[llvm] [VP] IR expansion to Int Func Call (PR #122867)

via llvm-commits llvm-commits at lists.llvm.org
Mon Jan 13 23:47:46 PST 2025


https://github.com/LiqinWeng created https://github.com/llvm/llvm-project/pull/122867

Add basic handling for VP ops that can expand to Int intrinsics, which includes: ctpop/cttz/ctlz/sadd.sat/uadd.sat/ssub.sat/usub.sat/fshl/fshr

>From 3bb2a4fe3013811374e8fa98162565d527d83bda Mon Sep 17 00:00:00 2001
From: LiqinWeng <liqin.weng at spacemit.com>
Date: Tue, 14 Jan 2025 15:37:22 +0800
Subject: [PATCH] [VP] IR expansion to Int Func Call

Add basic handling for VP ops that can expand to Int intrinsics, which includes: ctpop/cttz/ctlz/sadd.sat/uadd.sat/ssub.sat/usub.sat/fshl/fshr
---
 llvm/lib/CodeGen/ExpandVectorPredication.cpp  |   9 +
 .../CodeGen/X86/expand-vp-int-intrinsics.ll   | 395 ++++++++++++++++++
 2 files changed, 404 insertions(+)

diff --git a/llvm/lib/CodeGen/ExpandVectorPredication.cpp b/llvm/lib/CodeGen/ExpandVectorPredication.cpp
index 5ca223852cbde3..6f4a2c9ca7604f 100644
--- a/llvm/lib/CodeGen/ExpandVectorPredication.cpp
+++ b/llvm/lib/CodeGen/ExpandVectorPredication.cpp
@@ -609,6 +609,15 @@ Value *CachingVPExpander::expandPredication(VPIntrinsic &VPI) {
   case Intrinsic::vp_umin:
   case Intrinsic::vp_bswap:
   case Intrinsic::vp_bitreverse:
+  case Intrinsic::vp_ctpop:
+  case Intrinsic::vp_ctlz:
+  case Intrinsic::vp_cttz:
+  case Intrinsic::vp_sadd_sat:
+  case Intrinsic::vp_uadd_sat:
+  case Intrinsic::vp_ssub_sat:
+  case Intrinsic::vp_usub_sat:
+  case Intrinsic::vp_fshl:
+  case Intrinsic::vp_fshr:
     return expandPredicationToIntCall(Builder, VPI);
   case Intrinsic::vp_fabs:
   case Intrinsic::vp_sqrt:
diff --git a/llvm/test/CodeGen/X86/expand-vp-int-intrinsics.ll b/llvm/test/CodeGen/X86/expand-vp-int-intrinsics.ll
index f26368c02de2b7..bdc25cdec06b10 100644
--- a/llvm/test/CodeGen/X86/expand-vp-int-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/expand-vp-int-intrinsics.ll
@@ -1347,3 +1347,398 @@ define <4 x i32> @vp_bswap_v4i32(<4 x i32> %va, <4 x i1> %m, i32 zeroext %evl) {
   ret <4 x i32> %v
 }
 declare <4 x i32> @llvm.vp.bswap.v4i32(<4 x i32>, <4 x i1>, i32)
+
+define <4 x i32> @vp_ctpop_v4i32(<4 x i32> %va, <4 x i1> %m, i32 zeroext %evl) {
+; X86-LABEL: vp_ctpop_v4i32:
+; X86:       # %bb.0:
+; X86-NEXT:    vbroadcastss {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X86-NEXT:    vpand %xmm1, %xmm0, %xmm2
+; X86-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X86-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; X86-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; X86-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
+; X86-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; X86-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X86-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; X86-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
+; X86-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; X86-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; X86-NEXT:    retl
+;
+; SSE-LABEL: vp_ctpop_v4i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $1, %xmm1
+; SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE-NEXT:    psubb %xmm1, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    pand %xmm1, %xmm2
+; SSE-NEXT:    psrlw $2, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    paddb %xmm2, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $4, %xmm1
+; SSE-NEXT:    paddb %xmm1, %xmm0
+; SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; SSE-NEXT:    psadbw %xmm1, %xmm2
+; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE-NEXT:    psadbw %xmm1, %xmm0
+; SSE-NEXT:    packuswb %xmm2, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: vp_ctpop_v4i32:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
+; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX1-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; AVX1-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: vp_ctpop_v4i32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastb {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
+; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; AVX2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: vp_ctpop_v4i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm2
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX512-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
+; AVX512-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
+; AVX512-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; AVX512-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    retq
+  %v = call <4 x i32> @llvm.vp.ctpop.v4i32(<4 x i32> %va, <4 x i1> %m, i32 %evl)
+  ret <4 x i32> %v
+}
+declare <4 x i32> @llvm.vp.ctpop.v4i32(<4 x i32>, <4 x i1>, i32)
+
+define <4 x i32> @vp_ctlz_v4i32(<4 x i32> %va, <4 x i1> %m, i32 zeroext %evl) {
+; X86-LABEL: vp_ctlz_v4i32:
+; X86:       # %bb.0:
+; X86-NEXT:    vmovq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X86-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; X86-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; X86-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm3, %xmm3
+; X86-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; X86-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
+; X86-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; X86-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; X86-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
+; X86-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
+; X86-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; X86-NEXT:    vpand %xmm2, %xmm1, %xmm2
+; X86-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; X86-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
+; X86-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
+; X86-NEXT:    vpsrld $16, %xmm0, %xmm0
+; X86-NEXT:    vpand %xmm0, %xmm1, %xmm0
+; X86-NEXT:    vpsrld $16, %xmm1, %xmm1
+; X86-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
+; X86-NEXT:    retl
+;
+; SSE-LABEL: vp_ctlz_v4i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $1, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $2, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $4, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $8, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $1, %xmm1
+; SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE-NEXT:    psubb %xmm1, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    pand %xmm1, %xmm2
+; SSE-NEXT:    psrlw $2, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    paddb %xmm2, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $4, %xmm1
+; SSE-NEXT:    paddb %xmm1, %xmm0
+; SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; SSE-NEXT:    psadbw %xmm1, %xmm2
+; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE-NEXT:    psadbw %xmm1, %xmm0
+; SSE-NEXT:    packuswb %xmm2, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: vp_ctlz_v4i32:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vmovq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX1-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
+; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
+; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
+; AVX1-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
+; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm2
+; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpsrld $16, %xmm1, %xmm1
+; AVX1-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: vp_ctlz_v4i32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vmovq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX2-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
+; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX2-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
+; AVX2-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX2-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX2-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
+; AVX2-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
+; AVX2-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm2
+; AVX2-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX2-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
+; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm0
+; AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpsrld $16, %xmm1, %xmm1
+; AVX2-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: vp_ctlz_v4i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vmovq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX512-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX512-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm3, %xmm3
+; AVX512-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
+; AVX512-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX512-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX512-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
+; AVX512-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
+; AVX512-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; AVX512-NEXT:    vpand %xmm2, %xmm1, %xmm2
+; AVX512-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX512-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
+; AVX512-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
+; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm0
+; AVX512-NEXT:    vpand %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vpsrld $16, %xmm1, %xmm1
+; AVX512-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    retq
+  %v = call <4 x i32> @llvm.vp.ctlz.v4i32(<4 x i32> %va, i1 false, <4 x i1> %m, i32 %evl)
+  ret <4 x i32> %v
+}
+declare <4 x i32> @llvm.vp.ctlz.v4i32(<4 x i32>, i1, <4 x i1>, i32)
+
+define <4 x i32> @vp_cttz_v4i32(<4 x i32> %va, <4 x i1> %m, i32 zeroext %evl) {
+; X86-LABEL: vp_cttz_v4i32:
+; X86:       # %bb.0:
+; X86-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; X86-NEXT:    vpaddd %xmm1, %xmm0, %xmm1
