aboutsummaryrefslogtreecommitdiff
path: root/llvm/test/CodeGen/AArch64/machine-combiner.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/AArch64/machine-combiner.ll')
-rw-r--r--llvm/test/CodeGen/AArch64/machine-combiner.ll887
1 files changed, 524 insertions, 363 deletions
diff --git a/llvm/test/CodeGen/AArch64/machine-combiner.ll b/llvm/test/CodeGen/AArch64/machine-combiner.ll
index ec61fee..65afd92 100644
--- a/llvm/test/CodeGen/AArch64/machine-combiner.ll
+++ b/llvm/test/CodeGen/AArch64/machine-combiner.ll
@@ -1,29 +1,21 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=neoverse-n2 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-STD
-; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=neoverse-n2 -enable-unsafe-fp-math < %s | FileCheck %s --check-prefixes=CHECK,CHECK-UNSAFE
+; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=neoverse-n2 < %s | FileCheck %s
; Incremental updates of the instruction depths should be enough for this test
; case.
-; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=neoverse-n2 -enable-unsafe-fp-math \
-; RUN: -machine-combiner-inc-threshold=0 -machine-combiner-verify-pattern-order=true < %s | FileCheck %s --check-prefixes=CHECK,CHECK-UNSAFE
+; RUN: llc -mtriple=aarch64-gnu-linux -mcpu=neoverse-n2 \
+; RUN: -machine-combiner-inc-threshold=0 -machine-combiner-verify-pattern-order=true < %s | FileCheck %s
; Verify that the first two adds are independent regardless of how the inputs are
; commuted. The destination registers are used as source registers for the third add.
define float @reassociate_adds1(float %x0, float %x1, float %x2, float %x3) {
-; CHECK-STD-LABEL: reassociate_adds1:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd s0, s0, s1
-; CHECK-STD-NEXT: fadd s0, s0, s2
-; CHECK-STD-NEXT: fadd s0, s0, s3
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds1:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s1
-; CHECK-UNSAFE-NEXT: fadd s1, s2, s3
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s1
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds1:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s0, s0, s2
+; CHECK-NEXT: fadd s0, s0, s3
+; CHECK-NEXT: ret
%t0 = fadd float %x0, %x1
%t1 = fadd float %t0, %x2
%t2 = fadd float %t1, %x3
@@ -44,110 +36,110 @@ define float @reassociate_adds1_fast(float %x0, float %x1, float %x2, float %x3)
}
define float @reassociate_adds1_reassoc(float %x0, float %x1, float %x2, float %x3) {
-; CHECK-STD-LABEL: reassociate_adds1_reassoc:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd s0, s0, s1
-; CHECK-STD-NEXT: fadd s0, s0, s2
-; CHECK-STD-NEXT: fadd s0, s0, s3
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds1_reassoc:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s1
-; CHECK-UNSAFE-NEXT: fadd s1, s2, s3
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s1
-; CHECK-UNSAFE-NEXT: ret
- %t0 = fadd reassoc float %x0, %x1
- %t1 = fadd reassoc float %t0, %x2
- %t2 = fadd reassoc float %t1, %x3
+; CHECK-LABEL: reassociate_adds1_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s1, s2, s3
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz float %x0, %x1
+ %t1 = fadd reassoc nsz float %t0, %x2
+ %t2 = fadd reassoc nsz float %t1, %x3
ret float %t2
}
define float @reassociate_adds2(float %x0, float %x1, float %x2, float %x3) {
-; CHECK-STD-LABEL: reassociate_adds2:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd s0, s0, s1
-; CHECK-STD-NEXT: fadd s0, s2, s0
-; CHECK-STD-NEXT: fadd s0, s0, s3
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds2:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s1
-; CHECK-UNSAFE-NEXT: fadd s1, s2, s3
-; CHECK-UNSAFE-NEXT: fadd s0, s1, s0
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds2:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s0, s2, s0
+; CHECK-NEXT: fadd s0, s0, s3
+; CHECK-NEXT: ret
%t0 = fadd float %x0, %x1
%t1 = fadd float %x2, %t0
%t2 = fadd float %t1, %x3
ret float %t2
}
+define float @reassociate_adds2_reassoc(float %x0, float %x1, float %x2, float %x3) {
+; CHECK-LABEL: reassociate_adds2_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s1, s2, s3
+; CHECK-NEXT: fadd s0, s1, s0
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz float %x0, %x1
+ %t1 = fadd reassoc nsz float %x2, %t0
+ %t2 = fadd reassoc nsz float %t1, %x3
+ ret float %t2
+}
+
define float @reassociate_adds3(float %x0, float %x1, float %x2, float %x3) {
-; CHECK-STD-LABEL: reassociate_adds3:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd s0, s0, s1
-; CHECK-STD-NEXT: fadd s0, s0, s2
-; CHECK-STD-NEXT: fadd s0, s3, s0
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds3:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s1
-; CHECK-UNSAFE-NEXT: fadd s1, s3, s2
-; CHECK-UNSAFE-NEXT: fadd s0, s1, s0
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds3:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s0, s0, s2
+; CHECK-NEXT: fadd s0, s3, s0
+; CHECK-NEXT: ret
%t0 = fadd float %x0, %x1
%t1 = fadd float %t0, %x2
%t2 = fadd float %x3, %t1
ret float %t2
}
+define float @reassociate_adds3_reassoc(float %x0, float %x1, float %x2, float %x3) {
+; CHECK-LABEL: reassociate_adds3_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s1, s3, s2
+; CHECK-NEXT: fadd s0, s1, s0
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz float %x0, %x1
+ %t1 = fadd reassoc nsz float %t0, %x2
+ %t2 = fadd reassoc nsz float %x3, %t1
+ ret float %t2
+}
+
define float @reassociate_adds4(float %x0, float %x1, float %x2, float %x3) {
-; CHECK-STD-LABEL: reassociate_adds4:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd s0, s0, s1
-; CHECK-STD-NEXT: fadd s0, s2, s0
-; CHECK-STD-NEXT: fadd s0, s3, s0
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds4:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s1
-; CHECK-UNSAFE-NEXT: fadd s1, s3, s2
-; CHECK-UNSAFE-NEXT: fadd s0, s1, s0
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds4:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s0, s2, s0
+; CHECK-NEXT: fadd s0, s3, s0
+; CHECK-NEXT: ret
%t0 = fadd float %x0, %x1
%t1 = fadd float %x2, %t0
%t2 = fadd float %x3, %t1
ret float %t2
}
+define float @reassociate_adds4_reassoc(float %x0, float %x1, float %x2, float %x3) {
+; CHECK-LABEL: reassociate_adds4_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s1, s3, s2
+; CHECK-NEXT: fadd s0, s1, s0
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz float %x0, %x1
+ %t1 = fadd reassoc nsz float %x2, %t0
+ %t2 = fadd reassoc nsz float %x3, %t1
+ ret float %t2
+}
+
; Verify that we reassociate some of these ops. The optimal balanced tree of adds is not
; produced because that would cost more compile time.
