[X86] Fix the bug of pr62625

We should not call tryOptimizeLEAtoMOV() in eliminateFrameIndex() when
the base register is a virtual register, because tryOptimizeLEAtoMOV
would assume the base register be physical register. Although we can
also optimize LEA to MOV with virtual register, I'd like to leave the
optimization in another patch.

Differential Revision: https://reviews.llvm.org/D150521

diff --git a/llvm/lib/Target/X86/X86RegisterInfo.cpp b/llvm/lib/Target/X86/X86RegisterInfo.cpp
index 2677e5f9446e94b9e83474de7700e209bab1c1d3..0796ac65d7eecd412a55689a343d9533aef37c39 100644 (file)
--- a/llvm/lib/Target/X86/X86RegisterInfo.cpp
+++ b/llvm/lib/Target/X86/X86RegisterInfo.cpp
@@ -812,7 +812,7 @@ void X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
     int Offset = FIOffset + Imm;
     assert((!Is64Bit || isInt<32>((long long)FIOffset + Imm)) &&
            "Requesting 64-bit offset in 32-bit immediate!");
-    if (Offset != 0 || !tryOptimizeLEAtoMOV(II))
+    if (Offset != 0)
       MI.getOperand(FIOperandNum + 3).ChangeToImmediate(Offset);
   } else {
     // Offset is symbolic. This is extremely rare.

diff --git a/llvm/test/CodeGen/X86/x86-64-baseptr.ll b/llvm/test/CodeGen/X86/x86-64-baseptr.ll
index 21befe7edb11798e920c2211bd2113700306f9aa..8cda4ba2814bbab0be631e50e3a8caf7cbd7f298 100644 (file)
--- a/llvm/test/CodeGen/X86/x86-64-baseptr.ll
+++ b/llvm/test/CodeGen/X86/x86-64-baseptr.ll
@@ -306,6 +306,102 @@ entry:
   ret void
 }
 
