[llvm] r365704 - [X86] Add a few more TBM and BLSI test cases that show the same issue that r352306 fixed for BLSR.

Craig Topper via llvm-commits llvm-commits at lists.llvm.org
Wed Jul 10 15:44:25 PDT 2019


Author: ctopper
Date: Wed Jul 10 15:44:24 2019
New Revision: 365704

URL: http://llvm.org/viewvc/llvm-project?rev=365704&view=rev
Log:
[X86] Add a few more TBM and BLSI test cases that show the same issue that r352306 fixed for BLSR.

Modified:
    llvm/trunk/test/CodeGen/X86/bmi.ll
    llvm/trunk/test/CodeGen/X86/tbm_patterns.ll

Modified: llvm/trunk/test/CodeGen/X86/bmi.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/bmi.ll?rev=365704&r1=365703&r2=365704&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/bmi.ll (original)
+++ llvm/trunk/test/CodeGen/X86/bmi.ll Wed Jul 10 15:44:24 2019
@@ -1148,25 +1148,127 @@ define i64 @blsr64_branch(i64 %x) {
   ret i64 %tmp2
 }
 
+define i32 @blsi32_branch(i32 %x) {
+; X86-LABEL: blsi32_branch:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    .cfi_def_cfa_offset 8
+; X86-NEXT:    .cfi_offset %esi, -8
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl %eax, %esi
+; X86-NEXT:    negl %esi
+; X86-NEXT:    andl %eax, %esi
+; X86-NEXT:    jne .LBB48_2
+; X86-NEXT:  # %bb.1:
+; X86-NEXT:    calll bar
+; X86-NEXT:  .LBB48_2:
+; X86-NEXT:    movl %esi, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    .cfi_def_cfa_offset 4
+; X86-NEXT:    retl
+;
+; X64-LABEL: blsi32_branch:
+; X64:       # %bb.0:
+; X64-NEXT:    pushq %rbx
+; X64-NEXT:    .cfi_def_cfa_offset 16
+; X64-NEXT:    .cfi_offset %rbx, -16
+; X64-NEXT:    movl %edi, %ebx
+; X64-NEXT:    negl %ebx
+; X64-NEXT:    andl %edi, %ebx
+; X64-NEXT:    jne .LBB48_2
+; X64-NEXT:  # %bb.1:
+; X64-NEXT:    callq bar
+; X64-NEXT:  .LBB48_2:
+; X64-NEXT:    movl %ebx, %eax
+; X64-NEXT:    popq %rbx
+; X64-NEXT:    .cfi_def_cfa_offset 8
+; X64-NEXT:    retq
+  %tmp = sub i32 0, %x
+  %tmp2 = and i32 %x, %tmp
+  %cmp = icmp eq i32 %tmp2, 0
+  br i1 %cmp, label %1, label %2
+
+  tail call void @bar()
+  br label %2
+  ret i32 %tmp2
+}
+
+define i64 @blsi64_branch(i64 %x) {
+; X86-LABEL: blsi64_branch:
+; X86:       # %bb.0:
+; X86-NEXT:    pushl %edi
+; X86-NEXT:    .cfi_def_cfa_offset 8
+; X86-NEXT:    pushl %esi
+; X86-NEXT:    .cfi_def_cfa_offset 12
+; X86-NEXT:    .cfi_offset %esi, -12
+; X86-NEXT:    .cfi_offset %edi, -8
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    xorl %esi, %esi
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    negl %edi
+; X86-NEXT:    sbbl %ecx, %esi
+; X86-NEXT:    andl %ecx, %esi
+; X86-NEXT:    andl %eax, %edi
+; X86-NEXT:    movl %edi, %eax
+; X86-NEXT:    orl %esi, %eax
+; X86-NEXT:    jne .LBB49_2
+; X86-NEXT:  # %bb.1:
+; X86-NEXT:    calll bar
+; X86-NEXT:  .LBB49_2:
+; X86-NEXT:    movl %edi, %eax
+; X86-NEXT:    movl %esi, %edx
+; X86-NEXT:    popl %esi
+; X86-NEXT:    .cfi_def_cfa_offset 8
+; X86-NEXT:    popl %edi
+; X86-NEXT:    .cfi_def_cfa_offset 4
+; X86-NEXT:    retl
+;
+; X64-LABEL: blsi64_branch:
+; X64:       # %bb.0:
+; X64-NEXT:    pushq %rbx
+; X64-NEXT:    .cfi_def_cfa_offset 16
+; X64-NEXT:    .cfi_offset %rbx, -16
+; X64-NEXT:    movq %rdi, %rbx
+; X64-NEXT:    negq %rbx
+; X64-NEXT:    andq %rdi, %rbx
+; X64-NEXT:    jne .LBB49_2
+; X64-NEXT:  # %bb.1:
+; X64-NEXT:    callq bar
+; X64-NEXT:  .LBB49_2:
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    popq %rbx
+; X64-NEXT:    .cfi_def_cfa_offset 8
+; X64-NEXT:    retq
+  %tmp = sub i64 0, %x
+  %tmp2 = and i64 %x, %tmp
+  %cmp = icmp eq i64 %tmp2, 0
+  br i1 %cmp, label %1, label %2
+
+  tail call void @bar()
+  br label %2
+  ret i64 %tmp2
+}
+
 declare void @bar()
 