define float @reassociate_adds5(float %x0, float %x1, float %x2, float %x3, float %x4, float %x5, float %x6, float %x7) {
-; CHECK-STD-LABEL: reassociate_adds5:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd s0, s0, s1
-; CHECK-STD-NEXT: fadd s0, s0, s2
-; CHECK-STD-NEXT: fadd s0, s0, s3
-; CHECK-STD-NEXT: fadd s0, s0, s4
-; CHECK-STD-NEXT: fadd s0, s0, s5
-; CHECK-STD-NEXT: fadd s0, s0, s6
-; CHECK-STD-NEXT: fadd s0, s0, s7
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds5:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s1
-; CHECK-UNSAFE-NEXT: fadd s1, s2, s3
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s1
-; CHECK-UNSAFE-NEXT: fadd s1, s4, s5
-; CHECK-UNSAFE-NEXT: fadd s1, s1, s6
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s1
-; CHECK-UNSAFE-NEXT: fadd s0, s0, s7
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds5:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s0, s0, s2
+; CHECK-NEXT: fadd s0, s0, s3
+; CHECK-NEXT: fadd s0, s0, s4
+; CHECK-NEXT: fadd s0, s0, s5
+; CHECK-NEXT: fadd s0, s0, s6
+; CHECK-NEXT: fadd s0, s0, s7
+; CHECK-NEXT: ret
%t0 = fadd float %x0, %x1
%t1 = fadd float %t0, %x2
%t2 = fadd float %t1, %x3
@@ -158,141 +150,198 @@ define float @reassociate_adds5(float %x0, float %x1, float %x2, float %x3, floa
ret float %t6
}
+define float @reassociate_adds5_reassoc(float %x0, float %x1, float %x2, float %x3, float %x4, float %x5, float %x6, float %x7) {
+; CHECK-LABEL: reassociate_adds5_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s1, s2, s3
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s1, s4, s5
+; CHECK-NEXT: fadd s1, s1, s6
+; CHECK-NEXT: fadd s0, s0, s1
+; CHECK-NEXT: fadd s0, s0, s7
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz float %x0, %x1
+ %t1 = fadd reassoc nsz float %t0, %x2
+ %t2 = fadd reassoc nsz float %t1, %x3
+ %t3 = fadd reassoc nsz float %t2, %x4
+ %t4 = fadd reassoc nsz float %t3, %x5
+ %t5 = fadd reassoc nsz float %t4, %x6
+ %t6 = fadd reassoc nsz float %t5, %x7
+ ret float %t6
+}
+
; Verify that we only need two associative operations to reassociate the operands.
; Also, we should reassociate such that the result of the high latency division
; is used by the final 'add' rather than reassociating the %x3 operand with the
; division. The latter reassociation would not improve anything.
define float @reassociate_adds6(float %x0, float %x1, float %x2, float %x3) {
-; CHECK-STD-LABEL: reassociate_adds6:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fdiv s0, s0, s1
-; CHECK-STD-NEXT: fadd s0, s2, s0
-; CHECK-STD-NEXT: fadd s0, s3, s0
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds6:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fdiv s0, s0, s1
-; CHECK-UNSAFE-NEXT: fadd s1, s3, s2
-; CHECK-UNSAFE-NEXT: fadd s0, s1, s0
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds6:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv s0, s0, s1
+; CHECK-NEXT: fadd s0, s2, s0
+; CHECK-NEXT: fadd s0, s3, s0
+; CHECK-NEXT: ret
%t0 = fdiv float %x0, %x1
%t1 = fadd float %x2, %t0
%t2 = fadd float %x3, %t1
ret float %t2
}
+define float @reassociate_adds6_reassoc(float %x0, float %x1, float %x2, float %x3) {
+; CHECK-LABEL: reassociate_adds6_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv s0, s0, s1
+; CHECK-NEXT: fadd s1, s3, s2
+; CHECK-NEXT: fadd s0, s1, s0
+; CHECK-NEXT: ret
+ %t0 = fdiv reassoc nsz float %x0, %x1
+ %t1 = fadd reassoc nsz float %x2, %t0
+ %t2 = fadd reassoc nsz float %x3, %t1
+ ret float %t2
+}
+
; Verify that scalar single-precision multiplies are reassociated.
define float @reassociate_muls1(float %x0, float %x1, float %x2, float %x3) {
-; CHECK-STD-LABEL: reassociate_muls1:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fdiv s0, s0, s1
-; CHECK-STD-NEXT: fmul s0, s2, s0
-; CHECK-STD-NEXT: fmul s0, s3, s0
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_muls1:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fdiv s0, s0, s1
-; CHECK-UNSAFE-NEXT: fmul s1, s3, s2
-; CHECK-UNSAFE-NEXT: fmul s0, s1, s0
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_muls1:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv s0, s0, s1
+; CHECK-NEXT: fmul s0, s2, s0
+; CHECK-NEXT: fmul s0, s3, s0
+; CHECK-NEXT: ret
%t0 = fdiv float %x0, %x1
%t1 = fmul float %x2, %t0
%t2 = fmul float %x3, %t1
ret float %t2
}
+define float @reassociate_muls1_reassoc(float %x0, float %x1, float %x2, float %x3) {
+; CHECK-LABEL: reassociate_muls1_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv s0, s0, s1
+; CHECK-NEXT: fmul s1, s3, s2
+; CHECK-NEXT: fmul s0, s1, s0
+; CHECK-NEXT: ret
+ %t0 = fdiv reassoc nsz float %x0, %x1
+ %t1 = fmul reassoc nsz float %x2, %t0
+ %t2 = fmul reassoc nsz float %x3, %t1
+ ret float %t2
+}
+
; Verify that scalar double-precision adds are reassociated.
