[llvm] r276563 - [X86][AVX512VL] Added AVX512VL half2float vector conversions tests to demonstrate PR23941
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Sun Jul 24 06:01:52 PDT 2016
Author: rksimon
Date: Sun Jul 24 08:01:51 2016
New Revision: 276563
URL: http://llvm.org/viewvc/llvm-project?rev=276563&view=rev
Log:
[X86][AVX512VL] Added AVX512VL half2float vector conversions tests to demonstrate PR23941
Modified:
llvm/trunk/test/CodeGen/X86/vector-half-conversions.ll
Modified: llvm/trunk/test/CodeGen/X86/vector-half-conversions.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-half-conversions.ll?rev=276563&r1=276562&r2=276563&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-half-conversions.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-half-conversions.ll Sun Jul 24 08:01:51 2016
@@ -1,89 +1,290 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+f16c | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+f16c | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+f16c | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+f16c | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX512 --check-prefix=AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX512 --check-prefix=AVX512VL
;
; Half to Float
;
-define float @cvt_i16_to_f32(i16 %a0) {
-; ALL-LABEL: cvt_i16_to_f32:
-; ALL: # BB#0:
-; ALL-NEXT: movswl %di, %eax
-; ALL-NEXT: vmovd %eax, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: retq
+define float @cvt_i16_to_f32(i16 %a0) nounwind {
+; AVX1-LABEL: cvt_i16_to_f32:
+; AVX1: # BB#0:
+; AVX1-NEXT: movswl %di, %eax
+; AVX1-NEXT: vmovd %eax, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_i16_to_f32:
+; AVX2: # BB#0:
+; AVX2-NEXT: movswl %di, %eax
+; AVX2-NEXT: vmovd %eax, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_i16_to_f32:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movswl %di, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_i16_to_f32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: movzwl %di, %edi
+; AVX512VL-NEXT: jmp __gnu_h2f_ieee # TAILCALL
%1 = bitcast i16 %a0 to half
%2 = fpext half %1 to float
ret float %2
}
-define <4 x float> @cvt_4i16_to_4f32(<4 x i16> %a0) {
-; ALL-LABEL: cvt_4i16_to_4f32:
-; ALL: # BB#0:
-; ALL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; ALL-NEXT: vmovq %xmm0, %rax
-; ALL-NEXT: movq %rax, %rcx
-; ALL-NEXT: movq %rax, %rdx
-; ALL-NEXT: movswl %ax, %esi
-; ALL-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
-; ALL-NEXT: shrl $16, %eax
-; ALL-NEXT: shrq $32, %rcx
-; ALL-NEXT: shrq $48, %rdx
-; ALL-NEXT: movswl %dx, %edx
-; ALL-NEXT: vmovd %edx, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: movswl %cx, %ecx
-; ALL-NEXT: vmovd %ecx, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: cwtl
-; ALL-NEXT: vmovd %eax, %xmm2
-; ALL-NEXT: vcvtph2ps %xmm2, %xmm2
-; ALL-NEXT: vmovd %esi, %xmm3
-; ALL-NEXT: vcvtph2ps %xmm3, %xmm3
-; ALL-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
-; ALL-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
-; ALL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
-; ALL-NEXT: retq
+define <4 x float> @cvt_4i16_to_4f32(<4 x i16> %a0) nounwind {
+; AVX1-LABEL: cvt_4i16_to_4f32:
+; AVX1: # BB#0:
+; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX1-NEXT: vmovq %xmm0, %rax
+; AVX1-NEXT: movq %rax, %rcx
+; AVX1-NEXT: movq %rax, %rdx
+; AVX1-NEXT: movswl %ax, %esi
+; AVX1-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: shrq $32, %rcx
+; AVX1-NEXT: shrq $48, %rdx
+; AVX1-NEXT: movswl %dx, %edx
+; AVX1-NEXT: vmovd %edx, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: movswl %cx, %ecx
+; AVX1-NEXT: vmovd %ecx, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: cwtl
+; AVX1-NEXT: vmovd %eax, %xmm2
+; AVX1-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX1-NEXT: vmovd %esi, %xmm3
+; AVX1-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX1-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
+; AVX1-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX1-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_4i16_to_4f32:
+; AVX2: # BB#0:
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX2-NEXT: vmovq %xmm0, %rax
+; AVX2-NEXT: movq %rax, %rcx
+; AVX2-NEXT: movq %rax, %rdx
+; AVX2-NEXT: movswl %ax, %esi
+; AVX2-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: shrq $32, %rcx
+; AVX2-NEXT: shrq $48, %rdx
+; AVX2-NEXT: movswl %dx, %edx
+; AVX2-NEXT: vmovd %edx, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: movswl %cx, %ecx
+; AVX2-NEXT: vmovd %ecx, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: cwtl
+; AVX2-NEXT: vmovd %eax, %xmm2
+; AVX2-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX2-NEXT: vmovd %esi, %xmm3
+; AVX2-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX2-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
+; AVX2-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX2-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_4i16_to_4f32:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX512F-NEXT: vmovq %xmm0, %rax
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: movq %rax, %rdx
+; AVX512F-NEXT: movswl %ax, %esi
+; AVX512F-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX512F-NEXT: shrl $16, %eax
+; AVX512F-NEXT: shrq $32, %rcx
+; AVX512F-NEXT: shrq $48, %rdx
+; AVX512F-NEXT: movswl %dx, %edx
+; AVX512F-NEXT: vmovd %edx, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: vmovd %esi, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_4i16_to_4f32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $64, %rsp
+; AVX512VL-NEXT: vpmovdw %xmm0, {{[0-9]+}}(%rsp)
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rbp
+; AVX512VL-NEXT: movl %ebp, %r14d
+; AVX512VL-NEXT: shrl $16, %r14d
+; AVX512VL-NEXT: movq %rbp, %rbx
+; AVX512VL-NEXT: shrq $32, %rbx
+; AVX512VL-NEXT: movq %rbp, %rdi
+; AVX512VL-NEXT: shrq $48, %rdi
+; AVX512VL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<kill>
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bp, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %r14d, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1,2],mem[0]
+; AVX512VL-NEXT: addq $64, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = bitcast <4 x i16> %a0 to <4 x half>
%2 = fpext <4 x half> %1 to <4 x float>
ret <4 x float> %2
}
-define <4 x float> @cvt_8i16_to_4f32(<8 x i16> %a0) {
-; ALL-LABEL: cvt_8i16_to_4f32:
-; ALL: # BB#0:
-; ALL-NEXT: vmovq %xmm0, %rax
-; ALL-NEXT: movq %rax, %rcx
-; ALL-NEXT: movq %rax, %rdx
-; ALL-NEXT: movswl %ax, %esi
-; ALL-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
-; ALL-NEXT: shrl $16, %eax
-; ALL-NEXT: shrq $32, %rcx
-; ALL-NEXT: shrq $48, %rdx
-; ALL-NEXT: movswl %dx, %edx
-; ALL-NEXT: vmovd %edx, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: movswl %cx, %ecx
-; ALL-NEXT: vmovd %ecx, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: cwtl
-; ALL-NEXT: vmovd %eax, %xmm2
-; ALL-NEXT: vcvtph2ps %xmm2, %xmm2
-; ALL-NEXT: vmovd %esi, %xmm3
-; ALL-NEXT: vcvtph2ps %xmm3, %xmm3
-; ALL-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
-; ALL-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
-; ALL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
-; ALL-NEXT: retq
+define <4 x float> @cvt_8i16_to_4f32(<8 x i16> %a0) nounwind {
+; AVX1-LABEL: cvt_8i16_to_4f32:
+; AVX1: # BB#0:
+; AVX1-NEXT: vmovq %xmm0, %rax
+; AVX1-NEXT: movq %rax, %rcx
+; AVX1-NEXT: movq %rax, %rdx
+; AVX1-NEXT: movswl %ax, %esi
+; AVX1-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: shrq $32, %rcx
+; AVX1-NEXT: shrq $48, %rdx
+; AVX1-NEXT: movswl %dx, %edx
+; AVX1-NEXT: vmovd %edx, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: movswl %cx, %ecx
+; AVX1-NEXT: vmovd %ecx, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: cwtl
+; AVX1-NEXT: vmovd %eax, %xmm2
+; AVX1-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX1-NEXT: vmovd %esi, %xmm3
+; AVX1-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX1-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
+; AVX1-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX1-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_8i16_to_4f32:
+; AVX2: # BB#0:
+; AVX2-NEXT: vmovq %xmm0, %rax
+; AVX2-NEXT: movq %rax, %rcx
+; AVX2-NEXT: movq %rax, %rdx
+; AVX2-NEXT: movswl %ax, %esi
+; AVX2-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: shrq $32, %rcx
+; AVX2-NEXT: shrq $48, %rdx
+; AVX2-NEXT: movswl %dx, %edx
+; AVX2-NEXT: vmovd %edx, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: movswl %cx, %ecx
+; AVX2-NEXT: vmovd %ecx, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: cwtl
+; AVX2-NEXT: vmovd %eax, %xmm2
+; AVX2-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX2-NEXT: vmovd %esi, %xmm3
+; AVX2-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX2-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
+; AVX2-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX2-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_8i16_to_4f32:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovq %xmm0, %rax
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: movq %rax, %rdx
+; AVX512F-NEXT: movswl %ax, %esi
+; AVX512F-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX512F-NEXT: shrl $16, %eax
+; AVX512F-NEXT: shrq $32, %rcx
+; AVX512F-NEXT: shrq $48, %rdx
+; AVX512F-NEXT: movswl %dx, %edx
+; AVX512F-NEXT: vmovd %edx, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: vmovd %esi, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_8i16_to_4f32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $64, %rsp
+; AVX512VL-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX512VL-NEXT: vpmovdw %xmm0, {{[0-9]+}}(%rsp)
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rbp
+; AVX512VL-NEXT: movl %ebp, %r14d
+; AVX512VL-NEXT: shrl $16, %r14d
+; AVX512VL-NEXT: movq %rbp, %rbx
+; AVX512VL-NEXT: shrq $32, %rbx
+; AVX512VL-NEXT: movq %rbp, %rdi
+; AVX512VL-NEXT: shrq $48, %rdi
+; AVX512VL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<kill>
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bp, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %r14d, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1,2],mem[0]
+; AVX512VL-NEXT: addq $64, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%2 = bitcast <4 x i16> %1 to <4 x half>
%3 = fpext <4 x half> %2 to <4 x float>
ret <4 x float> %3
}
-define <8 x float> @cvt_8i16_to_8f32(<8 x i16> %a0) {
+define <8 x float> @cvt_8i16_to_8f32(<8 x i16> %a0) nounwind {
; AVX1-LABEL: cvt_8i16_to_8f32:
; AVX1: # BB#0:
; AVX1-NEXT: vpextrq $1, %xmm0, %rdx
@@ -182,60 +383,130 @@ define <8 x float> @cvt_8i16_to_8f32(<8
; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0
; AVX2-NEXT: retq
;
-; AVX512-LABEL: cvt_8i16_to_8f32:
-; AVX512: # BB#0:
-; AVX512-NEXT: vpextrq $1, %xmm0, %rdx
-; AVX512-NEXT: movq %rdx, %r8
-; AVX512-NEXT: movq %rdx, %r10
-; AVX512-NEXT: movswl %dx, %r9d
-; AVX512-NEXT: # kill: %EDX<def> %EDX<kill> %RDX<kill>
-; AVX512-NEXT: shrl $16, %edx
-; AVX512-NEXT: shrq $32, %r8
-; AVX512-NEXT: shrq $48, %r10
-; AVX512-NEXT: vmovq %xmm0, %rdi
-; AVX512-NEXT: movq %rdi, %rax
-; AVX512-NEXT: movq %rdi, %rsi
-; AVX512-NEXT: movswl %di, %ecx
-; AVX512-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<kill>
-; AVX512-NEXT: shrl $16, %edi
-; AVX512-NEXT: shrq $32, %rax
-; AVX512-NEXT: shrq $48, %rsi
-; AVX512-NEXT: movswl %si, %esi
-; AVX512-NEXT: vmovd %esi, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm0
-; AVX512-NEXT: cwtl
-; AVX512-NEXT: vmovd %eax, %xmm1
-; AVX512-NEXT: vcvtph2ps %xmm1, %xmm1
-; AVX512-NEXT: movswl %di, %eax
-; AVX512-NEXT: vmovd %eax, %xmm2
-; AVX512-NEXT: vcvtph2ps %xmm2, %xmm2
-; AVX512-NEXT: vmovd %ecx, %xmm3
-; AVX512-NEXT: vcvtph2ps %xmm3, %xmm3
-; AVX512-NEXT: movswl %r10w, %eax
-; AVX512-NEXT: vmovd %eax, %xmm4
-; AVX512-NEXT: vcvtph2ps %xmm4, %xmm4
-; AVX512-NEXT: movswl %r8w, %eax
-; AVX512-NEXT: vmovd %eax, %xmm5
-; AVX512-NEXT: vcvtph2ps %xmm5, %xmm5
-; AVX512-NEXT: movswl %dx, %eax
-; AVX512-NEXT: vmovd %eax, %xmm6
-; AVX512-NEXT: vcvtph2ps %xmm6, %xmm6
-; AVX512-NEXT: vmovd %r9d, %xmm7
-; AVX512-NEXT: vcvtph2ps %xmm7, %xmm7
-; AVX512-NEXT: vinsertps {{.*#+}} xmm6 = xmm7[0],xmm6[0],xmm7[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm5 = xmm6[0,1],xmm5[0],xmm6[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[0]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
-; AVX512-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm0
-; AVX512-NEXT: retq
+; AVX512F-LABEL: cvt_8i16_to_8f32:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vpextrq $1, %xmm0, %rdx
+; AVX512F-NEXT: movq %rdx, %r8
+; AVX512F-NEXT: movq %rdx, %r10
+; AVX512F-NEXT: movswl %dx, %r9d
+; AVX512F-NEXT: # kill: %EDX<def> %EDX<kill> %RDX<kill>
+; AVX512F-NEXT: shrl $16, %edx
+; AVX512F-NEXT: shrq $32, %r8
+; AVX512F-NEXT: shrq $48, %r10
+; AVX512F-NEXT: vmovq %xmm0, %rdi
+; AVX512F-NEXT: movq %rdi, %rax
+; AVX512F-NEXT: movq %rdi, %rsi
+; AVX512F-NEXT: movswl %di, %ecx
+; AVX512F-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<kill>
+; AVX512F-NEXT: shrl $16, %edi
+; AVX512F-NEXT: shrq $32, %rax
+; AVX512F-NEXT: shrq $48, %rsi
+; AVX512F-NEXT: movswl %si, %esi
+; AVX512F-NEXT: vmovd %esi, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: movswl %di, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: vmovd %ecx, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: movswl %r10w, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm4
+; AVX512F-NEXT: vcvtph2ps %xmm4, %xmm4
+; AVX512F-NEXT: movswl %r8w, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm5
+; AVX512F-NEXT: vcvtph2ps %xmm5, %xmm5
+; AVX512F-NEXT: movswl %dx, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm6
+; AVX512F-NEXT: vcvtph2ps %xmm6, %xmm6
+; AVX512F-NEXT: vmovd %r9d, %xmm7
+; AVX512F-NEXT: vcvtph2ps %xmm7, %xmm7
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm6 = xmm7[0],xmm6[0],xmm7[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm5 = xmm6[0,1],xmm5[0],xmm6[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[0]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_8i16_to_8f32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %r13
+; AVX512VL-NEXT: pushq %r12
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $136, %rsp
+; AVX512VL-NEXT: vpextrq $1, %xmm0, %r15
+; AVX512VL-NEXT: movl %r15d, %eax
+; AVX512VL-NEXT: shrl $16, %eax
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: movq %r15, %r12
+; AVX512VL-NEXT: shrq $32, %r12
+; AVX512VL-NEXT: movq %r15, %r13
+; AVX512VL-NEXT: shrq $48, %r13
+; AVX512VL-NEXT: vmovq %xmm0, %r14
+; AVX512VL-NEXT: movl %r14d, %ebp
+; AVX512VL-NEXT: shrl $16, %ebp
+; AVX512VL-NEXT: movq %r14, %rbx
+; AVX512VL-NEXT: shrq $32, %rbx
+; AVX512VL-NEXT: movq %r14, %rdi
+; AVX512VL-NEXT: shrq $48, %rdi
+; AVX512VL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<kill>
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %r14w, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %ebp, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %r13d, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %r12w, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %r15w, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl {{[0-9]+}}(%rsp), %edi # 4-byte Reload
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1,2],mem[0]
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps $16, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1],mem[0],xmm1[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1,2],mem[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: addq $136, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r12
+; AVX512VL-NEXT: popq %r13
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = bitcast <8 x i16> %a0 to <8 x half>
%2 = fpext <8 x half> %1 to <8 x float>
ret <8 x float> %2
}
-define <16 x float> @cvt_16i16_to_16f32(<16 x i16> %a0) {
+define <16 x float> @cvt_16i16_to_16f32(<16 x i16> %a0) nounwind {
; AVX1-LABEL: cvt_16i16_to_16f32:
; AVX1: # BB#0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4
@@ -426,101 +697,238 @@ define <16 x float> @cvt_16i16_to_16f32(
; AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm2, %ymm1
; AVX2-NEXT: retq
;
-; AVX512-LABEL: cvt_16i16_to_16f32:
-; AVX512: # BB#0:
-; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm10
-; AVX512-NEXT: vmovq %xmm0, %rax
-; AVX512-NEXT: movq %rax, %rcx
-; AVX512-NEXT: shrq $48, %rcx
-; AVX512-NEXT: movswl %cx, %ecx
-; AVX512-NEXT: vmovd %ecx, %xmm8
-; AVX512-NEXT: movq %rax, %rcx
-; AVX512-NEXT: shrq $32, %rcx
-; AVX512-NEXT: movswl %cx, %ecx
-; AVX512-NEXT: vmovd %ecx, %xmm9
-; AVX512-NEXT: movswl %ax, %ecx
-; AVX512-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
-; AVX512-NEXT: shrl $16, %eax
-; AVX512-NEXT: cwtl
-; AVX512-NEXT: vmovd %eax, %xmm11
-; AVX512-NEXT: vpextrq $1, %xmm0, %rax
-; AVX512-NEXT: vmovd %ecx, %xmm12
-; AVX512-NEXT: movq %rax, %rcx
-; AVX512-NEXT: shrq $48, %rcx
-; AVX512-NEXT: movswl %cx, %ecx
-; AVX512-NEXT: vmovd %ecx, %xmm13
-; AVX512-NEXT: movq %rax, %rcx
-; AVX512-NEXT: shrq $32, %rcx
-; AVX512-NEXT: movswl %cx, %ecx
-; AVX512-NEXT: vmovd %ecx, %xmm14
-; AVX512-NEXT: movswl %ax, %ecx
-; AVX512-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
-; AVX512-NEXT: shrl $16, %eax
-; AVX512-NEXT: cwtl
-; AVX512-NEXT: vmovd %eax, %xmm15
-; AVX512-NEXT: vmovq %xmm10, %rax
-; AVX512-NEXT: vmovd %ecx, %xmm2
-; AVX512-NEXT: movq %rax, %rcx
-; AVX512-NEXT: shrq $48, %rcx
-; AVX512-NEXT: movswl %cx, %ecx
-; AVX512-NEXT: vmovd %ecx, %xmm3
-; AVX512-NEXT: movq %rax, %rcx
-; AVX512-NEXT: shrq $32, %rcx
-; AVX512-NEXT: movswl %cx, %ecx
-; AVX512-NEXT: vmovd %ecx, %xmm1
-; AVX512-NEXT: movswl %ax, %ecx
-; AVX512-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
-; AVX512-NEXT: shrl $16, %eax
-; AVX512-NEXT: cwtl
-; AVX512-NEXT: vmovd %eax, %xmm4
-; AVX512-NEXT: vpextrq $1, %xmm10, %rax
-; AVX512-NEXT: vmovd %ecx, %xmm10
-; AVX512-NEXT: movq %rax, %rcx
-; AVX512-NEXT: shrq $48, %rcx
-; AVX512-NEXT: movswl %cx, %ecx
-; AVX512-NEXT: vmovd %ecx, %xmm5
-; AVX512-NEXT: movq %rax, %rcx
-; AVX512-NEXT: shrq $32, %rcx
-; AVX512-NEXT: movswl %cx, %ecx
-; AVX512-NEXT: vmovd %ecx, %xmm6
-; AVX512-NEXT: movl %eax, %ecx
-; AVX512-NEXT: shrl $16, %ecx
-; AVX512-NEXT: movswl %cx, %ecx
-; AVX512-NEXT: vmovd %ecx, %xmm7
-; AVX512-NEXT: cwtl
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm8, %xmm8
-; AVX512-NEXT: vcvtph2ps %xmm9, %xmm9
-; AVX512-NEXT: vcvtph2ps %xmm11, %xmm11
-; AVX512-NEXT: vcvtph2ps %xmm12, %xmm12
-; AVX512-NEXT: vcvtph2ps %xmm13, %xmm13
-; AVX512-NEXT: vcvtph2ps %xmm14, %xmm14
-; AVX512-NEXT: vcvtph2ps %xmm15, %xmm15
-; AVX512-NEXT: vcvtph2ps %xmm2, %xmm2
-; AVX512-NEXT: vcvtph2ps %xmm3, %xmm3
-; AVX512-NEXT: vcvtph2ps %xmm1, %xmm1
-; AVX512-NEXT: vcvtph2ps %xmm4, %xmm4
-; AVX512-NEXT: vcvtph2ps %xmm10, %xmm10
-; AVX512-NEXT: vcvtph2ps %xmm5, %xmm5
-; AVX512-NEXT: vcvtph2ps %xmm6, %xmm6
-; AVX512-NEXT: vcvtph2ps %xmm7, %xmm7
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm0
-; AVX512-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1],xmm6[0],xmm0[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],xmm5[0]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm4 = xmm10[0],xmm4[0],xmm10[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm4[0,1],xmm1[0],xmm4[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm3[0]
-; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm15[0],xmm2[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm14[0],xmm1[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm13[0]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm2 = xmm12[0],xmm11[0],xmm12[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],xmm9[0],xmm2[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],xmm8[0]
-; AVX512-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1
-; AVX512-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512-NEXT: retq
+; AVX512F-LABEL: cvt_16i16_to_16f32:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm10
+; AVX512F-NEXT: vmovq %xmm0, %rax
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: shrq $48, %rcx
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm8
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: shrq $32, %rcx
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm9
+; AVX512F-NEXT: movswl %ax, %ecx
+; AVX512F-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX512F-NEXT: shrl $16, %eax
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm11
+; AVX512F-NEXT: vpextrq $1, %xmm0, %rax
+; AVX512F-NEXT: vmovd %ecx, %xmm12
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: shrq $48, %rcx
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm13
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: shrq $32, %rcx
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm14
+; AVX512F-NEXT: movswl %ax, %ecx
+; AVX512F-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX512F-NEXT: shrl $16, %eax
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm15
+; AVX512F-NEXT: vmovq %xmm10, %rax
+; AVX512F-NEXT: vmovd %ecx, %xmm2
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: shrq $48, %rcx
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm3
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: shrq $32, %rcx
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm1
+; AVX512F-NEXT: movswl %ax, %ecx
+; AVX512F-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX512F-NEXT: shrl $16, %eax
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm4
+; AVX512F-NEXT: vpextrq $1, %xmm10, %rax
+; AVX512F-NEXT: vmovd %ecx, %xmm10
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: shrq $48, %rcx
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm5
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: shrq $32, %rcx
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm6
+; AVX512F-NEXT: movl %eax, %ecx
+; AVX512F-NEXT: shrl $16, %ecx
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm7
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm8, %xmm8
+; AVX512F-NEXT: vcvtph2ps %xmm9, %xmm9
+; AVX512F-NEXT: vcvtph2ps %xmm11, %xmm11
+; AVX512F-NEXT: vcvtph2ps %xmm12, %xmm12
+; AVX512F-NEXT: vcvtph2ps %xmm13, %xmm13
+; AVX512F-NEXT: vcvtph2ps %xmm14, %xmm14