 define void @pr42118_i32(i32 %x) {
 ; X86-LABEL: pr42118_i32:
 ; X86:       # %bb.0:
 ; X86-NEXT:    blsrl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    jne .LBB48_1
+; X86-NEXT:    jne .LBB50_1
 ; X86-NEXT:  # %bb.2:
 ; X86-NEXT:    jmp bar # TAILCALL
-; X86-NEXT:  .LBB48_1:
+; X86-NEXT:  .LBB50_1:
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: pr42118_i32:
 ; X64:       # %bb.0:
 ; X64-NEXT:    blsrl %edi, %eax
-; X64-NEXT:    jne .LBB48_1
+; X64-NEXT:    jne .LBB50_1
 ; X64-NEXT:  # %bb.2:
 ; X64-NEXT:    jmp bar # TAILCALL
-; X64-NEXT:  .LBB48_1:
+; X64-NEXT:  .LBB50_1:
 ; X64-NEXT:    retq
   %tmp = sub i32 0, %x
   %tmp1 = and i32 %tmp, %x
@@ -1194,12 +1296,12 @@ define void @pr42118_i64(i64 %x) {
 ; X86-NEXT:    andl %eax, %edx
 ; X86-NEXT:    andl %ecx, %esi
 ; X86-NEXT:    orl %edx, %esi
-; X86-NEXT:    jne .LBB49_1
+; X86-NEXT:    jne .LBB51_1
 ; X86-NEXT:  # %bb.2:
 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    .cfi_def_cfa_offset 4
 ; X86-NEXT:    jmp bar # TAILCALL
-; X86-NEXT:  .LBB49_1:
+; X86-NEXT:  .LBB51_1:
 ; X86-NEXT:    .cfi_def_cfa_offset 8
 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    .cfi_def_cfa_offset 4
@@ -1208,10 +1310,10 @@ define void @pr42118_i64(i64 %x) {
 ; X64-LABEL: pr42118_i64:
 ; X64:       # %bb.0:
 ; X64-NEXT:    blsrq %rdi, %rax
-; X64-NEXT:    jne .LBB49_1
+; X64-NEXT:    jne .LBB51_1
 ; X64-NEXT:  # %bb.2:
 ; X64-NEXT:    jmp bar # TAILCALL
-; X64-NEXT:  .LBB49_1:
+; X64-NEXT:  .LBB51_1:
 ; X64-NEXT:    retq
   %tmp = sub i64 0, %x
   %tmp1 = and i64 %tmp, %x
@@ -1229,11 +1331,11 @@ define i32 @blsi_cflag_32(i32 %x, i32 %y
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    testl %eax, %eax
-; X86-NEXT:    jne .LBB50_1
+; X86-NEXT:    jne .LBB52_1
 ; X86-NEXT:  # %bb.2:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    retl
-; X86-NEXT:  .LBB50_1:
+; X86-NEXT:  .LBB52_1:
 ; X86-NEXT:    blsil %eax, %eax
 ; X86-NEXT:    retl
 ;
@@ -1262,15 +1364,15 @@ define i64 @blsi_cflag_64(i64 %x, i64 %y
 ; X86-NEXT:    sbbl %esi, %edx
 ; X86-NEXT:    movl %ecx, %edi
 ; X86-NEXT:    orl %esi, %edi
-; X86-NEXT:    jne .LBB51_1
+; X86-NEXT:    jne .LBB53_1
 ; X86-NEXT:  # %bb.2:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    jmp .LBB51_3
-; X86-NEXT:  .LBB51_1:
+; X86-NEXT:    jmp .LBB53_3
+; X86-NEXT:  .LBB53_1:
 ; X86-NEXT:    andl %esi, %edx
 ; X86-NEXT:    andl %ecx, %eax
-; X86-NEXT:  .LBB51_3:
+; X86-NEXT:  .LBB53_3:
 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    retl

