[X86][TBM] Add tests showing failure to fold RFLAGS result into TBM instructions.

And fails to select TBM instructions at all.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@310790 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/test/CodeGen/X86/tbm-intrinsics-x86_64.ll b/test/CodeGen/X86/tbm-intrinsics-x86_64.ll
index 5a8c1bb5a107a34587a1f4504f3e9457bd940856..73d334452d46b98712e134ca621b8cae096e09b1 100644 (file)
--- a/test/CodeGen/X86/tbm-intrinsics-x86_64.ll
+++ b/test/CodeGen/X86/tbm-intrinsics-x86_64.ll
@@ -24,6 +24,20 @@ entry:
   ret i32 %0
 }
 
+define i32 @test_x86_tbm_bextri_u32_z(i32 %a, i32 %b) nounwind readonly {
+; CHECK-LABEL: test_x86_tbm_bextri_u32_z:
+; CHECK:       # BB#0: # %entry
+; CHECK-NEXT:    bextr $2814, %edi, %eax # imm = 0xAFE
+; CHECK-NEXT:    testl %eax, %eax
+; CHECK-NEXT:    cmovel %esi, %eax
+; CHECK-NEXT:    retq
+entry:
+  %0 = tail call i32 @llvm.x86.tbm.bextri.u32(i32 %a, i32 2814)
+  %1 = icmp eq i32 %0, 0
+  %2 = select i1 %1, i32 %b, i32 %0
+  ret i32 %2
+}
+
 define i64 @test_x86_tbm_bextri_u64(i64 %a) nounwind readnone {
 ; CHECK-LABEL: test_x86_tbm_bextri_u64:
 ; CHECK:       # BB#0: # %entry
@@ -46,3 +60,17 @@ entry:
   %0 = tail call i64 @llvm.x86.tbm.bextri.u64(i64 %tmp1, i64 2814)
   ret i64 %0
 }
+
+define i64 @test_x86_tbm_bextri_u64_z(i64 %a, i64 %b) nounwind readnone {
+; CHECK-LABEL: test_x86_tbm_bextri_u64_z:
+; CHECK:       # BB#0: # %entry
+; CHECK-NEXT:    bextr $2814, %rdi, %rax # imm = 0xAFE
+; CHECK-NEXT:    testq %rax, %rax
+; CHECK-NEXT:    cmoveq %rsi, %rax
+; CHECK-NEXT:    retq
+entry:
+  %0 = tail call i64 @llvm.x86.tbm.bextri.u64(i64 %a, i64 2814)
+  %1 = icmp eq i64 %0, 0
+  %2 = select i1 %1, i64 %b, i64 %0
+  ret i64 %2
+}

diff --git a/test/CodeGen/X86/tbm_patterns.ll b/test/CodeGen/X86/tbm_patterns.ll
index f110bd538bf5f84575dfefdb8644bfdc0db814ab..0bfd2c80d56ef949f246638867b3fb4e608cf75b 100644 (file)
--- a/test/CodeGen/X86/tbm_patterns.ll
+++ b/test/CodeGen/X86/tbm_patterns.ll
@@ -1,6 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+tbm < %s | FileCheck %s
 
+; TODO - Patterns fail to fold with ZF flags and prevents TBM instruction selection.
+
 define i32 @test_x86_tbm_bextri_u32(i32 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_bextri_u32:
 ; CHECK:       # BB#0:
@@ -22,6 +24,21 @@ define i32 @test_x86_tbm_bextri_u32_m(i32* nocapture %a) nounwind {
   ret i32 %t2
 }
 
+define i32 @test_x86_tbm_bextri_u32_z(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_bextri_u32_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    shrl $4, %edi
+; CHECK-NEXT:    andl $4095, %edi # imm = 0xFFF
+; CHECK-NEXT:    cmovel %esi, %edi
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    retq
+  %t0 = lshr i32 %a, 4
+  %t1 = and i32 %t0, 4095
+  %t2 = icmp eq i32 %t1, 0
+  %t3 = select i1 %t2, i32 %b, i32 %t1
+  ret i32 %t3
+}
+
 define i64 @test_x86_tbm_bextri_u64(i64 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_bextri_u64:
 ; CHECK:       # BB#0:
@@ -43,6 +60,21 @@ define i64 @test_x86_tbm_bextri_u64_m(i64* nocapture %a) nounwind {
   ret i64 %t2
 }
 
