[X86][AVX512] Improve support and testing for CTLZ of 512-bit vectors without CDI

llvm-svn: 302233

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 0bbb656..b5d1962 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -20972,7 +20972,7 @@ static SDValue Lower512IntUnary(SDValue Op, SelectionDAG &DAG) {
 // ( sub(trunc(lzcnt(zext32(x)))) ). In case zext32(x) is illegal,
 // split the vector, perform operation on it's Lo a Hi part and
 // concatenate the results.
-static SDValue LowerVectorCTLZ_AVX512(SDValue Op, SelectionDAG &DAG) {
+static SDValue LowerVectorCTLZ_AVX512CDI(SDValue Op, SelectionDAG &DAG) {
   assert(Op.getOpcode() == ISD::CTLZ);
   SDLoc dl(Op);
   MVT VT = Op.getSimpleValueType();
@@ -21074,13 +21074,17 @@ static SDValue LowerVectorCTLZ(SDValue Op, const SDLoc &DL,
                                SelectionDAG &DAG) {
   MVT VT = Op.getSimpleValueType();
 
-  if (Subtarget.hasAVX512())
-    return LowerVectorCTLZ_AVX512(Op, DAG);
+  if (Subtarget.hasCDI())
+    return LowerVectorCTLZ_AVX512CDI(Op, DAG);
 
   // Decompose 256-bit ops into smaller 128-bit ops.
   if (VT.is256BitVector() && !Subtarget.hasInt256())
     return Lower256IntUnary(Op, DAG);
 
+  // Decompose 512-bit ops into smaller 256-bit ops.
+  if (VT.is512BitVector() && !Subtarget.hasBWI())
+    return Lower512IntUnary(Op, DAG);
+
   assert(Subtarget.hasSSSE3() && "Expected SSSE3 support for PSHUFB");
   return LowerVectorCTLZInRegLUT(Op, DL, Subtarget, DAG);
 }

diff --git a/llvm/test/CodeGen/X86/vector-lzcnt-512.ll b/llvm/test/CodeGen/X86/vector-lzcnt-512.ll
index 79d133b..88378eb 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-lzcnt-512.ll
+++ b/llvm/test/CodeGen/X86/vector-lzcnt-512.ll
@@ -1,39 +1,337 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512cd | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512CD
-; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512cd,-avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512CD
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512cd,+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512CDBW
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=-avx512cd,+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512BW
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=-avx512cd,-avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512DQ
 
 define <8 x i64> @testv8i64(<8 x i64> %in) nounwind {
-; ALL-LABEL: testv8i64:
-; ALL:       ## BB#0:
-; ALL-NEXT:    vplzcntq %zmm0, %zmm0
-; ALL-NEXT:    retq
+; AVX512CD-LABEL: testv8i64:
+; AVX512CD:       ## BB#0:
+; AVX512CD-NEXT:    vplzcntq %zmm0, %zmm0
+; AVX512CD-NEXT:    retq
+;
+; AVX512CDBW-LABEL: testv8i64:
+; AVX512CDBW:       ## BB#0:
+; AVX512CDBW-NEXT:    vplzcntq %zmm0, %zmm0
+; AVX512CDBW-NEXT:    retq
+;
+; AVX512BW-LABEL: testv8i64:
+; AVX512BW:       ## BB#0:
+; AVX512BW-NEXT:    vpsrlq $1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $8, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $16, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $32, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandnq %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT:    vmovdqu8 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpxorq %zmm4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpxord %zmm1, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512DQ-LABEL: testv8i64:
+; AVX512DQ:       ## BB#0:
+; AVX512DQ-NEXT:    vpsrlq $1, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrlq $2, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrlq $4, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrlq $8, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrlq $16, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrlq $32, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1
+; AVX512DQ-NEXT:    vpxorq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
+; AVX512DQ-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpxor %ymm3, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm5
+; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512DQ-NEXT:    vpaddb %ymm5, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    retq
   %out = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %in, i1 0)
   ret <8 x i64> %out
 }
 
