[llvm] [X86] setoeq.ll - add f<->i64, uint and AVX1/AVX512 test coverage (PR #161197)
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 29 06:50:21 PDT 2025
https://github.com/RKSimon created https://github.com/llvm/llvm-project/pull/161197
None
>From 141bc27357accb761e5eba2143cdf28068a241dd Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev at redking.me.uk>
Date: Mon, 29 Sep 2025 14:48:37 +0100
Subject: [PATCH] [X86] setoeq.ll - add f<->i64, uint and AVX1/AVX512 test
coverage
---
llvm/test/CodeGen/X86/setoeq.ll | 566 +++++++++++++++++++++++++++++---
1 file changed, 529 insertions(+), 37 deletions(-)
diff --git a/llvm/test/CodeGen/X86/setoeq.ll b/llvm/test/CodeGen/X86/setoeq.ll
index f0addf4b64599..131e279aa645c 100644
--- a/llvm/test/CodeGen/X86/setoeq.ll
+++ b/llvm/test/CodeGen/X86/setoeq.ll
@@ -1,40 +1,532 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=i686-- -mattr=+sse2 | FileCheck %s
-
-define zeroext i8 @t(double %x) nounwind readnone {
-; CHECK-LABEL: t:
-; CHECK: # %bb.0: # %entry
-; CHECK-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
-; CHECK-NEXT: cvttpd2dq %xmm0, %xmm1
-; CHECK-NEXT: cvtdq2pd %xmm1, %xmm1
-; CHECK-NEXT: cmpeqsd %xmm0, %xmm1
-; CHECK-NEXT: movd %xmm1, %eax
-; CHECK-NEXT: andl $1, %eax
-; CHECK-NEXT: # kill: def $al killed $al killed $eax
-; CHECK-NEXT: retl
-entry:
- %0 = fptosi double %x to i32 ; <i32> [#uses=1]
- %1 = sitofp i32 %0 to double ; <double> [#uses=1]
- %2 = fcmp oeq double %1, %x ; <i1> [#uses=1]
- %retval12 = zext i1 %2 to i8 ; <i8> [#uses=1]
- ret i8 %retval12
-}
-
-define zeroext i8 @u(double %x) nounwind readnone {
-; CHECK-LABEL: u:
-; CHECK: # %bb.0: # %entry
-; CHECK-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
-; CHECK-NEXT: cvttpd2dq %xmm0, %xmm1
-; CHECK-NEXT: cvtdq2pd %xmm1, %xmm1
-; CHECK-NEXT: cmpneqsd %xmm0, %xmm1
-; CHECK-NEXT: movd %xmm1, %eax
-; CHECK-NEXT: andl $1, %eax
-; CHECK-NEXT: # kill: def $al killed $al killed $eax
-; CHECK-NEXT: retl
-entry:
- %0 = fptosi double %x to i32 ; <i32> [#uses=1]
- %1 = sitofp i32 %0 to double ; <double> [#uses=1]
- %2 = fcmp une double %1, %x ; <i1> [#uses=1]
- %retval12 = zext i1 %2 to i8 ; <i8> [#uses=1]
+; RUN: llc < %s -mtriple=i686-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE
+; RUN: llc < %s -mtriple=i686-- -mattr=+avx | FileCheck %s --check-prefixes=AVX
+; RUN: llc < %s -mtriple=i686-- -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX512
+
+define zeroext i8 @oeq_f64_i32(double %x) nounwind readnone {
+; SSE-LABEL: oeq_f64_i32:
+; SSE: # %bb.0: # %entry
+; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE-NEXT: cvttpd2dq %xmm0, %xmm1
+; SSE-NEXT: cvtdq2pd %xmm1, %xmm1
+; SSE-NEXT: cmpeqsd %xmm0, %xmm1
+; SSE-NEXT: movd %xmm1, %eax
+; SSE-NEXT: andl $1, %eax
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: retl
+;
+; AVX-LABEL: oeq_f64_i32:
+; AVX: # %bb.0: # %entry
+; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT: vcvttpd2dq %xmm0, %xmm1
+; AVX-NEXT: vcvtdq2pd %xmm1, %xmm1
+; AVX-NEXT: vcmpeqsd %xmm0, %xmm1, %xmm0
+; AVX-NEXT: vmovd %xmm0, %eax
+; AVX-NEXT: andl $1, %eax
+; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: retl
+;
+; AVX512-LABEL: oeq_f64_i32:
+; AVX512: # %bb.0: # %entry
+; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX512-NEXT: vcvttpd2dq %xmm0, %xmm1
+; AVX512-NEXT: vcvtdq2pd %xmm1, %xmm1
+; AVX512-NEXT: vcmpeqsd %xmm0, %xmm1, %k0
+; AVX512-NEXT: kmovd %k0, %eax
+; AVX512-NEXT: # kill: def $al killed $al killed $eax
+; AVX512-NEXT: retl
+entry:
+ %0 = fptosi double %x to i32
+ %1 = sitofp i32 %0 to double
+ %2 = fcmp oeq double %1, %x
+ %retval12 = zext i1 %2 to i8
+ ret i8 %retval12
+}
+
+define zeroext i8 @oeq_f64_u32(double %x) nounwind readnone {
+; SSE-LABEL: oeq_f64_u32:
+; SSE: # %bb.0: # %entry
+; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE-NEXT: cvttsd2si %xmm0, %eax
+; SSE-NEXT: movl %eax, %ecx
+; SSE-NEXT: sarl $31, %ecx
+; SSE-NEXT: movapd %xmm0, %xmm1
+; SSE-NEXT: subsd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
+; SSE-NEXT: cvttsd2si %xmm1, %edx
+; SSE-NEXT: andl %ecx, %edx
+; SSE-NEXT: orl %eax, %edx
+; SSE-NEXT: movd %edx, %xmm1
+; SSE-NEXT: por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
+; SSE-NEXT: subsd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
+; SSE-NEXT: cmpeqsd %xmm0, %xmm1
+; SSE-NEXT: movd %xmm1, %eax
+; SSE-NEXT: andl $1, %eax
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: retl
+;
+; AVX-LABEL: oeq_f64_u32:
+; AVX: # %bb.0: # %entry
+; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT: vcvttsd2si %xmm0, %eax
+; AVX-NEXT: movl %eax, %ecx
+; AVX-NEXT: sarl $31, %ecx
+; AVX-NEXT: vsubsd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm1
+; AVX-NEXT: vcvttsd2si %xmm1, %edx
+; AVX-NEXT: andl %ecx, %edx
+; AVX-NEXT: orl %eax, %edx
+; AVX-NEXT: vmovd %edx, %xmm1
+; AVX-NEXT: vpor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm1
+; AVX-NEXT: vsubsd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm1
+; AVX-NEXT: vcmpeqsd %xmm0, %xmm1, %xmm0
+; AVX-NEXT: vmovd %xmm0, %eax
+; AVX-NEXT: andl $1, %eax
+; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: retl
+;
+; AVX512-LABEL: oeq_f64_u32:
+; AVX512: # %bb.0: # %entry
+; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX512-NEXT: vcvttsd2usi %xmm0, %eax
+; AVX512-NEXT: vcvtusi2sd %eax, %xmm7, %xmm1
+; AVX512-NEXT: vcmpeqsd %xmm0, %xmm1, %k0
+; AVX512-NEXT: kmovd %k0, %eax
+; AVX512-NEXT: # kill: def $al killed $al killed $eax
+; AVX512-NEXT: retl
+entry:
+ %0 = fptoui double %x to i32
+ %1 = uitofp i32 %0 to double
+ %2 = fcmp oeq double %1, %x
+ %retval12 = zext i1 %2 to i8
+ ret i8 %retval12
+}
+
+define zeroext i8 @oeq_f64_i64(double %x) nounwind readnone {
+; SSE-LABEL: oeq_f64_i64:
+; SSE: # %bb.0: # %entry
+; SSE-NEXT: pushl %ebp