define double @reassociate_adds_double(double %x0, double %x1, double %x2, double %x3) {
-; CHECK-STD-LABEL: reassociate_adds_double:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fdiv d0, d0, d1
-; CHECK-STD-NEXT: fadd d0, d2, d0
-; CHECK-STD-NEXT: fadd d0, d3, d0
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds_double:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fdiv d0, d0, d1
-; CHECK-UNSAFE-NEXT: fadd d1, d3, d2
-; CHECK-UNSAFE-NEXT: fadd d0, d1, d0
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds_double:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv d0, d0, d1
+; CHECK-NEXT: fadd d0, d2, d0
+; CHECK-NEXT: fadd d0, d3, d0
+; CHECK-NEXT: ret
%t0 = fdiv double %x0, %x1
%t1 = fadd double %x2, %t0
%t2 = fadd double %x3, %t1
ret double %t2
}
+define double @reassociate_adds_double_reassoc(double %x0, double %x1, double %x2, double %x3) {
+; CHECK-LABEL: reassociate_adds_double_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv d0, d0, d1
+; CHECK-NEXT: fadd d1, d3, d2
+; CHECK-NEXT: fadd d0, d1, d0
+; CHECK-NEXT: ret
+ %t0 = fdiv reassoc nsz double %x0, %x1
+ %t1 = fadd reassoc nsz double %x2, %t0
+ %t2 = fadd reassoc nsz double %x3, %t1
+ ret double %t2
+}
+
; Verify that scalar double-precision multiplies are reassociated.
define double @reassociate_muls_double(double %x0, double %x1, double %x2, double %x3) {
-; CHECK-STD-LABEL: reassociate_muls_double:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fdiv d0, d0, d1
-; CHECK-STD-NEXT: fmul d0, d2, d0
-; CHECK-STD-NEXT: fmul d0, d3, d0
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_muls_double:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fdiv d0, d0, d1
-; CHECK-UNSAFE-NEXT: fmul d1, d3, d2
-; CHECK-UNSAFE-NEXT: fmul d0, d1, d0
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_muls_double:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv d0, d0, d1
+; CHECK-NEXT: fmul d0, d2, d0
+; CHECK-NEXT: fmul d0, d3, d0
+; CHECK-NEXT: ret
%t0 = fdiv double %x0, %x1
%t1 = fmul double %x2, %t0
%t2 = fmul double %x3, %t1
ret double %t2
}
+define double @reassociate_muls_double_reassoc(double %x0, double %x1, double %x2, double %x3) {
+; CHECK-LABEL: reassociate_muls_double_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv d0, d0, d1
+; CHECK-NEXT: fmul d1, d3, d2
+; CHECK-NEXT: fmul d0, d1, d0
+; CHECK-NEXT: ret
+ %t0 = fdiv reassoc nsz double %x0, %x1
+ %t1 = fmul reassoc nsz double %x2, %t0
+ %t2 = fmul reassoc nsz double %x3, %t1
+ ret double %t2
+}
+
; Verify that scalar half-precision adds are reassociated.
define half @reassociate_adds_half(half %x0, half %x1, half %x2, half %x3) {
-; CHECK-STD-LABEL: reassociate_adds_half:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fdiv h0, h0, h1
-; CHECK-STD-NEXT: fadd h0, h2, h0
-; CHECK-STD-NEXT: fadd h0, h3, h0
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds_half:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fdiv h0, h0, h1
-; CHECK-UNSAFE-NEXT: fadd h1, h3, h2
-; CHECK-UNSAFE-NEXT: fadd h0, h1, h0
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds_half:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv h0, h0, h1
+; CHECK-NEXT: fadd h0, h2, h0
+; CHECK-NEXT: fadd h0, h3, h0
+; CHECK-NEXT: ret
%t0 = fdiv half %x0, %x1
%t1 = fadd half %x2, %t0
%t2 = fadd half %x3, %t1
ret half %t2
}
+define half @reassociate_adds_half_reassoc(half %x0, half %x1, half %x2, half %x3) {
+; CHECK-LABEL: reassociate_adds_half_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv h0, h0, h1
+; CHECK-NEXT: fadd h1, h3, h2
+; CHECK-NEXT: fadd h0, h1, h0
+; CHECK-NEXT: ret
+ %t0 = fdiv reassoc nsz half %x0, %x1
+ %t1 = fadd reassoc nsz half %x2, %t0
+ %t2 = fadd reassoc nsz half %x3, %t1
+ ret half %t2
+}
+
; Verify that scalar half-precision multiplies are reassociated.
define half @reassociate_muls_half(half %x0, half %x1, half %x2, half %x3) {
-; CHECK-STD-LABEL: reassociate_muls_half:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fdiv h0, h0, h1
-; CHECK-STD-NEXT: fmul h0, h2, h0
-; CHECK-STD-NEXT: fmul h0, h3, h0
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_muls_half:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fdiv h0, h0, h1
-; CHECK-UNSAFE-NEXT: fmul h1, h3, h2
-; CHECK-UNSAFE-NEXT: fmul h0, h1, h0
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_muls_half:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv h0, h0, h1
+; CHECK-NEXT: fmul h0, h2, h0
+; CHECK-NEXT: fmul h0, h3, h0
+; CHECK-NEXT: ret
%t0 = fdiv half %x0, %x1
%t1 = fmul half %x2, %t0
%t2 = fmul half %x3, %t1
ret half %t2
}
+define half @reassociate_muls_half_reassoc(half %x0, half %x1, half %x2, half %x3) {
+; CHECK-LABEL: reassociate_muls_half_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fdiv h0, h0, h1
+; CHECK-NEXT: fmul h1, h3, h2
+; CHECK-NEXT: fmul h0, h1, h0
+; CHECK-NEXT: ret
+ %t0 = fdiv reassoc nsz half %x0, %x1
+ %t1 = fmul reassoc nsz half %x2, %t0
+ %t2 = fmul reassoc nsz half %x3, %t1
+ ret half %t2
+}
+
; Verify that scalar integer adds are reassociated.