 define <8 x i64> @testv8i64u(<8 x i64> %in) nounwind {
-; ALL-LABEL: testv8i64u:
-; ALL:       ## BB#0:
-; ALL-NEXT:    vplzcntq %zmm0, %zmm0
-; ALL-NEXT:    retq
+; AVX512CD-LABEL: testv8i64u:
+; AVX512CD:       ## BB#0:
+; AVX512CD-NEXT:    vplzcntq %zmm0, %zmm0
+; AVX512CD-NEXT:    retq
+;
+; AVX512CDBW-LABEL: testv8i64u:
+; AVX512CDBW:       ## BB#0:
+; AVX512CDBW-NEXT:    vplzcntq %zmm0, %zmm0
+; AVX512CDBW-NEXT:    retq
+;
+; AVX512BW-LABEL: testv8i64u:
+; AVX512BW:       ## BB#0:
+; AVX512BW-NEXT:    vpsrlq $1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $8, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $16, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $32, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandnq %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT:    vmovdqu8 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpxorq %zmm4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpxord %zmm1, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512DQ-LABEL: testv8i64u:
+; AVX512DQ:       ## BB#0:
+; AVX512DQ-NEXT:    vpsrlq $1, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrlq $2, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrlq $4, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrlq $8, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrlq $16, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrlq $32, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1
+; AVX512DQ-NEXT:    vpxorq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
+; AVX512DQ-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpxor %ymm3, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm5
+; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512DQ-NEXT:    vpaddb %ymm5, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    retq
   %out = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %in, i1 -1)
   ret <8 x i64> %out
 }
 
 define <16 x i32> @testv16i32(<16 x i32> %in) nounwind {
-; ALL-LABEL: testv16i32:
-; ALL:       ## BB#0:
-; ALL-NEXT:    vplzcntd %zmm0, %zmm0
-; ALL-NEXT:    retq
+; AVX512CD-LABEL: testv16i32:
+; AVX512CD:       ## BB#0:
+; AVX512CD-NEXT:    vplzcntd %zmm0, %zmm0
+; AVX512CD-NEXT:    retq
+;
+; AVX512CDBW-LABEL: testv16i32:
+; AVX512CDBW:       ## BB#0:
+; AVX512CDBW-NEXT:    vplzcntd %zmm0, %zmm0
+; AVX512CDBW-NEXT:    retq
+;
+; AVX512BW-LABEL: testv16i32:
+; AVX512BW:       ## BB#0:
+; AVX512BW-NEXT:    vpsrld $1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrld $2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrld $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrld $8, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrld $16, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandnq %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT:    vmovdqu8 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpxorq %zmm4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpxord %zmm1, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
+; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
+; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512DQ-LABEL: testv16i32:
+; AVX512DQ:       ## BB#0:
+; AVX512DQ-NEXT:    vpsrld $1, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrld $2, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrld $4, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrld $8, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrld $16, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1
+; AVX512DQ-NEXT:    vpxorq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
+; AVX512DQ-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpxor %ymm3, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpunpckhdq {{.*#+}} ymm5 = ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[6],ymm3[6],ymm1[7],ymm3[7]
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm5, %ymm5
+; AVX512DQ-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[4],ymm3[4],ymm1[5],ymm3[5]
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpackuswb %ymm5, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm5
+; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512DQ-NEXT:    vpaddb %ymm5, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm3[2],ymm0[3],ymm3[3],ymm0[6],ymm3[6],ymm0[7],ymm3[7]
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm3[0],ymm0[1],ymm3[1],ymm0[4],ymm3[4],ymm0[5],ymm3[5]
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    retq
   %out = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %in, i1 0)
   ret <16 x i32> %out
 }
 