+; AVX512F-NEXT: vcvtph2ps %xmm15, %xmm15
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm4, %xmm4
+; AVX512F-NEXT: vcvtph2ps %xmm10, %xmm10
+; AVX512F-NEXT: vcvtph2ps %xmm5, %xmm5
+; AVX512F-NEXT: vcvtph2ps %xmm6, %xmm6
+; AVX512F-NEXT: vcvtph2ps %xmm7, %xmm7
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1],xmm6[0],xmm0[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],xmm5[0]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm4 = xmm10[0],xmm4[0],xmm10[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm4[0,1],xmm1[0],xmm4[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm3[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm15[0],xmm2[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm14[0],xmm1[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm13[0]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm12[0],xmm11[0],xmm12[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],xmm9[0],xmm2[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],xmm8[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1
+; AVX512F-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_16i16_to_16f32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %r13
+; AVX512VL-NEXT: pushq %r12
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $264, %rsp # imm = 0x108
+; AVX512VL-NEXT: vextracti32x4 $1, %ymm0, %xmm1
+; AVX512VL-NEXT: vpextrq $1, %xmm1, %rcx
+; AVX512VL-NEXT: movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; AVX512VL-NEXT: movl %ecx, %eax
+; AVX512VL-NEXT: shrl $16, %eax
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: movq %rcx, %rax
+; AVX512VL-NEXT: shrq $32, %rax
+; AVX512VL-NEXT: movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; AVX512VL-NEXT: shrq $48, %rcx
+; AVX512VL-NEXT: movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; AVX512VL-NEXT: vmovq %xmm1, %rcx
+; AVX512VL-NEXT: movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; AVX512VL-NEXT: movl %ecx, %eax
+; AVX512VL-NEXT: shrl $16, %eax
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: movq %rcx, %rax
+; AVX512VL-NEXT: shrq $32, %rax
+; AVX512VL-NEXT: movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; AVX512VL-NEXT: shrq $48, %rcx
+; AVX512VL-NEXT: movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; AVX512VL-NEXT: vpextrq $1, %xmm0, %r15
+; AVX512VL-NEXT: movl %r15d, %eax
+; AVX512VL-NEXT: shrl $16, %eax
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: movq %r15, %r13
+; AVX512VL-NEXT: shrq $32, %r13
+; AVX512VL-NEXT: movq %r15, %rbx
+; AVX512VL-NEXT: shrq $48, %rbx
+; AVX512VL-NEXT: vmovq %xmm0, %rbp
+; AVX512VL-NEXT: movl %ebp, %r14d
+; AVX512VL-NEXT: shrl $16, %r14d
+; AVX512VL-NEXT: movq %rbp, %r12
+; AVX512VL-NEXT: shrq $32, %r12
+; AVX512VL-NEXT: movq %rbp, %rdi
+; AVX512VL-NEXT: shrq $48, %rdi
+; AVX512VL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<kill>
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %r12w, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bp, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %r14d, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %ebx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %r13w, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %r15w, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl {{[0-9]+}}(%rsp), %edi # 4-byte Reload
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; AVX512VL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<kill>
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; AVX512VL-NEXT: movzwl %ax, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; AVX512VL-NEXT: movzwl %ax, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl {{[0-9]+}}(%rsp), %edi # 4-byte Reload
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; AVX512VL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<kill>
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; AVX512VL-NEXT: movzwl %ax, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; AVX512VL-NEXT: movzwl %ax, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl {{[0-9]+}}(%rsp), %edi # 4-byte Reload
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1,2],mem[0]
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps $16, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1],mem[0],xmm1[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1,2],mem[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps $16, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1],mem[0],xmm1[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1,2],mem[0]
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps $16, {{[0-9]+}}(%rsp), %xmm2, %xmm2 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm2 = xmm2[0],mem[0],xmm2[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm2, %xmm2 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm2 = xmm2[0,1],mem[0],xmm2[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm2, %xmm2 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm2 = xmm2[0,1,2],mem[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VL-NEXT: addq $264, %rsp # imm = 0x108
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r12
+; AVX512VL-NEXT: popq %r13
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = bitcast <16 x i16> %a0 to <16 x half>
%2 = fpext <16 x half> %1 to <16 x float>
ret <16 x float> %2
@@ -530,70 +938,244 @@ define <16 x float> @cvt_16i16_to_16f32(
; Half to Float (Load)
;
-define float @load_cvt_i16_to_f32(i16* %a0) {
-; ALL-LABEL: load_cvt_i16_to_f32:
-; ALL: # BB#0:
-; ALL-NEXT: movswl (%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: retq
+define float @load_cvt_i16_to_f32(i16* %a0) nounwind {
+; AVX1-LABEL: load_cvt_i16_to_f32:
+; AVX1: # BB#0:
+; AVX1-NEXT: movswl (%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: load_cvt_i16_to_f32:
+; AVX2: # BB#0:
+; AVX2-NEXT: movswl (%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: load_cvt_i16_to_f32:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movswl (%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_cvt_i16_to_f32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: movzwl (%rdi), %edi
+; AVX512VL-NEXT: jmp __gnu_h2f_ieee # TAILCALL
%1 = load i16, i16* %a0
%2 = bitcast i16 %1 to half
%3 = fpext half %2 to float
ret float %3
}
-define <4 x float> @load_cvt_4i16_to_4f32(<4 x i16>* %a0) {
-; ALL-LABEL: load_cvt_4i16_to_4f32:
-; ALL: # BB#0:
-; ALL-NEXT: movswl 6(%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: movswl 4(%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: movswl (%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm2
-; ALL-NEXT: vcvtph2ps %xmm2, %xmm2
-; ALL-NEXT: movswl 2(%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm3
-; ALL-NEXT: vcvtph2ps %xmm3, %xmm3
-; ALL-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[2,3]
-; ALL-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
-; ALL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
-; ALL-NEXT: retq
+define <4 x float> @load_cvt_4i16_to_4f32(<4 x i16>* %a0) nounwind {
+; AVX1-LABEL: load_cvt_4i16_to_4f32:
+; AVX1: # BB#0:
+; AVX1-NEXT: movswl 6(%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: movswl 4(%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: movswl (%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm2
+; AVX1-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX1-NEXT: movswl 2(%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm3
+; AVX1-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX1-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[2,3]
+; AVX1-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX1-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: load_cvt_4i16_to_4f32:
+; AVX2: # BB#0:
+; AVX2-NEXT: movswl 6(%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: movswl 4(%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: movswl (%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm2
+; AVX2-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX2-NEXT: movswl 2(%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm3
+; AVX2-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX2-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[2,3]
+; AVX2-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX2-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: load_cvt_4i16_to_4f32:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movswl 6(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: movswl 4(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: movswl (%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: movswl 2(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_cvt_4i16_to_4f32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $48, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: movzwl 6(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 4(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl (%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 2(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1,2],mem[0]
+; AVX512VL-NEXT: addq $48, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: retq
%1 = load <4 x i16>, <4 x i16>* %a0
%2 = bitcast <4 x i16> %1 to <4 x half>
%3 = fpext <4 x half> %2 to <4 x float>
ret <4 x float> %3
}
-define <4 x float> @load_cvt_8i16_to_4f32(<8 x i16>* %a0) {
-; ALL-LABEL: load_cvt_8i16_to_4f32:
-; ALL: # BB#0:
-; ALL-NEXT: movq (%rdi), %rax
-; ALL-NEXT: movq %rax, %rcx
-; ALL-NEXT: movq %rax, %rdx
-; ALL-NEXT: movswl %ax, %esi
-; ALL-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
-; ALL-NEXT: shrl $16, %eax
-; ALL-NEXT: shrq $32, %rcx
-; ALL-NEXT: shrq $48, %rdx
-; ALL-NEXT: movswl %dx, %edx
-; ALL-NEXT: vmovd %edx, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: movswl %cx, %ecx
-; ALL-NEXT: vmovd %ecx, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: cwtl
-; ALL-NEXT: vmovd %eax, %xmm2
-; ALL-NEXT: vcvtph2ps %xmm2, %xmm2
-; ALL-NEXT: vmovd %esi, %xmm3
-; ALL-NEXT: vcvtph2ps %xmm3, %xmm3
-; ALL-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
-; ALL-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
-; ALL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
-; ALL-NEXT: retq
+define <4 x float> @load_cvt_8i16_to_4f32(<8 x i16>* %a0) nounwind {
+; AVX1-LABEL: load_cvt_8i16_to_4f32:
+; AVX1: # BB#0:
+; AVX1-NEXT: movq (%rdi), %rax
+; AVX1-NEXT: movq %rax, %rcx
+; AVX1-NEXT: movq %rax, %rdx
+; AVX1-NEXT: movswl %ax, %esi
+; AVX1-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: shrq $32, %rcx
+; AVX1-NEXT: shrq $48, %rdx
+; AVX1-NEXT: movswl %dx, %edx
+; AVX1-NEXT: vmovd %edx, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: movswl %cx, %ecx
+; AVX1-NEXT: vmovd %ecx, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: cwtl
+; AVX1-NEXT: vmovd %eax, %xmm2
+; AVX1-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX1-NEXT: vmovd %esi, %xmm3
+; AVX1-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX1-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
+; AVX1-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX1-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: load_cvt_8i16_to_4f32:
+; AVX2: # BB#0:
+; AVX2-NEXT: movq (%rdi), %rax
+; AVX2-NEXT: movq %rax, %rcx
+; AVX2-NEXT: movq %rax, %rdx
+; AVX2-NEXT: movswl %ax, %esi
+; AVX2-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: shrq $32, %rcx
+; AVX2-NEXT: shrq $48, %rdx
+; AVX2-NEXT: movswl %dx, %edx
+; AVX2-NEXT: vmovd %edx, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: movswl %cx, %ecx
+; AVX2-NEXT: vmovd %ecx, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: cwtl
+; AVX2-NEXT: vmovd %eax, %xmm2
+; AVX2-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX2-NEXT: vmovd %esi, %xmm3
+; AVX2-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX2-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
+; AVX2-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX2-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: load_cvt_8i16_to_4f32:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movq (%rdi), %rax
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: movq %rax, %rdx
+; AVX512F-NEXT: movswl %ax, %esi
+; AVX512F-NEXT: # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; AVX512F-NEXT: shrl $16, %eax
+; AVX512F-NEXT: shrq $32, %rcx
+; AVX512F-NEXT: shrq $48, %rdx
+; AVX512F-NEXT: movswl %dx, %edx
+; AVX512F-NEXT: vmovd %edx, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: vmovd %esi, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_cvt_8i16_to_4f32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $64, %rsp
+; AVX512VL-NEXT: vmovdqa64 (%rdi), %xmm0
+; AVX512VL-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX512VL-NEXT: vpmovdw %xmm0, {{[0-9]+}}(%rsp)
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rbp
+; AVX512VL-NEXT: movl %ebp, %r14d
+; AVX512VL-NEXT: shrl $16, %r14d
+; AVX512VL-NEXT: movq %rbp, %rbx
+; AVX512VL-NEXT: shrq $32, %rbx
+; AVX512VL-NEXT: movq %rbp, %rdi
+; AVX512VL-NEXT: shrq $48, %rdi
+; AVX512VL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<kill>
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bp, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %r14d, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1,2],mem[0]
+; AVX512VL-NEXT: addq $64, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = load <8 x i16>, <8 x i16>* %a0
%2 = shufflevector <8 x i16> %1, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%3 = bitcast <4 x i16> %2 to <4 x half>
@@ -601,7 +1183,7 @@ define <4 x float> @load_cvt_8i16_to_4f3
ret <4 x float> %4
}
-define <8 x float> @load_cvt_8i16_to_8f32(<8 x i16>* %a0) {
+define <8 x float> @load_cvt_8i16_to_8f32(<8 x i16>* %a0) nounwind {
; AVX1-LABEL: load_cvt_8i16_to_8f32:
; AVX1: # BB#0:
; AVX1-NEXT: movswl 6(%rdi), %eax
@@ -672,47 +1254,93 @@ define <8 x float> @load_cvt_8i16_to_8f3
; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_cvt_8i16_to_8f32:
-; AVX512: # BB#0:
-; AVX512-NEXT: movswl 6(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm0
-; AVX512-NEXT: movswl 4(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm1
-; AVX512-NEXT: vcvtph2ps %xmm1, %xmm1
-; AVX512-NEXT: movswl (%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm2
-; AVX512-NEXT: vcvtph2ps %xmm2, %xmm2
-; AVX512-NEXT: movswl 2(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm3
-; AVX512-NEXT: vcvtph2ps %xmm3, %xmm3
-; AVX512-NEXT: movswl 14(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm4
-; AVX512-NEXT: vcvtph2ps %xmm4, %xmm4
-; AVX512-NEXT: movswl 12(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm5
-; AVX512-NEXT: vcvtph2ps %xmm5, %xmm5
-; AVX512-NEXT: movswl 8(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm6
-; AVX512-NEXT: vcvtph2ps %xmm6, %xmm6
-; AVX512-NEXT: movswl 10(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm7
-; AVX512-NEXT: vcvtph2ps %xmm7, %xmm7
-; AVX512-NEXT: vinsertps {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm5 = xmm6[0,1],xmm5[0],xmm6[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[0]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
-; AVX512-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm0
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_cvt_8i16_to_8f32:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movswl 6(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: movswl 4(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: movswl (%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: movswl 2(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: movswl 14(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm4
+; AVX512F-NEXT: vcvtph2ps %xmm4, %xmm4
+; AVX512F-NEXT: movswl 12(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm5
+; AVX512F-NEXT: vcvtph2ps %xmm5, %xmm5
+; AVX512F-NEXT: movswl 8(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm6
+; AVX512F-NEXT: vcvtph2ps %xmm6, %xmm6
+; AVX512F-NEXT: movswl 10(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm7
+; AVX512F-NEXT: vcvtph2ps %xmm7, %xmm7
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm5 = xmm6[0,1],xmm5[0],xmm6[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[0]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_cvt_8i16_to_8f32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $112, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: movzwl 6(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 4(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl (%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 2(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 14(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 12(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 8(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 10(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1,2],mem[0]
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps $16, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1],mem[0],xmm1[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1,2],mem[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: addq $112, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: retq
%1 = load <8 x i16>, <8 x i16>* %a0
%2 = bitcast <8 x i16> %1 to <8 x half>
%3 = fpext <8 x half> %2 to <8 x float>
ret <8 x float> %3
}
-define <16 x float> @load_cvt_16i16_to_16f32(<16 x i16>* %a0) {
+define <16 x float> @load_cvt_16i16_to_16f32(<16 x i16>* %a0) nounwind {
; AVX1-LABEL: load_cvt_16i16_to_16f32:
; AVX1: # BB#0:
; AVX1-NEXT: movswl 22(%rdi), %eax
@@ -845,72 +1473,158 @@ define <16 x float> @load_cvt_16i16_to_1
; AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm2, %ymm1
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_cvt_16i16_to_16f32:
-; AVX512: # BB#0:
-; AVX512-NEXT: movswl 6(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm8
-; AVX512-NEXT: movswl 4(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm9
-; AVX512-NEXT: movswl (%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm10
-; AVX512-NEXT: movswl 2(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm11
-; AVX512-NEXT: movswl 14(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm12
-; AVX512-NEXT: movswl 12(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm13
-; AVX512-NEXT: movswl 8(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm14
-; AVX512-NEXT: movswl 10(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm15
-; AVX512-NEXT: movswl 22(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm0
-; AVX512-NEXT: movswl 20(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm1
-; AVX512-NEXT: vcvtph2ps %xmm1, %xmm1
-; AVX512-NEXT: movswl 16(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm2
-; AVX512-NEXT: vcvtph2ps %xmm2, %xmm2
-; AVX512-NEXT: movswl 18(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm3
-; AVX512-NEXT: vcvtph2ps %xmm3, %xmm3
-; AVX512-NEXT: movswl 30(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm4
-; AVX512-NEXT: vcvtph2ps %xmm4, %xmm4
-; AVX512-NEXT: movswl 28(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm5
-; AVX512-NEXT: vcvtph2ps %xmm5, %xmm5
-; AVX512-NEXT: movswl 24(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm6
-; AVX512-NEXT: vcvtph2ps %xmm6, %xmm6
-; AVX512-NEXT: movswl 26(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm7
-; AVX512-NEXT: vcvtph2ps %xmm7, %xmm7
-; AVX512-NEXT: vinsertps {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm5 = xmm6[0,1],xmm5[0],xmm6[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[0]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
-; AVX512-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm0
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm14[0],xmm15[0],xmm14[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm13[0],xmm1[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm12[0]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm2 = xmm10[0],xmm11[0],xmm10[2,3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],xmm9[0],xmm2[3]
-; AVX512-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],xmm8[0]
-; AVX512-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1
-; AVX512-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_cvt_16i16_to_16f32:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movswl 6(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm8
+; AVX512F-NEXT: movswl 4(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm9
+; AVX512F-NEXT: movswl (%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm10
+; AVX512F-NEXT: movswl 2(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm11
+; AVX512F-NEXT: movswl 14(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm12
+; AVX512F-NEXT: movswl 12(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm13
+; AVX512F-NEXT: movswl 8(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm14
+; AVX512F-NEXT: movswl 10(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm15
+; AVX512F-NEXT: movswl 22(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: movswl 20(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: movswl 16(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: movswl 18(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: movswl 30(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm4
+; AVX512F-NEXT: vcvtph2ps %xmm4, %xmm4
+; AVX512F-NEXT: movswl 28(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm5
+; AVX512F-NEXT: vcvtph2ps %xmm5, %xmm5
+; AVX512F-NEXT: movswl 24(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm6
+; AVX512F-NEXT: vcvtph2ps %xmm6, %xmm6
+; AVX512F-NEXT: movswl 26(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm7
+; AVX512F-NEXT: vcvtph2ps %xmm7, %xmm7
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm5 = xmm6[0,1],xmm5[0],xmm6[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[0]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1],xmm1[0],xmm2[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm0
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm14[0],xmm15[0],xmm14[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm13[0],xmm1[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm12[0]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm10[0],xmm11[0],xmm10[2,3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],xmm9[0],xmm2[3]