Modified: llvm/trunk/test/CodeGen/X86/tbm_patterns.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/tbm_patterns.ll?rev=365704&r1=365703&r2=365704&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/tbm_patterns.ll (original)
+++ llvm/trunk/test/CodeGen/X86/tbm_patterns.ll Wed Jul 10 15:44:24 2019
@@ -1,8 +1,6 @@
 ; 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:
@@ -873,3 +871,113 @@ define  i64 @masked_blcic(i64) {
   %5 = and i64 %4, %3
   ret i64 %5
 }
+
+define i32 @blcic32_branch(i32 %x) nounwind {
+; CHECK-LABEL: blcic32_branch:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    pushq %rbx
+; CHECK-NEXT:    movl %edi, %ebx
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    notl %eax
+; CHECK-NEXT:    incl %ebx
+; CHECK-NEXT:    andl %eax, %ebx
+; CHECK-NEXT:    jne .LBB69_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    callq bar
+; CHECK-NEXT:  .LBB69_2:
+; CHECK-NEXT:    movl %ebx, %eax
+; CHECK-NEXT:    popq %rbx
+; CHECK-NEXT:    retq
+  %tmp = xor i32 %x, -1
+  %tmp2 = add i32 %x, 1
+  %tmp3 = and i32 %tmp, %tmp2
+  %cmp = icmp eq i32 %tmp3, 0
+  br i1 %cmp, label %1, label %2
+
+  tail call void @bar()
+  br label %2
+  ret i32 %tmp3
+}
+
+define i64 @blcic64_branch(i64 %x) nounwind {
+; CHECK-LABEL: blcic64_branch:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    pushq %rbx
+; CHECK-NEXT:    movq %rdi, %rbx
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    notq %rax
+; CHECK-NEXT:    incq %rbx
+; CHECK-NEXT:    andq %rax, %rbx
+; CHECK-NEXT:    jne .LBB70_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    callq bar
+; CHECK-NEXT:  .LBB70_2:
+; CHECK-NEXT:    movq %rbx, %rax
+; CHECK-NEXT:    popq %rbx
+; CHECK-NEXT:    retq
+  %tmp = xor i64 %x, -1
+  %tmp2 = add i64 %x, 1
+  %tmp3 = and i64 %tmp, %tmp2
+  %cmp = icmp eq i64 %tmp3, 0
+  br i1 %cmp, label %1, label %2
+
+  tail call void @bar()
+  br label %2
+  ret i64 %tmp3
+}
+
+define i32 @tzmsk32_branch(i32 %x) nounwind {
+; CHECK-LABEL: tzmsk32_branch:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    pushq %rbx
+; CHECK-NEXT:    movl %edi, %ebx
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    notl %eax
+; CHECK-NEXT:    decl %ebx
+; CHECK-NEXT:    andl %eax, %ebx
+; CHECK-NEXT:    jne .LBB71_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    callq bar
+; CHECK-NEXT:  .LBB71_2:
+; CHECK-NEXT:    movl %ebx, %eax
+; CHECK-NEXT:    popq %rbx
+; CHECK-NEXT:    retq
+  %tmp = xor i32 %x, -1
+  %tmp2 = add i32 %x, -1
+  %tmp3 = and i32 %tmp, %tmp2
+  %cmp = icmp eq i32 %tmp3, 0
+  br i1 %cmp, label %1, label %2
+
+  tail call void @bar()
+  br label %2
+  ret i32 %tmp3
+}
+
+define i64 @tzmsk64_branch(i64 %x) nounwind {
+; CHECK-LABEL: tzmsk64_branch:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    pushq %rbx
+; CHECK-NEXT:    movq %rdi, %rbx
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    notq %rax
+; CHECK-NEXT:    decq %rbx
+; CHECK-NEXT:    andq %rax, %rbx
+; CHECK-NEXT:    jne .LBB72_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    callq bar
+; CHECK-NEXT:  .LBB72_2:
+; CHECK-NEXT:    movq %rbx, %rax
+; CHECK-NEXT:    popq %rbx
+; CHECK-NEXT:    retq
+  %tmp = xor i64 %x, -1
+  %tmp2 = add i64 %x, -1
+  %tmp3 = and i64 %tmp, %tmp2
+  %cmp = icmp eq i64 %tmp3, 0
+  br i1 %cmp, label %1, label %2
+
+  tail call void @bar()
+  br label %2
+  ret i64 %tmp3
+}
+
+declare void @bar()




More information about the llvm-commits mailing list