[llvm] r297486 - [X86][SSE] Added tests showing missed truncations for sitofp conversion

[Simon Pilgrim via llvm-commits]

Author: rksimon
Date: Fri Mar 10 12:01:53 2017
New Revision: 297486

URL: http://llvm.org/viewvc/llvm-project?rev=297486&view=rev
Log:
[X86][SSE] Added tests showing missed truncations for sitofp conversion

SelectionDAG::ComputeNumSignBits is poor at build_vector handling, meaning that we can't see that all the vXi64 sources are in fact sign extended i32 or smaller.

Added:
    llvm/trunk/test/CodeGen/X86/known-signbits-vector.ll

Added: llvm/trunk/test/CodeGen/X86/known-signbits-vector.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/known-signbits-vector.ll?rev=297486&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/known-signbits-vector.ll (added)
+++ llvm/trunk/test/CodeGen/X86/known-signbits-vector.ll Fri Mar 10 12:01:53 2017
@@ -0,0 +1,109 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X32
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X64
+
+define <2 x double> @signbits_sext_v2i64_sitofp_v2f64(i32 %a0, i32 %a1) nounwind {
+; X32-LABEL: signbits_sext_v2i64_sitofp_v2f64:
+; X32:       # BB#0:
+; X32-NEXT:    pushl %ebp
+; X32-NEXT:    movl %esp, %ebp
+; X32-NEXT:    andl $-8, %esp
+; X32-NEXT:    subl $32, %esp
+; X32-NEXT:    movl 8(%ebp), %eax
+; X32-NEXT:    movl 12(%ebp), %ecx
+; X32-NEXT:    vmovd %eax, %xmm0
+; X32-NEXT:    sarl $31, %eax
+; X32-NEXT:    vpinsrd $1, %eax, %xmm0, %xmm0
+; X32-NEXT:    vmovq %xmm0, {{[0-9]+}}(%esp)
+; X32-NEXT:    vmovd %ecx, %xmm0
+; X32-NEXT:    sarl $31, %ecx
+; X32-NEXT:    vpinsrd $1, %ecx, %xmm0, %xmm0
+; X32-NEXT:    vmovq %xmm0, {{[0-9]+}}(%esp)
+; X32-NEXT:    fildll {{[0-9]+}}(%esp)
+; X32-NEXT:    fstpl {{[0-9]+}}(%esp)
+; X32-NEXT:    fildll {{[0-9]+}}(%esp)
+; X32-NEXT:    fstpl (%esp)
+; X32-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X32-NEXT:    vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
+; X32-NEXT:    movl %ebp, %esp
+; X32-NEXT:    popl %ebp
+; X32-NEXT:    retl
+;
+; X64-LABEL: signbits_sext_v2i64_sitofp_v2f64:
+; X64:       # BB#0:
+; X64-NEXT:    vcvtsi2sdl %esi, %xmm0, %xmm0
+; X64-NEXT:    vcvtsi2sdl %edi, %xmm1, %xmm1
+; X64-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; X64-NEXT:    retq
+  %1 = sext i32 %a0 to i64
+  %2 = sext i32 %a1 to i64
+  %3 = insertelement <2 x i64> undef, i64 %1, i32 0
+  %4 = insertelement <2 x i64> %3, i64 %2, i32 1
+  %5 = sitofp <2 x i64> %4 to <2 x double>
+  ret <2 x double> %5
+}
+
+define <4 x float> @signbits_sext_v4i64_sitofp_v4f32(i8 signext %a0, i16 signext %a1, i32 %a2, i32 %a3) nounwind {
+; X32-LABEL: signbits_sext_v4i64_sitofp_v4f32:
+; X32:       # BB#0:
+; X32-NEXT:    pushl %ebp
+; X32-NEXT:    movl %esp, %ebp
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    andl $-8, %esp
+; X32-NEXT:    subl $56, %esp
+; X32-NEXT:    movsbl 8(%ebp), %eax
+; X32-NEXT:    movswl 12(%ebp), %ecx
+; X32-NEXT:    movl 16(%ebp), %edx
+; X32-NEXT:    movl 20(%ebp), %esi
+; X32-NEXT:    vmovd %eax, %xmm0
+; X32-NEXT:    sarl $31, %eax
+; X32-NEXT:    vpinsrd $1, %eax, %xmm0, %xmm0
+; X32-NEXT:    vmovq %xmm0, {{[0-9]+}}(%esp)
+; X32-NEXT:    vmovd %ecx, %xmm0
+; X32-NEXT:    sarl $31, %ecx
+; X32-NEXT:    vpinsrd $1, %ecx, %xmm0, %xmm0
+; X32-NEXT:    vmovq %xmm0, {{[0-9]+}}(%esp)
+; X32-NEXT:    vmovd %edx, %xmm0
+; X32-NEXT:    sarl $31, %edx
+; X32-NEXT:    vpinsrd $1, %edx, %xmm0, %xmm0
+; X32-NEXT:    vmovq %xmm0, {{[0-9]+}}(%esp)
+; X32-NEXT:    vmovd %esi, %xmm0
+; X32-NEXT:    sarl $31, %esi
+; X32-NEXT:    vpinsrd $1, %esi, %xmm0, %xmm0
+; X32-NEXT:    vmovq %xmm0, {{[0-9]+}}(%esp)
+; X32-NEXT:    fildll {{[0-9]+}}(%esp)
+; X32-NEXT:    fstps {{[0-9]+}}(%esp)
+; X32-NEXT:    fildll {{[0-9]+}}(%esp)
+; X32-NEXT:    fstps {{[0-9]+}}(%esp)
+; X32-NEXT:    fildll {{[0-9]+}}(%esp)
+; X32-NEXT:    fstps {{[0-9]+}}(%esp)
+; X32-NEXT:    fildll {{[0-9]+}}(%esp)
+; X32-NEXT:    fstps (%esp)
+; X32-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X32-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3]
+; X32-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; X32-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0]
+; X32-NEXT:    leal -4(%ebp), %esp
+; X32-NEXT:    popl %esi
+; X32-NEXT:    popl %ebp
+; X32-NEXT:    retl
+;
+; X64-LABEL: signbits_sext_v4i64_sitofp_v4f32:
+; X64:       # BB#0:
+; X64-NEXT:    vmovd %edi, %xmm0
+; X64-NEXT:    vpinsrd $1, %esi, %xmm0, %xmm0
+; X64-NEXT:    vpinsrd $2, %edx, %xmm0, %xmm0
+; X64-NEXT:    vpinsrd $3, %ecx, %xmm0, %xmm0
+; X64-NEXT:    vcvtdq2ps %xmm0, %xmm0
+; X64-NEXT:    retq
+  %1 = sext i8 %a0 to i64
+  %2 = sext i16 %a1 to i64
+  %3 = sext i32 %a2 to i64
+  %4 = sext i32 %a3 to i64
+  %5 = insertelement <4 x i64> undef, i64 %1, i32 0
+  %6 = insertelement <4 x i64> %5, i64 %2, i32 1
+  %7 = insertelement <4 x i64> %6, i64 %3, i32 2
+  %8 = insertelement <4 x i64> %7, i64 %4, i32 3
+  %9 = sitofp <4 x i64> %8 to <4 x float>
+  ret <4 x float> %9
+}