+; X86-NEXT:    vpandn %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vbroadcastss {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X86-NEXT:    vpand %xmm1, %xmm0, %xmm2
+; X86-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X86-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; X86-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; X86-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
+; X86-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; X86-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X86-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; X86-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
+; X86-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; X86-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; X86-NEXT:    retl
+;
+; SSE-LABEL: vp_cttz_v4i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE-NEXT:    paddd %xmm0, %xmm1
+; SSE-NEXT:    pandn %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $1, %xmm1
+; SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE-NEXT:    psubb %xmm1, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    pand %xmm1, %xmm2
+; SSE-NEXT:    psrlw $2, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    paddb %xmm2, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrlw $4, %xmm1
+; SSE-NEXT:    paddb %xmm1, %xmm0
+; SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; SSE-NEXT:    psadbw %xmm1, %xmm2
+; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE-NEXT:    psadbw %xmm1, %xmm0
+; SSE-NEXT:    packuswb %xmm2, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: vp_cttz_v4i32:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vpandn %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
+; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX1-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; AVX1-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: vp_cttz_v4i32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpaddd %xmm1, %xmm0, %xmm1
+; AVX2-NEXT:    vpandn %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpbroadcastb {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
+; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; AVX2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: vp_cttz_v4i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX512-NEXT:    vpaddd %xmm1, %xmm0, %xmm1
+; AVX512-NEXT:    vpandn %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm2
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX512-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
+; AVX512-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
+; AVX512-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; AVX512-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    retq
+  %v = call <4 x i32> @llvm.vp.cttz.v4i32(<4 x i32> %va, i1 false, <4 x i1> %m, i32 %evl)
+  ret <4 x i32> %v
+}
+declare <4 x i32> @llvm.vp.cttz.v4i32(<4 x i32>, i1, <4 x i1>, i32)
+
+
+define <4 x i32> @vp_sadd_sat_v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i1> %m, i32 zeroext %evl) {
+  %v = call <4 x i32> @llvm.vp.sadd.sat.v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i1> %m, i32 %evl)
+  ret <4 x i32> %v
+}
+declare <4 x i32> @llvm.vp.sadd.sat.v4i32(<4 x i32>, <4 x i32>, <4 x i1>, i32)
+
+define <4 x i32> @vp_sadd_uat_v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i1> %m, i32 zeroext %evl) {
+  %v = call <4 x i32> @llvm.vp.uadd.sat.v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i1> %m, i32 %evl)
+  ret <4 x i32> %v
+}
+declare <4 x i32> @llvm.vp.uadd.sat.v4i32(<4 x i32>, <4 x i32>, <4 x i1>, i32)
+
+define <4 x i32> @vp_ssub_sat_v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i1> %m, i32 zeroext %evl) {
+  %v = call <4 x i32> @llvm.vp.ssub.sat.v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i1> %m, i32 %evl)
+  ret <4 x i32> %v
+}
+declare <4 x i32> @llvm.vp.ssub.sat.v4i32(<4 x i32>, <4 x i32>, <4 x i1>, i32)
+
+define <4 x i32> @vp_usub_sat_v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i1> %m, i32 zeroext %evl) {
+  %v = call <4 x i32> @llvm.vp.usub.sat.v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i1> %m, i32 %evl)
+  ret <4 x i32> %v
+}
+declare <4 x i32> @llvm.vp.usub.sat.v4i32(<4 x i32>, <4 x i32>, <4 x i1>, i32)
+
+define <4 x i32> @vp_fshl_v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i32> %vc, <4 x i1> %m, i32 zeroext %evl) {
+  %v = call <4 x i32> @llvm.vp.fshl.v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i32> %vc, <4 x i1> %m, i32 %evl)
+  ret <4 x i32> %v
+}
+declare <4 x i32> @llvm.vp.fshl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>, <4 x i1>, i32)
+
+define <4 x i32> @vp_fshr_v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i32> %vc, <4 x i1> %m, i32 zeroext %evl) {
+  %v = call <4 x i32> @llvm.vp.fshr.v4i32(<4 x i32> %va, <4 x i32> %vb, <4 x i32> %vc, <4 x i1> %m, i32 %evl)
+  ret <4 x i32> %v
+}
+declare <4 x i32> @llvm.vp.fshr.v4i32(<4 x i32>, <4 x i32>, <4 x i32>, <4 x i1>, i32)



More information about the llvm-commits mailing list