+define i64 @test_x86_tbm_bextri_u64_z(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_bextri_u64_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    shrl $4, %edi
+; CHECK-NEXT:    andl $4095, %edi # imm = 0xFFF
+; CHECK-NEXT:    cmoveq %rsi, %rdi
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    retq
+  %t0 = lshr i64 %a, 4
+  %t1 = and i64 %t0, 4095
+  %t2 = icmp eq i64 %t1, 0
+  %t3 = select i1 %t2, i64 %b, i64 %t1
+  ret i64 %t3
+}
+
 define i32 @test_x86_tbm_blcfill_u32(i32 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blcfill_u32:
 ; CHECK:       # BB#0:
@@ -53,6 +85,21 @@ define i32 @test_x86_tbm_blcfill_u32(i32 %a) nounwind {
   ret i32 %t1
 }
 
+define i32 @test_x86_tbm_blcfill_u32_z(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blcfill_u32_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; CHECK-NEXT:    leal 1(%rdi), %eax
+; CHECK-NEXT:    andl %edi, %eax
+; CHECK-NEXT:    cmovel %esi, %eax
+; CHECK-NEXT:    retq
+  %t0 = add i32 %a, 1
+  %t1 = and i32 %t0, %a
+  %t2 = icmp eq i32 %t1, 0
+  %t3 = select i1 %t2, i32 %b, i32 %t1
+  ret i32 %t3
+}
+
 define i64 @test_x86_tbm_blcfill_u64(i64 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blcfill_u64:
 ; CHECK:       # BB#0:
@@ -63,6 +110,20 @@ define i64 @test_x86_tbm_blcfill_u64(i64 %a) nounwind {
   ret i64 %t1
 }
 
+define i64 @test_x86_tbm_blcfill_u64_z(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blcfill_u64_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    leaq 1(%rdi), %rax
+; CHECK-NEXT:    andq %rdi, %rax
+; CHECK-NEXT:    cmoveq %rsi, %rax
+; CHECK-NEXT:    retq
+  %t0 = add i64 %a, 1
+  %t1 = and i64 %t0, %a
+  %t2 = icmp eq i64 %t1, 0
+  %t3 = select i1 %t2, i64 %b, i64 %t1
+  ret i64 %t3
+}
+
 define i32 @test_x86_tbm_blci_u32(i32 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blci_u32:
 ; CHECK:       # BB#0:
@@ -74,6 +135,23 @@ define i32 @test_x86_tbm_blci_u32(i32 %a) nounwind {
   ret i32 %t2
 }
 
+define i32 @test_x86_tbm_blci_u32_z(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blci_u32_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; CHECK-NEXT:    leal 1(%rdi), %eax
+; CHECK-NEXT:    notl %eax
+; CHECK-NEXT:    orl %edi, %eax
+; CHECK-NEXT:    cmovel %esi, %eax
+; CHECK-NEXT:    retq
+  %t0 = add i32 1, %a
+  %t1 = xor i32 %t0, -1
+  %t2 = or i32 %t1, %a
+  %t3 = icmp eq i32 %t2, 0
+  %t4 = select i1 %t3, i32 %b, i32 %t2
+  ret i32 %t4
+}
+
 define i64 @test_x86_tbm_blci_u64(i64 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blci_u64:
 ; CHECK:       # BB#0:
@@ -85,6 +163,22 @@ define i64 @test_x86_tbm_blci_u64(i64 %a) nounwind {
   ret i64 %t2
 }
 
+define i64 @test_x86_tbm_blci_u64_z(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blci_u64_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    leaq 1(%rdi), %rax
+; CHECK-NEXT:    notq %rax
+; CHECK-NEXT:    orq %rdi, %rax
+; CHECK-NEXT:    cmoveq %rsi, %rax
+; CHECK-NEXT:    retq
+  %t0 = add i64 1, %a
+  %t1 = xor i64 %t0, -1
+  %t2 = or i64 %t1, %a
+  %t3 = icmp eq i64 %t2, 0
+  %t4 = select i1 %t3, i64 %b, i64 %t2
+  ret i64 %t4
+}
+
 define i32 @test_x86_tbm_blci_u32_b(i32 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blci_u32_b:
 ; CHECK:       # BB#0:
@@ -116,6 +210,24 @@ define i32 @test_x86_tbm_blcic_u32(i32 %a) nounwind {
   ret i32 %t2
 }
 