+; SSE-NEXT: movl %esp, %ebp
+; SSE-NEXT: andl $-8, %esp
+; SSE-NEXT: subl $32, %esp
+; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE-NEXT: movsd %xmm0, {{[0-9]+}}(%esp)
+; SSE-NEXT: fldl {{[0-9]+}}(%esp)
+; SSE-NEXT: fnstcw {{[0-9]+}}(%esp)
+; SSE-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; SSE-NEXT: orl $3072, %eax # imm = 0xC00
+; SSE-NEXT: movw %ax, {{[0-9]+}}(%esp)
+; SSE-NEXT: fldcw {{[0-9]+}}(%esp)
+; SSE-NEXT: fistpll {{[0-9]+}}(%esp)
+; SSE-NEXT: fldcw {{[0-9]+}}(%esp)
+; SSE-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
+; SSE-NEXT: movlps %xmm1, {{[0-9]+}}(%esp)
+; SSE-NEXT: fildll {{[0-9]+}}(%esp)
+; SSE-NEXT: fstpl {{[0-9]+}}(%esp)
+; SSE-NEXT: cmpeqsd {{[0-9]+}}(%esp), %xmm0
+; SSE-NEXT: movd %xmm0, %eax
+; SSE-NEXT: andl $1, %eax
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: movl %ebp, %esp
+; SSE-NEXT: popl %ebp
+; SSE-NEXT: retl
+;
+; AVX-LABEL: oeq_f64_i64:
+; AVX: # %bb.0: # %entry
+; AVX-NEXT: pushl %ebp
+; AVX-NEXT: movl %esp, %ebp
+; AVX-NEXT: andl $-8, %esp
+; AVX-NEXT: subl $24, %esp
+; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT: vmovsd %xmm0, (%esp)
+; AVX-NEXT: fldl (%esp)
+; AVX-NEXT: fisttpll (%esp)
+; AVX-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX-NEXT: vmovlps %xmm1, {{[0-9]+}}(%esp)
+; AVX-NEXT: fildll {{[0-9]+}}(%esp)
+; AVX-NEXT: fstpl {{[0-9]+}}(%esp)
+; AVX-NEXT: vcmpeqsd {{[0-9]+}}(%esp), %xmm0, %xmm0
+; AVX-NEXT: vmovd %xmm0, %eax
+; AVX-NEXT: andl $1, %eax
+; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: movl %ebp, %esp
+; AVX-NEXT: popl %ebp
+; AVX-NEXT: retl
+;
+; AVX512-LABEL: oeq_f64_i64:
+; AVX512: # %bb.0: # %entry
+; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX512-NEXT: vcvttpd2qq %xmm0, %xmm1
+; AVX512-NEXT: vcvtqq2pd %ymm1, %ymm1
+; AVX512-NEXT: vcmpeqsd %xmm0, %xmm1, %k0
+; AVX512-NEXT: kmovd %k0, %eax
+; AVX512-NEXT: # kill: def $al killed $al killed $eax
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retl
+entry:
+ %0 = fptosi double %x to i64
+ %1 = sitofp i64 %0 to double
+ %2 = fcmp oeq double %1, %x
+ %retval12 = zext i1 %2 to i8
+ ret i8 %retval12
+}
+
+define zeroext i8 @oeq_f64_u64(double %x) nounwind readnone {
+; SSE-LABEL: oeq_f64_u64:
+; SSE: # %bb.0: # %entry
+; SSE-NEXT: pushl %ebp
+; SSE-NEXT: movl %esp, %ebp
+; SSE-NEXT: andl $-8, %esp
+; SSE-NEXT: subl $16, %esp
+; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE-NEXT: movsd {{.*#+}} xmm1 = [9.2233720368547758E+18,0.0E+0]
+; SSE-NEXT: ucomisd %xmm0, %xmm1
+; SSE-NEXT: jbe .LBB3_2
+; SSE-NEXT: # %bb.1: # %entry
+; SSE-NEXT: xorpd %xmm1, %xmm1
+; SSE-NEXT: .LBB3_2: # %entry
+; SSE-NEXT: movapd %xmm0, %xmm2
+; SSE-NEXT: subsd %xmm1, %xmm2
+; SSE-NEXT: movsd %xmm2, {{[0-9]+}}(%esp)
+; SSE-NEXT: setbe %al
+; SSE-NEXT: fldl {{[0-9]+}}(%esp)
+; SSE-NEXT: fnstcw {{[0-9]+}}(%esp)
+; SSE-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