define i32 @reassociate_adds_i32(i32 %x0, i32 %x1, i32 %x2, i32 %x3) {
@@ -365,173 +414,222 @@ define i32 @reassociate_xors_i32(i32 %x0, i32 %x1, i32 %x2, i32 %x3) {
; Verify that we reassociate vector instructions too.
define <4 x float> @vector_reassociate_adds1(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
-; CHECK-STD-LABEL: vector_reassociate_adds1:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-STD-NEXT: fadd v0.4s, v0.4s, v2.4s
-; CHECK-STD-NEXT: fadd v0.4s, v0.4s, v3.4s
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: vector_reassociate_adds1:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-UNSAFE-NEXT: fadd v1.4s, v2.4s, v3.4s
-; CHECK-UNSAFE-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: vector_reassociate_adds1:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: fadd v0.4s, v0.4s, v2.4s
+; CHECK-NEXT: fadd v0.4s, v0.4s, v3.4s
+; CHECK-NEXT: ret
%t0 = fadd <4 x float> %x0, %x1
%t1 = fadd <4 x float> %t0, %x2
%t2 = fadd <4 x float> %t1, %x3
ret <4 x float> %t2
}
+define <4 x float> @vector_reassociate_adds1_reassoc(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
+; CHECK-LABEL: vector_reassociate_adds1_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: fadd v1.4s, v2.4s, v3.4s
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz <4 x float> %x0, %x1
+ %t1 = fadd reassoc nsz <4 x float> %t0, %x2
+ %t2 = fadd reassoc nsz <4 x float> %t1, %x3
+ ret <4 x float> %t2
+}
+
define <4 x float> @vector_reassociate_adds2(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
-; CHECK-STD-LABEL: vector_reassociate_adds2:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-STD-NEXT: fadd v0.4s, v2.4s, v0.4s
-; CHECK-STD-NEXT: fadd v0.4s, v0.4s, v3.4s
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: vector_reassociate_adds2:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-UNSAFE-NEXT: fadd v1.4s, v2.4s, v3.4s
-; CHECK-UNSAFE-NEXT: fadd v0.4s, v1.4s, v0.4s
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: vector_reassociate_adds2:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: fadd v0.4s, v2.4s, v0.4s
+; CHECK-NEXT: fadd v0.4s, v0.4s, v3.4s
+; CHECK-NEXT: ret
%t0 = fadd <4 x float> %x0, %x1
%t1 = fadd <4 x float> %x2, %t0
%t2 = fadd <4 x float> %t1, %x3
ret <4 x float> %t2
}
+define <4 x float> @vector_reassociate_adds2_reassoc(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
+; CHECK-LABEL: vector_reassociate_adds2_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: fadd v1.4s, v2.4s, v3.4s
+; CHECK-NEXT: fadd v0.4s, v1.4s, v0.4s
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz <4 x float> %x0, %x1
+ %t1 = fadd reassoc nsz <4 x float> %x2, %t0
+ %t2 = fadd reassoc nsz <4 x float> %t1, %x3
+ ret <4 x float> %t2
+}
+
define <4 x float> @vector_reassociate_adds3(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
-; CHECK-STD-LABEL: vector_reassociate_adds3:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-STD-NEXT: fadd v0.4s, v0.4s, v2.4s
-; CHECK-STD-NEXT: fadd v0.4s, v3.4s, v0.4s
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: vector_reassociate_adds3:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-UNSAFE-NEXT: fadd v1.4s, v3.4s, v2.4s
-; CHECK-UNSAFE-NEXT: fadd v0.4s, v1.4s, v0.4s
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: vector_reassociate_adds3:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: fadd v0.4s, v0.4s, v2.4s
+; CHECK-NEXT: fadd v0.4s, v3.4s, v0.4s
+; CHECK-NEXT: ret
%t0 = fadd <4 x float> %x0, %x1
%t1 = fadd <4 x float> %t0, %x2
%t2 = fadd <4 x float> %x3, %t1
ret <4 x float> %t2
}
+define <4 x float> @vector_reassociate_adds3_reassoc(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
+; CHECK-LABEL: vector_reassociate_adds3_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: fadd v1.4s, v3.4s, v2.4s
+; CHECK-NEXT: fadd v0.4s, v1.4s, v0.4s
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz <4 x float> %x0, %x1
+ %t1 = fadd reassoc nsz <4 x float> %t0, %x2
+ %t2 = fadd reassoc nsz <4 x float> %x3, %t1
+ ret <4 x float> %t2
+}
+
define <4 x float> @vector_reassociate_adds4(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
-; CHECK-STD-LABEL: vector_reassociate_adds4:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-STD-NEXT: fadd v0.4s, v2.4s, v0.4s
-; CHECK-STD-NEXT: fadd v0.4s, v3.4s, v0.4s
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: vector_reassociate_adds4:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-UNSAFE-NEXT: fadd v1.4s, v3.4s, v2.4s
-; CHECK-UNSAFE-NEXT: fadd v0.4s, v1.4s, v0.4s
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: vector_reassociate_adds4:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: fadd v0.4s, v2.4s, v0.4s
+; CHECK-NEXT: fadd v0.4s, v3.4s, v0.4s
+; CHECK-NEXT: ret
%t0 = fadd <4 x float> %x0, %x1
%t1 = fadd <4 x float> %x2, %t0
%t2 = fadd <4 x float> %x3, %t1
ret <4 x float> %t2
}
+define <4 x float> @vector_reassociate_adds4_reassoc(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
+; CHECK-LABEL: vector_reassociate_adds4_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: fadd v1.4s, v3.4s, v2.4s
+; CHECK-NEXT: fadd v0.4s, v1.4s, v0.4s
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz <4 x float> %x0, %x1
+ %t1 = fadd reassoc nsz <4 x float> %x2, %t0
+ %t2 = fadd reassoc nsz <4 x float> %x3, %t1
+ ret <4 x float> %t2
+}
+
; Verify that 64-bit vector half-precision adds are reassociated.
