[CodeGen] Add missing vector type legalization for ctlz_zero_undef

[RolandF]

Summary

The change to PPC vector cost model for D60160 exposed a regression where the support for legalizing a vector ctlz_zero_undef with illegal type was missing.

[craig.topper]

Looks like we can hit the same issue in X86 with -x86-experimental-vector-widening-legalization on the command line. Do you mind adding a test for that?

[RolandF]

Add X86 test.

[RKSimon]

llvm/test/CodeGen/X86/vec_clz.ll

2

Please can you use update_llc_test_checks.py to regenerate this? (and maybe the ppc file as well)

[RolandF]

Generated test checks with script.

[RKSimon]

LGTM

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

#endif
   // (they are probably not destined to become libcalls), then widen those like
   // any other unary ops.
   LLVM_FALLTHROUGH;
 
   case ISD::ABS:
   case ISD::BITREVERSE:
   case ISD::BSWAP:
   case ISD::CTLZ:
+  case ISD::CTLZ_ZERO_UNDEF:
   case ISD::CTPOP:
   case ISD::CTTZ:
+  case ISD::CTTZ_ZERO_UNDEF:
   case ISD::FNEG:
   case ISD::FCANONICALIZE:
     Res = WidenVecRes_Unary(N);
     break;
   case ISD::FMA:
     Res = WidenVecRes_Ternary(N);
     break;
   }

llvm/test/CodeGen/PowerPC/vec_clz.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; Check the vctlz* instructions that were added in P8
 ; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr8 < %s | FileCheck %s
-; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr8 -mattr=-vsx < %s | FileCheck %s
+; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr8 -mattr=-vsx < %s | FileCheck %s --check-prefix=CHECK-NOVSX
 
 declare <16 x i8> @llvm.ctlz.v16i8(<16 x i8>) nounwind readnone
 declare <8 x i16> @llvm.ctlz.v8i16(<8 x i16>) nounwind readnone
 declare <4 x i32> @llvm.ctlz.v4i32(<4 x i32>) nounwind readnone
 declare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>) nounwind readnone
 
 define <16 x i8> @test_v16i8(<16 x i8> %x) nounwind readnone {
+; CHECK-LABEL: test_v16i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vclzb 2, 2
+; CHECK-NEXT:    blr
+;
+; CHECK-NOVSX-LABEL: test_v16i8:
+; CHECK-NOVSX:       # %bb.0:
+; CHECK-NOVSX-NEXT:    vclzb 2, 2
+; CHECK-NOVSX-NEXT:    blr
        %vcnt = tail call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %x)
        ret <16 x i8> %vcnt
-; CHECK: @test_v16i8
-; CHECK: vclzb 2, 2
-; CHECK: blr
 }
 
 define <8 x i16> @test_v8i16(<8 x i16> %x) nounwind readnone {
+; CHECK-LABEL: test_v8i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vclzh 2, 2
+; CHECK-NEXT:    blr
+;
+; CHECK-NOVSX-LABEL: test_v8i16:
+; CHECK-NOVSX:       # %bb.0:
+; CHECK-NOVSX-NEXT:    vclzh 2, 2
+; CHECK-NOVSX-NEXT:    blr
        %vcnt = tail call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %x)
        ret <8 x i16> %vcnt
-; CHECK: @test_v8i16
-; CHECK: vclzh 2, 2
-; CHECK: blr
 }
 
 define <4 x i32> @test_v4i32(<4 x i32> %x) nounwind readnone {
+; CHECK-LABEL: test_v4i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vclzw 2, 2
+; CHECK-NEXT:    blr
+;
+; CHECK-NOVSX-LABEL: test_v4i32:
+; CHECK-NOVSX:       # %bb.0:
+; CHECK-NOVSX-NEXT:    vclzw 2, 2
+; CHECK-NOVSX-NEXT:    blr
        %vcnt = tail call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %x)
        ret <4 x i32> %vcnt
-; CHECK: @test_v4i32
-; CHECK: vclzw 2, 2
-; CHECK: blr
 }
 
