8238580aae
In r337348, I changed lowering to prefer X86ISD::UNPCKL/UNPCKH opcodes over MOVLHPS/MOVHLPS for v2f64 {0,0} and {1,1} shuffles when we have SSE2. This enabled the removal of a bunch of weirdly bitcasted isel patterns in r337349. To avoid changing the tests I placed a gross hack in isel to still emit movhlps instructions for fake unary unpckh nodes. A similar hack was not needed for unpckl and movlhps because we do execution domain switching for those. But unpckh and movhlps have swapped operand order.
This patch removes the hack.
This is a code size increase since unpckhpd requires a 0x66 prefix and movhlps does not. But if that's a big concern we should be using movhlps for all unpckhpd opcodes and let commuteInstruction turnit into unpckhpd when its an advantage.
Differential Revision: https://reviews.llvm.org/D49499
llvm-svn: 341973
216 lines
8.1 KiB
LLVM
216 lines
8.1 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefix=ALL --check-prefix=SSE
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fma | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=FMA
|
|
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=FMA
|
|
|
|
; PR31866
|
|
; complex float complex_square_f32(complex float x) {
|
|
; return x*x;
|
|
; }
|
|
|
|
define <2 x float> @complex_square_f32(<2 x float>) #0 {
|
|
; SSE-LABEL: complex_square_f32:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
|
|
; SSE-NEXT: movaps %xmm0, %xmm2
|
|
; SSE-NEXT: addss %xmm0, %xmm2
|
|
; SSE-NEXT: mulss %xmm1, %xmm2
|
|
; SSE-NEXT: mulss %xmm0, %xmm0
|
|
; SSE-NEXT: mulss %xmm1, %xmm1
|
|
; SSE-NEXT: subss %xmm1, %xmm0
|
|
; SSE-NEXT: insertps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[2,3]
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: complex_square_f32:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
|
|
; AVX1-NEXT: vaddss %xmm0, %xmm0, %xmm2
|
|
; AVX1-NEXT: vmulss %xmm2, %xmm1, %xmm2
|
|
; AVX1-NEXT: vmulss %xmm0, %xmm0, %xmm0
|
|
; AVX1-NEXT: vmulss %xmm1, %xmm1, %xmm1
|
|
; AVX1-NEXT: vsubss %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[2,3]
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; FMA-LABEL: complex_square_f32:
|
|
; FMA: # %bb.0:
|
|
; FMA-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3]
|
|
; FMA-NEXT: vaddss %xmm0, %xmm0, %xmm2
|
|
; FMA-NEXT: vmulss %xmm2, %xmm1, %xmm2
|
|
; FMA-NEXT: vmulss %xmm1, %xmm1, %xmm1
|
|
; FMA-NEXT: vfmsub231ss %xmm0, %xmm0, %xmm1
|
|
; FMA-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm2[0],xmm1[2,3]
|
|
; FMA-NEXT: retq
|
|
%2 = extractelement <2 x float> %0, i32 0
|
|
%3 = extractelement <2 x float> %0, i32 1
|
|
%4 = fmul fast float %3, 2.000000e+00
|
|
%5 = fmul fast float %4, %2
|
|
%6 = fmul fast float %2, %2
|
|
%7 = fmul fast float %3, %3
|
|
%8 = fsub fast float %6, %7
|
|
%9 = insertelement <2 x float> undef, float %8, i32 0
|
|
%10 = insertelement <2 x float> %9, float %5, i32 1
|
|
ret <2 x float> %10
|
|
}
|
|
|
|
define <2 x double> @complex_square_f64(<2 x double>) #0 {
|
|
; SSE-LABEL: complex_square_f64:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: movapd %xmm0, %xmm1