+; SSE-NEXT: orl $3072, %ecx # imm = 0xC00
+; SSE-NEXT: movw %cx, {{[0-9]+}}(%esp)
+; SSE-NEXT: fldcw {{[0-9]+}}(%esp)
+; SSE-NEXT: fistpll {{[0-9]+}}(%esp)
+; SSE-NEXT: fldcw {{[0-9]+}}(%esp)
+; SSE-NEXT: movzbl %al, %eax
+; SSE-NEXT: shll $31, %eax
+; SSE-NEXT: xorl {{[0-9]+}}(%esp), %eax
+; SSE-NEXT: movd %eax, %xmm1
+; SSE-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
+; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[1],mem[1]
+; SSE-NEXT: subpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm2
+; SSE-NEXT: movapd %xmm2, %xmm1
+; SSE-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm2[1]
+; SSE-NEXT: addsd %xmm2, %xmm1
+; SSE-NEXT: cmpeqsd %xmm0, %xmm1
+; SSE-NEXT: movd %xmm1, %eax
+; SSE-NEXT: andl $1, %eax
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: movl %ebp, %esp
+; SSE-NEXT: popl %ebp
+; SSE-NEXT: retl
+;
+; AVX-LABEL: oeq_f64_u64:
+; AVX: # %bb.0: # %entry
+; AVX-NEXT: pushl %ebp
+; AVX-NEXT: movl %esp, %ebp
+; AVX-NEXT: andl $-8, %esp
+; AVX-NEXT: subl $8, %esp
+; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT: vmovsd {{.*#+}} xmm1 = [9.2233720368547758E+18,0.0E+0]
+; AVX-NEXT: vucomisd %xmm0, %xmm1
+; AVX-NEXT: jbe .LBB3_2
+; AVX-NEXT: # %bb.1: # %entry
+; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
+; AVX-NEXT: .LBB3_2: # %entry
+; AVX-NEXT: vsubsd %xmm1, %xmm0, %xmm1
+; AVX-NEXT: vmovsd %xmm1, (%esp)
+; AVX-NEXT: fldl (%esp)
+; AVX-NEXT: fisttpll (%esp)
+; AVX-NEXT: setbe %al
+; AVX-NEXT: movzbl %al, %eax
+; AVX-NEXT: shll $31, %eax
+; AVX-NEXT: xorl {{[0-9]+}}(%esp), %eax
+; AVX-NEXT: vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX-NEXT: vpinsrd $1, %eax, %xmm1, %xmm1
+; AVX-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
+; AVX-NEXT: vsubpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm1
+; AVX-NEXT: vshufpd {{.*#+}} xmm2 = xmm1[1,0]
+; AVX-NEXT: vaddsd %xmm1, %xmm2, %xmm1
+; AVX-NEXT: vcmpeqsd %xmm0, %xmm1, %xmm0
+; AVX-NEXT: vmovd %xmm0, %eax
+; AVX-NEXT: andl $1, %eax
+; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: movl %ebp, %esp
+; AVX-NEXT: popl %ebp
+; AVX-NEXT: retl
+;
+; AVX512-LABEL: oeq_f64_u64:
+; AVX512: # %bb.0: # %entry
+; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX512-NEXT: vcvttpd2uqq %xmm0, %xmm1
+; AVX512-NEXT: vcvtuqq2pd %ymm1, %ymm1
+; AVX512-NEXT: vcmpeqsd %xmm0, %xmm1, %k0
+; AVX512-NEXT: kmovd %k0, %eax
+; AVX512-NEXT: # kill: def $al killed $al killed $eax
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retl
+entry:
+ %0 = fptoui double %x to i64
+ %1 = uitofp i64 %0 to double
+ %2 = fcmp oeq double %1, %x
+ %retval12 = zext i1 %2 to i8
+ ret i8 %retval12
+}
+
+define zeroext i8 @une_f64_i32(double %x) nounwind readnone {
+; SSE-LABEL: une_f64_i32:
+; SSE: # %bb.0: # %entry
+; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE-NEXT: cvttpd2dq %xmm0, %xmm1
+; SSE-NEXT: cvtdq2pd %xmm1, %xmm1
+; SSE-NEXT: cmpneqsd %xmm0, %xmm1
