Skip to content

Reland [VectorUtils] Trivially vectorize ldexp, [l]lround #152476

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Aug 7, 2025
Merged

Reland [VectorUtils] Trivially vectorize ldexp, [l]lround #152476

merged 1 commit into from
Aug 7, 2025

Conversation

artagnon
Copy link
Contributor

@artagnon artagnon commented Aug 7, 2025

Changes: The original patch, landed as 1336675, was reverted due to a bug in LoopVectorize resulting in a crash. The bug has now been fixed by 95c32bf ([VPlan] Return invalid cost if any skeleton block has invalid costs), and this reland is identical to the original patch.

@artagnon
Reland [VectorUtils] Trivially vectorize ldexp, [l]lround
6dcacc0 
Changes: The original patch, landed as 1336675, was reverted due to a
bug in LoopVectorize resulting in a crash. The bug has now been fixed by
95c32bf ([VPlan] Return invalid cost if any skeleton block has invalid
costs), and this reland is identical to the original patch.
@artagnon artagnon requested review from fhahn and davemgreen August 7, 2025 10:19
@llvmbot llvmbot added llvm:analysis Includes value tracking, cost tables and constant folding llvm:transforms labels Aug 7, 2025
@llvmbot
Copy link
Member

llvmbot commented Aug 7, 2025 •
edited
Loading

@llvm/pr-subscribers-llvm-analysis

@llvm/pr-subscribers-llvm-transforms

Author: Ramkumar Ramachandra (artagnon)

Changes

Changes: The original patch, landed as 1336675, was reverted due to a bug in LoopVectorize resulting in a crash. The bug has now been fixed by 95c32bf ([VPlan] Return invalid cost if any skeleton block has invalid costs), and this reland is identical to the original patch.

Patch is 43.85 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/152476.diff

5 Files Affected:

  • (modified) llvm/lib/Analysis/VectorUtils.cpp (+6)
  • (modified) llvm/test/Transforms/LoopVectorize/intrinsic.ll (+201)
  • (added) llvm/test/Transforms/SLPVectorizer/AArch64/exp.ll (+279)
  • (added) llvm/test/Transforms/SLPVectorizer/AArch64/fround.ll (+280)
  • (modified) llvm/test/Transforms/Scalarizer/intrinsics.ll (+47)
diff --git a/llvm/lib/Analysis/VectorUtils.cpp b/llvm/lib/Analysis/VectorUtils.cpp
index b3b4c37475eef..425ea311d653a 100644
--- a/llvm/lib/Analysis/VectorUtils.cpp
+++ b/llvm/lib/Analysis/VectorUtils.cpp
@@ -81,6 +81,7 @@ bool llvm::isTriviallyVectorizable(Intrinsic::ID ID) {
   case Intrinsic::exp:
   case Intrinsic::exp10:
   case Intrinsic::exp2:
+  case Intrinsic::ldexp:
   case Intrinsic::log:
   case Intrinsic::log10:
   case Intrinsic::log2:
@@ -108,6 +109,8 @@ bool llvm::isTriviallyVectorizable(Intrinsic::ID ID) {
   case Intrinsic::canonicalize:
   case Intrinsic::fptosi_sat:
   case Intrinsic::fptoui_sat:
+  case Intrinsic::lround:
+  case Intrinsic::llround:
   case Intrinsic::lrint:
   case Intrinsic::llrint:
   case Intrinsic::ucmp:
@@ -189,6 +192,8 @@ bool llvm::isVectorIntrinsicWithOverloadTypeAtArg(
   switch (ID) {
   case Intrinsic::fptosi_sat:
   case Intrinsic::fptoui_sat:
+  case Intrinsic::lround:
+  case Intrinsic::llround:
   case Intrinsic::lrint:
   case Intrinsic::llrint:
   case Intrinsic::vp_lrint:
@@ -203,6 +208,7 @@ bool llvm::isVectorIntrinsicWithOverloadTypeAtArg(
   case Intrinsic::vp_is_fpclass:
     return OpdIdx == 0;
   case Intrinsic::powi:
+  case Intrinsic::ldexp:
     return OpdIdx == -1 || OpdIdx == 1;
   default:
     return OpdIdx == -1;
diff --git a/llvm/test/Transforms/LoopVectorize/intrinsic.ll b/llvm/test/Transforms/LoopVectorize/intrinsic.ll
index 9c910d70807a1..10d83a456d0e2 100644
--- a/llvm/test/Transforms/LoopVectorize/intrinsic.ll
+++ b/llvm/test/Transforms/LoopVectorize/intrinsic.ll
@@ -324,6 +324,56 @@ for.end:                                          ; preds = %for.body, %entry
 
 declare double @llvm.exp2.f64(double)
 