+define i32 @test_x86_tbm_blcic_u32_z(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blcic_u32_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; CHECK-NEXT:    leal 1(%rdi), %eax
+; CHECK-NEXT:    movl %edi, %ecx
+; CHECK-NEXT:    notl %ecx
+; CHECK-NEXT:    andl %ecx, %eax
+; CHECK-NEXT:    cmovel %esi, %eax
+; CHECK-NEXT:    retq
+  %t0 = xor i32 %a, -1
+  %t1 = add i32 %a, 1
+  %t2 = and i32 %t1, %t0
+  %t3 = icmp eq i32 %t2, 0
+  %t4 = select i1 %t3, i32 %b, i32 %t2
+  ret i32 %t4
+}
+
 define i64 @test_x86_tbm_blcic_u64(i64 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blcic_u64:
 ; CHECK:       # BB#0:
@@ -127,6 +239,22 @@ define i64 @test_x86_tbm_blcic_u64(i64 %a) nounwind {
   ret i64 %t2
 }
 
+define i64 @test_x86_tbm_blcic_u64_z(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blcic_u64_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    leaq 1(%rdi), %rax
+; CHECK-NEXT:    notq %rdi
+; CHECK-NEXT:    andq %rdi, %rax
+; CHECK-NEXT:    cmoveq %rsi, %rax
+; CHECK-NEXT:    retq
+  %t0 = xor i64 %a, -1
+  %t1 = add i64 %a, 1
+  %t2 = and i64 %t1, %t0
+  %t3 = icmp eq i64 %t2, 0
+  %t4 = select i1 %t3, i64 %b, i64 %t2
+  ret i64 %t4
+}
+
 define i32 @test_x86_tbm_blcmsk_u32(i32 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blcmsk_u32:
 ; CHECK:       # BB#0:
@@ -137,6 +265,21 @@ define i32 @test_x86_tbm_blcmsk_u32(i32 %a) nounwind {
   ret i32 %t1
 }
 
+define i32 @test_x86_tbm_blcmsk_u32_z(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blcmsk_u32_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; CHECK-NEXT:    leal 1(%rdi), %eax
+; CHECK-NEXT:    xorl %edi, %eax
+; CHECK-NEXT:    cmovel %esi, %eax
+; CHECK-NEXT:    retq
+  %t0 = add i32 %a, 1
+  %t1 = xor i32 %t0, %a
+  %t2 = icmp eq i32 %t1, 0
+  %t3 = select i1 %t2, i32 %b, i32 %t1
+  ret i32 %t3
+}
+
 define i64 @test_x86_tbm_blcmsk_u64(i64 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blcmsk_u64:
 ; CHECK:       # BB#0:
@@ -147,6 +290,20 @@ define i64 @test_x86_tbm_blcmsk_u64(i64 %a) nounwind {
   ret i64 %t1
 }
 
+define i64 @test_x86_tbm_blcmsk_u64_z(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blcmsk_u64_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    leaq 1(%rdi), %rax
+; CHECK-NEXT:    xorq %rdi, %rax
+; CHECK-NEXT:    cmoveq %rsi, %rax
+; CHECK-NEXT:    retq
+  %t0 = add i64 %a, 1
+  %t1 = xor i64 %t0, %a
+  %t2 = icmp eq i64 %t1, 0
+  %t3 = select i1 %t2, i64 %b, i64 %t1
+  ret i64 %t3
+}
+
 define i32 @test_x86_tbm_blcs_u32(i32 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blcs_u32:
 ; CHECK:       # BB#0:
@@ -157,6 +314,21 @@ define i32 @test_x86_tbm_blcs_u32(i32 %a) nounwind {
   ret i32 %t1
 }
 
+define i32 @test_x86_tbm_blcs_u32_z(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blcs_u32_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; CHECK-NEXT:    leal 1(%rdi), %eax
+; CHECK-NEXT:    orl %edi, %eax
+; CHECK-NEXT:    cmovel %esi, %eax
+; CHECK-NEXT:    retq
+  %t0 = add i32 %a, 1
+  %t1 = or i32 %t0, %a
+  %t2 = icmp eq i32 %t1, 0
+  %t3 = select i1 %t2, i32 %b, i32 %t1
+  ret i32 %t3
+}
+
 define i64 @test_x86_tbm_blcs_u64(i64 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blcs_u64:
 ; CHECK:       # BB#0:
@@ -167,6 +339,20 @@ define i64 @test_x86_tbm_blcs_u64(i64 %a) nounwind {
   ret i64 %t1
 }
 