|
|
; SSE-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
|
|
; SSE-NEXT: movapd %xmm0, %xmm2
|
|
; SSE-NEXT: addsd %xmm0, %xmm2
|
|
; SSE-NEXT: mulsd %xmm1, %xmm2
|
|
; SSE-NEXT: mulsd %xmm0, %xmm0
|
|
; SSE-NEXT: mulsd %xmm1, %xmm1
|
|
; SSE-NEXT: subsd %xmm1, %xmm0
|
|
; SSE-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0]
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: complex_square_f64:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
|
|
; AVX1-NEXT: vaddsd %xmm0, %xmm0, %xmm2
|
|
; AVX1-NEXT: vmulsd %xmm2, %xmm1, %xmm2
|
|
; AVX1-NEXT: vmulsd %xmm0, %xmm0, %xmm0
|
|
; AVX1-NEXT: vmulsd %xmm1, %xmm1, %xmm1
|
|
; AVX1-NEXT: vsubsd %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0]
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; FMA-LABEL: complex_square_f64:
|
|
; FMA: # %bb.0:
|
|
; FMA-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
|
|
; FMA-NEXT: vaddsd %xmm0, %xmm0, %xmm2
|
|
; FMA-NEXT: vmulsd %xmm2, %xmm1, %xmm2
|
|
; FMA-NEXT: vmulsd %xmm1, %xmm1, %xmm1
|
|
; FMA-NEXT: vfmsub231sd %xmm0, %xmm0, %xmm1
|
|
; FMA-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm2[0]
|
|
; FMA-NEXT: retq
|
|
%2 = extractelement <2 x double> %0, i32 0
|
|
%3 = extractelement <2 x double> %0, i32 1
|
|
%4 = fmul fast double %3, 2.000000e+00
|
|
%5 = fmul fast double %4, %2
|
|
%6 = fmul fast double %2, %2
|
|
%7 = fmul fast double %3, %3
|
|
%8 = fsub fast double %6, %7
|
|
%9 = insertelement <2 x double> undef, double %8, i32 0
|
|
%10 = insertelement <2 x double> %9, double %5, i32 1
|
|
ret <2 x double> %10
|
|
}
|
|
|
|
; complex float complex_mul_f32(complex float x, complex float y) {
|
|
; return x*y;
|
|
; }
|
|
|
|
define <2 x float> @complex_mul_f32(<2 x float>, <2 x float>) #0 {
|
|
; SSE-LABEL: complex_mul_f32:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: movshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
|
|
; SSE-NEXT: movshdup {{.*#+}} xmm3 = xmm1[1,1,3,3]
|
|
; SSE-NEXT: movaps %xmm3, %xmm4
|
|
; SSE-NEXT: mulss %xmm0, %xmm4
|
|
; SSE-NEXT: mulss %xmm1, %xmm0
|
|
; SSE-NEXT: mulss %xmm2, %xmm1
|
|
; SSE-NEXT: addss %xmm4, %xmm1
|
|
; SSE-NEXT: mulss %xmm2, %xmm3
|
|
; SSE-NEXT: subss %xmm3, %xmm0
|
|
; SSE-NEXT: insertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[2,3]
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: complex_mul_f32:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
|
|
; AVX1-NEXT: vmovshdup {{.*#+}} xmm3 = xmm1[1,1,3,3]
|
|
; AVX1-NEXT: vmulss %xmm0, %xmm3, %xmm4
|
|
; AVX1-NEXT: vmulss %xmm2, %xmm1, %xmm5
|
|
; AVX1-NEXT: vaddss %xmm5, %xmm4, %xmm4
|
|
; AVX1-NEXT: vmulss %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vmulss %xmm2, %xmm3, %xmm1
|
|
; AVX1-NEXT: vsubss %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[2,3]
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; FMA-LABEL: complex_mul_f32:
|
|
; FMA: # %bb.0:
|
|
; FMA-NEXT: vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
|
|
; FMA-NEXT: vmovshdup {{.*#+}} xmm3 = xmm1[1,1,3,3]
|
|
; FMA-NEXT: vmulss %xmm2, %xmm1, %xmm4
|
|
; FMA-NEXT: vfmadd231ss %xmm0, %xmm3, %xmm4