 define <16 x i32> @testv16i32u(<16 x i32> %in) nounwind {
-; ALL-LABEL: testv16i32u:
-; ALL:       ## BB#0:
-; ALL-NEXT:    vplzcntd %zmm0, %zmm0
-; ALL-NEXT:    retq
+; AVX512CD-LABEL: testv16i32u:
+; AVX512CD:       ## BB#0:
+; AVX512CD-NEXT:    vplzcntd %zmm0, %zmm0
+; AVX512CD-NEXT:    retq
+;
+; AVX512CDBW-LABEL: testv16i32u:
+; AVX512CDBW:       ## BB#0:
+; AVX512CDBW-NEXT:    vplzcntd %zmm0, %zmm0
+; AVX512CDBW-NEXT:    retq
+;
+; AVX512BW-LABEL: testv16i32u:
+; AVX512BW:       ## BB#0:
+; AVX512BW-NEXT:    vpsrld $1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrld $2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrld $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrld $8, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrld $16, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandnq %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT:    vmovdqu8 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpxorq %zmm4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpxord %zmm1, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
+; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
+; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512DQ-LABEL: testv16i32u:
+; AVX512DQ:       ## BB#0:
+; AVX512DQ-NEXT:    vpsrld $1, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrld $2, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrld $4, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrld $8, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpsrld $16, %zmm0, %zmm1
+; AVX512DQ-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1
+; AVX512DQ-NEXT:    vpxorq %zmm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
+; AVX512DQ-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpxor %ymm3, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpunpckhdq {{.*#+}} ymm5 = ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[6],ymm3[6],ymm1[7],ymm3[7]
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm5, %ymm5
+; AVX512DQ-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[4],ymm3[4],ymm1[5],ymm3[5]
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpackuswb %ymm5, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm5
+; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512DQ-NEXT:    vpaddb %ymm5, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm3[2],ymm0[3],ymm3[3],ymm0[6],ymm3[6],ymm0[7],ymm3[7]
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm3[0],ymm0[1],ymm3[1],ymm0[4],ymm3[4],ymm0[5],ymm3[5]
+; AVX512DQ-NEXT:    vpsadbw %ymm3, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512DQ-NEXT:    retq
   %out = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %in, i1 -1)
   ret <16 x i32> %out
 }
@@ -52,20 +350,78 @@ define <32 x i16> @testv32i16(<32 x i16> %in) nounwind {
 ; AVX512CD-NEXT:    vpsubw %ymm2, %ymm1, %ymm1
 ; AVX512CD-NEXT:    retq
 ;
+; AVX512CDBW-LABEL: testv32i16:
+; AVX512CDBW:       ## BB#0:
+; AVX512CDBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512CDBW-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm1, %zmm1
+; AVX512CDBW-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512CDBW-NEXT:    vmovdqa {{.*#+}} ymm2 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512CDBW-NEXT:    vpsubw %ymm2, %ymm1, %ymm1
+; AVX512CDBW-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm0, %zmm0
+; AVX512CDBW-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512CDBW-NEXT:    vpsubw %ymm2, %ymm0, %ymm0
+; AVX512CDBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT:    retq
+;
 ; AVX512BW-LABEL: testv32i16:
 ; AVX512BW:       ## BB#0:
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BW-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
-; AVX512BW-NEXT:    vplzcntd %zmm1, %zmm1
-; AVX512BW-NEXT:    vpmovdw %zmm1, %ymm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %ymm2, %ymm1, %ymm1
-; AVX512BW-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512BW-NEXT:    vplzcntd %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsubw %ymm2, %ymm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandnq %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT:    vmovdqu8 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpxorq %zmm4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllw $8, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpaddb %zmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
+;
+; AVX512DQ-LABEL: testv32i16:
+; AVX512DQ:       ## BB#0:
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm5
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm5, %ymm5
+; AVX512DQ-NEXT:    vpxor %ymm6, %ymm6, %ymm6
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm5, %ymm7
+; AVX512DQ-NEXT:    vpand %ymm7, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
+; AVX512DQ-NEXT:    vpaddb %ymm5, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm0, %ymm3, %ymm0
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpaddw %ymm0, %ymm3, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm5