+define i64 @test_x86_tbm_blcs_u64_z(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blcs_u64_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    leaq 1(%rdi), %rax
+; CHECK-NEXT:    orq %rdi, %rax
+; CHECK-NEXT:    cmoveq %rsi, %rax
+; CHECK-NEXT:    retq
+  %t0 = add i64 %a, 1
+  %t1 = or i64 %t0, %a
+  %t2 = icmp eq i64 %t1, 0
+  %t3 = select i1 %t2, i64 %b, i64 %t1
+  ret i64 %t3
+}
+
 define i32 @test_x86_tbm_blsfill_u32(i32 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blsfill_u32:
 ; CHECK:       # BB#0:
@@ -177,6 +363,21 @@ define i32 @test_x86_tbm_blsfill_u32(i32 %a) nounwind {
   ret i32 %t1
 }
 
+define i32 @test_x86_tbm_blsfill_u32_z(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blsfill_u32_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; CHECK-NEXT:    leal -1(%rdi), %eax
+; CHECK-NEXT:    orl %edi, %eax
+; CHECK-NEXT:    cmovel %esi, %eax
+; CHECK-NEXT:    retq
+  %t0 = add i32 %a, -1
+  %t1 = or i32 %t0, %a
+  %t2 = icmp eq i32 %t1, 0
+  %t3 = select i1 %t2, i32 %b, i32 %t1
+  ret i32 %t3
+}
+
 define i64 @test_x86_tbm_blsfill_u64(i64 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blsfill_u64:
 ; CHECK:       # BB#0:
@@ -187,6 +388,20 @@ define i64 @test_x86_tbm_blsfill_u64(i64 %a) nounwind {
   ret i64 %t1
 }
 
+define i64 @test_x86_tbm_blsfill_u64_z(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blsfill_u64_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    leaq -1(%rdi), %rax
+; CHECK-NEXT:    orq %rdi, %rax
+; CHECK-NEXT:    cmoveq %rsi, %rax
+; CHECK-NEXT:    retq
+  %t0 = add i64 %a, -1
+  %t1 = or i64 %t0, %a
+  %t2 = icmp eq i64 %t1, 0
+  %t3 = select i1 %t2, i64 %b, i64 %t1
+  ret i64 %t3
+}
+
 define i32 @test_x86_tbm_blsic_u32(i32 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blsic_u32:
 ; CHECK:       # BB#0:
@@ -198,6 +413,24 @@ define i32 @test_x86_tbm_blsic_u32(i32 %a) nounwind {
   ret i32 %t2
 }
 
+define i32 @test_x86_tbm_blsic_u32_z(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blsic_u32_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    notl %eax
+; CHECK-NEXT:    decl %edi
+; CHECK-NEXT:    orl %eax, %edi
+; CHECK-NEXT:    cmovel %esi, %edi
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    retq
+  %t0 = xor i32 %a, -1
+  %t1 = add i32 %a, -1
+  %t2 = or i32 %t0, %t1
+  %t3 = icmp eq i32 %t2, 0
+  %t4 = select i1 %t3, i32 %b, i32 %t2
+  ret i32 %t4
+}
+
 define i64 @test_x86_tbm_blsic_u64(i64 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_blsic_u64:
 ; CHECK:       # BB#0:
@@ -209,6 +442,24 @@ define i64 @test_x86_tbm_blsic_u64(i64 %a) nounwind {
   ret i64 %t2
 }
 
+define i64 @test_x86_tbm_blsic_u64_z(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_blsic_u64_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    notq %rax
+; CHECK-NEXT:    decq %rdi
+; CHECK-NEXT:    orq %rax, %rdi
+; CHECK-NEXT:    cmoveq %rsi, %rdi
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    retq
+  %t0 = xor i64 %a, -1
+  %t1 = add i64 %a, -1
+  %t2 = or i64 %t0, %t1
+  %t3 = icmp eq i64 %t2, 0
+  %t4 = select i1 %t3, i64 %b, i64 %t2
+  ret i64 %t4
+}
+
 define i32 @test_x86_tbm_t1mskc_u32(i32 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_t1mskc_u32:
 ; CHECK:       # BB#0:
@@ -220,8 +471,26 @@ define i32 @test_x86_tbm_t1mskc_u32(i32 %a) nounwind {
   ret i32 %t2
 }
 