define <4 x half> @reassociate_adds_v4f16(<4 x half> %x0, <4 x half> %x1, <4 x half> %x2, <4 x half> %x3) {
-; CHECK-STD-LABEL: reassociate_adds_v4f16:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd v0.4h, v0.4h, v1.4h
-; CHECK-STD-NEXT: fadd v0.4h, v2.4h, v0.4h
-; CHECK-STD-NEXT: fadd v0.4h, v3.4h, v0.4h
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds_v4f16:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd v0.4h, v0.4h, v1.4h
-; CHECK-UNSAFE-NEXT: fadd v1.4h, v3.4h, v2.4h
-; CHECK-UNSAFE-NEXT: fadd v0.4h, v1.4h, v0.4h
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds_v4f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4h, v0.4h, v1.4h
+; CHECK-NEXT: fadd v0.4h, v2.4h, v0.4h
+; CHECK-NEXT: fadd v0.4h, v3.4h, v0.4h
+; CHECK-NEXT: ret
%t0 = fadd <4 x half> %x0, %x1
%t1 = fadd <4 x half> %x2, %t0
%t2 = fadd <4 x half> %x3, %t1
ret <4 x half> %t2
}
+define <4 x half> @reassociate_adds_v4f16_reassoc(<4 x half> %x0, <4 x half> %x1, <4 x half> %x2, <4 x half> %x3) {
+; CHECK-LABEL: reassociate_adds_v4f16_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4h, v0.4h, v1.4h
+; CHECK-NEXT: fadd v1.4h, v3.4h, v2.4h
+; CHECK-NEXT: fadd v0.4h, v1.4h, v0.4h
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz <4 x half> %x0, %x1
+ %t1 = fadd reassoc nsz <4 x half> %x2, %t0
+ %t2 = fadd reassoc nsz <4 x half> %x3, %t1
+ ret <4 x half> %t2
+}
+
; Verify that 128-bit vector half-precision multiplies are reassociated.
define <8 x half> @reassociate_muls_v8f16(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, <8 x half> %x3) {
-; CHECK-STD-LABEL: reassociate_muls_v8f16:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd v0.8h, v0.8h, v1.8h
-; CHECK-STD-NEXT: fmul v0.8h, v2.8h, v0.8h
-; CHECK-STD-NEXT: fmul v0.8h, v3.8h, v0.8h
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_muls_v8f16:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd v0.8h, v0.8h, v1.8h
-; CHECK-UNSAFE-NEXT: fmul v1.8h, v3.8h, v2.8h
-; CHECK-UNSAFE-NEXT: fmul v0.8h, v1.8h, v0.8h
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_muls_v8f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.8h, v0.8h, v1.8h
+; CHECK-NEXT: fmul v0.8h, v2.8h, v0.8h
+; CHECK-NEXT: fmul v0.8h, v3.8h, v0.8h
+; CHECK-NEXT: ret
%t0 = fadd <8 x half> %x0, %x1
%t1 = fmul <8 x half> %x2, %t0
%t2 = fmul <8 x half> %x3, %t1
ret <8 x half> %t2
}
+define <8 x half> @reassociate_muls_v8f16_reassoc(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, <8 x half> %x3) {
+; CHECK-LABEL: reassociate_muls_v8f16_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.8h, v0.8h, v1.8h
+; CHECK-NEXT: fmul v1.8h, v3.8h, v2.8h
+; CHECK-NEXT: fmul v0.8h, v1.8h, v0.8h
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz <8 x half> %x0, %x1
+ %t1 = fmul reassoc nsz <8 x half> %x2, %t0
+ %t2 = fmul reassoc nsz <8 x half> %x3, %t1
+ ret <8 x half> %t2
+}
+
; Verify that 128-bit vector single-precision multiplies are reassociated.
define <4 x float> @reassociate_muls_v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
-; CHECK-STD-LABEL: reassociate_muls_v4f32:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-STD-NEXT: fmul v0.4s, v2.4s, v0.4s
-; CHECK-STD-NEXT: fmul v0.4s, v3.4s, v0.4s
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_muls_v4f32:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd v0.4s, v0.4s, v1.4s
-; CHECK-UNSAFE-NEXT: fmul v1.4s, v3.4s, v2.4s
-; CHECK-UNSAFE-NEXT: fmul v0.4s, v1.4s, v0.4s
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_muls_v4f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: fmul v0.4s, v2.4s, v0.4s
+; CHECK-NEXT: fmul v0.4s, v3.4s, v0.4s
+; CHECK-NEXT: ret
%t0 = fadd <4 x float> %x0, %x1
%t1 = fmul <4 x float> %x2, %t0
%t2 = fmul <4 x float> %x3, %t1
ret <4 x float> %t2
}
+define <4 x float> @reassociate_muls_v4f32_reassoc(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, <4 x float> %x3) {
+; CHECK-LABEL: reassociate_muls_v4f32_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: fmul v1.4s, v3.4s, v2.4s
+; CHECK-NEXT: fmul v0.4s, v1.4s, v0.4s
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz <4 x float> %x0, %x1
+ %t1 = fmul reassoc nsz <4 x float> %x2, %t0
+ %t2 = fmul reassoc nsz <4 x float> %x3, %t1
+ ret <4 x float> %t2
+}
+
; Verify that 128-bit vector double-precision multiplies are reassociated.
