[llvm] 5eb2ec2 - InstCombine: Fold ldexp with constant exponent to fmul (#171731)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Dec 11 10:20:49 PST 2025
Author: Matt Arsenault
Date: 2025-12-11T19:20:45+01:00
New Revision: 5eb2ec2179b2b95be16c7f855fe8460b4a98b5b1
URL: https://github.com/llvm/llvm-project/commit/5eb2ec2179b2b95be16c7f855fe8460b4a98b5b1
DIFF: https://github.com/llvm/llvm-project/commit/5eb2ec2179b2b95be16c7f855fe8460b4a98b5b1.diff
LOG: InstCombine: Fold ldexp with constant exponent to fmul (#171731)
If we can represent this with an fmul, prefer it as a canonical
form. More optimizations will understand fmul, and allows contract to
fma.
Added:
Modified:
llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
llvm/test/Transforms/InstCombine/fold-select-fmul-if-zero.ll
llvm/test/Transforms/InstCombine/ldexp.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index 85602a5a7575a..b498bafb3caaa 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -3089,6 +3089,19 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst &CI) {
// exponent. Also could broaden sign check to cover == 0 case.
Value *Src = II->getArgOperand(0);
Value *Exp = II->getArgOperand(1);
+
+ uint64_t ConstExp;
+ if (match(Exp, m_ConstantInt(ConstExp))) {
+ // ldexp(x, K) -> fmul x, 2^K
+ const fltSemantics &FPTy =
+ Src->getType()->getScalarType()->getFltSemantics();
+ Constant *FPConst =
+ ConstantFP::get(Src->getType(), scalbn(APFloat::getOne(FPTy),
+ static_cast<int>(ConstExp),
+ APFloat::rmNearestTiesToEven));
+ return BinaryOperator::CreateFMulFMF(Src, FPConst, II);
+ }
+
Value *InnerSrc;
Value *InnerExp;
if (match(Src, m_OneUse(m_Intrinsic<Intrinsic::ldexp>(
diff --git a/llvm/test/Transforms/InstCombine/fold-select-fmul-if-zero.ll b/llvm/test/Transforms/InstCombine/fold-select-fmul-if-zero.ll
index 1ba7005f99e3d..4495b0f26042b 100644
--- a/llvm/test/Transforms/InstCombine/fold-select-fmul-if-zero.ll
+++ b/llvm/test/Transforms/InstCombine/fold-select-fmul-if-zero.ll
@@ -49,10 +49,7 @@ define float @fmul_by_32_if_0_oeq_zero_f32(float %x) {
define float @ldexp_by_5_if_0_oeq_zero_f32(float %x) {
; CHECK-LABEL: @ldexp_by_5_if_0_oeq_zero_f32(
-; CHECK-NEXT: [[X_IS_ZERO:%.*]] = fcmp oeq float [[X:%.*]], 0.000000e+00
-; CHECK-NEXT: [[SCALED_X:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 5)
-; CHECK-NEXT: [[SCALED_IF_DENORMAL:%.*]] = select i1 [[X_IS_ZERO]], float [[SCALED_X]], float [[X]]
-; CHECK-NEXT: ret float [[SCALED_IF_DENORMAL]]
+; CHECK-NEXT: ret float [[X:%.*]]
;
%x.is.zero = fcmp oeq float %x, 0.0
%scaled.x = call float @llvm.ldexp.f32.i32(float %x, i32 5)
@@ -62,10 +59,7 @@ define float @ldexp_by_5_if_0_oeq_zero_f32(float %x) {
define <2 x float> @ldexp_by_5_if_0_oeq_zero_v2f32(<2 x float> %x) {
; CHECK-LABEL: @ldexp_by_5_if_0_oeq_zero_v2f32(
-; CHECK-NEXT: [[X_IS_ZERO:%.*]] = fcmp oeq <2 x float> [[X:%.*]], zeroinitializer
-; CHECK-NEXT: [[SCALED_X:%.*]] = call <2 x float> @llvm.ldexp.v2f32.v2i32(<2 x float> [[X]], <2 x i32> splat (i32 5))
-; CHECK-NEXT: [[SCALED_IF_DENORMAL:%.*]] = select <2 x i1> [[X_IS_ZERO]], <2 x float> [[SCALED_X]], <2 x float> [[X]]
-; CHECK-NEXT: ret <2 x float> [[SCALED_IF_DENORMAL]]
+; CHECK-NEXT: ret <2 x float> [[SCALED_IF_DENORMAL:%.*]]
;
%x.is.zero = fcmp oeq <2 x float> %x, zeroinitializer
%scaled.x = call <2 x float> @llvm.ldexp.v2f32.v2i32(<2 x float> %x, <2 x i32> <i32 5, i32 5>)
diff --git a/llvm/test/Transforms/InstCombine/ldexp.ll b/llvm/test/Transforms/InstCombine/ldexp.ll
index 8908d476b4a2c..43e80b1a1e588 100644
--- a/llvm/test/Transforms/InstCombine/ldexp.ll
+++ b/llvm/test/Transforms/InstCombine/ldexp.ll
@@ -444,7 +444,7 @@ define float @ldexp_ldexp_
diff erent_exp_type(float %x, i32 %a, i64 %b) {
define float @ldexp_ldexp_constants(float %x) {
; CHECK-LABEL: define float @ldexp_ldexp_constants
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP1:%.*]] = call reassoc float @llvm.ldexp.f32.i32(float [[X]], i32 32)
+; CHECK-NEXT: [[LDEXP1:%.*]] = fmul reassoc float [[X]], 0x41F0000000000000
; CHECK-NEXT: ret float [[LDEXP1]]
;
%ldexp0 = call reassoc float @llvm.ldexp.f32.i32(float %x, i32 8)
@@ -455,7 +455,7 @@ define float @ldexp_ldexp_constants(float %x) {
define float @ldexp_ldexp_constants_nsz(float %x) {
; CHECK-LABEL: define float @ldexp_ldexp_constants_nsz
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP1:%.*]] = call reassoc nsz float @llvm.ldexp.f32.i32(float [[X]], i32 32)
+; CHECK-NEXT: [[LDEXP1:%.*]] = fmul reassoc nsz float [[X]], 0x41F0000000000000
; CHECK-NEXT: ret float [[LDEXP1]]
;
