[FastISel][AArch64] Teach the address computation to also fold sub instructions.

Tiny enhancement to the address computation code to also fold sub instructions
if the rhs is constant and can be folded into the offset.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219186 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Juergen Ributzka 2014-10-07 03:40:03 +00:00
parent ca07e256f6
commit 3692081566
2 changed files with 22 additions and 11 deletions

View File

@ -590,7 +590,7 @@ bool AArch64FastISel::computeAddress(const Value *Obj, Address &Addr, Type *Ty)
std::swap(LHS, RHS);
if (const ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
Addr.setOffset(Addr.getOffset() + (uint64_t)CI->getSExtValue());
Addr.setOffset(Addr.getOffset() + CI->getSExtValue());
return computeAddress(LHS, Addr, Ty);
}
@ -601,6 +601,17 @@ bool AArch64FastISel::computeAddress(const Value *Obj, Address &Addr, Type *Ty)
break;
}
case Instruction::Sub: {
// Subs of constants are common and easy enough.
const Value *LHS = U->getOperand(0);
const Value *RHS = U->getOperand(1);
if (const ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
Addr.setOffset(Addr.getOffset() - CI->getSExtValue());
return computeAddress(LHS, Addr, Ty);
}
break;
}
case Instruction::Shl: {
if (Addr.getOffsetReg())
break;

View File

@ -151,7 +151,7 @@ define i32 @load_unscaled_zext_i8_to_i32(i64 %a) {
; CHECK-LABEL: load_unscaled_zext_i8_to_i32
; CHECK: ldurb w0, [x0, #-8]
; CHECK-NOT: uxtb
%1 = add i64 %a, -8
%1 = sub i64 %a, 8
%2 = inttoptr i64 %1 to i8*
%3 = load i8* %2
%4 = zext i8 %3 to i32
@ -162,7 +162,7 @@ define i32 @load_unscaled_zext_i16_to_i32(i64 %a) {
; CHECK-LABEL: load_unscaled_zext_i16_to_i32
; CHECK: ldurh w0, [x0, #-8]
; CHECK-NOT: uxth
%1 = add i64 %a, -8
%1 = sub i64 %a, 8
%2 = inttoptr i64 %1 to i16*
%3 = load i16* %2
%4 = zext i16 %3 to i32
@ -173,7 +173,7 @@ define i64 @load_unscaled_zext_i8_to_i64(i64 %a) {
; CHECK-LABEL: load_unscaled_zext_i8_to_i64
; CHECK: ldurb w0, [x0, #-8]
; CHECK-NOT: uxtb
%1 = add i64 %a, -8
%1 = sub i64 %a, 8
%2 = inttoptr i64 %1 to i8*
%3 = load i8* %2
%4 = zext i8 %3 to i64
@ -184,7 +184,7 @@ define i64 @load_unscaled_zext_i16_to_i64(i64 %a) {
; CHECK-LABEL: load_unscaled_zext_i16_to_i64
; CHECK: ldurh w0, [x0, #-8]
; CHECK-NOT: uxth
%1 = add i64 %a, -8
%1 = sub i64 %a, 8
%2 = inttoptr i64 %1 to i16*
%3 = load i16* %2
%4 = zext i16 %3 to i64
@ -195,7 +195,7 @@ define i64 @load_unscaled_zext_i32_to_i64(i64 %a) {
; CHECK-LABEL: load_unscaled_zext_i32_to_i64
; CHECK: ldur w0, [x0, #-8]
; CHECK-NOT: uxtw
%1 = add i64 %a, -8
%1 = sub i64 %a, 8
%2 = inttoptr i64 %1 to i32*
%3 = load i32* %2
%4 = zext i32 %3 to i64
@ -206,7 +206,7 @@ define i32 @load_unscaled_sext_i8_to_i32(i64 %a) {
; CHECK-LABEL: load_unscaled_sext_i8_to_i32
; CHECK: ldursb w0, [x0, #-8]
; CHECK-NOT: sxtb
%1 = add i64 %a, -8
%1 = sub i64 %a, 8
%2 = inttoptr i64 %1 to i8*
%3 = load i8* %2
%4 = sext i8 %3 to i32
@ -217,7 +217,7 @@ define i32 @load_unscaled_sext_i16_to_i32(i64 %a) {
; CHECK-LABEL: load_unscaled_sext_i16_to_i32
; CHECK: ldursh w0, [x0, #-8]
; CHECK-NOT: sxth
%1 = add i64 %a, -8
%1 = sub i64 %a, 8
%2 = inttoptr i64 %1 to i16*
%3 = load i16* %2
%4 = sext i16 %3 to i32
@ -228,7 +228,7 @@ define i64 @load_unscaled_sext_i8_to_i64(i64 %a) {
; CHECK-LABEL: load_unscaled_sext_i8_to_i64
; CHECK: ldursb x0, [x0, #-8]
; CHECK-NOT: sxtb
%1 = add i64 %a, -8
%1 = sub i64 %a, 8
%2 = inttoptr i64 %1 to i8*
%3 = load i8* %2
%4 = sext i8 %3 to i64
@ -239,7 +239,7 @@ define i64 @load_unscaled_sext_i16_to_i64(i64 %a) {
; CHECK-LABEL: load_unscaled_sext_i16_to_i64
; CHECK: ldursh x0, [x0, #-8]
; CHECK-NOT: sxth
%1 = add i64 %a, -8
%1 = sub i64 %a, 8
%2 = inttoptr i64 %1 to i16*
%3 = load i16* %2
%4 = sext i16 %3 to i64
@ -250,7 +250,7 @@ define i64 @load_unscaled_sext_i32_to_i64(i64 %a) {
; CHECK-LABEL: load_unscaled_sext_i32_to_i64
; CHECK: ldursw x0, [x0, #-8]
; CHECK-NOT: sxtw
%1 = add i64 %a, -8
%1 = sub i64 %a, 8
%2 = inttoptr i64 %1 to i32*
%3 = load i32* %2
%4 = sext i32 %3 to i64