define <2 x double> @reassociate_muls_v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, <2 x double> %x3) {
-; CHECK-STD-LABEL: reassociate_muls_v2f64:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd v0.2d, v0.2d, v1.2d
-; CHECK-STD-NEXT: fmul v0.2d, v2.2d, v0.2d
-; CHECK-STD-NEXT: fmul v0.2d, v3.2d, v0.2d
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_muls_v2f64:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd v0.2d, v0.2d, v1.2d
-; CHECK-UNSAFE-NEXT: fmul v1.2d, v3.2d, v2.2d
-; CHECK-UNSAFE-NEXT: fmul v0.2d, v1.2d, v0.2d
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_muls_v2f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.2d, v0.2d, v1.2d
+; CHECK-NEXT: fmul v0.2d, v2.2d, v0.2d
+; CHECK-NEXT: fmul v0.2d, v3.2d, v0.2d
+; CHECK-NEXT: ret
%t0 = fadd <2 x double> %x0, %x1
%t1 = fmul <2 x double> %x2, %t0
%t2 = fmul <2 x double> %x3, %t1
ret <2 x double> %t2
}
+define <2 x double> @reassociate_muls_v2f64_reassoc(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, <2 x double> %x3) {
+; CHECK-LABEL: reassociate_muls_v2f64_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd v0.2d, v0.2d, v1.2d
+; CHECK-NEXT: fmul v1.2d, v3.2d, v2.2d
+; CHECK-NEXT: fmul v0.2d, v1.2d, v0.2d
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz <2 x double> %x0, %x1
+ %t1 = fmul reassoc nsz <2 x double> %x2, %t0
+ %t2 = fmul reassoc nsz <2 x double> %x3, %t1
+ ret <2 x double> %t2
+}
+
+
; Verify that vector integer arithmetic operations are reassociated.
define <2 x i32> @reassociate_muls_v2i32(<2 x i32> %x0, <2 x i32> %x1, <2 x i32> %x2, <2 x i32> %x3) {
@@ -606,65 +704,83 @@ define <4 x i32> @reassociate_xors_v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32>
; Verify that scalable vector FP arithmetic operations are reassociated.
define <vscale x 8 x half> @reassociate_adds_nxv4f16(<vscale x 8 x half> %x0, <vscale x 8 x half> %x1, <vscale x 8 x half> %x2, <vscale x 8 x half> %x3) {
-; CHECK-STD-LABEL: reassociate_adds_nxv4f16:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd z0.h, z0.h, z1.h
-; CHECK-STD-NEXT: fadd z0.h, z2.h, z0.h
-; CHECK-STD-NEXT: fadd z0.h, z3.h, z0.h
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds_nxv4f16:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd z0.h, z0.h, z1.h
-; CHECK-UNSAFE-NEXT: fadd z1.h, z3.h, z2.h
-; CHECK-UNSAFE-NEXT: fadd z0.h, z1.h, z0.h
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds_nxv4f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd z0.h, z0.h, z1.h
+; CHECK-NEXT: fadd z0.h, z2.h, z0.h
+; CHECK-NEXT: fadd z0.h, z3.h, z0.h
+; CHECK-NEXT: ret
%t0 = fadd reassoc <vscale x 8 x half> %x0, %x1
%t1 = fadd reassoc <vscale x 8 x half> %x2, %t0
%t2 = fadd reassoc <vscale x 8 x half> %x3, %t1
ret <vscale x 8 x half> %t2
}
+define <vscale x 8 x half> @reassociate_adds_nxv4f16_nsz(<vscale x 8 x half> %x0, <vscale x 8 x half> %x1, <vscale x 8 x half> %x2, <vscale x 8 x half> %x3) {
+; CHECK-LABEL: reassociate_adds_nxv4f16_nsz:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd z0.h, z0.h, z1.h
+; CHECK-NEXT: fadd z1.h, z3.h, z2.h
+; CHECK-NEXT: fadd z0.h, z1.h, z0.h
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz <vscale x 8 x half> %x0, %x1
+ %t1 = fadd reassoc nsz <vscale x 8 x half> %x2, %t0
+ %t2 = fadd reassoc nsz <vscale x 8 x half> %x3, %t1
+ ret <vscale x 8 x half> %t2
+}
+
define <vscale x 4 x float> @reassociate_adds_nxv4f32(<vscale x 4 x float> %x0, <vscale x 4 x float> %x1, <vscale x 4 x float> %x2, <vscale x 4 x float> %x3) {
-; CHECK-STD-LABEL: reassociate_adds_nxv4f32:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fadd z0.s, z0.s, z1.s
-; CHECK-STD-NEXT: fadd z0.s, z2.s, z0.s
-; CHECK-STD-NEXT: fadd z0.s, z3.s, z0.s
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds_nxv4f32:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fadd z0.s, z0.s, z1.s
-; CHECK-UNSAFE-NEXT: fadd z1.s, z3.s, z2.s
-; CHECK-UNSAFE-NEXT: fadd z0.s, z1.s, z0.s
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds_nxv4f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd z0.s, z0.s, z1.s
+; CHECK-NEXT: fadd z0.s, z2.s, z0.s
+; CHECK-NEXT: fadd z0.s, z3.s, z0.s
+; CHECK-NEXT: ret
%t0 = fadd reassoc <vscale x 4 x float> %x0, %x1
%t1 = fadd reassoc <vscale x 4 x float> %x2, %t0
%t2 = fadd reassoc <vscale x 4 x float> %x3, %t1