|
|
; FMA-NEXT: vmulss %xmm2, %xmm3, %xmm2
|
|
; FMA-NEXT: vfmsub231ss %xmm0, %xmm1, %xmm2
|
|
; FMA-NEXT: vinsertps {{.*#+}} xmm0 = xmm2[0],xmm4[0],xmm2[2,3]
|
|
; FMA-NEXT: retq
|
|
%3 = extractelement <2 x float> %0, i32 0
|
|
%4 = extractelement <2 x float> %0, i32 1
|
|
%5 = extractelement <2 x float> %1, i32 0
|
|
%6 = extractelement <2 x float> %1, i32 1
|
|
%7 = fmul fast float %6, %3
|
|
%8 = fmul fast float %5, %4
|
|
%9 = fadd fast float %7, %8
|
|
%10 = fmul fast float %5, %3
|
|
%11 = fmul fast float %6, %4
|
|
%12 = fsub fast float %10, %11
|
|
%13 = insertelement <2 x float> undef, float %12, i32 0
|
|
%14 = insertelement <2 x float> %13, float %9, i32 1
|
|
ret <2 x float> %14
|
|
}
|
|
|
|
define <2 x double> @complex_mul_f64(<2 x double>, <2 x double>) #0 {
|
|
; SSE-LABEL: complex_mul_f64:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: movapd %xmm0, %xmm2
|
|
; SSE-NEXT: unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm0[1]
|
|
; SSE-NEXT: movapd %xmm1, %xmm3
|
|
; SSE-NEXT: unpckhpd {{.*#+}} xmm3 = xmm3[1],xmm1[1]
|
|
; SSE-NEXT: movapd %xmm3, %xmm4
|
|
; SSE-NEXT: mulsd %xmm0, %xmm4
|
|
; SSE-NEXT: mulsd %xmm1, %xmm0
|
|
; SSE-NEXT: mulsd %xmm2, %xmm1
|
|
; SSE-NEXT: addsd %xmm4, %xmm1
|
|
; SSE-NEXT: mulsd %xmm2, %xmm3
|
|
; SSE-NEXT: subsd %xmm3, %xmm0
|
|
; SSE-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX1-LABEL: complex_mul_f64:
|
|
; AVX1: # %bb.0:
|
|
; AVX1-NEXT: vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
|
|
; AVX1-NEXT: vpermilpd {{.*#+}} xmm3 = xmm1[1,0]
|
|
; AVX1-NEXT: vmulsd %xmm0, %xmm3, %xmm4
|
|
; AVX1-NEXT: vmulsd %xmm2, %xmm1, %xmm5
|
|
; AVX1-NEXT: vaddsd %xmm5, %xmm4, %xmm4
|
|
; AVX1-NEXT: vmulsd %xmm0, %xmm1, %xmm0
|
|
; AVX1-NEXT: vmulsd %xmm2, %xmm3, %xmm1
|
|
; AVX1-NEXT: vsubsd %xmm1, %xmm0, %xmm0
|
|
; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm4[0]
|
|
; AVX1-NEXT: retq
|
|
;
|
|
; FMA-LABEL: complex_mul_f64:
|
|
; FMA: # %bb.0:
|
|
; FMA-NEXT: vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
|
|
; FMA-NEXT: vpermilpd {{.*#+}} xmm3 = xmm1[1,0]
|
|
; FMA-NEXT: vmulsd %xmm2, %xmm1, %xmm4
|
|
; FMA-NEXT: vfmadd231sd %xmm0, %xmm3, %xmm4
|
|
; FMA-NEXT: vmulsd %xmm2, %xmm3, %xmm2
|
|
; FMA-NEXT: vfmsub231sd %xmm0, %xmm1, %xmm2
|
|
; FMA-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm2[0],xmm4[0]
|
|
; FMA-NEXT: retq
|
|
%3 = extractelement <2 x double> %0, i32 0
|
|
%4 = extractelement <2 x double> %0, i32 1
|
|
%5 = extractelement <2 x double> %1, i32 0
|
|
%6 = extractelement <2 x double> %1, i32 1
|
|
%7 = fmul fast double %6, %3
|
|
%8 = fmul fast double %5, %4
|
|
%9 = fadd fast double %7, %8
|
|
%10 = fmul fast double %5, %3
|
|
%11 = fmul fast double %6, %4
|
|
%12 = fsub fast double %10, %11
|
|
%13 = insertelement <2 x double> undef, double %12, i32 0
|
|
%14 = insertelement <2 x double> %13, double %9, i32 1
|
|
ret <2 x double> %14
|
|
}
|
|
|
|
attributes #0 = { nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "less-precise-fpmad"="false" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "unsafe-fp-math"="true" }
|