-NOF16C-NEXT:    addl $12, %esp
 ; X32-NOF16C-NEXT:    retl
 ;
 ; X32-F16C-LABEL: add:
 ; X32-F16C:       ## %bb.0:
-; X32-F16C-NEXT:    subl $12, %esp
-; X32-F16C-NEXT:    .cfi_def_cfa_offset 16
 ; X32-F16C-NEXT:    movzwl _a, %eax
-; X32-F16C-NEXT:    movl %eax, (%esp)
-; X32-F16C-NEXT:    calll ___extendhfsf2
-; X32-F16C-NEXT:    fstps {{[-0-9]+}}(%e{{[sb]}}p) ## 4-byte Folded Spill
-; X32-F16C-NEXT:    wait
+; X32-F16C-NEXT:    vmovd %eax, %xmm0
+; X32-F16C-NEXT:    vcvtph2ps %xmm0, %xmm0
 ; X32-F16C-NEXT:    movzwl _b, %eax
-; X32-F16C-NEXT:    movl %eax, (%esp)
-; X32-F16C-NEXT:    calll ___extendhfsf2
-; X32-F16C-NEXT:    flds {{[-0-9]+}}(%e{{[sb]}}p) ## 4-byte Folded Reload
-; X32-F16C-NEXT:    faddp %st, %st(1)
-; X32-F16C-NEXT:    fstps (%esp)
-; X32-F16C-NEXT:    wait
-; X32-F16C-NEXT:    calll ___truncsfhf2
-; X32-F16C-NEXT:    movw %ax, _c
-; X32-F16C-NEXT:    addl $12, %esp
+; X32-F16C-NEXT:    vmovd %eax, %xmm1
+; X32-F16C-NEXT:    vcvtph2ps %xmm1, %xmm1
+; X32-F16C-NEXT:    vaddss %xmm1, %xmm0, %xmm0
+; X32-F16C-NEXT:    vxorps %xmm1, %xmm1, %xmm1
+; X32-F16C-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
+; X32-F16C-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
+; X32-F16C-NEXT:    vpextrw $0, %xmm0, _c
 ; X32-F16C-NEXT:    retl
 ;
 ; X64-NOF16C-LABEL: add:
 ; X64-NOF16C:       ## %bb.0:
 ; X64-NOF16C-NEXT:    pushq %rax
 ; X64-NOF16C-NEXT:    .cfi_def_cfa_offset 16
 ; X64-NOF16C-NEXT:    movzwl _a(%rip), %edi
 ; X64-NOF16C-NEXT:    callq ___extendhfsf2