76e14deb4a
BreakFalseDeps picks the best register for undef operands if instructions have false dependency. The problem is if the instruction is close to the beginning of the function, ReachingDefAnalysis is over optimism to the unused registers, which results in collision with registers just defined in the caller. This patch changes the selection of undef register in an reverse order, which reduces the probability of register collisions between caller and callee. It brings improvement in some of our internal benchmarks with negligible effect on other benchmarks.
225 lines
9.1 KiB
LLVM
225 lines
9.1 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: llc < %s -mtriple=x86_64 -mattr=+avx512fp16 | FileCheck %s
|
|
|
|
define <2 x half> @vector_sint64ToHalf(<2 x i64> %int64) {
|
|
; CHECK-LABEL: vector_sint64ToHalf:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; CHECK-NEXT: vcvtqq2ph %zmm0, %xmm0
|
|
; CHECK-NEXT: vzeroupper
|
|
; CHECK-NEXT: retq
|
|
%fp16 = sitofp <2 x i64> %int64 to <2 x half>
|
|
ret <2 x half> %fp16
|
|
}
|
|
|
|
define <4 x half> @vector_sint32ToHalf(<4 x i32> %int32) {
|
|
; CHECK-LABEL: vector_sint32ToHalf:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vextractps $3, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm1
|
|
; CHECK-NEXT: vextractps $2, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm2
|
|
; CHECK-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
|
|
; CHECK-NEXT: vextractps $1, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm2
|
|
; CHECK-NEXT: vmovd %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm0
|
|
; CHECK-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
|
|
; CHECK-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],zero,zero
|
|
; CHECK-NEXT: retq
|
|
%fp16 = sitofp <4 x i32> %int32 to <4 x half>
|
|
ret <4 x half> %fp16
|
|
}
|
|
|
|
define <8 x half> @vector_sint16ToHalf(<8 x i16> %int16) {
|
|
; CHECK-LABEL: vector_sint16ToHalf:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vpextrw $7, %xmm0, %eax
|
|
; CHECK-NEXT: cwtl
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm1
|
|
; CHECK-NEXT: vpextrw $6, %xmm0, %eax
|
|
; CHECK-NEXT: cwtl
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm2
|
|
; CHECK-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
|
|
; CHECK-NEXT: vpextrw $5, %xmm0, %eax
|
|
; CHECK-NEXT: cwtl
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm2
|
|
; CHECK-NEXT: vpextrw $4, %xmm0, %eax
|
|
; CHECK-NEXT: cwtl
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm3
|
|
; CHECK-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
|
|
; CHECK-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
|
|
; CHECK-NEXT: vpextrw $3, %xmm0, %eax
|
|
; CHECK-NEXT: cwtl
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm2
|
|
; CHECK-NEXT: vpextrw $2, %xmm0, %eax
|
|
; CHECK-NEXT: cwtl
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm3
|
|
; CHECK-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
|
|
; CHECK-NEXT: vpextrw $1, %xmm0, %eax
|
|
; CHECK-NEXT: cwtl
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm3
|
|
; CHECK-NEXT: vmovw %xmm0, %eax
|
|
; CHECK-NEXT: cwtl
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm0
|
|
; CHECK-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
|
|
; CHECK-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
|
|
; CHECK-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
|
|
; CHECK-NEXT: retq
|
|
%fp16 = sitofp <8 x i16> %int16 to <8 x half>
|
|
ret <8 x half> %fp16
|
|
}
|
|
|
|
define <2 x half> @vector_uint64ToHalf(<2 x i64> %int64) {
|
|
; CHECK-LABEL: vector_uint64ToHalf:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; CHECK-NEXT: vcvtuqq2ph %zmm0, %xmm0
|
|
; CHECK-NEXT: vzeroupper
|
|
; CHECK-NEXT: retq
|
|
%fp16 = uitofp <2 x i64> %int64 to <2 x half>
|
|
ret <2 x half> %fp16
|
|
}
|
|
|
|
define <4 x half> @vector_uint32ToHalf(<4 x i32> %int32) {
|
|
; CHECK-LABEL: vector_uint32ToHalf:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; CHECK-NEXT: vcvtudq2ph %zmm0, %ymm0
|
|
; CHECK-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
|
|
; CHECK-NEXT: vzeroupper
|
|
; CHECK-NEXT: retq
|
|
%fp16 = uitofp <4 x i32> %int32 to <4 x half>
|
|
ret <4 x half> %fp16
|
|
}
|
|
|
|
define <8 x half> @vector_uint16ToHalf(<8 x i16> %int16) {
|
|
; CHECK-LABEL: vector_uint16ToHalf:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vpextrw $7, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm1
|
|
; CHECK-NEXT: vpextrw $6, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm2
|
|
; CHECK-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
|
|
; CHECK-NEXT: vpextrw $5, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm2
|
|
; CHECK-NEXT: vpextrw $4, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm3
|
|
; CHECK-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