+; AVX512F-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],xmm8[0]
+; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1
+; AVX512F-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_cvt_16i16_to_16f32:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $240, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: movzwl 6(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 4(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl (%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 2(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 14(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 12(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 8(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 10(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 22(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 20(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 16(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 18(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 30(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 28(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 24(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 26(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm0 = xmm0[0,1,2],mem[0]
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps $16, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1],mem[0],xmm1[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1,2],mem[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps $16, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0],mem[0],xmm1[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1],mem[0],xmm1[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm1 = xmm1[0,1,2],mem[0]
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vinsertps $16, {{[0-9]+}}(%rsp), %xmm2, %xmm2 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm2 = xmm2[0],mem[0],xmm2[2,3]
+; AVX512VL-NEXT: vinsertps $32, {{[0-9]+}}(%rsp), %xmm2, %xmm2 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm2 = xmm2[0,1],mem[0],xmm2[3]
+; AVX512VL-NEXT: vinsertps $48, {{[0-9]+}}(%rsp), %xmm2, %xmm2 # 16-byte Folded Reload
+; AVX512VL-NEXT: # xmm2 = xmm2[0,1,2],mem[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VL-NEXT: addq $240, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: retq
%1 = load <16 x i16>, <16 x i16>* %a0
%2 = bitcast <16 x i16> %1 to <16 x half>
%3 = fpext <16 x half> %2 to <16 x float>
@@ -921,139 +1635,472 @@ define <16 x float> @load_cvt_16i16_to_1
; Half to Double
;
-define double @cvt_i16_to_f64(i16 %a0) {
-; ALL-LABEL: cvt_i16_to_f64:
-; ALL: # BB#0:
-; ALL-NEXT: movswl %di, %eax
-; ALL-NEXT: vmovd %eax, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; ALL-NEXT: retq
+define double @cvt_i16_to_f64(i16 %a0) nounwind {
+; AVX1-LABEL: cvt_i16_to_f64:
+; AVX1: # BB#0:
+; AVX1-NEXT: movswl %di, %eax
+; AVX1-NEXT: vmovd %eax, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_i16_to_f64:
+; AVX2: # BB#0:
+; AVX2-NEXT: movswl %di, %eax
+; AVX2-NEXT: vmovd %eax, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_i16_to_f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movswl %di, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_i16_to_f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rax
+; AVX512VL-NEXT: movzwl %di, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: popq %rax
+; AVX512VL-NEXT: retq
%1 = bitcast i16 %a0 to half
%2 = fpext half %1 to double
ret double %2
}
-define <2 x double> @cvt_2i16_to_2f64(<2 x i16> %a0) {
-; ALL-LABEL: cvt_2i16_to_2f64:
-; ALL: # BB#0:
-; ALL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; ALL-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; ALL-NEXT: vmovd %xmm0, %eax
-; ALL-NEXT: movswl %ax, %ecx
-; ALL-NEXT: shrl $16, %eax
-; ALL-NEXT: cwtl
-; ALL-NEXT: vmovd %eax, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: vmovd %ecx, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
-; ALL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; ALL-NEXT: retq
+define <2 x double> @cvt_2i16_to_2f64(<2 x i16> %a0) nounwind {
+; AVX1-LABEL: cvt_2i16_to_2f64:
+; AVX1: # BB#0:
+; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX1-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX1-NEXT: vmovd %xmm0, %eax
+; AVX1-NEXT: movswl %ax, %ecx
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: cwtl
+; AVX1-NEXT: vmovd %eax, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: vmovd %ecx, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_2i16_to_2f64:
+; AVX2: # BB#0:
+; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX2-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX2-NEXT: vmovd %xmm0, %eax
+; AVX2-NEXT: movswl %ax, %ecx
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: cwtl
+; AVX2-NEXT: vmovd %eax, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: vmovd %ecx, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_2i16_to_2f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512F-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: movswl %ax, %ecx
+; AVX512F-NEXT: shrl $16, %eax
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %ecx, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_2i16_to_2f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $32, %rsp
+; AVX512VL-NEXT: vpmovqw %xmm0, {{[0-9]+}}(%rsp)
+; AVX512VL-NEXT: movl {{[0-9]+}}(%rsp), %eax
+; AVX512VL-NEXT: movl %eax, %ebx
+; AVX512VL-NEXT: shrl $16, %ebx
+; AVX512VL-NEXT: movzwl %ax, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %ebx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: addq $32, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: retq
%1 = bitcast <2 x i16> %a0 to <2 x half>
%2 = fpext <2 x half> %1 to <2 x double>
ret <2 x double> %2
}
-define <4 x double> @cvt_4i16_to_4f64(<4 x i16> %a0) {
-; ALL-LABEL: cvt_4i16_to_4f64:
-; ALL: # BB#0:
-; ALL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; ALL-NEXT: vmovq %xmm0, %rax
-; ALL-NEXT: movq %rax, %rcx
-; ALL-NEXT: movl %eax, %edx
-; ALL-NEXT: movswl %ax, %esi
-; ALL-NEXT: shrq $48, %rax
-; ALL-NEXT: shrq $32, %rcx
-; ALL-NEXT: shrl $16, %edx
-; ALL-NEXT: movswl %dx, %edx
-; ALL-NEXT: vmovd %edx, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: vmovd %esi, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: movswl %cx, %ecx
-; ALL-NEXT: vmovd %ecx, %xmm2
-; ALL-NEXT: vcvtph2ps %xmm2, %xmm2
-; ALL-NEXT: cwtl
-; ALL-NEXT: vmovd %eax, %xmm3
-; ALL-NEXT: vcvtph2ps %xmm3, %xmm3
-; ALL-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
-; ALL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; ALL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
-; ALL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; ALL-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; ALL-NEXT: retq
- %1 = bitcast <4 x i16> %a0 to <4 x half>
- %2 = fpext <4 x half> %1 to <4 x double>
- ret <4 x double> %2
-}
-
-define <2 x double> @cvt_8i16_to_2f64(<8 x i16> %a0) {
-; ALL-LABEL: cvt_8i16_to_2f64:
-; ALL: # BB#0:
-; ALL-NEXT: vmovd %xmm0, %eax
-; ALL-NEXT: movswl %ax, %ecx
-; ALL-NEXT: shrl $16, %eax
-; ALL-NEXT: cwtl
-; ALL-NEXT: vmovd %eax, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: vmovd %ecx, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
-; ALL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; ALL-NEXT: retq
- %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <2 x i32> <i32 0, i32 1>
- %2 = bitcast <2 x i16> %1 to <2 x half>
- %3 = fpext <2 x half> %2 to <2 x double>
- ret <2 x double> %3
-}
-
-define <4 x double> @cvt_8i16_to_4f64(<8 x i16> %a0) {
-; ALL-LABEL: cvt_8i16_to_4f64:
-; ALL: # BB#0:
-; ALL-NEXT: vmovq %xmm0, %rax
-; ALL-NEXT: movq %rax, %rcx
-; ALL-NEXT: movl %eax, %edx
-; ALL-NEXT: movswl %ax, %esi
-; ALL-NEXT: shrq $48, %rax
-; ALL-NEXT: shrq $32, %rcx
-; ALL-NEXT: shrl $16, %edx
-; ALL-NEXT: movswl %dx, %edx
-; ALL-NEXT: vmovd %edx, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: vmovd %esi, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: movswl %cx, %ecx
-; ALL-NEXT: vmovd %ecx, %xmm2
-; ALL-NEXT: vcvtph2ps %xmm2, %xmm2
-; ALL-NEXT: cwtl
-; ALL-NEXT: vmovd %eax, %xmm3
-; ALL-NEXT: vcvtph2ps %xmm3, %xmm3
-; ALL-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
-; ALL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; ALL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
-; ALL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; ALL-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; ALL-NEXT: retq
- %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
- %2 = bitcast <4 x i16> %1 to <4 x half>
- %3 = fpext <4 x half> %2 to <4 x double>
- ret <4 x double> %3
-}
-
-define <8 x double> @cvt_8i16_to_8f64(<8 x i16> %a0) {
-; AVX1-LABEL: cvt_8i16_to_8f64:
+define <4 x double> @cvt_4i16_to_4f64(<4 x i16> %a0) nounwind {
+; AVX1-LABEL: cvt_4i16_to_4f64:
; AVX1: # BB#0:
-; AVX1-NEXT: vmovq %xmm0, %rdx
-; AVX1-NEXT: movq %rdx, %r9
-; AVX1-NEXT: movl %edx, %r10d
-; AVX1-NEXT: movswl %dx, %r8d
+; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX1-NEXT: vmovq %xmm0, %rax
+; AVX1-NEXT: movq %rax, %rcx
+; AVX1-NEXT: movl %eax, %edx
+; AVX1-NEXT: movswl %ax, %esi
+; AVX1-NEXT: shrq $48, %rax
+; AVX1-NEXT: shrq $32, %rcx
+; AVX1-NEXT: shrl $16, %edx
+; AVX1-NEXT: movswl %dx, %edx
+; AVX1-NEXT: vmovd %edx, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: vmovd %esi, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: movswl %cx, %ecx
+; AVX1-NEXT: vmovd %ecx, %xmm2
+; AVX1-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX1-NEXT: cwtl
+; AVX1-NEXT: vmovd %eax, %xmm3
+; AVX1-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX1-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX1-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX1-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_4i16_to_4f64:
+; AVX2: # BB#0:
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX2-NEXT: vmovq %xmm0, %rax
+; AVX2-NEXT: movq %rax, %rcx
+; AVX2-NEXT: movl %eax, %edx
+; AVX2-NEXT: movswl %ax, %esi
+; AVX2-NEXT: shrq $48, %rax
+; AVX2-NEXT: shrq $32, %rcx
+; AVX2-NEXT: shrl $16, %edx
+; AVX2-NEXT: movswl %dx, %edx
+; AVX2-NEXT: vmovd %edx, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: vmovd %esi, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: movswl %cx, %ecx
+; AVX2-NEXT: vmovd %ecx, %xmm2
+; AVX2-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX2-NEXT: cwtl
+; AVX2-NEXT: vmovd %eax, %xmm3
+; AVX2-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX2-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX2-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX2-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX2-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_4i16_to_4f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX512F-NEXT: vmovq %xmm0, %rax
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: movl %eax, %edx
+; AVX512F-NEXT: movswl %ax, %esi
+; AVX512F-NEXT: shrq $48, %rax
+; AVX512F-NEXT: shrq $32, %rcx
+; AVX512F-NEXT: shrl $16, %edx
+; AVX512F-NEXT: movswl %dx, %edx
+; AVX512F-NEXT: vmovd %edx, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %esi, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX512F-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX512F-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_4i16_to_4f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $64, %rsp
+; AVX512VL-NEXT: vpmovdw %xmm0, {{[0-9]+}}(%rsp)
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rax
+; AVX512VL-NEXT: movq %rax, %r14
+; AVX512VL-NEXT: shrq $48, %r14
+; AVX512VL-NEXT: movq %rax, %rbx
+; AVX512VL-NEXT: shrq $32, %rbx
+; AVX512VL-NEXT: movl %eax, %ebp
+; AVX512VL-NEXT: shrl $16, %ebp
+; AVX512VL-NEXT: movzwl %ax, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %ebp, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %r14d, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: addq $64, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
+ %1 = bitcast <4 x i16> %a0 to <4 x half>
+ %2 = fpext <4 x half> %1 to <4 x double>
+ ret <4 x double> %2
+}
+
+define <2 x double> @cvt_8i16_to_2f64(<8 x i16> %a0) nounwind {
+; AVX1-LABEL: cvt_8i16_to_2f64:
+; AVX1: # BB#0:
+; AVX1-NEXT: vmovd %xmm0, %eax
+; AVX1-NEXT: movswl %ax, %ecx
+; AVX1-NEXT: shrl $16, %eax
+; AVX1-NEXT: cwtl
+; AVX1-NEXT: vmovd %eax, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: vmovd %ecx, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_8i16_to_2f64:
+; AVX2: # BB#0:
+; AVX2-NEXT: vmovd %xmm0, %eax
+; AVX2-NEXT: movswl %ax, %ecx
+; AVX2-NEXT: shrl $16, %eax
+; AVX2-NEXT: cwtl
+; AVX2-NEXT: vmovd %eax, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: vmovd %ecx, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_8i16_to_2f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: movswl %ax, %ecx
+; AVX512F-NEXT: shrl $16, %eax
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %ecx, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_8i16_to_2f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $32, %rsp
+; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
+; AVX512VL-NEXT: vpmovqw %xmm0, {{[0-9]+}}(%rsp)
+; AVX512VL-NEXT: movl {{[0-9]+}}(%rsp), %eax
+; AVX512VL-NEXT: movl %eax, %ebx
+; AVX512VL-NEXT: shrl $16, %ebx
+; AVX512VL-NEXT: movzwl %ax, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %ebx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: addq $32, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: retq
+ %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <2 x i32> <i32 0, i32 1>
+ %2 = bitcast <2 x i16> %1 to <2 x half>
+ %3 = fpext <2 x half> %2 to <2 x double>
+ ret <2 x double> %3
+}
+
+define <4 x double> @cvt_8i16_to_4f64(<8 x i16> %a0) nounwind {
+; AVX1-LABEL: cvt_8i16_to_4f64:
+; AVX1: # BB#0:
+; AVX1-NEXT: vmovq %xmm0, %rax
+; AVX1-NEXT: movq %rax, %rcx
+; AVX1-NEXT: movl %eax, %edx
+; AVX1-NEXT: movswl %ax, %esi
+; AVX1-NEXT: shrq $48, %rax
+; AVX1-NEXT: shrq $32, %rcx
+; AVX1-NEXT: shrl $16, %edx
+; AVX1-NEXT: movswl %dx, %edx
+; AVX1-NEXT: vmovd %edx, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: vmovd %esi, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: movswl %cx, %ecx
+; AVX1-NEXT: vmovd %ecx, %xmm2
+; AVX1-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX1-NEXT: cwtl
+; AVX1-NEXT: vmovd %eax, %xmm3
+; AVX1-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX1-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX1-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX1-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_8i16_to_4f64:
+; AVX2: # BB#0:
+; AVX2-NEXT: vmovq %xmm0, %rax
+; AVX2-NEXT: movq %rax, %rcx
+; AVX2-NEXT: movl %eax, %edx
+; AVX2-NEXT: movswl %ax, %esi
+; AVX2-NEXT: shrq $48, %rax
+; AVX2-NEXT: shrq $32, %rcx
+; AVX2-NEXT: shrl $16, %edx
+; AVX2-NEXT: movswl %dx, %edx
+; AVX2-NEXT: vmovd %edx, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: vmovd %esi, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: movswl %cx, %ecx
+; AVX2-NEXT: vmovd %ecx, %xmm2
+; AVX2-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX2-NEXT: cwtl
+; AVX2-NEXT: vmovd %eax, %xmm3
+; AVX2-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX2-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX2-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX2-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX2-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_8i16_to_4f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovq %xmm0, %rax
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: movl %eax, %edx
+; AVX512F-NEXT: movswl %ax, %esi
+; AVX512F-NEXT: shrq $48, %rax
+; AVX512F-NEXT: shrq $32, %rcx
+; AVX512F-NEXT: shrl $16, %edx
+; AVX512F-NEXT: movswl %dx, %edx
+; AVX512F-NEXT: vmovd %edx, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %esi, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX512F-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX512F-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_8i16_to_4f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $64, %rsp
+; AVX512VL-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX512VL-NEXT: vpmovdw %xmm0, {{[0-9]+}}(%rsp)
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rax
+; AVX512VL-NEXT: movq %rax, %r14
+; AVX512VL-NEXT: shrq $48, %r14
+; AVX512VL-NEXT: movq %rax, %rbx
+; AVX512VL-NEXT: shrq $32, %rbx
+; AVX512VL-NEXT: movl %eax, %ebp
+; AVX512VL-NEXT: shrl $16, %ebp
+; AVX512VL-NEXT: movzwl %ax, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %ebp, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %r14d, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: addq $64, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
+ %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = bitcast <4 x i16> %1 to <4 x half>
+ %3 = fpext <4 x half> %2 to <4 x double>
+ ret <4 x double> %3
+}
+
+define <8 x double> @cvt_8i16_to_8f64(<8 x i16> %a0) nounwind {
+; AVX1-LABEL: cvt_8i16_to_8f64:
+; AVX1: # BB#0:
+; AVX1-NEXT: vmovq %xmm0, %rdx
+; AVX1-NEXT: movq %rdx, %r9
+; AVX1-NEXT: movl %edx, %r10d
+; AVX1-NEXT: movswl %dx, %r8d
; AVX1-NEXT: shrq $48, %rdx
; AVX1-NEXT: shrq $32, %r9
; AVX1-NEXT: shrl $16, %r10d
@@ -1156,60 +2203,139 @@ define <8 x double> @cvt_8i16_to_8f64(<8
; AVX2-NEXT: vinsertf128 $1, %xmm3, %ymm1, %ymm1
; AVX2-NEXT: retq
;
-; AVX512-LABEL: cvt_8i16_to_8f64:
-; AVX512: # BB#0:
-; AVX512-NEXT: vpextrq $1, %xmm0, %rdx
-; AVX512-NEXT: movq %rdx, %r8
-; AVX512-NEXT: movl %edx, %r10d
-; AVX512-NEXT: movswl %dx, %r9d
-; AVX512-NEXT: shrq $48, %rdx
-; AVX512-NEXT: shrq $32, %r8
-; AVX512-NEXT: shrl $16, %r10d
-; AVX512-NEXT: vmovq %xmm0, %rdi
-; AVX512-NEXT: movq %rdi, %rax
-; AVX512-NEXT: movl %edi, %esi
-; AVX512-NEXT: movswl %di, %ecx
-; AVX512-NEXT: shrq $48, %rdi
-; AVX512-NEXT: shrq $32, %rax
-; AVX512-NEXT: shrl $16, %esi
-; AVX512-NEXT: movswl %si, %esi
-; AVX512-NEXT: vmovd %esi, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm0
-; AVX512-NEXT: vmovd %ecx, %xmm1
-; AVX512-NEXT: vcvtph2ps %xmm1, %xmm1
-; AVX512-NEXT: cwtl
-; AVX512-NEXT: vmovd %eax, %xmm2
-; AVX512-NEXT: vcvtph2ps %xmm2, %xmm2
-; AVX512-NEXT: movswl %di, %eax
-; AVX512-NEXT: vmovd %eax, %xmm3
-; AVX512-NEXT: vcvtph2ps %xmm3, %xmm3
-; AVX512-NEXT: movswl %r10w, %eax
-; AVX512-NEXT: vmovd %eax, %xmm4
-; AVX512-NEXT: vcvtph2ps %xmm4, %xmm4
-; AVX512-NEXT: vmovd %r9d, %xmm5
-; AVX512-NEXT: vcvtph2ps %xmm5, %xmm5
-; AVX512-NEXT: movswl %r8w, %eax
-; AVX512-NEXT: vmovd %eax, %xmm6
-; AVX512-NEXT: vcvtph2ps %xmm6, %xmm6
-; AVX512-NEXT: movswl %dx, %eax
-; AVX512-NEXT: vmovd %eax, %xmm7
-; AVX512-NEXT: vcvtph2ps %xmm7, %xmm7
-; AVX512-NEXT: vcvtss2sd %xmm7, %xmm7, %xmm7
-; AVX512-NEXT: vcvtss2sd %xmm6, %xmm6, %xmm6
-; AVX512-NEXT: vunpcklpd {{.*#+}} xmm6 = xmm6[0],xmm7[0]
-; AVX512-NEXT: vcvtss2sd %xmm5, %xmm5, %xmm5
-; AVX512-NEXT: vcvtss2sd %xmm4, %xmm4, %xmm4
-; AVX512-NEXT: vunpcklpd {{.*#+}} xmm4 = xmm5[0],xmm4[0]
-; AVX512-NEXT: vinsertf128 $1, %xmm6, %ymm4, %ymm4
-; AVX512-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
-; AVX512-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
-; AVX512-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; AVX512-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
-; AVX512-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; AVX512-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX512-NEXT: vinsertf64x4 $1, %ymm4, %zmm0, %zmm0
-; AVX512-NEXT: retq
+; AVX512F-LABEL: cvt_8i16_to_8f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vpextrq $1, %xmm0, %rdx
+; AVX512F-NEXT: movq %rdx, %r8
+; AVX512F-NEXT: movl %edx, %r10d
+; AVX512F-NEXT: movswl %dx, %r9d
+; AVX512F-NEXT: shrq $48, %rdx
+; AVX512F-NEXT: shrq $32, %r8
+; AVX512F-NEXT: shrl $16, %r10d
+; AVX512F-NEXT: vmovq %xmm0, %rdi
+; AVX512F-NEXT: movq %rdi, %rax
+; AVX512F-NEXT: movl %edi, %esi
+; AVX512F-NEXT: movswl %di, %ecx
+; AVX512F-NEXT: shrq $48, %rdi
+; AVX512F-NEXT: shrq $32, %rax
+; AVX512F-NEXT: shrl $16, %esi
+; AVX512F-NEXT: movswl %si, %esi
+; AVX512F-NEXT: vmovd %esi, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %ecx, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: movswl %di, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: movswl %r10w, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm4
+; AVX512F-NEXT: vcvtph2ps %xmm4, %xmm4
+; AVX512F-NEXT: vmovd %r9d, %xmm5
+; AVX512F-NEXT: vcvtph2ps %xmm5, %xmm5
+; AVX512F-NEXT: movswl %r8w, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm6
+; AVX512F-NEXT: vcvtph2ps %xmm6, %xmm6
+; AVX512F-NEXT: movswl %dx, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm7
+; AVX512F-NEXT: vcvtph2ps %xmm7, %xmm7
+; AVX512F-NEXT: vcvtss2sd %xmm7, %xmm7, %xmm7
+; AVX512F-NEXT: vcvtss2sd %xmm6, %xmm6, %xmm6
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm6 = xmm6[0],xmm7[0]
+; AVX512F-NEXT: vcvtss2sd %xmm5, %xmm5, %xmm5
+; AVX512F-NEXT: vcvtss2sd %xmm4, %xmm4, %xmm4
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm4 = xmm5[0],xmm4[0]
+; AVX512F-NEXT: vinsertf128 $1, %xmm6, %ymm4, %ymm4
+; AVX512F-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX512F-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX512F-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT: vinsertf64x4 $1, %ymm4, %zmm0, %zmm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_8i16_to_8f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %r13
+; AVX512VL-NEXT: pushq %r12
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $136, %rsp
+; AVX512VL-NEXT: vpextrq $1, %xmm0, %r13
+; AVX512VL-NEXT: movq %r13, %rax
+; AVX512VL-NEXT: shrq $48, %rax
+; AVX512VL-NEXT: movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; AVX512VL-NEXT: movq %r13, %r12
+; AVX512VL-NEXT: shrq $32, %r12
+; AVX512VL-NEXT: movl %r13d, %r15d
+; AVX512VL-NEXT: shrl $16, %r15d
+; AVX512VL-NEXT: vmovq %xmm0, %rax
+; AVX512VL-NEXT: movq %rax, %rbx
+; AVX512VL-NEXT: shrq $48, %rbx
+; AVX512VL-NEXT: movq %rax, %r14
+; AVX512VL-NEXT: shrq $32, %r14
+; AVX512VL-NEXT: movl %eax, %ebp
+; AVX512VL-NEXT: shrl $16, %ebp