+define void @ldexp_f32i32(i32 %n, ptr %y, ptr %x, i32 %exp) {
+; CHECK-LABEL: @ldexp_f32i32(
+; CHECK: llvm.ldexp.v4f32.v4i32
+; CHECK: ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %iv = phi i32 [ 0, %entry ], [ %iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds float, ptr %y, i32 %iv
+  %0 = load float, ptr %arrayidx, align 4
+  %call = tail call float @llvm.ldexp.f32.i32(float %0, i32 %exp)
+  %arrayidx2 = getelementptr inbounds float, ptr %x, i32 %iv
+  store float %call, ptr %arrayidx2, align 4
+  %iv.next = add i32 %iv, 1
+  %exitcond = icmp eq i32 %iv.next, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+declare float @llvm.ldexp.f32.i32(float, i32)
+
+define void @ldexp_f64i32(i32 %n, ptr %y, ptr %x, i32 %exp) {
+; CHECK-LABEL: @ldexp_f64i32(
+; CHECK: llvm.ldexp.v4f64.v4i32
+; CHECK: ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %iv = phi i32 [ 0, %entry ], [ %iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds double, ptr %y, i32 %iv
+  %0 = load double, ptr %arrayidx, align 8
+  %call = tail call double @llvm.ldexp.f64.i32(double %0, i32 %exp)
+  %arrayidx2 = getelementptr inbounds double, ptr %x, i32 %iv
+  store double %call, ptr %arrayidx2, align 8
+  %iv.next = add i32 %iv, 1
+  %exitcond = icmp eq i32 %iv.next, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+declare double @llvm.ldexp.f64.i32(double, i32)
+
 define void @log_f32(i32 %n, ptr %y, ptr %x) {
 ; CHECK-LABEL: @log_f32(
 ; CHECK: llvm.log.v4f32
@@ -976,6 +1026,157 @@ for.end:                                          ; preds = %for.body, %entry
 
 declare double @llvm.roundeven.f64(double)
 