|
|
; CHECK-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
|
|
; CHECK-NEXT: vpextrw $3, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm2
|
|
; CHECK-NEXT: vpextrw $2, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm3
|
|
; CHECK-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
|
|
; CHECK-NEXT: vpextrw $1, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm3
|
|
; CHECK-NEXT: vpextrw $0, %xmm0, %eax
|
|
; CHECK-NEXT: vcvtsi2sh %eax, %xmm31, %xmm0
|
|
; CHECK-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
|
|
; CHECK-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
|
|
; CHECK-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
|
|
; CHECK-NEXT: retq
|
|
%fp16 = uitofp <8 x i16> %int16 to <8 x half>
|
|
ret <8 x half> %fp16
|
|
}
|
|
|
|
define <8 x half> @select(<8 x half> %x) {
|
|
; CHECK-LABEL: select:
|
|
; CHECK: # %bb.0: # %entry
|
|
; CHECK-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
|
|
; CHECK-NEXT: vmovsh {{.*#+}} xmm1 = [1.0E+0,0.0E+0,0.0E+0,0.0E+0,0.0E+0,0.0E+0,0.0E+0,0.0E+0]
|
|
; CHECK-NEXT: vucomish %xmm1, %xmm0
|
|
; CHECK-NEXT: seta %al
|
|
; CHECK-NEXT: andl $1, %eax
|
|
; CHECK-NEXT: kmovw %eax, %k0
|
|
; CHECK-NEXT: vpsrld $16, %xmm0, %xmm2
|
|
; CHECK-NEXT: vucomish %xmm1, %xmm2
|
|
; CHECK-NEXT: seta %al
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kshiftlw $15, %k1, %k1
|
|
; CHECK-NEXT: kshiftrw $14, %k1, %k1
|
|
; CHECK-NEXT: korw %k1, %k0, %k0
|
|
; CHECK-NEXT: movw $-5, %ax
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kandw %k1, %k0, %k0
|
|
; CHECK-NEXT: vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
|
|
; CHECK-NEXT: vucomish %xmm1, %xmm2
|
|
; CHECK-NEXT: seta %al
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kshiftlw $15, %k1, %k1
|
|
; CHECK-NEXT: kshiftrw $13, %k1, %k1
|
|
; CHECK-NEXT: korw %k1, %k0, %k0
|
|
; CHECK-NEXT: movw $-9, %ax
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kandw %k1, %k0, %k0
|
|
; CHECK-NEXT: vpsrlq $48, %xmm0, %xmm2
|
|
; CHECK-NEXT: vucomish %xmm1, %xmm2
|
|
; CHECK-NEXT: seta %al
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kshiftlw $15, %k1, %k1
|
|
; CHECK-NEXT: kshiftrw $12, %k1, %k1
|
|
; CHECK-NEXT: korw %k1, %k0, %k0
|
|
; CHECK-NEXT: movw $-17, %ax
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kandw %k1, %k0, %k0
|
|
; CHECK-NEXT: vshufpd {{.*#+}} xmm2 = xmm0[1,0]
|
|
; CHECK-NEXT: vucomish %xmm1, %xmm2
|
|
; CHECK-NEXT: seta %al
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kshiftlw $15, %k1, %k1
|
|
; CHECK-NEXT: kshiftrw $11, %k1, %k1
|
|
; CHECK-NEXT: korw %k1, %k0, %k0
|
|
; CHECK-NEXT: movw $-33, %ax
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kandw %k1, %k0, %k0
|
|
; CHECK-NEXT: vpsrldq {{.*#+}} xmm2 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; CHECK-NEXT: vucomish %xmm1, %xmm2
|
|
; CHECK-NEXT: seta %al
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kshiftlw $15, %k1, %k1
|
|
; CHECK-NEXT: kshiftrw $10, %k1, %k1
|
|
; CHECK-NEXT: korw %k1, %k0, %k0
|
|
; CHECK-NEXT: movw $-65, %ax
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kandw %k1, %k0, %k0
|
|
; CHECK-NEXT: vshufps {{.*#+}} xmm2 = xmm0[3,3,3,3]
|
|
; CHECK-NEXT: vucomish %xmm1, %xmm2
|
|
; CHECK-NEXT: seta %al
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kshiftlw $6, %k1, %k1
|
|
; CHECK-NEXT: korw %k1, %k0, %k0
|
|
; CHECK-NEXT: kshiftlw $9, %k0, %k0
|
|
; CHECK-NEXT: kshiftrw $9, %k0, %k0
|
|
; CHECK-NEXT: vpsrldq {{.*#+}} xmm2 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
|
|
; CHECK-NEXT: vucomish %xmm1, %xmm2
|
|
; CHECK-NEXT: seta %al
|
|
; CHECK-NEXT: kmovd %eax, %k1
|
|
; CHECK-NEXT: kshiftlw $7, %k1, %k1
|
|
; CHECK-NEXT: korw %k1, %k0, %k1
|
|
; CHECK-NEXT: vpbroadcastw {{.*#+}} xmm1 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
|
|
; CHECK-NEXT: vmovdqu16 %zmm1, %zmm0 {%k1}
|
|
; CHECK-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
|
|
; CHECK-NEXT: vzeroupper
|
|
; CHECK-NEXT: retq
|
|
entry:
|
|
%c = fcmp ogt <8 x half> %x, splat (half 0xH3C00)
|
|
%s = select <8 x i1> %c, <8 x half> splat (half 0xH3C00), <8 x half> %x
|
|
ret <8 x half> %s
|
|
}
|
|
|
|
define <4 x half> @select2(<4 x i32> %0, <4 x half> %1) {
|
|
; CHECK-LABEL: select2:
|
|
; CHECK: # %bb.0: # %entry
|
|
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
|
|
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm0, %xmm0
|
|
; CHECK-NEXT: vpackssdw %xmm0, %xmm0, %xmm0
|
|
; CHECK-NEXT: vmovq {{.*#+}} xmm2 = [9.6E+1,9.7E+1,9.8E+1,9.9E+1,0.0E+0,0.0E+0,0.0E+0,0.0E+0]
|
|
; CHECK-NEXT: vpblendvb %xmm0, %xmm1, %xmm2, %xmm0
|
|
; CHECK-NEXT: retq
|
|
entry:
|
|
%2 = icmp eq <4 x i32> %0, zeroinitializer
|
|
%3 = select <4 x i1> %2, <4 x half> %1, <4 x half> <half 0xH5600, half 0xH5610, half 0xH5620, half 0xH5630>
|
|
ret <4 x half> %3
|
|
}
|