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm5, %ymm2
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm2, %ymm5
+; AVX512DQ-NEXT:    vpand %ymm5, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm4, %ymm2
+; AVX512DQ-NEXT:    vpaddb %ymm2, %ymm3, %ymm2
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpaddw %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    retq
   %out = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %in, i1 0)
   ret <32 x i16> %out
 }
@@ -84,20 +440,78 @@ define <32 x i16> @testv32i16u(<32 x i16> %in) nounwind {
 ; AVX512CD-NEXT:    vpsubw %ymm2, %ymm1, %ymm1
 ; AVX512CD-NEXT:    retq
 ;
+; AVX512CDBW-LABEL: testv32i16u:
+; AVX512CDBW:       ## BB#0:
+; AVX512CDBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512CDBW-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm1, %zmm1
+; AVX512CDBW-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512CDBW-NEXT:    vmovdqa {{.*#+}} ymm2 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512CDBW-NEXT:    vpsubw %ymm2, %ymm1, %ymm1
+; AVX512CDBW-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm0, %zmm0
+; AVX512CDBW-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512CDBW-NEXT:    vpsubw %ymm2, %ymm0, %ymm0
+; AVX512CDBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT:    retq
+;
 ; AVX512BW-LABEL: testv32i16u:
 ; AVX512BW:       ## BB#0:
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BW-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
-; AVX512BW-NEXT:    vplzcntd %zmm1, %zmm1
-; AVX512BW-NEXT:    vpmovdw %zmm1, %ymm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %ymm2, %ymm1, %ymm1
-; AVX512BW-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512BW-NEXT:    vplzcntd %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsubw %ymm2, %ymm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandnq %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT:    vmovdqu8 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpxorq %zmm4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllw $8, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpaddb %zmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
+;
+; AVX512DQ-LABEL: testv32i16u:
+; AVX512DQ:       ## BB#0:
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm5
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm5, %ymm5
+; AVX512DQ-NEXT:    vpxor %ymm6, %ymm6, %ymm6
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm5, %ymm7
+; AVX512DQ-NEXT:    vpand %ymm7, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
+; AVX512DQ-NEXT:    vpaddb %ymm5, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm0, %ymm3, %ymm0
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpaddw %ymm0, %ymm3, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm5
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm5, %ymm2
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm2, %ymm5
+; AVX512DQ-NEXT:    vpand %ymm5, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm4, %ymm2
+; AVX512DQ-NEXT:    vpaddb %ymm2, %ymm3, %ymm2
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpaddw %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    retq
   %out = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %in, i1 -1)
   ret <32 x i16> %out
 }
@@ -128,32 +542,78 @@ define <64 x i8> @testv64i8(<64 x i8> %in) nounwind {
 ; AVX512CD-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
 ; AVX512CD-NEXT:    retq
 ;
+; AVX512CDBW-LABEL: testv64i8:
+; AVX512CDBW:       ## BB#0:
+; AVX512CDBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512CDBW-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX512CDBW-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm2, %zmm2
+; AVX512CDBW-NEXT:    vpmovdb %zmm2, %xmm2
+; AVX512CDBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [24,24,24,24,24,24,24,24,24,24,24,24,24,24,24,24]
+; AVX512CDBW-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
+; AVX512CDBW-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm1, %zmm1
+; AVX512CDBW-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512CDBW-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
+; AVX512CDBW-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512CDBW-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX512CDBW-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm2, %zmm2
+; AVX512CDBW-NEXT:    vpmovdb %zmm2, %xmm2
+; AVX512CDBW-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
+; AVX512CDBW-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm0, %zmm0
+; AVX512CDBW-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512CDBW-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
+; AVX512CDBW-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX512CDBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT:    retq
+;
 ; AVX512BW-LABEL: testv64i8:
 ; AVX512BW:       ## BB#0:
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512BW-NEXT:    vplzcntd %zmm2, %zmm2
-; AVX512BW-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [24,24,24,24,24,24,24,24,24,24,24,24,24,24,24,24]
