[llvm] [X86] Lower mathlib call ldexp into scalef when avx512 is enabled (PR #69710)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,361 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
+; RUN: llc -mtriple=x86_64 -mattr=+avx512f < %s -o - | FileCheck %s --check-prefixes=AVX512
+; RUN: llc -mtriple=x86_64 -mattr=+avx512vl < %s -o - | FileCheck %s --check-prefixes=AVX512VL
+
+define half @test_half(half %x, i32 %exp) {
+; AVX512-LABEL: test_half:
+; AVX512:       # %bb.0: # %entry
+; AVX512-NEXT:    vpextrw $0, %xmm0, %eax
+; AVX512-NEXT:    vcvtsi2ss %edi, %xmm1, %xmm0
+; AVX512-NEXT:    movzwl %ax, %eax
+; AVX512-NEXT:    vmovd %eax, %xmm1
+; AVX512-NEXT:    vcvtph2ps %xmm1, %xmm1
+; AVX512-NEXT:    vscalefss %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
+; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    vpinsrw $0, %eax, %xmm0, %xmm0
+; AVX512-NEXT:    retq
+;
+; AVX512VL-LABEL: test_half:
+; AVX512VL:       # %bb.0: # %entry
+; AVX512VL-NEXT:    vpextrw $0, %xmm0, %eax
+; AVX512VL-NEXT:    vcvtsi2ss %edi, %xmm1, %xmm0
+; AVX512VL-NEXT:    movzwl %ax, %eax
+; AVX512VL-NEXT:    vmovd %eax, %xmm1
+; AVX512VL-NEXT:    vcvtph2ps %xmm1, %xmm1
+; AVX512VL-NEXT:    vscalefss %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
+; AVX512VL-NEXT:    vmovd %xmm0, %eax
+; AVX512VL-NEXT:    vpinsrw $0, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    retq
+entry:
+  %r = tail call fast half @llvm.ldexp.f16.i32(half %x, i32 %exp)
----------------
arsenm wrote:

Drop unnecessary fast flags?

https://github.com/llvm/llvm-project/pull/69710