xref: /llvm-project/llvm/test/CodeGen/X86/apx/evex-to-vex.ll (revision a3b7b2d6354f622372ce6967e082869979c25ed8)
1; Check EVEX is not compressed into VEX when egpr is used.
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+egpr -show-mc-encoding | FileCheck %s
3
4define void @test_x86_vcvtps2ph_256_m(ptr nocapture %d, <8 x float> %a) nounwind {
5; CHECK:    vcvtps2ph $3, %ymm0, (%r16) # encoding: [0x62,0xfb,0x7d,0x28,0x1d,0x00,0x03]
6entry:
7  %0 = load i32, ptr %d, align 4
8  call void asm sideeffect "", "~{eax},~{ebx},~{ecx},~{edx},~{edi},~{esi},~{ebp},~{esp},~{r8d},~{r9d},~{r10d},~{r11d},~{r12d},~{r13d},~{r14d},~{r15d}"()
9  %1 = tail call <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float> %a, i32 3)
10  store <8 x i16> %1, ptr %d, align 16
11  ret void
12}
13
14declare <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float>, i32) nounwind readonly
15