+; SSE-NEXT: movd %xmm1, %eax
+; SSE-NEXT: andl $1, %eax
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: retl
+;
+; AVX-LABEL: une_f64_i32:
+; AVX: # %bb.0: # %entry
+; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT: vcvttpd2dq %xmm0, %xmm1
+; AVX-NEXT: vcvtdq2pd %xmm1, %xmm1
+; AVX-NEXT: vcmpneqsd %xmm0, %xmm1, %xmm0
+; AVX-NEXT: vmovd %xmm0, %eax
+; AVX-NEXT: andl $1, %eax
+; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: retl
+;
+; AVX512-LABEL: une_f64_i32:
+; AVX512: # %bb.0: # %entry
+; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX512-NEXT: vcvttpd2dq %xmm0, %xmm1
+; AVX512-NEXT: vcvtdq2pd %xmm1, %xmm1
+; AVX512-NEXT: vcmpneqsd %xmm0, %xmm1, %k0
+; AVX512-NEXT: kmovd %k0, %eax
+; AVX512-NEXT: # kill: def $al killed $al killed $eax
+; AVX512-NEXT: retl
+entry:
+ %0 = fptosi double %x to i32
+ %1 = sitofp i32 %0 to double
+ %2 = fcmp une double %1, %x
+ %retval12 = zext i1 %2 to i8
+ ret i8 %retval12
+}
+
+define zeroext i8 @une_f64_u32(double %x) nounwind readnone {
+; SSE-LABEL: une_f64_u32:
+; SSE: # %bb.0: # %entry
+; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE-NEXT: cvttsd2si %xmm0, %eax
+; SSE-NEXT: movl %eax, %ecx
+; SSE-NEXT: sarl $31, %ecx
+; SSE-NEXT: movapd %xmm0, %xmm1
+; SSE-NEXT: subsd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
+; SSE-NEXT: cvttsd2si %xmm1, %edx
+; SSE-NEXT: andl %ecx, %edx
+; SSE-NEXT: orl %eax, %edx
+; SSE-NEXT: movd %edx, %xmm1
+; SSE-NEXT: por {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
+; SSE-NEXT: subsd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
+; SSE-NEXT: cmpneqsd %xmm0, %xmm1
+; SSE-NEXT: movd %xmm1, %eax
+; SSE-NEXT: andl $1, %eax
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: retl
+;
+; AVX-LABEL: une_f64_u32:
+; AVX: # %bb.0: # %entry
+; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT: vcvttsd2si %xmm0, %eax
+; AVX-NEXT: movl %eax, %ecx
+; AVX-NEXT: sarl $31, %ecx
+; AVX-NEXT: vsubsd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm1
+; AVX-NEXT: vcvttsd2si %xmm1, %edx
+; AVX-NEXT: andl %ecx, %edx
+; AVX-NEXT: orl %eax, %edx
+; AVX-NEXT: vmovd %edx, %xmm1
+; AVX-NEXT: vpor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm1
+; AVX-NEXT: vsubsd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm1
+; AVX-NEXT: vcmpneqsd %xmm0, %xmm1, %xmm0
+; AVX-NEXT: vmovd %xmm0, %eax
+; AVX-NEXT: andl $1, %eax
+; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: retl
+;
+; AVX512-LABEL: une_f64_u32:
+; AVX512: # %bb.0: # %entry
+; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX512-NEXT: vcvttsd2usi %xmm0, %eax
+; AVX512-NEXT: vcvtusi2sd %eax, %xmm7, %xmm1
+; AVX512-NEXT: vcmpneqsd %xmm0, %xmm1, %k0
+; AVX512-NEXT: kmovd %k0, %eax
+; AVX512-NEXT: # kill: def $al killed $al killed $eax
+; AVX512-NEXT: retl
+entry:
+ %0 = fptoui double %x to i32
+ %1 = uitofp i32 %0 to double
+ %2 = fcmp une double %1, %x