+; AVX512VL-NEXT: movzwl %ax, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %ebp, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %r14w, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %ebx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %r13w, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %r15d, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %r12w, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; AVX512VL-NEXT: # kill: %EDI<def> %EDI<kill> %RDI<kill>
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm3 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VL-NEXT: addq $136, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r12
+; AVX512VL-NEXT: popq %r13
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = bitcast <8 x i16> %a0 to <8 x half>
%2 = fpext <8 x half> %1 to <8 x double>
ret <8 x double> %2
@@ -1219,97 +2345,343 @@ define <8 x double> @cvt_8i16_to_8f64(<8
; Half to Double (Load)
;
-define double @load_cvt_i16_to_f64(i16* %a0) {
-; ALL-LABEL: load_cvt_i16_to_f64:
-; ALL: # BB#0:
-; ALL-NEXT: movswl (%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; ALL-NEXT: retq
+define double @load_cvt_i16_to_f64(i16* %a0) nounwind {
+; AVX1-LABEL: load_cvt_i16_to_f64:
+; AVX1: # BB#0:
+; AVX1-NEXT: movswl (%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: load_cvt_i16_to_f64:
+; AVX2: # BB#0:
+; AVX2-NEXT: movswl (%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: load_cvt_i16_to_f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movswl (%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_cvt_i16_to_f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rax
+; AVX512VL-NEXT: movzwl (%rdi), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: popq %rax
+; AVX512VL-NEXT: retq
%1 = load i16, i16* %a0
%2 = bitcast i16 %1 to half
%3 = fpext half %2 to double
ret double %3
}
-define <2 x double> @load_cvt_2i16_to_2f64(<2 x i16>* %a0) {
-; ALL-LABEL: load_cvt_2i16_to_2f64:
-; ALL: # BB#0:
-; ALL-NEXT: movswl (%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: movswl 2(%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
-; ALL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; ALL-NEXT: retq
+define <2 x double> @load_cvt_2i16_to_2f64(<2 x i16>* %a0) nounwind {
+; AVX1-LABEL: load_cvt_2i16_to_2f64:
+; AVX1: # BB#0:
+; AVX1-NEXT: movswl (%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: movswl 2(%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: load_cvt_2i16_to_2f64:
+; AVX2: # BB#0:
+; AVX2-NEXT: movswl (%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: movswl 2(%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: load_cvt_2i16_to_2f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movswl (%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: movswl 2(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_cvt_2i16_to_2f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $16, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: movzwl (%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 2(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: addq $16, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: retq
%1 = load <2 x i16>, <2 x i16>* %a0
%2 = bitcast <2 x i16> %1 to <2 x half>
%3 = fpext <2 x half> %2 to <2 x double>
ret <2 x double> %3
}
-define <4 x double> @load_cvt_4i16_to_4f64(<4 x i16>* %a0) {
-; ALL-LABEL: load_cvt_4i16_to_4f64:
-; ALL: # BB#0:
-; ALL-NEXT: movswl (%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: movswl 2(%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: movswl 4(%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm2
-; ALL-NEXT: vcvtph2ps %xmm2, %xmm2
-; ALL-NEXT: movswl 6(%rdi), %eax
-; ALL-NEXT: vmovd %eax, %xmm3
-; ALL-NEXT: vcvtph2ps %xmm3, %xmm3
-; ALL-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
-; ALL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; ALL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
-; ALL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; ALL-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; ALL-NEXT: retq
+define <4 x double> @load_cvt_4i16_to_4f64(<4 x i16>* %a0) nounwind {
+; AVX1-LABEL: load_cvt_4i16_to_4f64:
+; AVX1: # BB#0:
+; AVX1-NEXT: movswl (%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: movswl 2(%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: movswl 4(%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm2
+; AVX1-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX1-NEXT: movswl 6(%rdi), %eax
+; AVX1-NEXT: vmovd %eax, %xmm3
+; AVX1-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX1-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX1-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX1-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: load_cvt_4i16_to_4f64:
+; AVX2: # BB#0:
+; AVX2-NEXT: movswl (%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: movswl 2(%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: movswl 4(%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm2
+; AVX2-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX2-NEXT: movswl 6(%rdi), %eax
+; AVX2-NEXT: vmovd %eax, %xmm3
+; AVX2-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX2-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX2-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX2-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: load_cvt_4i16_to_4f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movswl (%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: movswl 2(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: movswl 4(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: movswl 6(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX512F-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX512F-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX512F-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_cvt_4i16_to_4f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $48, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: movzwl (%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 2(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 4(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 6(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: addq $48, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: retq
%1 = load <4 x i16>, <4 x i16>* %a0
%2 = bitcast <4 x i16> %1 to <4 x half>
%3 = fpext <4 x half> %2 to <4 x double>
ret <4 x double> %3
}
-define <4 x double> @load_cvt_8i16_to_4f64(<8 x i16>* %a0) {
-; ALL-LABEL: load_cvt_8i16_to_4f64:
-; ALL: # BB#0:
-; ALL-NEXT: movq (%rdi), %rax
-; ALL-NEXT: movq %rax, %rcx
-; ALL-NEXT: movl %eax, %edx
-; ALL-NEXT: movswl %ax, %esi
-; ALL-NEXT: shrq $48, %rax
-; ALL-NEXT: shrq $32, %rcx
-; ALL-NEXT: shrl $16, %edx
-; ALL-NEXT: movswl %dx, %edx
-; ALL-NEXT: vmovd %edx, %xmm0
-; ALL-NEXT: vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT: vmovd %esi, %xmm1
-; ALL-NEXT: vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT: movswl %cx, %ecx
-; ALL-NEXT: vmovd %ecx, %xmm2
-; ALL-NEXT: vcvtph2ps %xmm2, %xmm2
-; ALL-NEXT: cwtl
-; ALL-NEXT: vmovd %eax, %xmm3
-; ALL-NEXT: vcvtph2ps %xmm3, %xmm3
-; ALL-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
-; ALL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; ALL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
-; ALL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; ALL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; ALL-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; ALL-NEXT: retq
+define <4 x double> @load_cvt_8i16_to_4f64(<8 x i16>* %a0) nounwind {
+; AVX1-LABEL: load_cvt_8i16_to_4f64:
+; AVX1: # BB#0:
+; AVX1-NEXT: movq (%rdi), %rax
+; AVX1-NEXT: movq %rax, %rcx
+; AVX1-NEXT: movl %eax, %edx
+; AVX1-NEXT: movswl %ax, %esi
+; AVX1-NEXT: shrq $48, %rax
+; AVX1-NEXT: shrq $32, %rcx
+; AVX1-NEXT: shrl $16, %edx
+; AVX1-NEXT: movswl %dx, %edx
+; AVX1-NEXT: vmovd %edx, %xmm0
+; AVX1-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX1-NEXT: vmovd %esi, %xmm1
+; AVX1-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX1-NEXT: movswl %cx, %ecx
+; AVX1-NEXT: vmovd %ecx, %xmm2
+; AVX1-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX1-NEXT: cwtl
+; AVX1-NEXT: vmovd %eax, %xmm3
+; AVX1-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX1-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX1-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX1-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: load_cvt_8i16_to_4f64:
+; AVX2: # BB#0:
+; AVX2-NEXT: movq (%rdi), %rax
+; AVX2-NEXT: movq %rax, %rcx
+; AVX2-NEXT: movl %eax, %edx
+; AVX2-NEXT: movswl %ax, %esi
+; AVX2-NEXT: shrq $48, %rax
+; AVX2-NEXT: shrq $32, %rcx
+; AVX2-NEXT: shrl $16, %edx
+; AVX2-NEXT: movswl %dx, %edx
+; AVX2-NEXT: vmovd %edx, %xmm0
+; AVX2-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX2-NEXT: vmovd %esi, %xmm1
+; AVX2-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX2-NEXT: movswl %cx, %ecx
+; AVX2-NEXT: vmovd %ecx, %xmm2
+; AVX2-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX2-NEXT: cwtl
+; AVX2-NEXT: vmovd %eax, %xmm3
+; AVX2-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX2-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX2-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX2-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX2-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX2-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: load_cvt_8i16_to_4f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movq (%rdi), %rax
+; AVX512F-NEXT: movq %rax, %rcx
+; AVX512F-NEXT: movl %eax, %edx
+; AVX512F-NEXT: movswl %ax, %esi
+; AVX512F-NEXT: shrq $48, %rax
+; AVX512F-NEXT: shrq $32, %rcx
+; AVX512F-NEXT: shrl $16, %edx
+; AVX512F-NEXT: movswl %dx, %edx
+; AVX512F-NEXT: vmovd %edx, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %esi, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: movswl %cx, %ecx
+; AVX512F-NEXT: vmovd %ecx, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: cwtl
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX512F-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX512F-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_cvt_8i16_to_4f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $64, %rsp
+; AVX512VL-NEXT: vmovdqa64 (%rdi), %xmm0
+; AVX512VL-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX512VL-NEXT: vpmovdw %xmm0, {{[0-9]+}}(%rsp)
+; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rax
+; AVX512VL-NEXT: movq %rax, %r14
+; AVX512VL-NEXT: shrq $48, %r14
+; AVX512VL-NEXT: movq %rax, %rbx
+; AVX512VL-NEXT: shrq $32, %rbx
+; AVX512VL-NEXT: movl %eax, %ebp
+; AVX512VL-NEXT: shrl $16, %ebp
+; AVX512VL-NEXT: movzwl %ax, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %ebp, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl %bx, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovdqa %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movl %r14d, %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: addq $64, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = load <8 x i16>, <8 x i16>* %a0
%2 = shufflevector <8 x i16> %1, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%3 = bitcast <4 x i16> %2 to <4 x half>
@@ -1317,7 +2689,7 @@ define <4 x double> @load_cvt_8i16_to_4f
ret <4 x double> %4
}
-define <8 x double> @load_cvt_8i16_to_8f64(<8 x i16>* %a0) {
+define <8 x double> @load_cvt_8i16_to_8f64(<8 x i16>* %a0) nounwind {
; AVX1-LABEL: load_cvt_8i16_to_8f64:
; AVX1: # BB#0:
; AVX1-NEXT: movswl 8(%rdi), %eax
@@ -1402,48 +2774,102 @@ define <8 x double> @load_cvt_8i16_to_8f
; AVX2-NEXT: vinsertf128 $1, %xmm3, %ymm1, %ymm1
; AVX2-NEXT: retq
;
-; AVX512-LABEL: load_cvt_8i16_to_8f64:
-; AVX512: # BB#0:
-; AVX512-NEXT: movswl (%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm0
-; AVX512-NEXT: vcvtph2ps %xmm0, %xmm0
-; AVX512-NEXT: movswl 2(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm1
-; AVX512-NEXT: vcvtph2ps %xmm1, %xmm1
-; AVX512-NEXT: movswl 4(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm2
-; AVX512-NEXT: vcvtph2ps %xmm2, %xmm2
-; AVX512-NEXT: movswl 6(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm3
-; AVX512-NEXT: vcvtph2ps %xmm3, %xmm3
-; AVX512-NEXT: movswl 8(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm4
-; AVX512-NEXT: vcvtph2ps %xmm4, %xmm4
-; AVX512-NEXT: movswl 10(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm5
-; AVX512-NEXT: vcvtph2ps %xmm5, %xmm5
-; AVX512-NEXT: movswl 12(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm6
-; AVX512-NEXT: vcvtph2ps %xmm6, %xmm6
-; AVX512-NEXT: movswl 14(%rdi), %eax
-; AVX512-NEXT: vmovd %eax, %xmm7
-; AVX512-NEXT: vcvtph2ps %xmm7, %xmm7
-; AVX512-NEXT: vcvtss2sd %xmm7, %xmm7, %xmm7
-; AVX512-NEXT: vcvtss2sd %xmm6, %xmm6, %xmm6
-; AVX512-NEXT: vunpcklpd {{.*#+}} xmm6 = xmm6[0],xmm7[0]
-; AVX512-NEXT: vcvtss2sd %xmm5, %xmm5, %xmm5
-; AVX512-NEXT: vcvtss2sd %xmm4, %xmm4, %xmm4
-; AVX512-NEXT: vunpcklpd {{.*#+}} xmm4 = xmm4[0],xmm5[0]
-; AVX512-NEXT: vinsertf128 $1, %xmm6, %ymm4, %ymm4
-; AVX512-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
-; AVX512-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
-; AVX512-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; AVX512-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
-; AVX512-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX512-NEXT: vinsertf64x4 $1, %ymm4, %zmm0, %zmm0
-; AVX512-NEXT: retq
+; AVX512F-LABEL: load_cvt_8i16_to_8f64:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: movswl (%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: vcvtph2ps %xmm0, %xmm0
+; AVX512F-NEXT: movswl 2(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm1
+; AVX512F-NEXT: vcvtph2ps %xmm1, %xmm1
+; AVX512F-NEXT: movswl 4(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm2
+; AVX512F-NEXT: vcvtph2ps %xmm2, %xmm2
+; AVX512F-NEXT: movswl 6(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vcvtph2ps %xmm3, %xmm3
+; AVX512F-NEXT: movswl 8(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm4
+; AVX512F-NEXT: vcvtph2ps %xmm4, %xmm4
+; AVX512F-NEXT: movswl 10(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm5
+; AVX512F-NEXT: vcvtph2ps %xmm5, %xmm5
+; AVX512F-NEXT: movswl 12(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm6
+; AVX512F-NEXT: vcvtph2ps %xmm6, %xmm6
+; AVX512F-NEXT: movswl 14(%rdi), %eax
+; AVX512F-NEXT: vmovd %eax, %xmm7
+; AVX512F-NEXT: vcvtph2ps %xmm7, %xmm7
+; AVX512F-NEXT: vcvtss2sd %xmm7, %xmm7, %xmm7
+; AVX512F-NEXT: vcvtss2sd %xmm6, %xmm6, %xmm6
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm6 = xmm6[0],xmm7[0]
+; AVX512F-NEXT: vcvtss2sd %xmm5, %xmm5, %xmm5
+; AVX512F-NEXT: vcvtss2sd %xmm4, %xmm4, %xmm4
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm4 = xmm4[0],xmm5[0]
+; AVX512F-NEXT: vinsertf128 $1, %xmm6, %ymm4, %ymm4
+; AVX512F-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX512F-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; AVX512F-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512F-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX512F-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT: vinsertf64x4 $1, %ymm4, %zmm0, %zmm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: load_cvt_8i16_to_8f64:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $112, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: movzwl (%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 2(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 4(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 6(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 8(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 10(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 12(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: movzwl 14(%rbx), %edi
+; AVX512VL-NEXT: callq __gnu_h2f_ieee
+; AVX512VL-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm1 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm3 # 16-byte Reload
+; AVX512VL-NEXT: vcvtss2sd %xmm3, %xmm3, %xmm3
+; AVX512VL-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm3[0],xmm2[0]
+; AVX512VL-NEXT: vinsertf32x4 $1, %xmm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VL-NEXT: addq $112, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: retq
%1 = load <8 x i16>, <8 x i16>* %a0
%2 = bitcast <8 x i16> %1 to <8 x half>
%3 = fpext <8 x half> %2 to <8 x double>
@@ -1454,110 +2880,380 @@ define <8 x double> @load_cvt_8i16_to_8f
; Float to Half
;
-define i16 @cvt_f32_to_i16(float %a0) {
-; ALL-LABEL: cvt_f32_to_i16:
-; ALL: # BB#0:
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; ALL-NEXT: vmovd %xmm0, %eax
-; ALL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
-; ALL-NEXT: retq
+define i16 @cvt_f32_to_i16(float %a0) nounwind {
+; AVX1-LABEL: cvt_f32_to_i16:
+; AVX1: # BB#0:
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX1-NEXT: vmovd %xmm0, %eax
+; AVX1-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_f32_to_i16:
+; AVX2: # BB#0:
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX2-NEXT: vmovd %xmm0, %eax
+; AVX2-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_f32_to_i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_f32_to_i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: jmp __gnu_f2h_ieee # TAILCALL
%1 = fptrunc float %a0 to half
%2 = bitcast half %1 to i16
ret i16 %2
}
-define <4 x i16> @cvt_4f32_to_4i16(<4 x float> %a0) {
-; ALL-LABEL: cvt_4f32_to_4i16:
-; ALL: # BB#0:
-; ALL-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: shll $16, %eax
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm1
-; ALL-NEXT: vmovd %xmm1, %ecx
-; ALL-NEXT: movzwl %cx, %ecx
-; ALL-NEXT: orl %eax, %ecx
-; ALL-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: shll $16, %eax
-; ALL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; ALL-NEXT: vmovd %xmm0, %edx
-; ALL-NEXT: movzwl %dx, %edx
-; ALL-NEXT: orl %eax, %edx
-; ALL-NEXT: shlq $32, %rdx
-; ALL-NEXT: orq %rcx, %rdx
-; ALL-NEXT: vmovq %rdx, %xmm0
-; ALL-NEXT: retq
+define <4 x i16> @cvt_4f32_to_4i16(<4 x float> %a0) nounwind {
+; AVX1-LABEL: cvt_4f32_to_4i16:
+; AVX1: # BB#0:
+; AVX1-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: shll $16, %eax
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %ecx
+; AVX1-NEXT: movzwl %cx, %ecx
+; AVX1-NEXT: orl %eax, %ecx
+; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: shll $16, %eax
+; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX1-NEXT: vmovd %xmm0, %edx
+; AVX1-NEXT: movzwl %dx, %edx
+; AVX1-NEXT: orl %eax, %edx
+; AVX1-NEXT: shlq $32, %rdx
+; AVX1-NEXT: orq %rcx, %rdx
+; AVX1-NEXT: vmovq %rdx, %xmm0
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_4f32_to_4i16:
+; AVX2: # BB#0:
+; AVX2-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: shll $16, %eax
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %ecx
+; AVX2-NEXT: movzwl %cx, %ecx
+; AVX2-NEXT: orl %eax, %ecx
+; AVX2-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: shll $16, %eax
+; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX2-NEXT: vmovd %xmm0, %edx
+; AVX2-NEXT: movzwl %dx, %edx
+; AVX2-NEXT: orl %eax, %edx
+; AVX2-NEXT: shlq $32, %rdx
+; AVX2-NEXT: orq %rcx, %rdx
+; AVX2-NEXT: vmovq %rdx, %xmm0
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_4f32_to_4i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %ecx
+; AVX512F-NEXT: movzwl %cx, %ecx
+; AVX512F-NEXT: orl %eax, %ecx
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %edx
+; AVX512F-NEXT: movzwl %dx, %edx
+; AVX512F-NEXT: orl %eax, %edx
+; AVX512F-NEXT: shlq $32, %rdx
+; AVX512F-NEXT: orq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_4f32_to_4i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $24, %rsp
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %r14d
+; AVX512VL-NEXT: orl %ebx, %r14d
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebx, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %r14, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: addq $24, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x float> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
ret <4 x i16> %2
}
-define <8 x i16> @cvt_4f32_to_8i16_undef(<4 x float> %a0) {
-; ALL-LABEL: cvt_4f32_to_8i16_undef:
-; ALL: # BB#0:
-; ALL-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: shll $16, %eax
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm1
-; ALL-NEXT: vmovd %xmm1, %ecx
-; ALL-NEXT: movzwl %cx, %ecx
-; ALL-NEXT: orl %eax, %ecx
-; ALL-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: shll $16, %eax
-; ALL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; ALL-NEXT: vmovd %xmm0, %edx
-; ALL-NEXT: movzwl %dx, %edx
-; ALL-NEXT: orl %eax, %edx
-; ALL-NEXT: shlq $32, %rdx
-; ALL-NEXT: orq %rcx, %rdx
-; ALL-NEXT: vmovq %rdx, %xmm0
-; ALL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; ALL-NEXT: retq
+define <8 x i16> @cvt_4f32_to_8i16_undef(<4 x float> %a0) nounwind {
+; AVX1-LABEL: cvt_4f32_to_8i16_undef:
+; AVX1: # BB#0:
+; AVX1-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: shll $16, %eax
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %ecx
+; AVX1-NEXT: movzwl %cx, %ecx
+; AVX1-NEXT: orl %eax, %ecx
+; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: shll $16, %eax
+; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX1-NEXT: vmovd %xmm0, %edx
+; AVX1-NEXT: movzwl %dx, %edx
+; AVX1-NEXT: orl %eax, %edx
+; AVX1-NEXT: shlq $32, %rdx
+; AVX1-NEXT: orq %rcx, %rdx
+; AVX1-NEXT: vmovq %rdx, %xmm0
+; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_4f32_to_8i16_undef:
+; AVX2: # BB#0:
+; AVX2-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: shll $16, %eax
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %ecx
+; AVX2-NEXT: movzwl %cx, %ecx
+; AVX2-NEXT: orl %eax, %ecx
+; AVX2-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: shll $16, %eax
+; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX2-NEXT: vmovd %xmm0, %edx
+; AVX2-NEXT: movzwl %dx, %edx
+; AVX2-NEXT: orl %eax, %edx
+; AVX2-NEXT: shlq $32, %rdx
+; AVX2-NEXT: orq %rcx, %rdx
+; AVX2-NEXT: vmovq %rdx, %xmm0
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_4f32_to_8i16_undef:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %ecx
+; AVX512F-NEXT: movzwl %cx, %ecx
+; AVX512F-NEXT: orl %eax, %ecx
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %edx
+; AVX512F-NEXT: movzwl %dx, %edx
