Value *OldLHS = Op->getOperand(0);
Value *OldRHS = Op->getOperand(1);
- if (NewLHS == OldLHS && NewRHS == OldRHS)
- // Nothing changed, leave it alone.
- break;
-
- if (NewLHS == OldRHS && NewRHS == OldLHS) {
- // The order of the operands was reversed. Swap them.
+ // The new operation differs trivially from the original.
+ if ((NewLHS == OldLHS && NewRHS == OldRHS) ||
+ (NewLHS == OldRHS && NewRHS == OldLHS)) {
DEBUG(dbgs() << "RA: " << *Op << '\n');
- Op->swapOperands();
+ canonicalizeOperands(Op);
DEBUG(dbgs() << "TO: " << *Op << '\n');
MadeChange = true;
++NumChanged;
NodesToRewrite.push_back(BO);
Op->setOperand(1, NewRHS);
}
+ // Put the operands in canonical form.
+ canonicalizeOperands(Op);
DEBUG(dbgs() << "TO: " << *Op << '\n');
ExpressionChanged = Op;
// into it.
BinaryOperator *BO = isReassociableOp(Op->getOperand(0), Opcode);
if (BO && !NotRewritable.count(BO)) {
+ canonicalizeOperands(Op);
Op = BO;
continue;
}
DEBUG(dbgs() << "RA: " << *Op << '\n');
Op->setOperand(0, NewOp);
+ canonicalizeOperands(Op);
DEBUG(dbgs() << "TO: " << *Op << '\n');
ExpressionChanged = Op;
MadeChange = true;
define i32 @test1(i32 %A, i32 %B) {
; CHECK-LABEL: test1
-; CHECK: %Z = add i32 %B, %A
+; CHECK: %Z = add i32 %A, %B
; CHECK: ret i32 %Z
%W = add i32 %B, -5
%Y = add i32 %A, 5
; With sub reassociation, constant folding can eliminate the two 12 constants.
define i32 @test2(i32 %A, i32 %B, i32 %C, i32 %D) {
; CHECK-LABEL: test2
-; CHECK-NEXT: %sum = add i32 %B, %A
+; CHECK-NEXT: %sum = add i32 %A, %B
; CHECK-NEXT: %sum1 = add i32 %sum, %C
; CHECK-NEXT: %Q = sub i32 %D, %sum1
; CHECK-NEXT: ret i32 %Q
ret i32 %reg117
; CHECK-LABEL: @test2
-; CHECK-NEXT: %reg117 = add i32 %reg1111, %reg109
+; CHECK-NEXT: %reg117 = add i32 %reg109, %reg1111
; CHECK-NEXT: ret i32 %reg117
}
ret i32 %r
; CHECK-LABEL: @test7
-; CHECK-NEXT: add i32 %C, %B
+; CHECK-NEXT: add i32 %B, %C
; CHECK-NEXT: mul i32
; CHECK-NEXT: mul i32
; CHECK-NEXT: ret i32
ret i32 %C
; CHECK-LABEL: @test8
-; CHECK-NEXT: %A = mul i32 %Y, %X
+; CHECK-NEXT: %A = mul i32 %X, %Y
; CHECK-NEXT: %C = sub i32 %Z, %A
; CHECK-NEXT: ret i32 %C
}
define i64 @test4(i64 %x, i64 %y) {
; CHECK-LABEL: @test4
; CHECK-NEXT: mul i64 %y, 1234
-; CHECK-NEXT: add i64 %mul, %x
+; CHECK-NEXT: add i64 %x, %mul
; CHECK-NEXT: ret i64 %sub
%mul = mul i64 %y, -1234
define void @test1(i32 %x, i32 %y) {
; CHECK-LABEL: test1
-; CHECK: mul i32 %y, %x
-; CHECK: mul i32 %y, %x
+; CHECK: mul i32 %x, %y
+; CHECK: mul i32 %x, %y
; CHECK: sub i32 %1, %2
; CHECK: call void @use(i32 %{{.*}})
; CHECK: call void @use(i32 %{{.*}})
; With sub reassociation, constant folding can eliminate the two 12 constants.