+ %retval12 = zext i1 %2 to i8
+ ret i8 %retval12
+}
+
+define zeroext i8 @une_f64_i64(double %x) nounwind readnone {
+; SSE-LABEL: une_f64_i64:
+; SSE: # %bb.0: # %entry
+; SSE-NEXT: pushl %ebp
+; SSE-NEXT: movl %esp, %ebp
+; SSE-NEXT: andl $-8, %esp
+; SSE-NEXT: subl $32, %esp
+; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE-NEXT: movsd %xmm0, {{[0-9]+}}(%esp)
+; SSE-NEXT: fldl {{[0-9]+}}(%esp)
+; SSE-NEXT: fnstcw {{[0-9]+}}(%esp)
+; SSE-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; SSE-NEXT: orl $3072, %eax # imm = 0xC00
+; SSE-NEXT: movw %ax, {{[0-9]+}}(%esp)
+; SSE-NEXT: fldcw {{[0-9]+}}(%esp)
+; SSE-NEXT: fistpll {{[0-9]+}}(%esp)
+; SSE-NEXT: fldcw {{[0-9]+}}(%esp)
+; SSE-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero
+; SSE-NEXT: movlps %xmm1, {{[0-9]+}}(%esp)
+; SSE-NEXT: fildll {{[0-9]+}}(%esp)
+; SSE-NEXT: fstpl {{[0-9]+}}(%esp)
+; SSE-NEXT: cmpneqsd {{[0-9]+}}(%esp), %xmm0
+; SSE-NEXT: movd %xmm0, %eax
+; SSE-NEXT: andl $1, %eax
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: movl %ebp, %esp
+; SSE-NEXT: popl %ebp
+; SSE-NEXT: retl
+;
+; AVX-LABEL: une_f64_i64:
+; AVX: # %bb.0: # %entry
+; AVX-NEXT: pushl %ebp
+; AVX-NEXT: movl %esp, %ebp
+; AVX-NEXT: andl $-8, %esp
+; AVX-NEXT: subl $24, %esp
+; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT: vmovsd %xmm0, (%esp)
+; AVX-NEXT: fldl (%esp)
+; AVX-NEXT: fisttpll (%esp)
+; AVX-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX-NEXT: vmovlps %xmm1, {{[0-9]+}}(%esp)
+; AVX-NEXT: fildll {{[0-9]+}}(%esp)
+; AVX-NEXT: fstpl {{[0-9]+}}(%esp)
+; AVX-NEXT: vcmpneqsd {{[0-9]+}}(%esp), %xmm0, %xmm0
+; AVX-NEXT: vmovd %xmm0, %eax
+; AVX-NEXT: andl $1, %eax
+; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: movl %ebp, %esp
+; AVX-NEXT: popl %ebp
+; AVX-NEXT: retl
+;
+; AVX512-LABEL: une_f64_i64:
+; AVX512: # %bb.0: # %entry
+; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX512-NEXT: vcvttpd2qq %xmm0, %xmm1
+; AVX512-NEXT: vcvtqq2pd %ymm1, %ymm1
+; AVX512-NEXT: vcmpneqsd %xmm0, %xmm1, %k0
+; AVX512-NEXT: kmovd %k0, %eax
+; AVX512-NEXT: # kill: def $al killed $al killed $eax
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retl
+entry:
+ %0 = fptosi double %x to i64
+ %1 = sitofp i64 %0 to double
+ %2 = fcmp une double %1, %x
+ %retval12 = zext i1 %2 to i8
+ ret i8 %retval12
+}
+
+define zeroext i8 @une_f64_u64(double %x) nounwind readnone {
+; SSE-LABEL: une_f64_u64:
+; SSE: # %bb.0: # %entry
+; SSE-NEXT: pushl %ebp
+; SSE-NEXT: movl %esp, %ebp
+; SSE-NEXT: andl $-8, %esp
+; SSE-NEXT: subl $16, %esp
+; SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
+; SSE-NEXT: movsd {{.*#+}} xmm1 = [9.2233720368547758E+18,0.0E+0]
+; SSE-NEXT: ucomisd %xmm0, %xmm1
+; SSE-NEXT: jbe .LBB7_2
+; SSE-NEXT: # %bb.1: # %entry
+; SSE-NEXT: xorpd %xmm1, %xmm1
+; SSE-NEXT: .LBB7_2: # %entry
+; SSE-NEXT: movapd %xmm0, %xmm2
+; SSE-NEXT: subsd %xmm1, %xmm2