 define <2 x i64> @test_v2i64(<2 x i64> %x) nounwind readnone {
+; CHECK-LABEL: test_v2i64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vclzd 2, 2
+; CHECK-NEXT:    blr
+;
+; CHECK-NOVSX-LABEL: test_v2i64:
+; CHECK-NOVSX:       # %bb.0:
+; CHECK-NOVSX-NEXT:    vclzd 2, 2
+; CHECK-NOVSX-NEXT:    blr
        %vcnt = tail call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %x)
        ret <2 x i64> %vcnt
-; CHECK: @test_v2i64
-; CHECK: vclzd 2, 2
-; CHECK: blr
+}
+
+declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1 immarg)
+
+define <2 x i32> @illegal_ctlz(<2 x i32> %v1) {
+; CHECK-LABEL: illegal_ctlz:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vclzw 2, 2
+; CHECK-NEXT:    blr
+;
+; CHECK-NOVSX-LABEL: illegal_ctlz:
+; CHECK-NOVSX:       # %bb.0:
+; CHECK-NOVSX-NEXT:    vclzw 2, 2
+; CHECK-NOVSX-NEXT:    blr
+  %v2 = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %v1, i1 true)
+  ret <2 x i32> %v2
+}
+
+declare <2 x i32> @llvm.cttz.v2i32(<2 x i32>, i1 immarg)
+
+define <2 x i32> @illegal_cttz(<2 x i32> %v1) {
+; CHECK-LABEL: illegal_cttz:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vspltisw 3, 1
+; CHECK-NEXT:    vsubuwm 3, 2, 3
+; CHECK-NEXT:    xxlandc 34, 35, 34
+; CHECK-NEXT:    vpopcntw 2, 2
+; CHECK-NEXT:    blr
+;
+; CHECK-NOVSX-LABEL: illegal_cttz:
+; CHECK-NOVSX:       # %bb.0:
+; CHECK-NOVSX-NEXT:    vspltisw 3, 1
+; CHECK-NOVSX-NEXT:    vsubuwm 3, 2, 3
+; CHECK-NOVSX-NEXT:    vandc 2, 3, 2
+; CHECK-NOVSX-NEXT:    vpopcntw 2, 2
+; CHECK-NOVSX-NEXT:    blr
+  %v2 = call <2 x i32> @llvm.cttz.v2i32(<2 x i32> %v1, i1 true)
+  ret <2 x i32> %v2
 }

llvm/test/CodeGen/X86/vec_clz.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -x86-experimental-vector-widening-legalization | FileCheck %s
+
+declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1 immarg)
+
+define <2 x i32> @illegal_ctlz(<2 x i32> %v1) {
+; CHECK-LABEL: illegal_ctlz:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    psrld $1, %xmm1
+; CHECK-NEXT:    por %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    psrld $2, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    psrld $4, %xmm1
+; CHECK-NEXT:    por %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    psrld $8, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    psrld $16, %xmm1
+; CHECK-NEXT:    por %xmm0, %xmm1
+; CHECK-NEXT:    pcmpeqd %xmm2, %xmm2
+; CHECK-NEXT:    pxor %xmm1, %xmm2
+; CHECK-NEXT:    movdqa %xmm2, %xmm0
+; CHECK-NEXT:    psrlw $1, %xmm0
+; CHECK-NEXT:    pand {{.*}}(%rip), %xmm0
+; CHECK-NEXT:    psubb %xmm0, %xmm2
+; CHECK-NEXT:    movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; CHECK-NEXT:    movdqa %xmm2, %xmm1
+; CHECK-NEXT:    pand %xmm0, %xmm1
+; CHECK-NEXT:    psrlw $2, %xmm2
+; CHECK-NEXT:    pand %xmm0, %xmm2
+; CHECK-NEXT:    paddb %xmm1, %xmm2
+; CHECK-NEXT:    movdqa %xmm2, %xmm0
+; CHECK-NEXT:    psrlw $4, %xmm0
+; CHECK-NEXT:    paddb %xmm2, %xmm0
+; CHECK-NEXT:    pand {{.*}}(%rip), %xmm0
+; CHECK-NEXT:    pxor %xmm1, %xmm1
+; CHECK-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; CHECK-NEXT:    psadbw %xmm1, %xmm2
+; CHECK-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; CHECK-NEXT:    psadbw %xmm1, %xmm0
+; CHECK-NEXT:    packuswb %xmm2, %xmm0
+; CHECK-NEXT:    retq
+  %v2 = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %v1, i1 true)
+  ret <2 x i32> %v2
+}
+
+declare <2 x i32> @llvm.cttz.v2i32(<2 x i32>, i1 immarg)
+
+define <2 x i32> @illegal_cttz(<2 x i32> %v1) {
+; CHECK-LABEL: illegal_cttz:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    pcmpeqd %xmm1, %xmm1
+; CHECK-NEXT:    paddd %xmm0, %xmm1
+; CHECK-NEXT:    pandn %xmm1, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    psrlw $1, %xmm1
+; CHECK-NEXT:    pand {{.*}}(%rip), %xmm1
+; CHECK-NEXT:    psubb %xmm1, %xmm0
+; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; CHECK-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-NEXT:    pand %xmm1, %xmm2
+; CHECK-NEXT:    psrlw $2, %xmm0
+; CHECK-NEXT:    pand %xmm1, %xmm0
+; CHECK-NEXT:    paddb %xmm2, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    psrlw $4, %xmm1
+; CHECK-NEXT:    paddb %xmm0, %xmm1
+; CHECK-NEXT:    pand {{.*}}(%rip), %xmm1
+; CHECK-NEXT:    pxor %xmm0, %xmm0
+; CHECK-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; CHECK-NEXT:    psadbw %xmm0, %xmm2
+; CHECK-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; CHECK-NEXT:    psadbw %xmm0, %xmm1
+; CHECK-NEXT:    packuswb %xmm2, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    retq
+  %v2 = call <2 x i32> @llvm.cttz.v2i32(<2 x i32> %v1, i1 true)
+  ret <2 x i32> %v2
+}