%ldexp0 = call reassoc nsz float @llvm.ldexp.f32.i32(float %x, i32 8)
@@ -466,7 +466,7 @@ define float @ldexp_ldexp_constants_nsz(float %x) {
define float @ldexp_ldexp_constants_nsz0(float %x) {
; CHECK-LABEL: define float @ldexp_ldexp_constants_nsz0
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP1:%.*]] = call reassoc nsz float @llvm.ldexp.f32.i32(float [[X]], i32 32)
+; CHECK-NEXT: [[LDEXP1:%.*]] = fmul reassoc float [[X]], 0x41F0000000000000
; CHECK-NEXT: ret float [[LDEXP1]]
;
%ldexp0 = call reassoc nsz float @llvm.ldexp.f32.i32(float %x, i32 8)
@@ -477,7 +477,7 @@ define float @ldexp_ldexp_constants_nsz0(float %x) {
define float @ldexp_ldexp_constants_nsz1(float %x) {
; CHECK-LABEL: define float @ldexp_ldexp_constants_nsz1
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP1:%.*]] = call reassoc nsz float @llvm.ldexp.f32.i32(float [[X]], i32 32)
+; CHECK-NEXT: [[LDEXP1:%.*]] = fmul reassoc nsz float [[X]], 0x41F0000000000000
; CHECK-NEXT: ret float [[LDEXP1]]
;
%ldexp0 = call reassoc float @llvm.ldexp.f32.i32(float %x, i32 8)
@@ -641,7 +641,7 @@ define float @ldexp_ldexp_0(float %x, i32 %y) {
define float @ldexp_neg150(float %x) {
; CHECK-LABEL: define float @ldexp_neg150
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -150)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0.000000e+00
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -150)
@@ -651,7 +651,7 @@ define float @ldexp_neg150(float %x) {
define float @ldexp_neg149(float %x) {
; CHECK-LABEL: define float @ldexp_neg149
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -149)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0x36A0000000000000
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -149)
@@ -661,7 +661,7 @@ define float @ldexp_neg149(float %x) {
define float @ldexp_neg148(float %x) {
; CHECK-LABEL: define float @ldexp_neg148
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -148)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0x36B0000000000000
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -148)
@@ -671,7 +671,7 @@ define float @ldexp_neg148(float %x) {
define float @ldexp_neg127(float %x) {
; CHECK-LABEL: define float @ldexp_neg127
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -127)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0x3800000000000000
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -127)
@@ -681,7 +681,7 @@ define float @ldexp_neg127(float %x) {
define float @ldexp_neg126(float %x) {
; CHECK-LABEL: define float @ldexp_neg126
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -126)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0x3810000000000000
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -126)
@@ -691,7 +691,7 @@ define float @ldexp_neg126(float %x) {
define float @ldexp_neg125(float %x) {
; CHECK-LABEL: define float @ldexp_neg125
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -125)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0x3820000000000000
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -125)
@@ -701,7 +701,7 @@ define float @ldexp_neg125(float %x) {
define float @ldexp_neg16(float %x) {
; CHECK-LABEL: define float @ldexp_neg16
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -16)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0x3EF0000000000000
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -16)
@@ -711,7 +711,7 @@ define float @ldexp_neg16(float %x) {
define float @ldexp_neg8(float %x) {
; CHECK-LABEL: define float @ldexp_neg8
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -8)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 3.906250e-03
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -8)
@@ -721,7 +721,7 @@ define float @ldexp_neg8(float %x) {
define float @ldexp_neg4(float %x) {
; CHECK-LABEL: define float @ldexp_neg4
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -4)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 6.250000e-02
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -4)
@@ -731,7 +731,7 @@ define float @ldexp_neg4(float %x) {
define float @ldexp_neg2(float %x) {
; CHECK-LABEL: define float @ldexp_neg2
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -2)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 2.500000e-01
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -2)
@@ -741,7 +741,7 @@ define float @ldexp_neg2(float %x) {
define float @ldexp_neg1(float %x) {
; CHECK-LABEL: define float @ldexp_neg1
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 -1)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 5.000000e-01