-define i64 @Ttest_x86_tbm_t1mskc_u64(i64 %a) nounwind {
-; CHECK-LABEL: Ttest_x86_tbm_t1mskc_u64:
+define i32 @test_x86_tbm_t1mskc_u32_z(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_t1mskc_u32_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    notl %eax
+; CHECK-NEXT:    incl %edi
+; CHECK-NEXT:    orl %eax, %edi
+; CHECK-NEXT:    cmovel %esi, %edi
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    retq
+  %t0 = xor i32 %a, -1
+  %t1 = add i32 %a, 1
+  %t2 = or i32 %t0, %t1
+  %t3 = icmp eq i32 %t2, 0
+  %t4 = select i1 %t3, i32 %b, i32 %t2
+  ret i32 %t4
+}
+
+define i64 @test_x86_tbm_t1mskc_u64(i64 %a) nounwind {
+; CHECK-LABEL: test_x86_tbm_t1mskc_u64:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    t1mskc %rdi, %rax
 ; CHECK-NEXT:    retq
@@ -231,6 +500,24 @@ define i64 @Ttest_x86_tbm_t1mskc_u64(i64 %a) nounwind {
   ret i64 %t2
 }
 
+define i64 @test_x86_tbm_t1mskc_u64_z(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_t1mskc_u64_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    notq %rax
+; CHECK-NEXT:    incq %rdi
+; CHECK-NEXT:    orq %rax, %rdi
+; CHECK-NEXT:    cmoveq %rsi, %rdi
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    retq
+  %t0 = xor i64 %a, -1
+  %t1 = add i64 %a, 1
+  %t2 = or i64 %t0, %t1
+  %t3 = icmp eq i64 %t2, 0
+  %t4 = select i1 %t3, i64 %b, i64 %t2
+  ret i64 %t4
+}
+
 define i32 @test_x86_tbm_tzmsk_u32(i32 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_tzmsk_u32:
 ; CHECK:       # BB#0:
@@ -242,6 +529,24 @@ define i32 @test_x86_tbm_tzmsk_u32(i32 %a) nounwind {
   ret i32 %t2
 }
 
+define i32 @test_x86_tbm_tzmsk_u32_z(i32 %a, i32 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_tzmsk_u32_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    notl %eax
+; CHECK-NEXT:    decl %edi
+; CHECK-NEXT:    andl %eax, %edi
+; CHECK-NEXT:    cmovel %esi, %edi
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    retq
+  %t0 = xor i32 %a, -1
+  %t1 = add i32 %a, -1
+  %t2 = and i32 %t0, %t1
+  %t3 = icmp eq i32 %t2, 0
+  %t4 = select i1 %t3, i32 %b, i32 %t2
+  ret i32 %t4
+}
+
 define i64 @test_x86_tbm_tzmsk_u64(i64 %a) nounwind {
 ; CHECK-LABEL: test_x86_tbm_tzmsk_u64:
 ; CHECK:       # BB#0:
@@ -253,6 +558,24 @@ define i64 @test_x86_tbm_tzmsk_u64(i64 %a) nounwind {
   ret i64 %t2
 }
 
+define i64 @test_x86_tbm_tzmsk_u64_z(i64 %a, i64 %b) nounwind {
+; CHECK-LABEL: test_x86_tbm_tzmsk_u64_z:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    notq %rax
+; CHECK-NEXT:    decq %rdi
+; CHECK-NEXT:    andq %rax, %rdi
+; CHECK-NEXT:    cmoveq %rsi, %rdi
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    retq
+  %t0 = xor i64 %a, -1
+  %t1 = add i64 %a, -1
+  %t2 = and i64 %t0, %t1
+  %t3 = icmp eq i64 %t2, 0
+  %t4 = select i1 %t3, i64 %b, i64 %t2
+  ret i64 %t4
+}
+
 define i64 @test_and_large_constant_mask(i64 %x) {
 ; CHECK-LABEL: test_and_large_constant_mask:
 ; CHECK:       # BB#0: # %entry