e9f9467da0
Similar to the original VPERMILPSri -> VSHUFPSrri mapping added in D143787, replacing VPERMILPDri -> VSHUFPDrri should never be any slower and saves an encoding byte. The sibling VPERMILPDmi -> VPSHUFDmi mapping is trickier as we need the same shuffle mask in every lane (and it needs to be adjusted) - I haven't attempted that yet but we can investigate it in the future if there's interest. Fixes #61060 Differential Revision: https://reviews.llvm.org/D148999
22 lines
860 B
LLVM
22 lines
860 B
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=avx2 | FileCheck %s
|
|
|
|
; The broadcast node takes a vector operand as input and changes its length.
|
|
|
|
define <4 x double> @PR43402(i64 %x) {
|
|
; CHECK-LABEL: PR43402:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
|
|
; CHECK-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
|
|
; CHECK-NEXT: vsubpd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0
|
|
; CHECK-NEXT: vshufpd {{.*#+}} xmm1 = xmm0[1,0]
|
|
; CHECK-NEXT: vaddsd %xmm0, %xmm1, %xmm0
|
|
; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0
|
|
; CHECK-NEXT: retl
|
|
%conv = uitofp i64 %x to double
|
|
%t2 = insertelement <4 x double> undef, double %conv, i32 0
|
|
%t3 = shufflevector <4 x double> %t2, <4 x double> undef, <4 x i32> zeroinitializer
|
|
ret <4 x double> %t3
|
|
}
|
|
|