+; AVX512F-NEXT: orl %eax, %edx
+; AVX512F-NEXT: shlq $32, %rdx
+; AVX512F-NEXT: orq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm0
+; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_4f32_to_8i16_undef:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $24, %rsp
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %r14d
+; AVX512VL-NEXT: orl %ebx, %r14d
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebx, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %r14, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512VL-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512VL-NEXT: addq $24, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x float> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
%3 = shufflevector <4 x i16> %2, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <8 x i16> %3
}
-define <8 x i16> @cvt_4f32_to_8i16_zero(<4 x float> %a0) {
-; ALL-LABEL: cvt_4f32_to_8i16_zero:
-; ALL: # BB#0:
-; ALL-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: shll $16, %eax
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm1
-; ALL-NEXT: vmovd %xmm1, %ecx
-; ALL-NEXT: movzwl %cx, %ecx
-; ALL-NEXT: orl %eax, %ecx
-; ALL-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: shll $16, %eax
-; ALL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; ALL-NEXT: vmovd %xmm0, %edx
-; ALL-NEXT: movzwl %dx, %edx
-; ALL-NEXT: orl %eax, %edx
-; ALL-NEXT: shlq $32, %rdx
-; ALL-NEXT: orq %rcx, %rdx
-; ALL-NEXT: vmovq %rdx, %xmm0
-; ALL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
-; ALL-NEXT: retq
+define <8 x i16> @cvt_4f32_to_8i16_zero(<4 x float> %a0) nounwind {
+; AVX1-LABEL: cvt_4f32_to_8i16_zero:
+; AVX1: # BB#0:
+; AVX1-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: shll $16, %eax
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %ecx
+; AVX1-NEXT: movzwl %cx, %ecx
+; AVX1-NEXT: orl %eax, %ecx
+; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: shll $16, %eax
+; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX1-NEXT: vmovd %xmm0, %edx
+; AVX1-NEXT: movzwl %dx, %edx
+; AVX1-NEXT: orl %eax, %edx
+; AVX1-NEXT: shlq $32, %rdx
+; AVX1-NEXT: orq %rcx, %rdx
+; AVX1-NEXT: vmovq %rdx, %xmm0
+; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_4f32_to_8i16_zero:
+; AVX2: # BB#0:
+; AVX2-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: shll $16, %eax
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %ecx
+; AVX2-NEXT: movzwl %cx, %ecx
+; AVX2-NEXT: orl %eax, %ecx
+; AVX2-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: shll $16, %eax
+; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX2-NEXT: vmovd %xmm0, %edx
+; AVX2-NEXT: movzwl %dx, %edx
+; AVX2-NEXT: orl %eax, %edx
+; AVX2-NEXT: shlq $32, %rdx
+; AVX2-NEXT: orq %rcx, %rdx
+; AVX2-NEXT: vmovq %rdx, %xmm0
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_4f32_to_8i16_zero:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %ecx
+; AVX512F-NEXT: movzwl %cx, %ecx
+; AVX512F-NEXT: orl %eax, %ecx
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %edx
+; AVX512F-NEXT: movzwl %dx, %edx
+; AVX512F-NEXT: orl %eax, %edx
+; AVX512F-NEXT: shlq $32, %rdx
+; AVX512F-NEXT: orq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm0
+; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_4f32_to_8i16_zero:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $24, %rsp
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %r14d
+; AVX512VL-NEXT: orl %ebx, %r14d
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebx, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %r14, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512VL-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,2]
+; AVX512VL-NEXT: vpxord %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX512VL-NEXT: addq $24, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x float> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
%3 = shufflevector <4 x i16> %2, <4 x i16> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <8 x i16> %3
}
-define <8 x i16> @cvt_8f32_to_8i16(<8 x float> %a0) {
+define <8 x i16> @cvt_8f32_to_8i16(<8 x float> %a0) nounwind {
; AVX1-LABEL: cvt_8f32_to_8i16:
; AVX1: # BB#0:
; AVX1-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
@@ -1652,57 +3348,117 @@ define <8 x i16> @cvt_8f32_to_8i16(<8 x
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; AVX512-LABEL: cvt_8f32_to_8i16:
-; AVX512: # BB#0:
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vmovd %xmm1, %eax
-; AVX512-NEXT: shll $16, %eax
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm1
-; AVX512-NEXT: vmovd %xmm1, %ecx
-; AVX512-NEXT: movzwl %cx, %ecx
-; AVX512-NEXT: orl %eax, %ecx
-; AVX512-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vmovd %xmm1, %edx
-; AVX512-NEXT: shll $16, %edx
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vmovd %xmm1, %eax
-; AVX512-NEXT: movzwl %ax, %eax
-; AVX512-NEXT: orl %edx, %eax
-; AVX512-NEXT: shlq $32, %rax
-; AVX512-NEXT: orq %rcx, %rax
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vmovd %xmm1, %ecx
-; AVX512-NEXT: shll $16, %ecx
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm1
-; AVX512-NEXT: vmovd %xmm1, %edx
-; AVX512-NEXT: movzwl %dx, %edx
-; AVX512-NEXT: orl %ecx, %edx
-; AVX512-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vmovd %xmm1, %ecx
-; AVX512-NEXT: shll $16, %ecx
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; AVX512-NEXT: vmovd %xmm0, %esi
-; AVX512-NEXT: movzwl %si, %esi
-; AVX512-NEXT: orl %ecx, %esi
-; AVX512-NEXT: shlq $32, %rsi
-; AVX512-NEXT: orq %rdx, %rsi
-; AVX512-NEXT: vmovq %rsi, %xmm0
-; AVX512-NEXT: vmovq %rax, %xmm1
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; AVX512-NEXT: retq
+; AVX512F-LABEL: cvt_8f32_to_8i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %ecx
+; AVX512F-NEXT: movzwl %cx, %ecx
+; AVX512F-NEXT: orl %eax, %ecx
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %edx
+; AVX512F-NEXT: shll $16, %edx
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: movzwl %ax, %eax
+; AVX512F-NEXT: orl %edx, %eax
+; AVX512F-NEXT: shlq $32, %rax
+; AVX512F-NEXT: orq %rcx, %rax
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %ecx
+; AVX512F-NEXT: shll $16, %ecx
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %edx
+; AVX512F-NEXT: movzwl %dx, %edx
+; AVX512F-NEXT: orl %ecx, %edx
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %ecx
+; AVX512F-NEXT: shll $16, %ecx
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %esi
+; AVX512F-NEXT: movzwl %si, %esi
+; AVX512F-NEXT: orl %ecx, %esi
+; AVX512F-NEXT: shlq $32, %rsi
+; AVX512F-NEXT: orq %rdx, %rsi
+; AVX512F-NEXT: vmovq %rsi, %xmm0
+; AVX512F-NEXT: vmovq %rax, %xmm1
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_8f32_to_8i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $32, %rsp
+; AVX512VL-NEXT: vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %r15d
+; AVX512VL-NEXT: orl %ebx, %r15d
+; AVX512VL-NEXT: vmovdqu (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovdqu (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %r14d
+; AVX512VL-NEXT: orl %ebx, %r14d
+; AVX512VL-NEXT: shlq $32, %r14
+; AVX512VL-NEXT: orq %r15, %r14
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %r15d
+; AVX512VL-NEXT: orl %ebx, %r15d
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebx, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %r15, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: vmovq %r14, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: addq $32, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: retq
%1 = fptrunc <8 x float> %a0 to <8 x half>
%2 = bitcast <8 x half> %1 to <8 x i16>
ret <8 x i16> %2
}
-define <16 x i16> @cvt_16f32_to_16i16(<16 x float> %a0) {
+define <16 x i16> @cvt_16f32_to_16i16(<16 x float> %a0) nounwind {
; AVX1-LABEL: cvt_16f32_to_16i16:
; AVX1: # BB#0:
; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm2
@@ -1837,73 +3593,186 @@ define <16 x i16> @cvt_16f32_to_16i16(<1
; AVX2-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
; AVX2-NEXT: retq
;
-; AVX512-LABEL: cvt_16f32_to_16i16:
-; AVX512: # BB#0:
-; AVX512-NEXT: vextractf64x4 $1, %zmm0, %ymm1
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm2
-; AVX512-NEXT: vmovd %xmm2, %eax
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm2
-; AVX512-NEXT: vmovd %eax, %xmm3
-; AVX512-NEXT: vmovd %xmm2, %eax
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm2 = xmm1[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm2
-; AVX512-NEXT: vpinsrw $1, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm2, %eax
-; AVX512-NEXT: vextractf128 $1, %ymm1, %xmm2
-; AVX512-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vpinsrw $2, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm1, %eax
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm1
-; AVX512-NEXT: vpinsrw $3, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm1, %eax
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm1 = xmm2[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vpinsrw $4, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm1, %eax
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm1 = xmm2[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vpinsrw $5, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm1, %eax
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm1
-; AVX512-NEXT: vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm2
-; AVX512-NEXT: vpinsrw $6, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm2, %eax
-; AVX512-NEXT: vpinsrw $7, %eax, %xmm3, %xmm2
-; AVX512-NEXT: vmovd %xmm1, %eax
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vmovd %eax, %xmm3
-; AVX512-NEXT: vmovd %xmm1, %eax
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vpinsrw $1, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm1, %eax
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX512-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; AVX512-NEXT: vpinsrw $2, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm0
-; AVX512-NEXT: vpinsrw $3, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm0 = xmm1[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; AVX512-NEXT: vpinsrw $4, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm1[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; AVX512-NEXT: vpinsrw $5, %eax, %xmm3, %xmm3
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vpermilps {{.*#+}} xmm0 = xmm1[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; AVX512-NEXT: vpinsrw $6, %eax, %xmm3, %xmm1
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: vpinsrw $7, %eax, %xmm1, %xmm0
-; AVX512-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
-; AVX512-NEXT: retq
+; AVX512F-LABEL: cvt_16f32_to_16i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vextractf64x4 $1, %zmm0, %ymm1
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm2
+; AVX512F-NEXT: vmovd %xmm2, %eax
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm2
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vmovd %xmm2, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm2 = xmm1[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm2
+; AVX512F-NEXT: vpinsrw $1, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm2, %eax
+; AVX512F-NEXT: vextractf128 $1, %ymm1, %xmm2
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vpinsrw $2, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm1
+; AVX512F-NEXT: vpinsrw $3, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm2[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vpinsrw $4, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm1 = xmm2[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vpinsrw $5, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm2
+; AVX512F-NEXT: vpinsrw $6, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm2, %eax
+; AVX512F-NEXT: vpinsrw $7, %eax, %xmm3, %xmm2
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %eax, %xmm3
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vpinsrw $1, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm1
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vpinsrw $2, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm0
+; AVX512F-NEXT: vpinsrw $3, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm0 = xmm1[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vpinsrw $4, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm1[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vpinsrw $5, %eax, %xmm3, %xmm3
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm0 = xmm1[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vpinsrw $6, %eax, %xmm3, %xmm1
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: vpinsrw $7, %eax, %xmm1, %xmm0
+; AVX512F-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_16f32_to_16i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %r13
+; AVX512VL-NEXT: pushq %r12
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $136, %rsp
+; AVX512VL-NEXT: vmovups %zmm0, {{[0-9]+}}(%rsp) # 64-byte Spill
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: # kill: %AX<def> %AX<kill> %EAX<def>
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: # kill: %AX<def> %AX<kill> %EAX<def>
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: # kill: %AX<def> %AX<kill> %EAX<def>
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: # kill: %AX<def> %AX<kill> %EAX<def>
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: vmovdqu64 {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: # kill: %AX<def> %AX<kill> %EAX<def>
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: vmovdqu64 {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: # kill: %AX<def> %AX<kill> %EAX<def>
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: # kill: %AX<def> %AX<kill> %EAX<def>
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: vmovdqu64 {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: # kill: %AX<def> %AX<kill> %EAX<def>
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: vmovupd {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
+; AVX512VL-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: # kill: %AX<def> %AX<kill> %EAX<def>
+; AVX512VL-NEXT: movl %eax, {{[0-9]+}}(%rsp) # 4-byte Spill
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %r13w
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bp
+; AVX512VL-NEXT: vmovdqu {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %r14w
+; AVX512VL-NEXT: vmovdqu {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %r15w
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %r12w
+; AVX512VL-NEXT: vmovdqu {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: # kill: %AX<def> %AX<kill> %EAX<def>
+; AVX512VL-NEXT: vmovd %r12d, %xmm0
+; AVX512VL-NEXT: vpinsrw $1, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT: vpinsrw $2, %r15d, %xmm0, %xmm0
+; AVX512VL-NEXT: vpinsrw $3, %r14d, %xmm0, %xmm0
+; AVX512VL-NEXT: vpinsrw $4, %ebp, %xmm0, %xmm0
+; AVX512VL-NEXT: vpinsrw $5, %ebx, %xmm0, %xmm0
+; AVX512VL-NEXT: vpinsrw $6, %r13d, %xmm0, %xmm0
+; AVX512VL-NEXT: vpinsrw $7, {{[0-9]+}}(%rsp), %xmm0, %xmm0 # 4-byte Folded Reload
+; AVX512VL-NEXT: movl {{[0-9]+}}(%rsp), %eax # 4-byte Reload
+; AVX512VL-NEXT: vmovd %eax, %xmm1
+; AVX512VL-NEXT: vpinsrw $1, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 4-byte Folded Reload
+; AVX512VL-NEXT: vpinsrw $2, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 4-byte Folded Reload
+; AVX512VL-NEXT: vpinsrw $3, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 4-byte Folded Reload
+; AVX512VL-NEXT: vpinsrw $4, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 4-byte Folded Reload
+; AVX512VL-NEXT: vpinsrw $5, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 4-byte Folded Reload
+; AVX512VL-NEXT: vpinsrw $6, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 4-byte Folded Reload
+; AVX512VL-NEXT: vpinsrw $7, {{[0-9]+}}(%rsp), %xmm1, %xmm1 # 4-byte Folded Reload
+; AVX512VL-NEXT: vinserti32x4 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT: addq $136, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r12
+; AVX512VL-NEXT: popq %r13
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <16 x float> %a0 to <16 x half>
%2 = bitcast <16 x half> %1 to <16 x i16>
ret <16 x i16> %2
@@ -1913,70 +3782,255 @@ define <16 x i16> @cvt_16f32_to_16i16(<1
; Float to Half (Store)
;
-define void @store_cvt_f32_to_i16(float %a0, i16* %a1) {
-; ALL-LABEL: store_cvt_f32_to_i16:
-; ALL: # BB#0:
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; ALL-NEXT: vmovd %xmm0, %eax
-; ALL-NEXT: movw %ax, (%rdi)
-; ALL-NEXT: retq
+define void @store_cvt_f32_to_i16(float %a0, i16* %a1) nounwind {
+; AVX1-LABEL: store_cvt_f32_to_i16:
+; AVX1: # BB#0:
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX1-NEXT: vmovd %xmm0, %eax
+; AVX1-NEXT: movw %ax, (%rdi)
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: store_cvt_f32_to_i16:
+; AVX2: # BB#0:
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX2-NEXT: vmovd %xmm0, %eax
+; AVX2-NEXT: movw %ax, (%rdi)
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: store_cvt_f32_to_i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: movw %ax, (%rdi)
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_f32_to_i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, (%rbx)
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: retq
%1 = fptrunc float %a0 to half
%2 = bitcast half %1 to i16
store i16 %2, i16* %a1
ret void
}
-define void @store_cvt_4f32_to_4i16(<4 x float> %a0, <4 x i16>* %a1) {
-; ALL-LABEL: store_cvt_4f32_to_4i16:
-; ALL: # BB#0:
-; ALL-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %ecx
-; ALL-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %edx
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; ALL-NEXT: vmovd %xmm0, %esi
-; ALL-NEXT: movw %si, (%rdi)
-; ALL-NEXT: movw %dx, 6(%rdi)
-; ALL-NEXT: movw %cx, 4(%rdi)
-; ALL-NEXT: movw %ax, 2(%rdi)
-; ALL-NEXT: retq
+define void @store_cvt_4f32_to_4i16(<4 x float> %a0, <4 x i16>* %a1) nounwind {
+; AVX1-LABEL: store_cvt_4f32_to_4i16:
+; AVX1: # BB#0:
+; AVX1-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %ecx
+; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %edx
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX1-NEXT: vmovd %xmm0, %esi
+; AVX1-NEXT: movw %si, (%rdi)
+; AVX1-NEXT: movw %dx, 6(%rdi)
+; AVX1-NEXT: movw %cx, 4(%rdi)
+; AVX1-NEXT: movw %ax, 2(%rdi)
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: store_cvt_4f32_to_4i16:
+; AVX2: # BB#0:
+; AVX2-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %ecx
+; AVX2-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %edx
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX2-NEXT: vmovd %xmm0, %esi
+; AVX2-NEXT: movw %si, (%rdi)
+; AVX2-NEXT: movw %dx, 6(%rdi)
+; AVX2-NEXT: movw %cx, 4(%rdi)
+; AVX2-NEXT: movw %ax, 2(%rdi)
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: store_cvt_4f32_to_4i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %ecx
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %edx
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %esi
+; AVX512F-NEXT: movw %si, (%rdi)
+; AVX512F-NEXT: movw %dx, 6(%rdi)
+; AVX512F-NEXT: movw %cx, 4(%rdi)
+; AVX512F-NEXT: movw %ax, 2(%rdi)
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_4f32_to_4i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $24, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %r14d
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %r15d
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %ebp
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, (%rbx)
+; AVX512VL-NEXT: movw %bp, 6(%rbx)
+; AVX512VL-NEXT: movw %r15w, 4(%rbx)
+; AVX512VL-NEXT: movw %r14w, 2(%rbx)
+; AVX512VL-NEXT: addq $24, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x float> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
store <4 x i16> %2, <4 x i16>* %a1
ret void
}
-define void @store_cvt_4f32_to_8i16_undef(<4 x float> %a0, <8 x i16>* %a1) {
-; ALL-LABEL: store_cvt_4f32_to_8i16_undef:
-; ALL: # BB#0:
-; ALL-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: shll $16, %eax
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm1
-; ALL-NEXT: vmovd %xmm1, %ecx
-; ALL-NEXT: movzwl %cx, %ecx
-; ALL-NEXT: orl %eax, %ecx
-; ALL-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: shll $16, %eax
-; ALL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; ALL-NEXT: vmovd %xmm0, %edx
-; ALL-NEXT: movzwl %dx, %edx
-; ALL-NEXT: orl %eax, %edx
-; ALL-NEXT: shlq $32, %rdx
-; ALL-NEXT: orq %rcx, %rdx
-; ALL-NEXT: vmovq %rdx, %xmm0
-; ALL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; ALL-NEXT: vmovdqa %xmm0, (%rdi)
-; ALL-NEXT: retq
+define void @store_cvt_4f32_to_8i16_undef(<4 x float> %a0, <8 x i16>* %a1) nounwind {
+; AVX1-LABEL: store_cvt_4f32_to_8i16_undef:
+; AVX1: # BB#0:
+; AVX1-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: shll $16, %eax
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %ecx
+; AVX1-NEXT: movzwl %cx, %ecx
+; AVX1-NEXT: orl %eax, %ecx
+; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: shll $16, %eax
+; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX1-NEXT: vmovd %xmm0, %edx
+; AVX1-NEXT: movzwl %dx, %edx
+; AVX1-NEXT: orl %eax, %edx
+; AVX1-NEXT: shlq $32, %rdx
+; AVX1-NEXT: orq %rcx, %rdx
+; AVX1-NEXT: vmovq %rdx, %xmm0
+; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX1-NEXT: vmovdqa %xmm0, (%rdi)
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: store_cvt_4f32_to_8i16_undef:
+; AVX2: # BB#0:
+; AVX2-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: shll $16, %eax
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %ecx
+; AVX2-NEXT: movzwl %cx, %ecx
+; AVX2-NEXT: orl %eax, %ecx
+; AVX2-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: shll $16, %eax
+; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX2-NEXT: vmovd %xmm0, %edx
+; AVX2-NEXT: movzwl %dx, %edx
+; AVX2-NEXT: orl %eax, %edx
+; AVX2-NEXT: shlq $32, %rdx
+; AVX2-NEXT: orq %rcx, %rdx
+; AVX2-NEXT: vmovq %rdx, %xmm0
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX2-NEXT: vmovdqa %xmm0, (%rdi)
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: store_cvt_4f32_to_8i16_undef:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %ecx