ret <vscale x 4 x float> %t2
}
+define <vscale x 4 x float> @reassociate_adds_nxv4f32_nsz(<vscale x 4 x float> %x0, <vscale x 4 x float> %x1, <vscale x 4 x float> %x2, <vscale x 4 x float> %x3) {
+; CHECK-LABEL: reassociate_adds_nxv4f32_nsz:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fadd z0.s, z0.s, z1.s
+; CHECK-NEXT: fadd z1.s, z3.s, z2.s
+; CHECK-NEXT: fadd z0.s, z1.s, z0.s
+; CHECK-NEXT: ret
+ %t0 = fadd reassoc nsz <vscale x 4 x float> %x0, %x1
+ %t1 = fadd reassoc nsz <vscale x 4 x float> %x2, %t0
+ %t2 = fadd reassoc nsz <vscale x 4 x float> %x3, %t1
+ ret <vscale x 4 x float> %t2
+}
+
define <vscale x 2 x double> @reassociate_muls_nxv2f64(<vscale x 2 x double> %x0, <vscale x 2 x double> %x1, <vscale x 2 x double> %x2, <vscale x 2 x double> %x3) {
-; CHECK-STD-LABEL: reassociate_muls_nxv2f64:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: fmul z0.d, z0.d, z1.d
-; CHECK-STD-NEXT: fmul z0.d, z2.d, z0.d
-; CHECK-STD-NEXT: fmul z0.d, z3.d, z0.d
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_muls_nxv2f64:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: fmul z0.d, z0.d, z1.d
-; CHECK-UNSAFE-NEXT: fmul z1.d, z3.d, z2.d
-; CHECK-UNSAFE-NEXT: fmul z0.d, z1.d, z0.d
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_muls_nxv2f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fmul z0.d, z0.d, z1.d
+; CHECK-NEXT: fmul z0.d, z2.d, z0.d
+; CHECK-NEXT: fmul z0.d, z3.d, z0.d
+; CHECK-NEXT: ret
%t0 = fmul reassoc <vscale x 2 x double> %x0, %x1
%t1 = fmul reassoc <vscale x 2 x double> %x2, %t0
%t2 = fmul reassoc <vscale x 2 x double> %x3, %t1
ret <vscale x 2 x double> %t2
}
+define <vscale x 2 x double> @reassociate_muls_nxv2f64_nsz(<vscale x 2 x double> %x0, <vscale x 2 x double> %x1, <vscale x 2 x double> %x2, <vscale x 2 x double> %x3) {
+; CHECK-LABEL: reassociate_muls_nxv2f64_nsz:
+; CHECK: // %bb.0:
+; CHECK-NEXT: fmul z0.d, z0.d, z1.d
+; CHECK-NEXT: fmul z1.d, z3.d, z2.d
+; CHECK-NEXT: fmul z0.d, z1.d, z0.d
+; CHECK-NEXT: ret
+ %t0 = fmul reassoc nsz <vscale x 2 x double> %x0, %x1
+ %t1 = fmul reassoc nsz <vscale x 2 x double> %x2, %t0
+ %t2 = fmul reassoc nsz <vscale x 2 x double> %x3, %t1
+ ret <vscale x 2 x double> %t2
+}
+
; Verify that scalable vector integer arithmetic operations are reassociated.
define <vscale x 16 x i8> @reassociate_muls_nxv16i8(<vscale x 16 x i8> %x0, <vscale x 16 x i8> %x1, <vscale x 16 x i8> %x2, <vscale x 16 x i8> %x3) {
@@ -753,55 +869,30 @@ define <vscale x 8 x i16> @reassociate_ors_nxv8i16(<vscale x 8 x i16> %x0, <vsca
declare double @bar()
define double @reassociate_adds_from_calls() {
-; CHECK-STD-LABEL: reassociate_adds_from_calls:
-; CHECK-STD: // %bb.0:
-; CHECK-STD-NEXT: str d10, [sp, #-32]! // 8-byte Folded Spill
-; CHECK-STD-NEXT: stp d9, d8, [sp, #8] // 16-byte Folded Spill
-; CHECK-STD-NEXT: str x30, [sp, #24] // 8-byte Folded Spill
-; CHECK-STD-NEXT: .cfi_def_cfa_offset 32
-; CHECK-STD-NEXT: .cfi_offset w30, -8
-; CHECK-STD-NEXT: .cfi_offset b8, -16
-; CHECK-STD-NEXT: .cfi_offset b9, -24
-; CHECK-STD-NEXT: .cfi_offset b10, -32
-; CHECK-STD-NEXT: bl bar
-; CHECK-STD-NEXT: fmov d8, d0
-; CHECK-STD-NEXT: bl bar
-; CHECK-STD-NEXT: fmov d9, d0
-; CHECK-STD-NEXT: bl bar
-; CHECK-STD-NEXT: fmov d10, d0
-; CHECK-STD-NEXT: bl bar
-; CHECK-STD-NEXT: fadd d1, d8, d9
-; CHECK-STD-NEXT: ldp d9, d8, [sp, #8] // 16-byte Folded Reload
-; CHECK-STD-NEXT: ldr x30, [sp, #24] // 8-byte Folded Reload
-; CHECK-STD-NEXT: fadd d1, d1, d10
-; CHECK-STD-NEXT: fadd d0, d1, d0
-; CHECK-STD-NEXT: ldr d10, [sp], #32 // 8-byte Folded Reload
-; CHECK-STD-NEXT: ret
-;
-; CHECK-UNSAFE-LABEL: reassociate_adds_from_calls:
-; CHECK-UNSAFE: // %bb.0:
-; CHECK-UNSAFE-NEXT: str d10, [sp, #-32]! // 8-byte Folded Spill
-; CHECK-UNSAFE-NEXT: stp d9, d8, [sp, #8] // 16-byte Folded Spill
-; CHECK-UNSAFE-NEXT: str x30, [sp, #24] // 8-byte Folded Spill
-; CHECK-UNSAFE-NEXT: .cfi_def_cfa_offset 32
-; CHECK-UNSAFE-NEXT: .cfi_offset w30, -8
-; CHECK-UNSAFE-NEXT: .cfi_offset b8, -16
-; CHECK-UNSAFE-NEXT: .cfi_offset b9, -24
-; CHECK-UNSAFE-NEXT: .cfi_offset b10, -32
-; CHECK-UNSAFE-NEXT: bl bar
-; CHECK-UNSAFE-NEXT: fmov d8, d0
-; CHECK-UNSAFE-NEXT: bl bar
-; CHECK-UNSAFE-NEXT: fmov d9, d0
-; CHECK-UNSAFE-NEXT: bl bar
-; CHECK-UNSAFE-NEXT: fmov d10, d0
-; CHECK-UNSAFE-NEXT: bl bar