define float @test4(float %A, float %B, float %C, float %D) {
; CHECK-LABEL: test4
-; CHECK-NEXT: %B.neg = fsub fast float -0.000000e+00, %B
-; CHECK-NEXT: %O.neg = fsub fast float %B.neg, %A
+; CHECK-NEXT: %A.neg = fsub fast float -0.000000e+00, %A
+; CHECK-NEXT: %O.neg = fsub fast float %A.neg, %B
; CHECK-NEXT: %P = fsub fast float %O.neg, %C
; CHECK-NEXT: %Q = fadd fast float %P, %D
; CHECK-NEXT: ret float %Q
define float @test3(float %reg109, float %reg1111) {
; CHECK-LABEL: @test3
-; CHECK-NEXT: %reg117 = fadd fast float %reg109, %reg1111
+; CHECK-NEXT: %reg117 = fadd fast float %reg1111, %reg109
; CHECK-NEXT: ret float %reg117
%reg115 = fadd fast float %reg109, -3.000000e+01
define float @test7(float %A, float %B, float %C) {
; CHECK-LABEL: @test7
-; CHECK-NEXT: fadd fast float %C, %B
+; CHECK-NEXT: fadd fast float %B, %C
; CHECK-NEXT: fmul fast float %A, %A
; CHECK-NEXT: fmul fast float %1, %tmp2
; CHECK-NEXT: ret float
define float @test8(float %X, float %Y, float %Z) {
; CHECK-LABEL: @test8
-; CHECK-NEXT: fmul fast float %Y, %X
+; CHECK-NEXT: fmul fast float %X, %Y
; CHECK-NEXT: fsub fast float %Z
; CHECK-NEXT: ret float
; With sub reassociation, constant folding can eliminate the uses of %a.
define float @test20(float %a, float %b, float %c) nounwind {
; CHECK-LABEL: @test20
-; CHECK-NEXT: fsub fast float -0.000000e+00, %b
-; CHECK-NEXT: fsub fast float %b.neg, %c
+; CHECK-NEXT: fsub fast float -0.000000e+00, %c
+; CHECK-NEXT: fsub fast float %c.neg, %b
; CHECK-NEXT: ret float
; FIXME: Should be able to generate the below, which may expose more
define void @test1(float %x, float %y) {
; CHECK-LABEL: test1
-; CHECK: fmul fast float %y, %x
-; CHECK: fmul fast float %y, %x
+; CHECK: fmul fast float %x, %y
+; CHECK: fmul fast float %x, %y
; CHECK: fsub fast float %1, %2
; CHECK: call void @use(float %{{.*}})
; CHECK: call void @use(float %{{.*}})
define float @test2(float %x, float %y) {
; CHECK-LABEL: test2
-; CHECK-NEXT: fmul fast float %y, %x
-; CHECK-NEXT: fmul fast float %y, %x
+; CHECK-NEXT: fmul fast float %x, %y
+; CHECK-NEXT: fmul fast float %x, %y
; CHECK-NEXT: fsub fast float %1, %2
; CHECK-NEXT: ret float %3
define float @test3(float %x, float %y) {
; CHECK-LABEL: test3
-; CHECK-NEXT: %factor = fmul fast float %y, 2.000000e+00
-; CHECK-NEXT: %tmp1 = fmul fast float %factor, %x
+; CHECK-NEXT: %factor = fmul fast float %x, 2.000000e+00
+; CHECK-NEXT: %tmp1 = fmul fast float %factor, %y
; CHECK-NEXT: ret float %tmp1
%1 = fmul fast float %x, %y
define float @fmultistep1(float %a, float %b, float %c) {
; Check that a*a*b+a*a*c is turned into a*(a*(b+c)).
; CHECK-LABEL: @fmultistep1
-; CHECK-NEXT: fadd fast float %c, %b
+; CHECK-NEXT: fadd fast float %b, %c
; CHECK-NEXT: fmul fast float %a, %tmp2
-; CHECK-NEXT: fmul fast float %tmp3, %a
+; CHECK-NEXT: fmul fast float %a, %tmp3
; CHECK-NEXT: ret float
%t0 = fmul fast float %a, %b
define float @fmultistep2(float %a, float %b, float %c, float %d) {
; Check that a*b+a*c+d is turned into a*(b+c)+d.