+; AVX512F-NEXT: movzwl %cx, %ecx
+; AVX512F-NEXT: orl %eax, %ecx
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %edx
+; AVX512F-NEXT: movzwl %dx, %edx
+; AVX512F-NEXT: orl %eax, %edx
+; AVX512F-NEXT: shlq $32, %rdx
+; AVX512F-NEXT: orq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm0
+; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX512F-NEXT: vmovdqa %xmm0, (%rdi)
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_4f32_to_8i16_undef:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $16, %rsp
+; AVX512VL-NEXT: movq %rdi, %r14
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bp
+; AVX512VL-NEXT: shll $16, %ebp
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %ebx
+; AVX512VL-NEXT: orl %ebp, %ebx
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bp
+; AVX512VL-NEXT: shll $16, %ebp
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebp, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %rbx, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512VL-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512VL-NEXT: vmovdqa32 %xmm0, (%r14)
+; AVX512VL-NEXT: addq $16, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x float> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
%3 = shufflevector <4 x i16> %2, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
@@ -1984,32 +4038,125 @@ define void @store_cvt_4f32_to_8i16_unde
ret void
}
-define void @store_cvt_4f32_to_8i16_zero(<4 x float> %a0, <8 x i16>* %a1) {
-; ALL-LABEL: store_cvt_4f32_to_8i16_zero:
-; ALL: # BB#0:
-; ALL-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: shll $16, %eax
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm1
-; ALL-NEXT: vmovd %xmm1, %ecx
-; ALL-NEXT: movzwl %cx, %ecx
-; ALL-NEXT: orl %eax, %ecx
-; ALL-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; ALL-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; ALL-NEXT: vmovd %xmm1, %eax
-; ALL-NEXT: shll $16, %eax
-; ALL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; ALL-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; ALL-NEXT: vmovd %xmm0, %edx
-; ALL-NEXT: movzwl %dx, %edx
-; ALL-NEXT: orl %eax, %edx
-; ALL-NEXT: shlq $32, %rdx
-; ALL-NEXT: orq %rcx, %rdx
-; ALL-NEXT: vmovq %rdx, %xmm0
-; ALL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
-; ALL-NEXT: vmovdqa %xmm0, (%rdi)
-; ALL-NEXT: retq
+define void @store_cvt_4f32_to_8i16_zero(<4 x float> %a0, <8 x i16>* %a1) nounwind {
+; AVX1-LABEL: store_cvt_4f32_to_8i16_zero:
+; AVX1: # BB#0:
+; AVX1-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: shll $16, %eax
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %ecx
+; AVX1-NEXT: movzwl %cx, %ecx
+; AVX1-NEXT: orl %eax, %ecx
+; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX1-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX1-NEXT: vmovd %xmm1, %eax
+; AVX1-NEXT: shll $16, %eax
+; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX1-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX1-NEXT: vmovd %xmm0, %edx
+; AVX1-NEXT: movzwl %dx, %edx
+; AVX1-NEXT: orl %eax, %edx
+; AVX1-NEXT: shlq $32, %rdx
+; AVX1-NEXT: orq %rcx, %rdx
+; AVX1-NEXT: vmovq %rdx, %xmm0
+; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT: vmovdqa %xmm0, (%rdi)
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: store_cvt_4f32_to_8i16_zero:
+; AVX2: # BB#0:
+; AVX2-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: shll $16, %eax
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %ecx
+; AVX2-NEXT: movzwl %cx, %ecx
+; AVX2-NEXT: orl %eax, %ecx
+; AVX2-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX2-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX2-NEXT: vmovd %xmm1, %eax
+; AVX2-NEXT: shll $16, %eax
+; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX2-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX2-NEXT: vmovd %xmm0, %edx
+; AVX2-NEXT: movzwl %dx, %edx
+; AVX2-NEXT: orl %eax, %edx
+; AVX2-NEXT: shlq $32, %rdx
+; AVX2-NEXT: orq %rcx, %rdx
+; AVX2-NEXT: vmovq %rdx, %xmm0
+; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-NEXT: vmovdqa %xmm0, (%rdi)
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: store_cvt_4f32_to_8i16_zero:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %ecx
+; AVX512F-NEXT: movzwl %cx, %ecx
+; AVX512F-NEXT: orl %eax, %ecx
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: shll $16, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %edx
+; AVX512F-NEXT: movzwl %dx, %edx
+; AVX512F-NEXT: orl %eax, %edx
+; AVX512F-NEXT: shlq $32, %rdx
+; AVX512F-NEXT: orq %rcx, %rdx
+; AVX512F-NEXT: vmovq %rdx, %xmm0
+; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT: vmovdqa %xmm0, (%rdi)
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_4f32_to_8i16_zero:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $16, %rsp
+; AVX512VL-NEXT: movq %rdi, %r14
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bp
+; AVX512VL-NEXT: shll $16, %ebp
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %ebx
+; AVX512VL-NEXT: orl %ebp, %ebx
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, %bp
+; AVX512VL-NEXT: shll $16, %ebp
+; AVX512VL-NEXT: vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebp, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %rbx, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512VL-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,2]
+; AVX512VL-NEXT: vpxord %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX512VL-NEXT: vmovdqa %xmm0, (%r14)
+; AVX512VL-NEXT: addq $16, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x float> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
%3 = shufflevector <4 x i16> %2, <4 x i16> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
@@ -2017,7 +4164,7 @@ define void @store_cvt_4f32_to_8i16_zero
ret void
}
-define void @store_cvt_8f32_to_8i16(<8 x float> %a0, <8 x i16>* %a1) {
+define void @store_cvt_8f32_to_8i16(<8 x float> %a0, <8 x i16>* %a1) nounwind {
; AVX1-LABEL: store_cvt_8f32_to_8i16:
; AVX1: # BB#0:
; AVX1-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
@@ -2090,47 +4237,108 @@ define void @store_cvt_8f32_to_8i16(<8 x
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; AVX512-LABEL: store_cvt_8f32_to_8i16:
-; AVX512: # BB#0:
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vmovd %xmm1, %r8d
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vmovd %xmm1, %r9d
-; AVX512-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: vmovd %xmm1, %r10d
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm2
-; AVX512-NEXT: vmovd %xmm2, %r11d
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm2 = xmm1[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm2
-; AVX512-NEXT: vmovd %xmm2, %eax
-; AVX512-NEXT: vpermilps {{.*#+}} xmm2 = xmm1[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm2
-; AVX512-NEXT: vmovd %xmm2, %ecx
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; AVX512-NEXT: vmovd %xmm0, %edx
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm0
-; AVX512-NEXT: vmovd %xmm0, %esi
-; AVX512-NEXT: movw %si, 8(%rdi)
-; AVX512-NEXT: movw %dx, (%rdi)
-; AVX512-NEXT: movw %cx, 14(%rdi)
-; AVX512-NEXT: movw %ax, 12(%rdi)
-; AVX512-NEXT: movw %r11w, 10(%rdi)
-; AVX512-NEXT: movw %r10w, 6(%rdi)
-; AVX512-NEXT: movw %r9w, 4(%rdi)
-; AVX512-NEXT: movw %r8w, 2(%rdi)
-; AVX512-NEXT: retq
+; AVX512F-LABEL: store_cvt_8f32_to_8i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %r8d
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %r9d
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: vmovd %xmm1, %r10d
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm1
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm2
+; AVX512F-NEXT: vmovd %xmm2, %r11d
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm2 = xmm1[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm2
+; AVX512F-NEXT: vmovd %xmm2, %eax
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm2 = xmm1[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm2
+; AVX512F-NEXT: vmovd %xmm2, %ecx
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %edx
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm0
+; AVX512F-NEXT: vmovd %xmm0, %esi
+; AVX512F-NEXT: movw %si, 8(%rdi)
+; AVX512F-NEXT: movw %dx, (%rdi)
+; AVX512F-NEXT: movw %cx, 14(%rdi)
+; AVX512F-NEXT: movw %ax, 12(%rdi)
+; AVX512F-NEXT: movw %r11w, 10(%rdi)
+; AVX512F-NEXT: movw %r10w, 6(%rdi)
+; AVX512F-NEXT: movw %r9w, 4(%rdi)
+; AVX512F-NEXT: movw %r8w, 2(%rdi)
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_8f32_to_8i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %r13
+; AVX512VL-NEXT: pushq %r12
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $72, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: vmovdqu %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovdqu {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovdqu {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %r12d
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %r13d
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %ebp
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %r14d
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %r15d
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, 8(%rbx)
+; AVX512VL-NEXT: movw %r15w, (%rbx)
+; AVX512VL-NEXT: movw %r14w, 14(%rbx)
+; AVX512VL-NEXT: movw %bp, 12(%rbx)
+; AVX512VL-NEXT: movw %r13w, 10(%rbx)
+; AVX512VL-NEXT: movw %r12w, 6(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 4(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 2(%rbx)
+; AVX512VL-NEXT: addq $72, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r12
+; AVX512VL-NEXT: popq %r13
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <8 x float> %a0 to <8 x half>
%2 = bitcast <8 x half> %1 to <8 x i16>
store <8 x i16> %2, <8 x i16>* %a1
ret void
}
-define void @store_cvt_16f32_to_16i16(<16 x float> %a0, <16 x i16>* %a1) {
+define void @store_cvt_16f32_to_16i16(<16 x float> %a0, <16 x i16>* %a1) nounwind {
; AVX1-LABEL: store_cvt_16f32_to_16i16:
; AVX1: # BB#0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
@@ -2265,72 +4473,184 @@ define void @store_cvt_16f32_to_16i16(<1
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
-; AVX512-LABEL: store_cvt_16f32_to_16i16:
-; AVX512: # BB#0:
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX512-NEXT: vextractf64x4 $1, %zmm0, %ymm2
-; AVX512-NEXT: vextractf128 $1, %ymm2, %xmm3
-; AVX512-NEXT: vcvtps2ph $4, %xmm3, %xmm4
-; AVX512-NEXT: vmovd %xmm4, %eax
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm4
-; AVX512-NEXT: movw %ax, 24(%rdi)
-; AVX512-NEXT: vmovd %xmm4, %eax
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm4
-; AVX512-NEXT: movw %ax, 16(%rdi)
-; AVX512-NEXT: vmovd %xmm4, %eax
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm4
-; AVX512-NEXT: movw %ax, 8(%rdi)
-; AVX512-NEXT: vmovd %xmm4, %eax
-; AVX512-NEXT: vpermilps {{.*#+}} xmm4 = xmm3[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm4, %xmm4
-; AVX512-NEXT: movw %ax, (%rdi)
-; AVX512-NEXT: vmovd %xmm4, %eax
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm4 = xmm3[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm4, %xmm4
-; AVX512-NEXT: movw %ax, 30(%rdi)
-; AVX512-NEXT: vmovd %xmm4, %eax
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm4 = xmm0[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm4, %xmm4
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm3, %xmm3
-; AVX512-NEXT: movw %ax, 28(%rdi)
-; AVX512-NEXT: vmovd %xmm3, %eax
-; AVX512-NEXT: vpermilps {{.*#+}} xmm3 = xmm2[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm3, %xmm3
-; AVX512-NEXT: movw %ax, 26(%rdi)
-; AVX512-NEXT: vmovd %xmm3, %eax
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm3, %xmm3
-; AVX512-NEXT: movw %ax, 22(%rdi)
-; AVX512-NEXT: vmovd %xmm3, %eax
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm3 = xmm0[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm3, %xmm3
-; AVX512-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm0
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm2
-; AVX512-NEXT: movw %ax, 20(%rdi)
-; AVX512-NEXT: vmovd %xmm2, %eax
-; AVX512-NEXT: vpermilps {{.*#+}} xmm2 = xmm1[3,1,2,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm2
-; AVX512-NEXT: movw %ax, 18(%rdi)
-; AVX512-NEXT: vmovd %xmm2, %eax
-; AVX512-NEXT: vmovshdup {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; AVX512-NEXT: vcvtps2ph $4, %xmm2, %xmm2
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm1 = xmm1[1,0]
-; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1
-; AVX512-NEXT: movw %ax, 14(%rdi)
-; AVX512-NEXT: vmovd %xmm1, %eax
-; AVX512-NEXT: movw %ax, 12(%rdi)
-; AVX512-NEXT: vmovd %xmm2, %eax
-; AVX512-NEXT: movw %ax, 10(%rdi)
-; AVX512-NEXT: vmovd %xmm0, %eax
-; AVX512-NEXT: movw %ax, 6(%rdi)
-; AVX512-NEXT: vmovd %xmm3, %eax
-; AVX512-NEXT: movw %ax, 4(%rdi)
-; AVX512-NEXT: vmovd %xmm4, %eax
-; AVX512-NEXT: movw %ax, 2(%rdi)
-; AVX512-NEXT: retq
+; AVX512F-LABEL: store_cvt_16f32_to_16i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm1
+; AVX512F-NEXT: vextractf64x4 $1, %zmm0, %ymm2
+; AVX512F-NEXT: vextractf128 $1, %ymm2, %xmm3
+; AVX512F-NEXT: vcvtps2ph $4, %xmm3, %xmm4
+; AVX512F-NEXT: vmovd %xmm4, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm4
+; AVX512F-NEXT: movw %ax, 24(%rdi)
+; AVX512F-NEXT: vmovd %xmm4, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm4
+; AVX512F-NEXT: movw %ax, 16(%rdi)
+; AVX512F-NEXT: vmovd %xmm4, %eax
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm4
+; AVX512F-NEXT: movw %ax, 8(%rdi)
+; AVX512F-NEXT: vmovd %xmm4, %eax
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm4 = xmm3[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm4, %xmm4
+; AVX512F-NEXT: movw %ax, (%rdi)
+; AVX512F-NEXT: vmovd %xmm4, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm4 = xmm3[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm4, %xmm4
+; AVX512F-NEXT: movw %ax, 30(%rdi)
+; AVX512F-NEXT: vmovd %xmm4, %eax
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm4 = xmm0[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm4, %xmm4
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm3, %xmm3
+; AVX512F-NEXT: movw %ax, 28(%rdi)
+; AVX512F-NEXT: vmovd %xmm3, %eax
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm3 = xmm2[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm3, %xmm3
+; AVX512F-NEXT: movw %ax, 26(%rdi)
+; AVX512F-NEXT: vmovd %xmm3, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm3, %xmm3
+; AVX512F-NEXT: movw %ax, 22(%rdi)
+; AVX512F-NEXT: vmovd %xmm3, %eax
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm3 = xmm0[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm3, %xmm3
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm0, %xmm0
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm2
+; AVX512F-NEXT: movw %ax, 20(%rdi)
+; AVX512F-NEXT: vmovd %xmm2, %eax
+; AVX512F-NEXT: vpermilps {{.*#+}} xmm2 = xmm1[3,1,2,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm2
+; AVX512F-NEXT: movw %ax, 18(%rdi)
+; AVX512F-NEXT: vmovd %xmm2, %eax
+; AVX512F-NEXT: vmovshdup {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm2, %xmm2
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm1 = xmm1[1,0]
+; AVX512F-NEXT: vcvtps2ph $4, %xmm1, %xmm1
+; AVX512F-NEXT: movw %ax, 14(%rdi)
+; AVX512F-NEXT: vmovd %xmm1, %eax
+; AVX512F-NEXT: movw %ax, 12(%rdi)
+; AVX512F-NEXT: vmovd %xmm2, %eax
+; AVX512F-NEXT: movw %ax, 10(%rdi)
+; AVX512F-NEXT: vmovd %xmm0, %eax
+; AVX512F-NEXT: movw %ax, 6(%rdi)
+; AVX512F-NEXT: vmovd %xmm3, %eax
+; AVX512F-NEXT: movw %ax, 4(%rdi)
+; AVX512F-NEXT: vmovd %xmm4, %eax
+; AVX512F-NEXT: movw %ax, 2(%rdi)
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_16f32_to_16i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %r13
+; AVX512VL-NEXT: pushq %r12
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $200, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: vmovdqu64 %zmm0, {{[0-9]+}}(%rsp) # 64-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovdqu64 {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovdqu64 {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovupd {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
+; AVX512VL-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovdqu {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovdqu {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %r14d
+; AVX512VL-NEXT: vmovdqa {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %r15d
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %r12d
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %r13d
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movl %eax, %ebp
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __gnu_f2h_ieee
+; AVX512VL-NEXT: movw %ax, 24(%rbx)
+; AVX512VL-NEXT: movw %bp, 16(%rbx)
+; AVX512VL-NEXT: movw %r13w, 8(%rbx)
+; AVX512VL-NEXT: movw %r12w, (%rbx)
+; AVX512VL-NEXT: movw %r15w, 30(%rbx)
+; AVX512VL-NEXT: movw %r14w, 28(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 26(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 22(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 20(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 18(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 14(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 12(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 10(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 6(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 4(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 2(%rbx)
+; AVX512VL-NEXT: addq $200, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r12
+; AVX512VL-NEXT: popq %r13
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <16 x float> %a0 to <16 x half>
%2 = bitcast <16 x half> %1 to <16 x i16>
store <16 x i16> %2, <16 x i16>* %a1
@@ -2341,7 +4661,7 @@ define void @store_cvt_16f32_to_16i16(<1
; Double to Half
;
-define i16 @cvt_f64_to_i16(double %a0) {
+define i16 @cvt_f64_to_i16(double %a0) nounwind {
; ALL-LABEL: cvt_f64_to_i16:
; ALL: # BB#0:
; ALL-NEXT: jmp __truncdfhf2 # TAILCALL
@@ -2350,51 +4670,89 @@ define i16 @cvt_f64_to_i16(double %a0) {
ret i16 %2
}
-define <2 x i16> @cvt_2f64_to_2i16(<2 x double> %a0) {
-; ALL-LABEL: cvt_2f64_to_2i16:
-; ALL: # BB#0:
-; ALL-NEXT: pushq %rbx
-; ALL-NEXT: .Ltmp0:
-; ALL-NEXT: .cfi_def_cfa_offset 16
-; ALL-NEXT: subq $16, %rsp
-; ALL-NEXT: .Ltmp1:
-; ALL-NEXT: .cfi_def_cfa_offset 32
-; ALL-NEXT: .Ltmp2:
-; ALL-NEXT: .cfi_offset %rbx, -16
-; ALL-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
-; ALL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; ALL-NEXT: callq __truncdfhf2
-; ALL-NEXT: movw %ax, %bx
-; ALL-NEXT: shll $16, %ebx
-; ALL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
-; ALL-NEXT: callq __truncdfhf2
-; ALL-NEXT: movzwl %ax, %eax
-; ALL-NEXT: orl %ebx, %eax
-; ALL-NEXT: vmovd %eax, %xmm0
-; ALL-NEXT: addq $16, %rsp
-; ALL-NEXT: popq %rbx
-; ALL-NEXT: retq
+define <2 x i16> @cvt_2f64_to_2i16(<2 x double> %a0) nounwind {
+; AVX1-LABEL: cvt_2f64_to_2i16:
+; AVX1: # BB#0:
+; AVX1-NEXT: pushq %rbx
+; AVX1-NEXT: subq $16, %rsp
+; AVX1-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX1-NEXT: callq __truncdfhf2
+; AVX1-NEXT: movw %ax, %bx
+; AVX1-NEXT: shll $16, %ebx
+; AVX1-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX1-NEXT: callq __truncdfhf2
+; AVX1-NEXT: movzwl %ax, %eax
+; AVX1-NEXT: orl %ebx, %eax
+; AVX1-NEXT: vmovd %eax, %xmm0
+; AVX1-NEXT: addq $16, %rsp
+; AVX1-NEXT: popq %rbx
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: cvt_2f64_to_2i16:
+; AVX2: # BB#0:
+; AVX2-NEXT: pushq %rbx
+; AVX2-NEXT: subq $16, %rsp
+; AVX2-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX2-NEXT: callq __truncdfhf2
+; AVX2-NEXT: movw %ax, %bx
+; AVX2-NEXT: shll $16, %ebx
+; AVX2-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX2-NEXT: callq __truncdfhf2
+; AVX2-NEXT: movzwl %ax, %eax
+; AVX2-NEXT: orl %ebx, %eax
+; AVX2-NEXT: vmovd %eax, %xmm0
+; AVX2-NEXT: addq $16, %rsp
+; AVX2-NEXT: popq %rbx
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: cvt_2f64_to_2i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: subq $16, %rsp
+; AVX512F-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %eax
+; AVX512F-NEXT: orl %ebx, %eax
+; AVX512F-NEXT: vmovd %eax, %xmm0
+; AVX512F-NEXT: addq $16, %rsp
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_2f64_to_2i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $16, %rsp
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebx, %eax
+; AVX512VL-NEXT: vmovd %eax, %xmm0
+; AVX512VL-NEXT: addq $16, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: retq
%1 = fptrunc <2 x double> %a0 to <2 x half>
%2 = bitcast <2 x half> %1 to <2 x i16>
ret <2 x i16> %2
}
-define <4 x i16> @cvt_4f64_to_4i16(<4 x double> %a0) {
+define <4 x i16> @cvt_4f64_to_4i16(<4 x double> %a0) nounwind {
; AVX1-LABEL: cvt_4f64_to_4i16:
; AVX1: # BB#0:
; AVX1-NEXT: pushq %r14
-; AVX1-NEXT: .Ltmp3:
-; AVX1-NEXT: .cfi_def_cfa_offset 16
; AVX1-NEXT: pushq %rbx
-; AVX1-NEXT: .Ltmp4:
-; AVX1-NEXT: .cfi_def_cfa_offset 24
; AVX1-NEXT: subq $40, %rsp
-; AVX1-NEXT: .Ltmp5:
-; AVX1-NEXT: .cfi_def_cfa_offset 64
-; AVX1-NEXT: .Ltmp6:
-; AVX1-NEXT: .cfi_offset %rbx, -24
-; AVX1-NEXT: .Ltmp7:
-; AVX1-NEXT: .cfi_offset %r14, -16
; AVX1-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT: vzeroupper
@@ -2430,18 +4788,8 @@ define <4 x i16> @cvt_4f64_to_4i16(<4 x
; AVX2-LABEL: cvt_4f64_to_4i16:
; AVX2: # BB#0:
; AVX2-NEXT: pushq %r14
-; AVX2-NEXT: .Ltmp3:
-; AVX2-NEXT: .cfi_def_cfa_offset 16
; AVX2-NEXT: pushq %rbx
-; AVX2-NEXT: .Ltmp4:
-; AVX2-NEXT: .cfi_def_cfa_offset 24
; AVX2-NEXT: subq $40, %rsp
-; AVX2-NEXT: .Ltmp5:
-; AVX2-NEXT: .cfi_def_cfa_offset 64
-; AVX2-NEXT: .Ltmp6:
-; AVX2-NEXT: .cfi_offset %rbx, -24
-; AVX2-NEXT: .Ltmp7:
-; AVX2-NEXT: .cfi_offset %r14, -16
; AVX2-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT: vzeroupper
@@ -2474,70 +4822,84 @@ define <4 x i16> @cvt_4f64_to_4i16(<4 x
; AVX2-NEXT: popq %r14
; AVX2-NEXT: retq
;
-; AVX512-LABEL: cvt_4f64_to_4i16:
-; AVX512: # BB#0:
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: .Ltmp3:
-; AVX512-NEXT: .cfi_def_cfa_offset 16
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: .Ltmp4:
-; AVX512-NEXT: .cfi_def_cfa_offset 24
-; AVX512-NEXT: subq $40, %rsp
-; AVX512-NEXT: .Ltmp5:
-; AVX512-NEXT: .cfi_def_cfa_offset 64
-; AVX512-NEXT: .Ltmp6:
-; AVX512-NEXT: .cfi_offset %rbx, -24
-; AVX512-NEXT: .Ltmp7:
-; AVX512-NEXT: .cfi_offset %r14, -16
-; AVX512-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bx
-; AVX512-NEXT: shll $16, %ebx
-; AVX512-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %r14d
-; AVX512-NEXT: orl %ebx, %r14d
-; AVX512-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bx
-; AVX512-NEXT: shll $16, %ebx
-; AVX512-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %eax
-; AVX512-NEXT: orl %ebx, %eax
-; AVX512-NEXT: shlq $32, %rax
-; AVX512-NEXT: orq %r14, %rax
-; AVX512-NEXT: vmovq %rax, %xmm0
-; AVX512-NEXT: addq $40, %rsp
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: retq
+; AVX512F-LABEL: cvt_4f64_to_4i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: subq $40, %rsp
+; AVX512F-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %r14d
+; AVX512F-NEXT: orl %ebx, %r14d
+; AVX512F-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %eax
+; AVX512F-NEXT: orl %ebx, %eax
+; AVX512F-NEXT: shlq $32, %rax
+; AVX512F-NEXT: orq %r14, %rax
+; AVX512F-NEXT: vmovq %rax, %xmm0
+; AVX512F-NEXT: addq $40, %rsp
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_4f64_to_4i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $40, %rsp
+; AVX512VL-NEXT: vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %r14d
+; AVX512VL-NEXT: orl %ebx, %r14d
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebx, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %r14, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: addq $40, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x double> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
ret <4 x i16> %2
}
-define <8 x i16> @cvt_4f64_to_8i16_undef(<4 x double> %a0) {
+define <8 x i16> @cvt_4f64_to_8i16_undef(<4 x double> %a0) nounwind {
; AVX1-LABEL: cvt_4f64_to_8i16_undef:
; AVX1: # BB#0:
; AVX1-NEXT: pushq %r14
-; AVX1-NEXT: .Ltmp8:
-; AVX1-NEXT: .cfi_def_cfa_offset 16
; AVX1-NEXT: pushq %rbx
-; AVX1-NEXT: .Ltmp9:
-; AVX1-NEXT: .cfi_def_cfa_offset 24
; AVX1-NEXT: subq $40, %rsp
-; AVX1-NEXT: .Ltmp10:
-; AVX1-NEXT: .cfi_def_cfa_offset 64
-; AVX1-NEXT: .Ltmp11:
-; AVX1-NEXT: .cfi_offset %rbx, -24
-; AVX1-NEXT: .Ltmp12:
-; AVX1-NEXT: .cfi_offset %r14, -16
; AVX1-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT: vzeroupper
@@ -2574,18 +4936,8 @@ define <8 x i16> @cvt_4f64_to_8i16_undef
; AVX2-LABEL: cvt_4f64_to_8i16_undef:
; AVX2: # BB#0:
; AVX2-NEXT: pushq %r14
-; AVX2-NEXT: .Ltmp8:
-; AVX2-NEXT: .cfi_def_cfa_offset 16
; AVX2-NEXT: pushq %rbx
-; AVX2-NEXT: .Ltmp9:
-; AVX2-NEXT: .cfi_def_cfa_offset 24
; AVX2-NEXT: subq $40, %rsp
-; AVX2-NEXT: .Ltmp10:
-; AVX2-NEXT: .cfi_def_cfa_offset 64
-; AVX2-NEXT: .Ltmp11:
-; AVX2-NEXT: .cfi_offset %rbx, -24
-; AVX2-NEXT: .Ltmp12:
-; AVX2-NEXT: .cfi_offset %r14, -16
; AVX2-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT: vzeroupper
@@ -2619,72 +4971,89 @@ define <8 x i16> @cvt_4f64_to_8i16_undef
; AVX2-NEXT: popq %r14
; AVX2-NEXT: retq
;
-; AVX512-LABEL: cvt_4f64_to_8i16_undef:
-; AVX512: # BB#0:
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: .Ltmp8:
-; AVX512-NEXT: .cfi_def_cfa_offset 16
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: .Ltmp9:
-; AVX512-NEXT: .cfi_def_cfa_offset 24
-; AVX512-NEXT: subq $40, %rsp
-; AVX512-NEXT: .Ltmp10:
-; AVX512-NEXT: .cfi_def_cfa_offset 64
-; AVX512-NEXT: .Ltmp11:
-; AVX512-NEXT: .cfi_offset %rbx, -24
-; AVX512-NEXT: .Ltmp12:
-; AVX512-NEXT: .cfi_offset %r14, -16
-; AVX512-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bx
-; AVX512-NEXT: shll $16, %ebx
-; AVX512-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %r14d
-; AVX512-NEXT: orl %ebx, %r14d
-; AVX512-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bx
-; AVX512-NEXT: shll $16, %ebx
-; AVX512-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %eax
-; AVX512-NEXT: orl %ebx, %eax
-; AVX512-NEXT: shlq $32, %rax
-; AVX512-NEXT: orq %r14, %rax
-; AVX512-NEXT: vmovq %rax, %xmm0
-; AVX512-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX512-NEXT: addq $40, %rsp
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: retq
+; AVX512F-LABEL: cvt_4f64_to_8i16_undef:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: subq $40, %rsp
+; AVX512F-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %r14d
+; AVX512F-NEXT: orl %ebx, %r14d
+; AVX512F-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %eax
+; AVX512F-NEXT: orl %ebx, %eax
+; AVX512F-NEXT: shlq $32, %rax
+; AVX512F-NEXT: orq %r14, %rax
+; AVX512F-NEXT: vmovq %rax, %xmm0
+; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX512F-NEXT: addq $40, %rsp
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_4f64_to_8i16_undef:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $40, %rsp
+; AVX512VL-NEXT: vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %r14d
+; AVX512VL-NEXT: orl %ebx, %r14d
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebx, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %r14, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512VL-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512VL-NEXT: addq $40, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x double> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
%3 = shufflevector <4 x i16> %2, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <8 x i16> %3
}
-define <8 x i16> @cvt_4f64_to_8i16_zero(<4 x double> %a0) {
+define <8 x i16> @cvt_4f64_to_8i16_zero(<4 x double> %a0) nounwind {
; AVX1-LABEL: cvt_4f64_to_8i16_zero:
; AVX1: # BB#0:
; AVX1-NEXT: pushq %r14
-; AVX1-NEXT: .Ltmp13:
-; AVX1-NEXT: .cfi_def_cfa_offset 16
; AVX1-NEXT: pushq %rbx
-; AVX1-NEXT: .Ltmp14:
-; AVX1-NEXT: .cfi_def_cfa_offset 24
; AVX1-NEXT: subq $40, %rsp
-; AVX1-NEXT: .Ltmp15:
-; AVX1-NEXT: .cfi_def_cfa_offset 64
-; AVX1-NEXT: .Ltmp16:
-; AVX1-NEXT: .cfi_offset %rbx, -24
-; AVX1-NEXT: .Ltmp17:
-; AVX1-NEXT: .cfi_offset %r14, -16
; AVX1-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT: vzeroupper
@@ -2721,18 +5090,8 @@ define <8 x i16> @cvt_4f64_to_8i16_zero(
; AVX2-LABEL: cvt_4f64_to_8i16_zero:
; AVX2: # BB#0:
; AVX2-NEXT: pushq %r14
-; AVX2-NEXT: .Ltmp13:
-; AVX2-NEXT: .cfi_def_cfa_offset 16
; AVX2-NEXT: pushq %rbx
-; AVX2-NEXT: .Ltmp14:
-; AVX2-NEXT: .cfi_def_cfa_offset 24
; AVX2-NEXT: subq $40, %rsp
-; AVX2-NEXT: .Ltmp15:
-; AVX2-NEXT: .cfi_def_cfa_offset 64
-; AVX2-NEXT: .Ltmp16:
-; AVX2-NEXT: .cfi_offset %rbx, -24
-; AVX2-NEXT: .Ltmp17:
-; AVX2-NEXT: .cfi_offset %r14, -16
; AVX2-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT: vzeroupper
@@ -2766,77 +5125,92 @@ define <8 x i16> @cvt_4f64_to_8i16_zero(
; AVX2-NEXT: popq %r14
; AVX2-NEXT: retq
;
-; AVX512-LABEL: cvt_4f64_to_8i16_zero:
-; AVX512: # BB#0:
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: .Ltmp13:
-; AVX512-NEXT: .cfi_def_cfa_offset 16
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: .Ltmp14:
-; AVX512-NEXT: .cfi_def_cfa_offset 24
-; AVX512-NEXT: subq $40, %rsp
-; AVX512-NEXT: .Ltmp15:
-; AVX512-NEXT: .cfi_def_cfa_offset 64
-; AVX512-NEXT: .Ltmp16:
-; AVX512-NEXT: .cfi_offset %rbx, -24
-; AVX512-NEXT: .Ltmp17:
-; AVX512-NEXT: .cfi_offset %r14, -16
-; AVX512-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bx
-; AVX512-NEXT: shll $16, %ebx
-; AVX512-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %r14d
-; AVX512-NEXT: orl %ebx, %r14d
-; AVX512-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bx
-; AVX512-NEXT: shll $16, %ebx
-; AVX512-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %eax
-; AVX512-NEXT: orl %ebx, %eax
-; AVX512-NEXT: shlq $32, %rax
-; AVX512-NEXT: orq %r14, %rax
-; AVX512-NEXT: vmovq %rax, %xmm0
-; AVX512-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
-; AVX512-NEXT: addq $40, %rsp
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: retq
+; AVX512F-LABEL: cvt_4f64_to_8i16_zero:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: subq $40, %rsp
+; AVX512F-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %r14d
+; AVX512F-NEXT: orl %ebx, %r14d
+; AVX512F-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %eax
+; AVX512F-NEXT: orl %ebx, %eax
+; AVX512F-NEXT: shlq $32, %rax
+; AVX512F-NEXT: orq %r14, %rax
+; AVX512F-NEXT: vmovq %rax, %xmm0
+; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT: addq $40, %rsp
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_4f64_to_8i16_zero:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $40, %rsp
+; AVX512VL-NEXT: vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %r14d
+; AVX512VL-NEXT: orl %ebx, %r14d
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebx, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %r14, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512VL-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,2]
+; AVX512VL-NEXT: vpxord %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX512VL-NEXT: addq $40, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x double> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
%3 = shufflevector <4 x i16> %2, <4 x i16> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <8 x i16> %3
}
-define <8 x i16> @cvt_8f64_to_8i16(<8 x double> %a0) {
+define <8 x i16> @cvt_8f64_to_8i16(<8 x double> %a0) nounwind {
; AVX1-LABEL: cvt_8f64_to_8i16:
; AVX1: # BB#0:
; AVX1-NEXT: pushq %r15
-; AVX1-NEXT: .Ltmp18:
-; AVX1-NEXT: .cfi_def_cfa_offset 16
; AVX1-NEXT: pushq %r14
-; AVX1-NEXT: .Ltmp19:
-; AVX1-NEXT: .cfi_def_cfa_offset 24
; AVX1-NEXT: pushq %rbx
-; AVX1-NEXT: .Ltmp20:
-; AVX1-NEXT: .cfi_def_cfa_offset 32
; AVX1-NEXT: subq $64, %rsp
-; AVX1-NEXT: .Ltmp21:
-; AVX1-NEXT: .cfi_def_cfa_offset 96
-; AVX1-NEXT: .Ltmp22:
-; AVX1-NEXT: .cfi_offset %rbx, -32
-; AVX1-NEXT: .Ltmp23:
-; AVX1-NEXT: .cfi_offset %r14, -24
-; AVX1-NEXT: .Ltmp24:
-; AVX1-NEXT: .cfi_offset %r15, -16
; AVX1-NEXT: vmovups %ymm1, (%rsp) # 32-byte Spill
; AVX1-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
@@ -2902,23 +5276,9 @@ define <8 x i16> @cvt_8f64_to_8i16(<8 x
; AVX2-LABEL: cvt_8f64_to_8i16:
; AVX2: # BB#0:
; AVX2-NEXT: pushq %r15
-; AVX2-NEXT: .Ltmp18:
-; AVX2-NEXT: .cfi_def_cfa_offset 16
; AVX2-NEXT: pushq %r14
-; AVX2-NEXT: .Ltmp19:
-; AVX2-NEXT: .cfi_def_cfa_offset 24
; AVX2-NEXT: pushq %rbx
-; AVX2-NEXT: .Ltmp20:
-; AVX2-NEXT: .cfi_def_cfa_offset 32
; AVX2-NEXT: subq $64, %rsp
-; AVX2-NEXT: .Ltmp21:
-; AVX2-NEXT: .cfi_def_cfa_offset 96
-; AVX2-NEXT: .Ltmp22:
-; AVX2-NEXT: .cfi_offset %rbx, -32
-; AVX2-NEXT: .Ltmp23:
-; AVX2-NEXT: .cfi_offset %r14, -24
-; AVX2-NEXT: .Ltmp24:
-; AVX2-NEXT: .cfi_offset %r15, -16
; AVX2-NEXT: vmovups %ymm1, (%rsp) # 32-byte Spill
; AVX2-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
@@ -2981,82 +5341,131 @@ define <8 x i16> @cvt_8f64_to_8i16(<8 x
; AVX2-NEXT: popq %r15
; AVX2-NEXT: retq
;
-; AVX512-LABEL: cvt_8f64_to_8i16:
-; AVX512: # BB#0:
-; AVX512-NEXT: pushq %r15
-; AVX512-NEXT: .Ltmp18:
-; AVX512-NEXT: .cfi_def_cfa_offset 16
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: .Ltmp19:
-; AVX512-NEXT: .cfi_def_cfa_offset 24
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: .Ltmp20:
-; AVX512-NEXT: .cfi_def_cfa_offset 32
-; AVX512-NEXT: subq $96, %rsp
-; AVX512-NEXT: .Ltmp21:
-; AVX512-NEXT: .cfi_def_cfa_offset 128
-; AVX512-NEXT: .Ltmp22:
-; AVX512-NEXT: .cfi_offset %rbx, -32
-; AVX512-NEXT: .Ltmp23:
-; AVX512-NEXT: .cfi_offset %r14, -24
-; AVX512-NEXT: .Ltmp24:
-; AVX512-NEXT: .cfi_offset %r15, -16
-; AVX512-NEXT: vmovupd %zmm0, (%rsp) # 64-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bx
-; AVX512-NEXT: shll $16, %ebx
-; AVX512-NEXT: vmovups (%rsp), %zmm0 # 64-byte Reload
-; AVX512-NEXT: # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %r15d
-; AVX512-NEXT: orl %ebx, %r15d
-; AVX512-NEXT: vmovupd (%rsp), %zmm0 # 64-byte Reload
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovapd %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bx
-; AVX512-NEXT: shll $16, %ebx
-; AVX512-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %r14d
-; AVX512-NEXT: orl %ebx, %r14d
-; AVX512-NEXT: shlq $32, %r14
-; AVX512-NEXT: orq %r15, %r14
-; AVX512-NEXT: vmovupd (%rsp), %zmm0 # 64-byte Reload
-; AVX512-NEXT: vextractf64x4 $1, %zmm0, %ymm0
-; AVX512-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bx
-; AVX512-NEXT: shll $16, %ebx
-; AVX512-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %r15d
-; AVX512-NEXT: orl %ebx, %r15d
-; AVX512-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bx
-; AVX512-NEXT: shll $16, %ebx
-; AVX512-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %eax
-; AVX512-NEXT: orl %ebx, %eax
-; AVX512-NEXT: shlq $32, %rax
-; AVX512-NEXT: orq %r15, %rax
-; AVX512-NEXT: vmovq %rax, %xmm0
-; AVX512-NEXT: vmovq %r14, %xmm1
-; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; AVX512-NEXT: addq $96, %rsp
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: popq %r15
-; AVX512-NEXT: retq
+; AVX512F-LABEL: cvt_8f64_to_8i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: pushq %r15
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: subq $96, %rsp
+; AVX512F-NEXT: vmovupd %zmm0, (%rsp) # 64-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovups (%rsp), %zmm0 # 64-byte Reload
+; AVX512F-NEXT: # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %r15d
+; AVX512F-NEXT: orl %ebx, %r15d
+; AVX512F-NEXT: vmovupd (%rsp), %zmm0 # 64-byte Reload
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovapd %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %r14d
+; AVX512F-NEXT: orl %ebx, %r14d
+; AVX512F-NEXT: shlq $32, %r14
+; AVX512F-NEXT: orq %r15, %r14
+; AVX512F-NEXT: vmovupd (%rsp), %zmm0 # 64-byte Reload
+; AVX512F-NEXT: vextractf64x4 $1, %zmm0, %ymm0
+; AVX512F-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %r15d
+; AVX512F-NEXT: orl %ebx, %r15d
+; AVX512F-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bx
+; AVX512F-NEXT: shll $16, %ebx
+; AVX512F-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %eax
+; AVX512F-NEXT: orl %ebx, %eax
+; AVX512F-NEXT: shlq $32, %rax
+; AVX512F-NEXT: orq %r15, %rax
+; AVX512F-NEXT: vmovq %rax, %xmm0
+; AVX512F-NEXT: vmovq %r14, %xmm1
+; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT: addq $96, %rsp
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: popq %r15
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: cvt_8f64_to_8i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $96, %rsp
+; AVX512VL-NEXT: vmovdqu64 %zmm0, (%rsp) # 64-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovups (%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %r15d
+; AVX512VL-NEXT: orl %ebx, %r15d
+; AVX512VL-NEXT: vmovups (%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %r14d
+; AVX512VL-NEXT: orl %ebx, %r14d
+; AVX512VL-NEXT: shlq $32, %r14
+; AVX512VL-NEXT: orq %r15, %r14
+; AVX512VL-NEXT: vmovupd (%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
+; AVX512VL-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %r15d
+; AVX512VL-NEXT: orl %ebx, %r15d
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bx
+; AVX512VL-NEXT: shll $16, %ebx
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebx, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %r15, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: vmovq %r14, %xmm1
+; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512VL-NEXT: addq $96, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: retq
%1 = fptrunc <8 x double> %a0 to <8 x half>
%2 = bitcast <8 x half> %1 to <8 x i16>
ret <8 x i16> %2
@@ -3066,14 +5475,10 @@ define <8 x i16> @cvt_8f64_to_8i16(<8 x
; Double to Half (Store)
;
-define void @store_cvt_f64_to_i16(double %a0, i16* %a1) {
+define void @store_cvt_f64_to_i16(double %a0, i16* %a1) nounwind {
; ALL-LABEL: store_cvt_f64_to_i16:
; ALL: # BB#0:
; ALL-NEXT: pushq %rbx
-; ALL-NEXT: .Ltmp25:
-; ALL-NEXT: .cfi_def_cfa_offset 16
-; ALL-NEXT: .Ltmp26:
-; ALL-NEXT: .cfi_offset %rbx, -16
; ALL-NEXT: movq %rdi, %rbx
; ALL-NEXT: callq __truncdfhf2
; ALL-NEXT: movw %ax, (%rbx)
@@ -3085,67 +5490,96 @@ define void @store_cvt_f64_to_i16(double
ret void
}
-define void @store_cvt_2f64_to_2i16(<2 x double> %a0, <2 x i16>* %a1) {
-; ALL-LABEL: store_cvt_2f64_to_2i16:
-; ALL: # BB#0:
-; ALL-NEXT: pushq %rbp
-; ALL-NEXT: .Ltmp27:
-; ALL-NEXT: .cfi_def_cfa_offset 16
-; ALL-NEXT: pushq %rbx
-; ALL-NEXT: .Ltmp28:
-; ALL-NEXT: .cfi_def_cfa_offset 24
-; ALL-NEXT: subq $24, %rsp
-; ALL-NEXT: .Ltmp29:
-; ALL-NEXT: .cfi_def_cfa_offset 48
-; ALL-NEXT: .Ltmp30:
-; ALL-NEXT: .cfi_offset %rbx, -24
-; ALL-NEXT: .Ltmp31:
-; ALL-NEXT: .cfi_offset %rbp, -16
-; ALL-NEXT: movq %rdi, %rbx
-; ALL-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
-; ALL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; ALL-NEXT: callq __truncdfhf2
-; ALL-NEXT: movl %eax, %ebp
-; ALL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
-; ALL-NEXT: callq __truncdfhf2
-; ALL-NEXT: movw %ax, (%rbx)
-; ALL-NEXT: movw %bp, 2(%rbx)
-; ALL-NEXT: addq $24, %rsp
-; ALL-NEXT: popq %rbx
-; ALL-NEXT: popq %rbp
-; ALL-NEXT: retq
+define void @store_cvt_2f64_to_2i16(<2 x double> %a0, <2 x i16>* %a1) nounwind {
+; AVX1-LABEL: store_cvt_2f64_to_2i16:
+; AVX1: # BB#0:
+; AVX1-NEXT: pushq %rbp
+; AVX1-NEXT: pushq %rbx
+; AVX1-NEXT: subq $24, %rsp
+; AVX1-NEXT: movq %rdi, %rbx
+; AVX1-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX1-NEXT: callq __truncdfhf2
+; AVX1-NEXT: movl %eax, %ebp
+; AVX1-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX1-NEXT: callq __truncdfhf2
+; AVX1-NEXT: movw %ax, (%rbx)
+; AVX1-NEXT: movw %bp, 2(%rbx)
+; AVX1-NEXT: addq $24, %rsp
+; AVX1-NEXT: popq %rbx
+; AVX1-NEXT: popq %rbp
+; AVX1-NEXT: retq
+;
+; AVX2-LABEL: store_cvt_2f64_to_2i16:
+; AVX2: # BB#0:
+; AVX2-NEXT: pushq %rbp
+; AVX2-NEXT: pushq %rbx
+; AVX2-NEXT: subq $24, %rsp
+; AVX2-NEXT: movq %rdi, %rbx
+; AVX2-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX2-NEXT: callq __truncdfhf2
+; AVX2-NEXT: movl %eax, %ebp
+; AVX2-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX2-NEXT: callq __truncdfhf2
+; AVX2-NEXT: movw %ax, (%rbx)
+; AVX2-NEXT: movw %bp, 2(%rbx)
+; AVX2-NEXT: addq $24, %rsp
+; AVX2-NEXT: popq %rbx
+; AVX2-NEXT: popq %rbp
+; AVX2-NEXT: retq
+;
+; AVX512F-LABEL: store_cvt_2f64_to_2i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: pushq %rbp
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: subq $24, %rsp
+; AVX512F-NEXT: movq %rdi, %rbx
+; AVX512F-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movl %eax, %ebp
+; AVX512F-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, (%rbx)
+; AVX512F-NEXT: movw %bp, 2(%rbx)
+; AVX512F-NEXT: addq $24, %rsp
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %rbp
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_2f64_to_2i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $24, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movl %eax, %ebp
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, (%rbx)
+; AVX512VL-NEXT: movw %bp, 2(%rbx)
+; AVX512VL-NEXT: addq $24, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <2 x double> %a0 to <2 x half>
%2 = bitcast <2 x half> %1 to <2 x i16>
store <2 x i16> %2, <2 x i16>* %a1
ret void
}
-define void @store_cvt_4f64_to_4i16(<4 x double> %a0, <4 x i16>* %a1) {
+define void @store_cvt_4f64_to_4i16(<4 x double> %a0, <4 x i16>* %a1) nounwind {
; AVX1-LABEL: store_cvt_4f64_to_4i16:
; AVX1: # BB#0:
; AVX1-NEXT: pushq %rbp
-; AVX1-NEXT: .Ltmp32:
-; AVX1-NEXT: .cfi_def_cfa_offset 16
; AVX1-NEXT: pushq %r15
-; AVX1-NEXT: .Ltmp33:
-; AVX1-NEXT: .cfi_def_cfa_offset 24
; AVX1-NEXT: pushq %r14
-; AVX1-NEXT: .Ltmp34:
-; AVX1-NEXT: .cfi_def_cfa_offset 32
; AVX1-NEXT: pushq %rbx
-; AVX1-NEXT: .Ltmp35:
-; AVX1-NEXT: .cfi_def_cfa_offset 40
; AVX1-NEXT: subq $88, %rsp
-; AVX1-NEXT: .Ltmp36:
-; AVX1-NEXT: .cfi_def_cfa_offset 128
-; AVX1-NEXT: .Ltmp37:
-; AVX1-NEXT: .cfi_offset %rbx, -40
-; AVX1-NEXT: .Ltmp38:
-; AVX1-NEXT: .cfi_offset %r14, -32
-; AVX1-NEXT: .Ltmp39:
-; AVX1-NEXT: .cfi_offset %r15, -24
-; AVX1-NEXT: .Ltmp40:
-; AVX1-NEXT: .cfi_offset %rbp, -16
; AVX1-NEXT: movq %rdi, %rbx
; AVX1-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
@@ -3180,28 +5614,10 @@ define void @store_cvt_4f64_to_4i16(<4 x
; AVX2-LABEL: store_cvt_4f64_to_4i16:
; AVX2: # BB#0:
; AVX2-NEXT: pushq %rbp
-; AVX2-NEXT: .Ltmp32:
-; AVX2-NEXT: .cfi_def_cfa_offset 16
; AVX2-NEXT: pushq %r15
-; AVX2-NEXT: .Ltmp33:
-; AVX2-NEXT: .cfi_def_cfa_offset 24
; AVX2-NEXT: pushq %r14
-; AVX2-NEXT: .Ltmp34:
-; AVX2-NEXT: .cfi_def_cfa_offset 32
; AVX2-NEXT: pushq %rbx
-; AVX2-NEXT: .Ltmp35:
-; AVX2-NEXT: .cfi_def_cfa_offset 40
; AVX2-NEXT: subq $88, %rsp
-; AVX2-NEXT: .Ltmp36:
-; AVX2-NEXT: .cfi_def_cfa_offset 128
-; AVX2-NEXT: .Ltmp37:
-; AVX2-NEXT: .cfi_offset %rbx, -40
-; AVX2-NEXT: .Ltmp38:
-; AVX2-NEXT: .cfi_offset %r14, -32
-; AVX2-NEXT: .Ltmp39:
-; AVX2-NEXT: .cfi_offset %r15, -24
-; AVX2-NEXT: .Ltmp40:
-; AVX2-NEXT: .cfi_offset %rbp, -16
; AVX2-NEXT: movq %rdi, %rbx
; AVX2-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
@@ -3233,85 +5649,88 @@ define void @store_cvt_4f64_to_4i16(<4 x
; AVX2-NEXT: popq %rbp
; AVX2-NEXT: retq
;
-; AVX512-LABEL: store_cvt_4f64_to_4i16:
-; AVX512: # BB#0:
-; AVX512-NEXT: pushq %rbp
-; AVX512-NEXT: .Ltmp32:
-; AVX512-NEXT: .cfi_def_cfa_offset 16
-; AVX512-NEXT: pushq %r15
-; AVX512-NEXT: .Ltmp33:
-; AVX512-NEXT: .cfi_def_cfa_offset 24
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: .Ltmp34:
-; AVX512-NEXT: .cfi_def_cfa_offset 32
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: .Ltmp35:
-; AVX512-NEXT: .cfi_def_cfa_offset 40
-; AVX512-NEXT: subq $88, %rsp
-; AVX512-NEXT: .Ltmp36:
-; AVX512-NEXT: .cfi_def_cfa_offset 128
-; AVX512-NEXT: .Ltmp37:
-; AVX512-NEXT: .cfi_offset %rbx, -40
-; AVX512-NEXT: .Ltmp38:
-; AVX512-NEXT: .cfi_offset %r14, -32
-; AVX512-NEXT: .Ltmp39:
-; AVX512-NEXT: .cfi_offset %r15, -24
-; AVX512-NEXT: .Ltmp40:
-; AVX512-NEXT: .cfi_offset %rbp, -16
-; AVX512-NEXT: movq %rdi, %rbx
-; AVX512-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movl %eax, %r14d
-; AVX512-NEXT: vmovupd {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovapd %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movl %eax, %r15d