+; SSE-NEXT: movsd %xmm2, {{[0-9]+}}(%esp)
+; SSE-NEXT: setbe %al
+; SSE-NEXT: fldl {{[0-9]+}}(%esp)
+; SSE-NEXT: fnstcw {{[0-9]+}}(%esp)
+; SSE-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
+; SSE-NEXT: orl $3072, %ecx # imm = 0xC00
+; SSE-NEXT: movw %cx, {{[0-9]+}}(%esp)
+; SSE-NEXT: fldcw {{[0-9]+}}(%esp)
+; SSE-NEXT: fistpll {{[0-9]+}}(%esp)
+; SSE-NEXT: fldcw {{[0-9]+}}(%esp)
+; SSE-NEXT: movzbl %al, %eax
+; SSE-NEXT: shll $31, %eax
+; SSE-NEXT: xorl {{[0-9]+}}(%esp), %eax
+; SSE-NEXT: movd %eax, %xmm1
+; SSE-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero
+; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[1],mem[1]
+; SSE-NEXT: subpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm2
+; SSE-NEXT: movapd %xmm2, %xmm1
+; SSE-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm2[1]
+; SSE-NEXT: addsd %xmm2, %xmm1
+; SSE-NEXT: cmpneqsd %xmm0, %xmm1
+; SSE-NEXT: movd %xmm1, %eax
+; SSE-NEXT: andl $1, %eax
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: movl %ebp, %esp
+; SSE-NEXT: popl %ebp
+; SSE-NEXT: retl
+;
+; AVX-LABEL: une_f64_u64:
+; AVX: # %bb.0: # %entry
+; AVX-NEXT: pushl %ebp
+; AVX-NEXT: movl %esp, %ebp
+; AVX-NEXT: andl $-8, %esp
+; AVX-NEXT: subl $8, %esp
+; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT: vmovsd {{.*#+}} xmm1 = [9.2233720368547758E+18,0.0E+0]
+; AVX-NEXT: vucomisd %xmm0, %xmm1
+; AVX-NEXT: jbe .LBB7_2
+; AVX-NEXT: # %bb.1: # %entry
+; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
+; AVX-NEXT: .LBB7_2: # %entry
+; AVX-NEXT: vsubsd %xmm1, %xmm0, %xmm1
+; AVX-NEXT: vmovsd %xmm1, (%esp)
+; AVX-NEXT: fldl (%esp)
+; AVX-NEXT: fisttpll (%esp)
+; AVX-NEXT: setbe %al
+; AVX-NEXT: movzbl %al, %eax
+; AVX-NEXT: shll $31, %eax
+; AVX-NEXT: xorl {{[0-9]+}}(%esp), %eax
+; AVX-NEXT: vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX-NEXT: vpinsrd $1, %eax, %xmm1, %xmm1
+; AVX-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
+; AVX-NEXT: vsubpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1, %xmm1
+; AVX-NEXT: vshufpd {{.*#+}} xmm2 = xmm1[1,0]
+; AVX-NEXT: vaddsd %xmm1, %xmm2, %xmm1
+; AVX-NEXT: vcmpneqsd %xmm0, %xmm1, %xmm0
+; AVX-NEXT: vmovd %xmm0, %eax
+; AVX-NEXT: andl $1, %eax
+; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: movl %ebp, %esp
+; AVX-NEXT: popl %ebp
+; AVX-NEXT: retl
+;
+; AVX512-LABEL: une_f64_u64:
+; AVX512: # %bb.0: # %entry
+; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX512-NEXT: vcvttpd2uqq %xmm0, %xmm1
+; AVX512-NEXT: vcvtuqq2pd %ymm1, %ymm1
+; AVX512-NEXT: vcmpneqsd %xmm0, %xmm1, %k0
+; AVX512-NEXT: kmovd %k0, %eax
+; AVX512-NEXT: # kill: def $al killed $al killed $eax
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retl
+entry:
+ %0 = fptoui double %x to i64
+ %1 = uitofp i64 %0 to double
+ %2 = fcmp une double %1, %x
+ %retval12 = zext i1 %2 to i8
ret i8 %retval12
}
More information about the llvm-commits
mailing list