; CHECK-LABEL: @fmultistep2
-; CHECK-NEXT: fadd fast float %c, %b
-; CHECK-NEXT: fmul fast float %tmp, %a
-; CHECK-NEXT: fadd fast float %tmp1, %d
+; CHECK-NEXT: fadd fast float %b, %c
+; CHECK-NEXT: fmul fast float %a, %tmp
+; CHECK-NEXT: fadd fast float %d, %tmp1
; CHECK-NEXT: ret float
%t0 = fmul fast float %a, %b
define float @foo(float %a,float %b, float %c) {
; CHECK: %mul3 = fmul float %a, %b
; CHECK-NEXT: fmul fast float %c, 2.000000e+00
-; CHECK-NEXT: fadd fast float %factor, %b
-; CHECK-NEXT: fmul fast float %tmp1, %a
-; CHECK-NEXT: fadd fast float %tmp2, %mul3
+; CHECK-NEXT: fadd fast float %b, %factor
+; CHECK-NEXT: fmul fast float %a, %tmp1
+; CHECK-NEXT: fadd fast float %mul3, %tmp2
; CHECK-NEXT: ret float
%mul1 = fmul fast float %a, %c
%mul2 = fmul fast float %a, %b
%t2 = mul i64 %a, %c
%t3 = mul i64 %a, %t2 ; a*(a*c)
%t4 = add i64 %t1, %t3
-; CHECK-NEXT: add i64 %c, %b
+; CHECK-NEXT: add i64 %b, %c
+; CHECK-NEXT: mul i64 %a, %tmp{{.*}}
; CHECK-NEXT: mul i64 %a, %tmp{{.*}}
-; CHECK-NEXT: mul i64 %tmp{{.*}}, %a
; CHECK-NEXT: ret
ret i64 %t4
}
%t1 = mul i64 %a, %c
%t2 = add i64 %t1, %d ; a*c+d
%t3 = add i64 %t0, %t2 ; a*b+(a*c+d)
-; CHECK-NEXT: add i64 %c, %b
-; CHECK-NEXT: mul i64 %tmp{{.*}}, %a
-; CHECK-NEXT: add i64 %tmp{{.*}}, %d
+; CHECK-NEXT: add i64 %b, %c
+; CHECK-NEXT: mul i64 %a, %tmp{{.*}}
+; CHECK-NEXT: add i64 %d, %tmp{{.*}}
; CHECK-NEXT: ret
ret i64 %t3
}
; The initial add doesn't change so should not lose the nsw flag.
; CHECK-LABEL: @test2(
%a0 = add nsw i32 %b, %a
-; CHECK-NEXT: %a0 = add nsw i32 %b, %a
+; CHECK-NEXT: %a0 = add nsw i32 %a, %b
%a1 = add nsw i32 %a0, %d
; CHECK-NEXT: %a1 = add i32 %a0, %c
%a2 = add nsw i32 %a1, %c
-; CHECK-NEXT: %a2 = add i32 %a1, %d
+; CHECK-NEXT: %a2 = add i32 %d, %a1
call void @use(i32 %a2)
; CHECK-NEXT: call void @use
ret void
; Reassociate should clear optional flags like nsw when reassociating.
; CHECK-LABEL: @test0(
-; CHECK: %y = add i64 %b, %a
-; CHECK: %z = add i64 %y, %c
+; CHECK: %y = add i64 %a, %b
+; CHECK: %z = add i64 %c, %y
define i64 @test0(i64 %a, i64 %b, i64 %c) {
%y = add nsw i64 %c, %b
%z = add i64 %y, %a
}
; CHECK-LABEL: @test1(
-; CHECK: %y = add i64 %b, %a
-; CHECK: %z = add i64 %y, %c
+; CHECK: %y = add i64 %a, %b
+; CHECK: %z = add i64 %c, %y
define i64 @test1(i64 %a, i64 %b, i64 %c) {
%y = add i64 %c, %b
%z = add nsw i64 %y, %a
}
; PR9215
-; CHECK: %s = add nsw i32 %y, %x
+; CHECK: %s = add nsw i32 %x, %y
define i32 @test2(i32 %x, i32 %y) {
%s = add nsw i32 %x, %y
ret i32 %s
define i32 @test1(i32 %X, i32 %Y) {
; CHECK-LABEL: test1
-; CHECK-NEXT: %tmp = add i32 %Y, %X
+; CHECK-NEXT: %tmp = add i32 %X, %Y
; CHECK-NEXT: %tmp1 = shl i32 %tmp, 1
; CHECK-NEXT: ret i32 %tmp1
; With sub reassociation, constant folding can eliminate the uses of %a.
define i32 @test2(i32 %a, i32 %b, i32 %c) nounwind {
; CHECK-LABEL: @test2
-; CHECK-NEXT: %sum = add i32 %c, %b
+; CHECK-NEXT: %sum = add i32 %b, %c
; CHECK-NEXT: %tmp7 = sub i32 0, %sum
; CHECK-NEXT: ret i32 %tmp7
;CHECK-LABEL: @xor3(
;CHECK: %and.ra = and i32 %x, -436
;CHECK: %xor = xor i32 %y, 123
-;CHECK: %xor1 = xor i32 %xor, %and.ra
+;CHECK: %xor1 = xor i32 %and.ra, %xor
}
; Test rule: (x | c1) ^ c2 = (x & ~c1) ^ (c1 ^ c2)
; CHECK-LABEL: @xor4(
; CHECK: %and = and i32 %x, -124
; CHECK: %xor = xor i32 %y, 435
-; CHECK: %xor1 = xor i32 %xor, %and
+; CHECK: %xor1 = xor i32 %and, %xor
}
; ==========================================================================
ret i32 %xor1
; CHECK-LABEL: @xor_special2(
; CHECK: %xor = xor i32 %y, 123
-; CHECK: %xor1 = xor i32 %xor, %x
+; CHECK: %xor1 = xor i32 %x, %xor
; CHECK: ret i32 %xor1
}