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 -1)
@@ -760,7 +760,7 @@ define float @ldexp_0(float %x) {
define float @ldexp_1(float %x) {
; CHECK-LABEL: define float @ldexp_1
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 1)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 2.000000e+00
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 1)
@@ -770,7 +770,7 @@ define float @ldexp_1(float %x) {
define float @ldexp_2(float %x) {
; CHECK-LABEL: define float @ldexp_2
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 2)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 4.000000e+00
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 2)
@@ -780,7 +780,7 @@ define float @ldexp_2(float %x) {
define float @ldexp_3(float %x) {
; CHECK-LABEL: define float @ldexp_3
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 3)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 8.000000e+00
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 3)
@@ -790,7 +790,7 @@ define float @ldexp_3(float %x) {
define float @ldexp_10(float %x) {
; CHECK-LABEL: define float @ldexp_10
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 10)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 1.024000e+03
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 10)
@@ -800,7 +800,7 @@ define float @ldexp_10(float %x) {
define float @ldexp_125(float %x) {
; CHECK-LABEL: define float @ldexp_125
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 125)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0x47C0000000000000
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 125)
@@ -810,7 +810,7 @@ define float @ldexp_125(float %x) {
define float @ldexp_126(float %x) {
; CHECK-LABEL: define float @ldexp_126
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 126)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0x47D0000000000000
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 126)
@@ -820,17 +820,27 @@ define float @ldexp_126(float %x) {
define float @ldexp_127(float %x) {
; CHECK-LABEL: define float @ldexp_127
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 127)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0x47E0000000000000
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 127)
ret float %ldexp
}
+define float @ldexp_128(float %x) {
+; CHECK-LABEL: define float @ldexp_128
+; CHECK-SAME: (float [[X:%.*]]) {
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 0x7FF0000000000000
+; CHECK-NEXT: ret float [[LDEXP]]
+;
+ %ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 128)
+ ret float %ldexp
+}
+
define <2 x float> @ldexp_3_vector(<2 x float> %x) {
; CHECK-LABEL: define <2 x float> @ldexp_3_vector
; CHECK-SAME: (<2 x float> [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call <2 x float> @llvm.ldexp.v2f32.v2i32(<2 x float> [[X]], <2 x i32> splat (i32 3))
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul <2 x float> [[X]], splat (float 8.000000e+00)
; CHECK-NEXT: ret <2 x float> [[LDEXP]]
;
%ldexp = call <2 x float> @llvm.ldexp.v2f32.v2i32(<2 x float> %x, <2 x i32> <i32 3, i32 3>)
@@ -860,7 +870,7 @@ define <2 x float> @ldexp_3_4_vector(<2 x float> %x) {
define float @ldexp_2_flags(float %x) {
; CHECK-LABEL: define float @ldexp_2_flags
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call nsz contract float @llvm.ldexp.f32.i32(float [[X]], i32 2)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul nsz contract float [[X]], 4.000000e+00
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call contract nsz float @llvm.ldexp.f32.i32(float %x, i32 2)
@@ -870,7 +880,7 @@ define float @ldexp_2_flags(float %x) {
define float @ldexp_metadata(float %x) {
; CHECK-LABEL: define float @ldexp_metadata
; CHECK-SAME: (float [[X:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call float @llvm.ldexp.f32.i32(float [[X]], i32 2), !foo [[META2:![0-9]+]]
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul float [[X]], 4.000000e+00
; CHECK-NEXT: ret float [[LDEXP]]
;
%ldexp = call float @llvm.ldexp.f32.i32(float %x, i32 2), !foo !2
@@ -880,7 +890,7 @@ define float @ldexp_metadata(float %x) {
define float @ldexp_8_contractable(float %x, float %y) {
; CHECK-LABEL: define float @ldexp_8_contractable
; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]]) {
-; CHECK-NEXT: [[LDEXP:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[X]], i32 2)
+; CHECK-NEXT: [[LDEXP:%.*]] = fmul contract float [[X]], 4.000000e+00
; CHECK-NEXT: [[FADD:%.*]] = fadd contract float [[LDEXP]], [[Y]]
; CHECK-NEXT: ret float [[FADD]]
;
More information about the llvm-commits
mailing list