-; AVX512BW-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
-; AVX512BW-NEXT:    vplzcntd %zmm1, %zmm1
-; AVX512BW-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512BW-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
-; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512BW-NEXT:    vplzcntd %zmm2, %zmm2
-; AVX512BW-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512BW-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512BW-NEXT:    vplzcntd %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512BW-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
-; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandnq %zmm2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vmovdqu8 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT:    vpshufb %zmm1, %zmm3, %zmm1
+; AVX512BW-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpxorq %zmm4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
+;
+; AVX512DQ-LABEL: testv64i8:
+; AVX512DQ:       ## BB#0:
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpxor %ymm5, %ymm5, %ymm5
+; AVX512DQ-NEXT:    vpcmpeqb %ymm5, %ymm0, %ymm6
+; AVX512DQ-NEXT:    vpand %ymm6, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512DQ-NEXT:    vpaddb %ymm0, %ymm3, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm2
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm3, %ymm2
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
+; AVX512DQ-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    retq
   %out = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %in, i1 0)
   ret <64 x i8> %out
 }
@@ -184,32 +644,78 @@ define <64 x i8> @testv64i8u(<64 x i8> %in) nounwind {
 ; AVX512CD-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
 ; AVX512CD-NEXT:    retq
 ;
+; AVX512CDBW-LABEL: testv64i8u:
+; AVX512CDBW:       ## BB#0:
+; AVX512CDBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512CDBW-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX512CDBW-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm2, %zmm2
+; AVX512CDBW-NEXT:    vpmovdb %zmm2, %xmm2
+; AVX512CDBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [24,24,24,24,24,24,24,24,24,24,24,24,24,24,24,24]
+; AVX512CDBW-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
+; AVX512CDBW-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm1, %zmm1
+; AVX512CDBW-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512CDBW-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
+; AVX512CDBW-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
+; AVX512CDBW-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX512CDBW-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm2, %zmm2
+; AVX512CDBW-NEXT:    vpmovdb %zmm2, %xmm2
+; AVX512CDBW-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
+; AVX512CDBW-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512CDBW-NEXT:    vplzcntd %zmm0, %zmm0
+; AVX512CDBW-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512CDBW-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
+; AVX512CDBW-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX512CDBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT:    retq
+;
 ; AVX512BW-LABEL: testv64i8u:
 ; AVX512BW:       ## BB#0:
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512BW-NEXT:    vplzcntd %zmm2, %zmm2
-; AVX512BW-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [24,24,24,24,24,24,24,24,24,24,24,24,24,24,24,24]
-; AVX512BW-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
-; AVX512BW-NEXT:    vplzcntd %zmm1, %zmm1
-; AVX512BW-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512BW-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
-; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512BW-NEXT:    vplzcntd %zmm2, %zmm2
-; AVX512BW-NEXT:    vpmovdb %zmm2, %xmm2
-; AVX512BW-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512BW-NEXT:    vplzcntd %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512BW-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
-; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandnq %zmm2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vmovdqu8 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT:    vpshufb %zmm1, %zmm3, %zmm1
+; AVX512BW-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpxorq %zmm4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
+;
+; AVX512DQ-LABEL: testv64i8u:
+; AVX512DQ:       ## BB#0:
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpxor %ymm5, %ymm5, %ymm5
+; AVX512DQ-NEXT:    vpcmpeqb %ymm5, %ymm0, %ymm6
+; AVX512DQ-NEXT:    vpand %ymm6, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512DQ-NEXT:    vpaddb %ymm0, %ymm3, %ymm0
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm2
+; AVX512DQ-NEXT:    vpand %ymm2, %ymm3, %ymm2
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
+; AVX512DQ-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    retq
   %out = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %in, i1 -1)
   ret <64 x i8> %out
 }