-; CHECK-UNSAFE-NEXT: fadd d1, d8, d9
-; CHECK-UNSAFE-NEXT: ldp d9, d8, [sp, #8] // 16-byte Folded Reload
-; CHECK-UNSAFE-NEXT: ldr x30, [sp, #24] // 8-byte Folded Reload
-; CHECK-UNSAFE-NEXT: fadd d0, d10, d0
-; CHECK-UNSAFE-NEXT: fadd d0, d1, d0
-; CHECK-UNSAFE-NEXT: ldr d10, [sp], #32 // 8-byte Folded Reload
-; CHECK-UNSAFE-NEXT: ret
+; CHECK-LABEL: reassociate_adds_from_calls:
+; CHECK: // %bb.0:
+; CHECK-NEXT: str d10, [sp, #-32]! // 8-byte Folded Spill
+; CHECK-NEXT: stp d9, d8, [sp, #8] // 16-byte Folded Spill
+; CHECK-NEXT: str x30, [sp, #24] // 8-byte Folded Spill
+; CHECK-NEXT: .cfi_def_cfa_offset 32
+; CHECK-NEXT: .cfi_offset w30, -8
+; CHECK-NEXT: .cfi_offset b8, -16
+; CHECK-NEXT: .cfi_offset b9, -24
+; CHECK-NEXT: .cfi_offset b10, -32
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fmov d8, d0
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fmov d9, d0
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fmov d10, d0
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fadd d1, d8, d9
+; CHECK-NEXT: ldp d9, d8, [sp, #8] // 16-byte Folded Reload
+; CHECK-NEXT: ldr x30, [sp, #24] // 8-byte Folded Reload
+; CHECK-NEXT: fadd d1, d1, d10
+; CHECK-NEXT: fadd d0, d1, d0
+; CHECK-NEXT: ldr d10, [sp], #32 // 8-byte Folded Reload
+; CHECK-NEXT: ret
%x0 = call double @bar()
%x1 = call double @bar()
%x2 = call double @bar()
@@ -812,6 +903,41 @@ define double @reassociate_adds_from_calls() {
ret double %t2
}
+define double @reassociate_adds_from_calls_reassoc() {
+; CHECK-LABEL: reassociate_adds_from_calls_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: str d10, [sp, #-32]! // 8-byte Folded Spill
+; CHECK-NEXT: stp d9, d8, [sp, #8] // 16-byte Folded Spill
+; CHECK-NEXT: str x30, [sp, #24] // 8-byte Folded Spill
+; CHECK-NEXT: .cfi_def_cfa_offset 32
+; CHECK-NEXT: .cfi_offset w30, -8
+; CHECK-NEXT: .cfi_offset b8, -16
+; CHECK-NEXT: .cfi_offset b9, -24
+; CHECK-NEXT: .cfi_offset b10, -32
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fmov d8, d0
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fmov d9, d0
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fmov d10, d0
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fadd d1, d8, d9
+; CHECK-NEXT: ldp d9, d8, [sp, #8] // 16-byte Folded Reload
+; CHECK-NEXT: ldr x30, [sp, #24] // 8-byte Folded Reload
+; CHECK-NEXT: fadd d0, d10, d0
+; CHECK-NEXT: fadd d0, d1, d0
+; CHECK-NEXT: ldr d10, [sp], #32 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+ %x0 = call reassoc nsz double @bar()
+ %x1 = call reassoc nsz double @bar()
+ %x2 = call reassoc nsz double @bar()
+ %x3 = call reassoc nsz double @bar()
+ %t0 = fadd reassoc nsz double %x0, %x1
+ %t1 = fadd reassoc nsz double %t0, %x2
+ %t2 = fadd reassoc nsz double %t1, %x3
+ ret double %t2
+}
+
define double @already_reassociated() {
; CHECK-LABEL: already_reassociated:
; CHECK: // %bb.0:
@@ -846,3 +972,38 @@ define double @already_reassociated() {
%t2 = fadd double %t0, %t1
ret double %t2
}
+
+define double @already_reassociated_reassoc() {
+; CHECK-LABEL: already_reassociated_reassoc:
+; CHECK: // %bb.0:
+; CHECK-NEXT: str d10, [sp, #-32]! // 8-byte Folded Spill
+; CHECK-NEXT: stp d9, d8, [sp, #8] // 16-byte Folded Spill
+; CHECK-NEXT: str x30, [sp, #24] // 8-byte Folded Spill
+; CHECK-NEXT: .cfi_def_cfa_offset 32
+; CHECK-NEXT: .cfi_offset w30, -8
+; CHECK-NEXT: .cfi_offset b8, -16
+; CHECK-NEXT: .cfi_offset b9, -24
+; CHECK-NEXT: .cfi_offset b10, -32
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fmov d8, d0
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fmov d9, d0
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fmov d10, d0
+; CHECK-NEXT: bl bar
+; CHECK-NEXT: fadd d1, d8, d9
+; CHECK-NEXT: ldp d9, d8, [sp, #8] // 16-byte Folded Reload
+; CHECK-NEXT: ldr x30, [sp, #24] // 8-byte Folded Reload
+; CHECK-NEXT: fadd d0, d10, d0
+; CHECK-NEXT: fadd d0, d1, d0
+; CHECK-NEXT: ldr d10, [sp], #32 // 8-byte Folded Reload
+; CHECK-NEXT: ret
+ %x0 = call reassoc nsz double @bar()
+ %x1 = call reassoc nsz double @bar()
+ %x2 = call reassoc nsz double @bar()
+ %x3 = call reassoc nsz double @bar()
+ %t0 = fadd reassoc nsz double %x0, %x1
+ %t1 = fadd reassoc nsz double %x2, %x3
+ %t2 = fadd reassoc nsz double %t0, %t1
+ ret double %t2
+}
generated by cgit v1.1 at 2025-07-29 21:54:49 +0000