-; AVX512-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movl %eax, %ebp
-; AVX512-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, 4(%rbx)
-; AVX512-NEXT: movw %bp, (%rbx)
-; AVX512-NEXT: movw %r15w, 6(%rbx)
-; AVX512-NEXT: movw %r14w, 2(%rbx)
-; AVX512-NEXT: addq $88, %rsp
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: popq %r15
-; AVX512-NEXT: popq %rbp
-; AVX512-NEXT: retq
+; AVX512F-LABEL: store_cvt_4f64_to_4i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: pushq %rbp
+; AVX512F-NEXT: pushq %r15
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: subq $88, %rsp
+; AVX512F-NEXT: movq %rdi, %rbx
+; AVX512F-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movl %eax, %r14d
+; AVX512F-NEXT: vmovupd {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovapd %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movl %eax, %r15d
+; AVX512F-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movl %eax, %ebp
+; AVX512F-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, 4(%rbx)
+; AVX512F-NEXT: movw %bp, (%rbx)
+; AVX512F-NEXT: movw %r15w, 6(%rbx)
+; AVX512F-NEXT: movw %r14w, 2(%rbx)
+; AVX512F-NEXT: addq $88, %rsp
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: popq %r15
+; AVX512F-NEXT: popq %rbp
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_4f64_to_4i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $88, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: vmovdqu %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movl %eax, %r14d
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movl %eax, %r15d
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movl %eax, %ebp
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, 4(%rbx)
+; AVX512VL-NEXT: movw %bp, (%rbx)
+; AVX512VL-NEXT: movw %r15w, 6(%rbx)
+; AVX512VL-NEXT: movw %r14w, 2(%rbx)
+; AVX512VL-NEXT: addq $88, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x double> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
store <4 x i16> %2, <4 x i16>* %a1
ret void
}
-define void @store_cvt_4f64_to_8i16_undef(<4 x double> %a0, <8 x i16>* %a1) {
+define void @store_cvt_4f64_to_8i16_undef(<4 x double> %a0, <8 x i16>* %a1) nounwind {
; AVX1-LABEL: store_cvt_4f64_to_8i16_undef:
; AVX1: # BB#0:
; AVX1-NEXT: pushq %rbp
-; AVX1-NEXT: .Ltmp41:
-; AVX1-NEXT: .cfi_def_cfa_offset 16
; AVX1-NEXT: pushq %r14
-; AVX1-NEXT: .Ltmp42:
-; AVX1-NEXT: .cfi_def_cfa_offset 24
; AVX1-NEXT: pushq %rbx
-; AVX1-NEXT: .Ltmp43:
-; AVX1-NEXT: .cfi_def_cfa_offset 32
; AVX1-NEXT: subq $32, %rsp
-; AVX1-NEXT: .Ltmp44:
-; AVX1-NEXT: .cfi_def_cfa_offset 64
-; AVX1-NEXT: .Ltmp45:
-; AVX1-NEXT: .cfi_offset %rbx, -32
-; AVX1-NEXT: .Ltmp46:
-; AVX1-NEXT: .cfi_offset %r14, -24
-; AVX1-NEXT: .Ltmp47:
-; AVX1-NEXT: .cfi_offset %rbp, -16
; AVX1-NEXT: movq %rdi, %r14
; AVX1-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
@@ -3351,23 +5770,9 @@ define void @store_cvt_4f64_to_8i16_unde
; AVX2-LABEL: store_cvt_4f64_to_8i16_undef:
; AVX2: # BB#0:
; AVX2-NEXT: pushq %rbp
-; AVX2-NEXT: .Ltmp41:
-; AVX2-NEXT: .cfi_def_cfa_offset 16
; AVX2-NEXT: pushq %r14
-; AVX2-NEXT: .Ltmp42:
-; AVX2-NEXT: .cfi_def_cfa_offset 24
; AVX2-NEXT: pushq %rbx
-; AVX2-NEXT: .Ltmp43:
-; AVX2-NEXT: .cfi_def_cfa_offset 32
; AVX2-NEXT: subq $32, %rsp
-; AVX2-NEXT: .Ltmp44:
-; AVX2-NEXT: .cfi_def_cfa_offset 64
-; AVX2-NEXT: .Ltmp45:
-; AVX2-NEXT: .cfi_offset %rbx, -32
-; AVX2-NEXT: .Ltmp46:
-; AVX2-NEXT: .cfi_offset %r14, -24
-; AVX2-NEXT: .Ltmp47:
-; AVX2-NEXT: .cfi_offset %rbp, -16
; AVX2-NEXT: movq %rdi, %r14
; AVX2-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
@@ -3404,58 +5809,85 @@ define void @store_cvt_4f64_to_8i16_unde
; AVX2-NEXT: popq %rbp
; AVX2-NEXT: retq
;
-; AVX512-LABEL: store_cvt_4f64_to_8i16_undef:
-; AVX512: # BB#0:
-; AVX512-NEXT: pushq %rbp
-; AVX512-NEXT: .Ltmp41:
-; AVX512-NEXT: .cfi_def_cfa_offset 16
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: .Ltmp42:
-; AVX512-NEXT: .cfi_def_cfa_offset 24
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: .Ltmp43:
-; AVX512-NEXT: .cfi_def_cfa_offset 32
-; AVX512-NEXT: subq $32, %rsp
-; AVX512-NEXT: .Ltmp44:
-; AVX512-NEXT: .cfi_def_cfa_offset 64
-; AVX512-NEXT: .Ltmp45:
-; AVX512-NEXT: .cfi_offset %rbx, -32
-; AVX512-NEXT: .Ltmp46:
-; AVX512-NEXT: .cfi_offset %r14, -24
-; AVX512-NEXT: .Ltmp47:
-; AVX512-NEXT: .cfi_offset %rbp, -16
-; AVX512-NEXT: movq %rdi, %r14
-; AVX512-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bp
-; AVX512-NEXT: shll $16, %ebp
-; AVX512-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %ebx
-; AVX512-NEXT: orl %ebp, %ebx
-; AVX512-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bp
-; AVX512-NEXT: shll $16, %ebp
-; AVX512-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %eax
-; AVX512-NEXT: orl %ebp, %eax
-; AVX512-NEXT: shlq $32, %rax
-; AVX512-NEXT: orq %rbx, %rax
-; AVX512-NEXT: vmovq %rax, %xmm0
-; AVX512-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX512-NEXT: vmovdqa %xmm0, (%r14)
-; AVX512-NEXT: addq $32, %rsp
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: popq %rbp
-; AVX512-NEXT: retq
+; AVX512F-LABEL: store_cvt_4f64_to_8i16_undef:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: pushq %rbp
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: subq $32, %rsp
+; AVX512F-NEXT: movq %rdi, %r14
+; AVX512F-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bp
+; AVX512F-NEXT: shll $16, %ebp
+; AVX512F-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %ebx
+; AVX512F-NEXT: orl %ebp, %ebx
+; AVX512F-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bp
+; AVX512F-NEXT: shll $16, %ebp
+; AVX512F-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %eax
+; AVX512F-NEXT: orl %ebp, %eax
+; AVX512F-NEXT: shlq $32, %rax
+; AVX512F-NEXT: orq %rbx, %rax
+; AVX512F-NEXT: vmovq %rax, %xmm0
+; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX512F-NEXT: vmovdqa %xmm0, (%r14)
+; AVX512F-NEXT: addq $32, %rsp
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: popq %rbp
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_4f64_to_8i16_undef:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $32, %rsp
+; AVX512VL-NEXT: movq %rdi, %r14
+; AVX512VL-NEXT: vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bp
+; AVX512VL-NEXT: shll $16, %ebp
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %ebx
+; AVX512VL-NEXT: orl %ebp, %ebx
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bp
+; AVX512VL-NEXT: shll $16, %ebp
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebp, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %rbx, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512VL-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512VL-NEXT: vmovdqa32 %xmm0, (%r14)
+; AVX512VL-NEXT: addq $32, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x double> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
%3 = shufflevector <4 x i16> %2, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
@@ -3463,27 +5895,13 @@ define void @store_cvt_4f64_to_8i16_unde
ret void
}
-define void @store_cvt_4f64_to_8i16_zero(<4 x double> %a0, <8 x i16>* %a1) {
+define void @store_cvt_4f64_to_8i16_zero(<4 x double> %a0, <8 x i16>* %a1) nounwind {
; AVX1-LABEL: store_cvt_4f64_to_8i16_zero:
; AVX1: # BB#0:
; AVX1-NEXT: pushq %rbp
-; AVX1-NEXT: .Ltmp48:
-; AVX1-NEXT: .cfi_def_cfa_offset 16
; AVX1-NEXT: pushq %r14
-; AVX1-NEXT: .Ltmp49:
-; AVX1-NEXT: .cfi_def_cfa_offset 24
; AVX1-NEXT: pushq %rbx
-; AVX1-NEXT: .Ltmp50:
-; AVX1-NEXT: .cfi_def_cfa_offset 32
; AVX1-NEXT: subq $32, %rsp
-; AVX1-NEXT: .Ltmp51:
-; AVX1-NEXT: .cfi_def_cfa_offset 64
-; AVX1-NEXT: .Ltmp52:
-; AVX1-NEXT: .cfi_offset %rbx, -32
-; AVX1-NEXT: .Ltmp53:
-; AVX1-NEXT: .cfi_offset %r14, -24
-; AVX1-NEXT: .Ltmp54:
-; AVX1-NEXT: .cfi_offset %rbp, -16
; AVX1-NEXT: movq %rdi, %r14
; AVX1-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
@@ -3523,23 +5941,9 @@ define void @store_cvt_4f64_to_8i16_zero
; AVX2-LABEL: store_cvt_4f64_to_8i16_zero:
; AVX2: # BB#0:
; AVX2-NEXT: pushq %rbp
-; AVX2-NEXT: .Ltmp48:
-; AVX2-NEXT: .cfi_def_cfa_offset 16
; AVX2-NEXT: pushq %r14
-; AVX2-NEXT: .Ltmp49:
-; AVX2-NEXT: .cfi_def_cfa_offset 24
; AVX2-NEXT: pushq %rbx
-; AVX2-NEXT: .Ltmp50:
-; AVX2-NEXT: .cfi_def_cfa_offset 32
; AVX2-NEXT: subq $32, %rsp
-; AVX2-NEXT: .Ltmp51:
-; AVX2-NEXT: .cfi_def_cfa_offset 64
-; AVX2-NEXT: .Ltmp52:
-; AVX2-NEXT: .cfi_offset %rbx, -32
-; AVX2-NEXT: .Ltmp53:
-; AVX2-NEXT: .cfi_offset %r14, -24
-; AVX2-NEXT: .Ltmp54:
-; AVX2-NEXT: .cfi_offset %rbp, -16
; AVX2-NEXT: movq %rdi, %r14
; AVX2-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX2-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
@@ -3576,58 +5980,87 @@ define void @store_cvt_4f64_to_8i16_zero
; AVX2-NEXT: popq %rbp
; AVX2-NEXT: retq
;
-; AVX512-LABEL: store_cvt_4f64_to_8i16_zero:
-; AVX512: # BB#0:
-; AVX512-NEXT: pushq %rbp
-; AVX512-NEXT: .Ltmp48:
-; AVX512-NEXT: .cfi_def_cfa_offset 16
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: .Ltmp49:
-; AVX512-NEXT: .cfi_def_cfa_offset 24
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: .Ltmp50:
-; AVX512-NEXT: .cfi_def_cfa_offset 32
-; AVX512-NEXT: subq $32, %rsp
-; AVX512-NEXT: .Ltmp51:
-; AVX512-NEXT: .cfi_def_cfa_offset 64
-; AVX512-NEXT: .Ltmp52:
-; AVX512-NEXT: .cfi_offset %rbx, -32
-; AVX512-NEXT: .Ltmp53:
-; AVX512-NEXT: .cfi_offset %r14, -24
-; AVX512-NEXT: .Ltmp54:
-; AVX512-NEXT: .cfi_offset %rbp, -16
-; AVX512-NEXT: movq %rdi, %r14
-; AVX512-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bp
-; AVX512-NEXT: shll $16, %ebp
-; AVX512-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %ebx
-; AVX512-NEXT: orl %ebp, %ebx
-; AVX512-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, %bp
-; AVX512-NEXT: shll $16, %ebp
-; AVX512-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movzwl %ax, %eax
-; AVX512-NEXT: orl %ebp, %eax
-; AVX512-NEXT: shlq $32, %rax
-; AVX512-NEXT: orq %rbx, %rax
-; AVX512-NEXT: vmovq %rax, %xmm0
-; AVX512-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
-; AVX512-NEXT: vmovdqa %xmm0, (%r14)
-; AVX512-NEXT: addq $32, %rsp
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: popq %rbp
-; AVX512-NEXT: retq
+; AVX512F-LABEL: store_cvt_4f64_to_8i16_zero:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: pushq %rbp
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: subq $32, %rsp
+; AVX512F-NEXT: movq %rdi, %r14
+; AVX512F-NEXT: vmovupd %ymm0, (%rsp) # 32-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bp
+; AVX512F-NEXT: shll $16, %ebp
+; AVX512F-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %ebx
+; AVX512F-NEXT: orl %ebp, %ebx
+; AVX512F-NEXT: vmovupd (%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovapd %xmm0, (%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, %bp
+; AVX512F-NEXT: shll $16, %ebp
+; AVX512F-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movzwl %ax, %eax
+; AVX512F-NEXT: orl %ebp, %eax
+; AVX512F-NEXT: shlq $32, %rax
+; AVX512F-NEXT: orq %rbx, %rax
+; AVX512F-NEXT: vmovq %rax, %xmm0
+; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT: vmovdqa %xmm0, (%r14)
+; AVX512F-NEXT: addq $32, %rsp
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: popq %rbp
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_4f64_to_8i16_zero:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $32, %rsp
+; AVX512VL-NEXT: movq %rdi, %r14
+; AVX512VL-NEXT: vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bp
+; AVX512VL-NEXT: shll $16, %ebp
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %ebx
+; AVX512VL-NEXT: orl %ebp, %ebx
+; AVX512VL-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, %bp
+; AVX512VL-NEXT: shll $16, %ebp
+; AVX512VL-NEXT: vmovaps (%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movzwl %ax, %eax
+; AVX512VL-NEXT: orl %ebp, %eax
+; AVX512VL-NEXT: shlq $32, %rax
+; AVX512VL-NEXT: orq %rbx, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm0
+; AVX512VL-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX512VL-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,2]
+; AVX512VL-NEXT: vpxord %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX512VL-NEXT: vmovdqa %xmm0, (%r14)
+; AVX512VL-NEXT: addq $32, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <4 x double> %a0 to <4 x half>
%2 = bitcast <4 x half> %1 to <4 x i16>
%3 = shufflevector <4 x i16> %2, <4 x i16> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
@@ -3635,42 +6068,16 @@ define void @store_cvt_4f64_to_8i16_zero
ret void
}
-define void @store_cvt_8f64_to_8i16(<8 x double> %a0, <8 x i16>* %a1) {
+define void @store_cvt_8f64_to_8i16(<8 x double> %a0, <8 x i16>* %a1) nounwind {
; AVX1-LABEL: store_cvt_8f64_to_8i16:
; AVX1: # BB#0:
; AVX1-NEXT: pushq %rbp
-; AVX1-NEXT: .Ltmp55:
-; AVX1-NEXT: .cfi_def_cfa_offset 16
; AVX1-NEXT: pushq %r15
-; AVX1-NEXT: .Ltmp56:
-; AVX1-NEXT: .cfi_def_cfa_offset 24
; AVX1-NEXT: pushq %r14
-; AVX1-NEXT: .Ltmp57:
-; AVX1-NEXT: .cfi_def_cfa_offset 32
; AVX1-NEXT: pushq %r13
-; AVX1-NEXT: .Ltmp58:
-; AVX1-NEXT: .cfi_def_cfa_offset 40
; AVX1-NEXT: pushq %r12
-; AVX1-NEXT: .Ltmp59:
-; AVX1-NEXT: .cfi_def_cfa_offset 48
; AVX1-NEXT: pushq %rbx
-; AVX1-NEXT: .Ltmp60:
-; AVX1-NEXT: .cfi_def_cfa_offset 56
; AVX1-NEXT: subq $136, %rsp
-; AVX1-NEXT: .Ltmp61:
-; AVX1-NEXT: .cfi_def_cfa_offset 192
-; AVX1-NEXT: .Ltmp62:
-; AVX1-NEXT: .cfi_offset %rbx, -56
-; AVX1-NEXT: .Ltmp63:
-; AVX1-NEXT: .cfi_offset %r12, -48
-; AVX1-NEXT: .Ltmp64:
-; AVX1-NEXT: .cfi_offset %r13, -40
-; AVX1-NEXT: .Ltmp65:
-; AVX1-NEXT: .cfi_offset %r14, -32
-; AVX1-NEXT: .Ltmp66:
-; AVX1-NEXT: .cfi_offset %r15, -24
-; AVX1-NEXT: .Ltmp67:
-; AVX1-NEXT: .cfi_offset %rbp, -16
; AVX1-NEXT: movq %rdi, %rbx
; AVX1-NEXT: vmovups %ymm1, {{[0-9]+}}(%rsp) # 32-byte Spill
; AVX1-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
@@ -3734,38 +6141,12 @@ define void @store_cvt_8f64_to_8i16(<8 x
; AVX2-LABEL: store_cvt_8f64_to_8i16:
; AVX2: # BB#0:
; AVX2-NEXT: pushq %rbp
-; AVX2-NEXT: .Ltmp55:
-; AVX2-NEXT: .cfi_def_cfa_offset 16
; AVX2-NEXT: pushq %r15
-; AVX2-NEXT: .Ltmp56:
-; AVX2-NEXT: .cfi_def_cfa_offset 24
; AVX2-NEXT: pushq %r14
-; AVX2-NEXT: .Ltmp57:
-; AVX2-NEXT: .cfi_def_cfa_offset 32
; AVX2-NEXT: pushq %r13
-; AVX2-NEXT: .Ltmp58:
-; AVX2-NEXT: .cfi_def_cfa_offset 40
; AVX2-NEXT: pushq %r12
-; AVX2-NEXT: .Ltmp59:
-; AVX2-NEXT: .cfi_def_cfa_offset 48
; AVX2-NEXT: pushq %rbx
-; AVX2-NEXT: .Ltmp60:
-; AVX2-NEXT: .cfi_def_cfa_offset 56
; AVX2-NEXT: subq $136, %rsp
-; AVX2-NEXT: .Ltmp61:
-; AVX2-NEXT: .cfi_def_cfa_offset 192
-; AVX2-NEXT: .Ltmp62:
-; AVX2-NEXT: .cfi_offset %rbx, -56
-; AVX2-NEXT: .Ltmp63:
-; AVX2-NEXT: .cfi_offset %r12, -48
-; AVX2-NEXT: .Ltmp64:
-; AVX2-NEXT: .cfi_offset %r13, -40
-; AVX2-NEXT: .Ltmp65:
-; AVX2-NEXT: .cfi_offset %r14, -32
-; AVX2-NEXT: .Ltmp66:
-; AVX2-NEXT: .cfi_offset %r15, -24
-; AVX2-NEXT: .Ltmp67:
-; AVX2-NEXT: .cfi_offset %rbp, -16
; AVX2-NEXT: movq %rdi, %rbx
; AVX2-NEXT: vmovups %ymm1, {{[0-9]+}}(%rsp) # 32-byte Spill
; AVX2-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
@@ -3826,95 +6207,133 @@ define void @store_cvt_8f64_to_8i16(<8 x
; AVX2-NEXT: popq %rbp
; AVX2-NEXT: retq
;
-; AVX512-LABEL: store_cvt_8f64_to_8i16:
-; AVX512: # BB#0:
-; AVX512-NEXT: pushq %rbp
-; AVX512-NEXT: .Ltmp55:
-; AVX512-NEXT: .cfi_def_cfa_offset 16
-; AVX512-NEXT: pushq %r15
-; AVX512-NEXT: .Ltmp56:
-; AVX512-NEXT: .cfi_def_cfa_offset 24
-; AVX512-NEXT: pushq %r14
-; AVX512-NEXT: .Ltmp57:
-; AVX512-NEXT: .cfi_def_cfa_offset 32
-; AVX512-NEXT: pushq %r13
-; AVX512-NEXT: .Ltmp58:
-; AVX512-NEXT: .cfi_def_cfa_offset 40
-; AVX512-NEXT: pushq %r12
-; AVX512-NEXT: .Ltmp59:
-; AVX512-NEXT: .cfi_def_cfa_offset 48
-; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: .Ltmp60:
-; AVX512-NEXT: .cfi_def_cfa_offset 56
-; AVX512-NEXT: subq $200, %rsp
-; AVX512-NEXT: .Ltmp61:
-; AVX512-NEXT: .cfi_def_cfa_offset 256
-; AVX512-NEXT: .Ltmp62:
-; AVX512-NEXT: .cfi_offset %rbx, -56
-; AVX512-NEXT: .Ltmp63:
-; AVX512-NEXT: .cfi_offset %r12, -48
-; AVX512-NEXT: .Ltmp64:
-; AVX512-NEXT: .cfi_offset %r13, -40
-; AVX512-NEXT: .Ltmp65:
-; AVX512-NEXT: .cfi_offset %r14, -32
-; AVX512-NEXT: .Ltmp66:
-; AVX512-NEXT: .cfi_offset %r15, -24
-; AVX512-NEXT: .Ltmp67:
-; AVX512-NEXT: .cfi_offset %rbp, -16
-; AVX512-NEXT: movq %rdi, %rbx
-; AVX512-NEXT: vmovupd %zmm0, {{[0-9]+}}(%rsp) # 64-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
-; AVX512-NEXT: vmovupd {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovapd %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
-; AVX512-NEXT: vmovupd {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
-; AVX512-NEXT: vextractf64x4 $1, %zmm0, %ymm0
-; AVX512-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movl %eax, %r12d
-; AVX512-NEXT: vmovupd {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vmovapd %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
-; AVX512-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movl %eax, %r13d
-; AVX512-NEXT: vmovups {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
-; AVX512-NEXT: # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movl %eax, %ebp
-; AVX512-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movl %eax, %r14d
-; AVX512-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
-; AVX512-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movl %eax, %r15d
-; AVX512-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
-; AVX512-NEXT: callq __truncdfhf2
-; AVX512-NEXT: movw %ax, 12(%rbx)
-; AVX512-NEXT: movw %r15w, 8(%rbx)
-; AVX512-NEXT: movw %r14w, 4(%rbx)
-; AVX512-NEXT: movw %bp, (%rbx)
-; AVX512-NEXT: movw %r13w, 14(%rbx)
-; AVX512-NEXT: movw %r12w, 10(%rbx)
-; AVX512-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
-; AVX512-NEXT: movw %ax, 6(%rbx)
-; AVX512-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
-; AVX512-NEXT: movw %ax, 2(%rbx)
-; AVX512-NEXT: addq $200, %rsp
-; AVX512-NEXT: popq %rbx
-; AVX512-NEXT: popq %r12
-; AVX512-NEXT: popq %r13
-; AVX512-NEXT: popq %r14
-; AVX512-NEXT: popq %r15
-; AVX512-NEXT: popq %rbp
-; AVX512-NEXT: retq
+; AVX512F-LABEL: store_cvt_8f64_to_8i16:
+; AVX512F: # BB#0:
+; AVX512F-NEXT: pushq %rbp
+; AVX512F-NEXT: pushq %r15
+; AVX512F-NEXT: pushq %r14
+; AVX512F-NEXT: pushq %r13
+; AVX512F-NEXT: pushq %r12
+; AVX512F-NEXT: pushq %rbx
+; AVX512F-NEXT: subq $200, %rsp
+; AVX512F-NEXT: movq %rdi, %rbx
+; AVX512F-NEXT: vmovupd %zmm0, {{[0-9]+}}(%rsp) # 64-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512F-NEXT: vmovupd {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovapd %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512F-NEXT: vmovupd {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512F-NEXT: vextractf64x4 $1, %zmm0, %ymm0
+; AVX512F-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movl %eax, %r12d
+; AVX512F-NEXT: vmovupd {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm0
+; AVX512F-NEXT: vmovapd %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512F-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movl %eax, %r13d
+; AVX512F-NEXT: vmovups {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512F-NEXT: # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movl %eax, %ebp
+; AVX512F-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movl %eax, %r14d
+; AVX512F-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512F-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movl %eax, %r15d
+; AVX512F-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512F-NEXT: callq __truncdfhf2
+; AVX512F-NEXT: movw %ax, 12(%rbx)
+; AVX512F-NEXT: movw %r15w, 8(%rbx)
+; AVX512F-NEXT: movw %r14w, 4(%rbx)
+; AVX512F-NEXT: movw %bp, (%rbx)
+; AVX512F-NEXT: movw %r13w, 14(%rbx)
+; AVX512F-NEXT: movw %r12w, 10(%rbx)
+; AVX512F-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512F-NEXT: movw %ax, 6(%rbx)
+; AVX512F-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512F-NEXT: movw %ax, 2(%rbx)
+; AVX512F-NEXT: addq $200, %rsp
+; AVX512F-NEXT: popq %rbx
+; AVX512F-NEXT: popq %r12
+; AVX512F-NEXT: popq %r13
+; AVX512F-NEXT: popq %r14
+; AVX512F-NEXT: popq %r15
+; AVX512F-NEXT: popq %rbp
+; AVX512F-NEXT: retq
+;
+; AVX512VL-LABEL: store_cvt_8f64_to_8i16:
+; AVX512VL: # BB#0:
+; AVX512VL-NEXT: pushq %rbp
+; AVX512VL-NEXT: pushq %r15
+; AVX512VL-NEXT: pushq %r14
+; AVX512VL-NEXT: pushq %r13
+; AVX512VL-NEXT: pushq %r12
+; AVX512VL-NEXT: pushq %rbx
+; AVX512VL-NEXT: subq $200, %rsp
+; AVX512VL-NEXT: movq %rdi, %rbx
+; AVX512VL-NEXT: vmovdqu64 %zmm0, {{[0-9]+}}(%rsp) # 64-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, {{[0-9]+}}(%rsp) # 2-byte Spill
+; AVX512VL-NEXT: vmovupd {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
+; AVX512VL-NEXT: vmovupd %ymm0, {{[0-9]+}}(%rsp) # 32-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movl %eax, %r12d
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm0
+; AVX512VL-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movl %eax, %r13d
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %zmm0 # 64-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movl %eax, %ebp
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movl %eax, %r14d
+; AVX512VL-NEXT: vmovups {{[0-9]+}}(%rsp), %ymm0 # 32-byte Reload
+; AVX512VL-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movl %eax, %r15d
+; AVX512VL-NEXT: vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; AVX512VL-NEXT: callq __truncdfhf2
+; AVX512VL-NEXT: movw %ax, 12(%rbx)
+; AVX512VL-NEXT: movw %r15w, 8(%rbx)
+; AVX512VL-NEXT: movw %r14w, 4(%rbx)
+; AVX512VL-NEXT: movw %bp, (%rbx)
+; AVX512VL-NEXT: movw %r13w, 14(%rbx)
+; AVX512VL-NEXT: movw %r12w, 10(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 6(%rbx)
+; AVX512VL-NEXT: movzwl {{[0-9]+}}(%rsp), %eax # 2-byte Folded Reload
+; AVX512VL-NEXT: movw %ax, 2(%rbx)
+; AVX512VL-NEXT: addq $200, %rsp
+; AVX512VL-NEXT: popq %rbx
+; AVX512VL-NEXT: popq %r12
+; AVX512VL-NEXT: popq %r13
+; AVX512VL-NEXT: popq %r14
+; AVX512VL-NEXT: popq %r15
+; AVX512VL-NEXT: popq %rbp
+; AVX512VL-NEXT: retq
%1 = fptrunc <8 x double> %a0 to <8 x half>
%2 = bitcast <8 x half> %1 to <8 x i16>
store <8 x i16> %2, <8 x i16>* %a1
More information about the llvm-commits
mailing list