+; pr62625
+define void @vmw_host_printf(ptr %fmt, ...) nounwind {
+; CHECK-LABEL: vmw_host_printf:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    leaq {{[0-9]+}}(%rsp), %r10
+; CHECK-NEXT:    andq $-16, %rsp
+; CHECK-NEXT:    pushq -8(%r10)
+; CHECK-NEXT:    pushq %rbp
+; CHECK-NEXT:    movq %rsp, %rbp
+; CHECK-NEXT:    pushq %rbx
+; CHECK-NEXT:    subq $200, %rsp
+; CHECK-NEXT:    movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; CHECK-NEXT:    movq %rsi, -184(%rbp)
+; CHECK-NEXT:    movq %rdx, -176(%rbp)
+; CHECK-NEXT:    movq %rcx, -168(%rbp)
+; CHECK-NEXT:    movq %r8, -160(%rbp)
+; CHECK-NEXT:    movq %r9, -152(%rbp)
+; CHECK-NEXT:    testb %al, %al
+; CHECK-NEXT:    je .LBB3_2
+; CHECK-NEXT:  # %bb.1: # %entry
+; CHECK-NEXT:    movaps %xmm0, -144(%rbp)
+; CHECK-NEXT:    movaps %xmm1, -128(%rbp)
+; CHECK-NEXT:    movaps %xmm2, -112(%rbp)
+; CHECK-NEXT:    movaps %xmm3, -96(%rbp)
+; CHECK-NEXT:    movaps %xmm4, -80(%rbp)
+; CHECK-NEXT:    movaps %xmm5, -64(%rbp)
+; CHECK-NEXT:    movaps %xmm6, -48(%rbp)
+; CHECK-NEXT:    movaps %xmm7, -32(%rbp)
+; CHECK-NEXT:  .LBB3_2: # %entry
+; CHECK-NEXT:    leaq -192(%rbp), %rax
+; CHECK-NEXT:    movq %rax, (%rax)
+; CHECK-NEXT:    leaq (%r10), %rax
+; CHECK-NEXT:    movq %rax, (%rax)
+; CHECK-NEXT:    movl $48, (%rax)
+; CHECK-NEXT:    movl $8, (%rax)
+; CHECK-NEXT:    xorl %eax, %eax
+; CHECK-NEXT:    xorl %ebx, %ebx
+; CHECK-NEXT:    xorl %ecx, %ecx
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
+; CHECK-NEXT:    leaq -8(%rbp), %rsp
+; CHECK-NEXT:    popq %rbx
+; CHECK-NEXT:    popq %rbp
+; CHECK-NEXT:    leaq -8(%r10), %rsp
+; CHECK-NEXT:    retq
+;
+; X32ABI-LABEL: vmw_host_printf:
+; X32ABI:       # %bb.0: # %entry
+; X32ABI-NEXT:    pushq %rbp
+; X32ABI-NEXT:    movl %esp, %ebp
+; X32ABI-NEXT:    pushq %rbx
+; X32ABI-NEXT:    andl $-16, %esp
+; X32ABI-NEXT:    subl $208, %esp
+; X32ABI-NEXT:    movl %esp, %ebx
+; X32ABI-NEXT:    movq %rsi, 24(%ebx)
+; X32ABI-NEXT:    movq %rdx, 32(%ebx)
+; X32ABI-NEXT:    movq %rcx, 40(%ebx)
+; X32ABI-NEXT:    movq %r8, 48(%ebx)
+; X32ABI-NEXT:    movq %r9, 56(%ebx)
+; X32ABI-NEXT:    testb %al, %al
+; X32ABI-NEXT:    je .LBB3_2
+; X32ABI-NEXT:  # %bb.1: # %entry
+; X32ABI-NEXT:    movaps %xmm0, 64(%ebx)
+; X32ABI-NEXT:    movaps %xmm1, 80(%ebx)
+; X32ABI-NEXT:    movaps %xmm2, 96(%ebx)
+; X32ABI-NEXT:    movaps %xmm3, 112(%ebx)
+; X32ABI-NEXT:    movaps %xmm4, 128(%ebx)
+; X32ABI-NEXT:    movaps %xmm5, 144(%ebx)
+; X32ABI-NEXT:    movaps %xmm6, 160(%ebx)
+; X32ABI-NEXT:    movaps %xmm7, 176(%ebx)
+; X32ABI-NEXT:  .LBB3_2: # %entry
+; X32ABI-NEXT:    leal 16(%rbx), %eax
+; X32ABI-NEXT:    movl %eax, (%eax)
+; X32ABI-NEXT:    leal 16(%rbp), %eax
+; X32ABI-NEXT:    movl %eax, (%eax)
+; X32ABI-NEXT:    movl $48, (%eax)
+; X32ABI-NEXT:    movl $8, (%eax)
+; X32ABI-NEXT:    xorl %eax, %eax
+; X32ABI-NEXT:    xorl %ebx, %ebx
+; X32ABI-NEXT:    xorl %ecx, %ecx
+; X32ABI-NEXT:    #APP
+; X32ABI-NEXT:    #NO_APP
+; X32ABI-NEXT:    leal -8(%ebp), %esp
+; X32ABI-NEXT:    popq %rbx
+; X32ABI-NEXT:    popq %rbp
+; X32ABI-NEXT:    retq
+entry:
+  %0 = alloca i8, i64 poison, align 8
+  call void @llvm.va_start(ptr nonnull poison)
+  %1 = call { i64, i64, i64, i64, i64, i64 } asm sideeffect "", "={ax},={bx},={cx},={dx},={si},={di},{ax},{bx},{cx},{dx},{si},{di},~{memory},~{dirflag},~{fpsr},~{flags}"(i32 0, i32 0, i32 0, i16 undef, i64 undef, i64 undef)
+  ret void
+}
+
+declare void @llvm.va_start(ptr)
+
 attributes #0 = {"frame-pointer"="all"}
 !llvm.module.flags = !{!0}
 !0 = !{i32 2, !"override-stack-alignment", i32 32}