+
+define void @lround_i32f32(i32 %n, ptr %y, ptr %x) {
+; CHECK-LABEL: @lround_i32f32(
+; CHECK: llvm.lround.v4i32.v4f32
+; CHECK: ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %iv = phi i32 [ 0, %entry ], [ %iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds float, ptr %y, i32 %iv
+  %0 = load float, ptr %arrayidx, align 4
+  %call = tail call i32 @llvm.lround.i32.f32(float %0)
+  %arrayidx2 = getelementptr inbounds i32, ptr %x, i32 %iv
+  store i32 %call, ptr %arrayidx2, align 4
+  %iv.next = add i32 %iv, 1
+  %exitcond = icmp eq i32 %iv.next, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+declare i32 @llvm.lround.i32.f32(float)
+
+define void @lround_i32f64(i32 %n, ptr %y, ptr %x) {
+; CHECK-LABEL: @lround_i32f64(
+; CHECK: llvm.lround.v4i32.v4f64
+; CHECK: ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %iv = phi i32 [ 0, %entry ], [ %iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds double, ptr %y, i32 %iv
+  %0 = load double, ptr %arrayidx, align 8
+  %call = tail call i32 @llvm.lround.i32.f64(double %0)
+  %arrayidx2 = getelementptr inbounds i32, ptr %x, i32 %iv
+  store i32 %call, ptr %arrayidx2, align 8
+  %iv.next = add i32 %iv, 1
+  %exitcond = icmp eq i32 %iv.next, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+declare i32 @llvm.lround.i32.f64(double)
+
+define void @lround_i64f32(i32 %n, ptr %y, ptr %x) {
+; CHECK-LABEL: @lround_i64f32(
+; CHECK: llvm.lround.v4i64.v4f32
+; CHECK: ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %iv = phi i32 [ 0, %entry ], [ %iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds float, ptr %y, i32 %iv
+  %0 = load float, ptr %arrayidx, align 4
+  %call = tail call i64 @llvm.lround.i64.f32(float %0)
+  %arrayidx2 = getelementptr inbounds i64, ptr %x, i32 %iv
+  store i64 %call, ptr %arrayidx2, align 4
+  %iv.next = add i32 %iv, 1
+  %exitcond = icmp eq i32 %iv.next, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+declare i64 @llvm.lround.i64.f32(float)
+
+define void @lround_i64f64(i32 %n, ptr %y, ptr %x) {
+; CHECK-LABEL: @lround_i64f64(
+; CHECK: llvm.lround.v4i64.v4f64
+; CHECK: ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %iv = phi i32 [ 0, %entry ], [ %iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds double, ptr %y, i32 %iv
+  %0 = load double, ptr %arrayidx, align 8
+  %call = tail call i64 @llvm.lround.i64.f64(double %0)
+  %arrayidx2 = getelementptr inbounds i64, ptr %x, i32 %iv
+  store i64 %call, ptr %arrayidx2, align 8
+  %iv.next = add i32 %iv, 1
+  %exitcond = icmp eq i32 %iv.next, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+declare i64 @llvm.lround.i64.f64(double)
+
+define void @llround_i64f32(i32 %n, ptr %y, ptr %x) {
+; CHECK-LABEL: @llround_i64f32(
+; CHECK: llvm.llround.v4i64.v4f32
+; CHECK: ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %iv = phi i32 [ 0, %entry ], [ %iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds float, ptr %y, i32 %iv
+  %0 = load float, ptr %arrayidx, align 4
+  %call = tail call i64 @llvm.llround.i64.f32(float %0)
+  %arrayidx2 = getelementptr inbounds i64, ptr %x, i32 %iv
+  store i64 %call, ptr %arrayidx2, align 4
+  %iv.next = add i32 %iv, 1
+  %exitcond = icmp eq i32 %iv.next, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+declare i64 @llvm.llround.i64.f32(float)
+
+define void @llround_i64f64(i32 %n, ptr %y, ptr %x) {
+; CHECK-LABEL: @llround_i64f64(
+; CHECK: llvm.llround.v4i64.v4f64
+; CHECK: ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %iv = phi i32 [ 0, %entry ], [ %iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds double, ptr %y, i32 %iv
+  %0 = load double, ptr %arrayidx, align 8
+  %call = tail call i64 @llvm.llround.i64.f64(double %0)
+  %arrayidx2 = getelementptr inbounds i64, ptr %x, i32 %iv
+  store i64 %call, ptr %arrayidx2, align 8
+  %iv.next = add i32 %iv, 1
+  %exitcond = icmp eq i32 %iv.next, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+declare i64 @llvm.llround.i64.f64(double)
+
 define void @fma_f32(i32 %n, ptr %y, ptr %x, ptr %z, ptr %w) {
 ; CHECK-LABEL: @fma_f32(
 ; CHECK: llvm.fma.v4f32
diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/exp.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/exp.ll
new file mode 100644
index 0000000000000..301e5da28f0d8
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/AArch64/exp.ll
@@ -0,0 +1,279 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -S -passes=slp-vectorizer -mtriple=aarch64 < %s | FileCheck %s
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+
+define void @ldexp_f32i32(ptr %x, ptr %y, i32 %exp) {
+; CHECK-LABEL: @ldexp_f32i32(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[L0:%.*]] = load float, ptr [[X:%.*]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds float, ptr [[X]], i64 1
+; CHECK-NEXT:    [[L2:%.*]] = load float, ptr [[ARRAYIDX_1]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds float, ptr [[X]], i64 2
+; CHECK-NEXT:    [[L4:%.*]] = load float, ptr [[ARRAYIDX_2]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds float, ptr [[X]], i64 3
+; CHECK-NEXT:    [[L6:%.*]] = load float, ptr [[ARRAYIDX_3]], align 4
+; CHECK-NEXT:    [[L1:%.*]] = tail call float @llvm.ldexp.f32.i32(float [[L0]], i32 [[EXP:%.*]])
+; CHECK-NEXT:    [[L3:%.*]] = tail call float @llvm.ldexp.f32.i32(float [[L2]], i32 [[EXP]])
+; CHECK-NEXT:    [[L5:%.*]] = tail call float @llvm.ldexp.f32.i32(float [[L4]], i32 [[EXP]])
+; CHECK-NEXT:    [[L7:%.*]] = tail call float @llvm.ldexp.f32.i32(float [[L6]], i32 [[EXP]])
+; CHECK-NEXT:    store float [[L1]], ptr [[Y:%.*]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_1:%.*]] = getelementptr inbounds float, ptr [[Y]], i64 1
+; CHECK-NEXT:    store float [[L3]], ptr [[ARRAYIDX2_1]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_2:%.*]] = getelementptr inbounds float, ptr [[Y]], i64 2
+; CHECK-NEXT:    store float [[L5]], ptr [[ARRAYIDX2_2]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_3:%.*]] = getelementptr inbounds float, ptr [[Y]], i64 3
+; CHECK-NEXT:    store float [[L7]], ptr [[ARRAYIDX2_3]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %l0 = load float, ptr %x, align 4
+  %arrayidx.1 = getelementptr inbounds float, ptr %x, i64 1
+  %l2 = load float, ptr %arrayidx.1, align 4
+  %arrayidx.2 = getelementptr inbounds float, ptr %x, i64 2
+  %l4 = load float, ptr %arrayidx.2, align 4
+  %arrayidx.3 = getelementptr inbounds float, ptr %x, i64 3
+  %l6 = load float, ptr %arrayidx.3, align 4
+  %l1 = tail call float @llvm.ldexp.f32.i32(float %l0, i32 %exp)
+  %l3 = tail call float @llvm.ldexp.f32.i32(float %l2, i32 %exp)
+  %l5 = tail call float @llvm.ldexp.f32.i32(float %l4, i32 %exp)
+  %l7 = tail call float @llvm.ldexp.f32.i32(float %l6, i32 %exp)
+  store float %l1, ptr %y, align 4
+  %arrayidx2.1 = getelementptr inbounds float, ptr %y, i64 1
+  store float %l3, ptr %arrayidx2.1, align 4
+  %arrayidx2.2 = getelementptr inbounds float, ptr %y, i64 2
+  store float %l5, ptr %arrayidx2.2, align 4
+  %arrayidx2.3 = getelementptr inbounds float, ptr %y, i64 3
+  store float %l7, ptr %arrayidx2.3, align 4
+  ret void
+}
+
+define void @ldexp_f64i32(ptr %x, ptr %y, i32 %exp) {
+; CHECK-LABEL: @ldexp_f64i32(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[L0:%.*]] = load double, ptr [[X:%.*]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds double, ptr [[X]], i64 1
+; CHECK-NEXT:    [[L2:%.*]] = load double, ptr [[ARRAYIDX_1]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds double, ptr [[X]], i64 2
+; CHECK-NEXT:    [[L4:%.*]] = load double, ptr [[ARRAYIDX_2]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds double, ptr [[X]], i64 3
+; CHECK-NEXT:    [[L6:%.*]] = load double, ptr [[ARRAYIDX_3]], align 4
+; CHECK-NEXT:    [[L1:%.*]] = tail call double @llvm.ldexp.f64.i32(double [[L0]], i32 [[EXP:%.*]])
+; CHECK-NEXT:    [[L3:%.*]] = tail call double @llvm.ldexp.f64.i32(double [[L2]], i32 [[EXP]])
+; CHECK-NEXT:    [[L5:%.*]] = tail call double @llvm.ldexp.f64.i32(double [[L4]], i32 [[EXP]])
+; CHECK-NEXT:    [[L7:%.*]] = tail call double @llvm.ldexp.f64.i32(double [[L6]], i32 [[EXP]])
+; CHECK-NEXT:    store double [[L1]], ptr [[Y:%.*]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_1:%.*]] = getelementptr inbounds double, ptr [[Y]], i64 1
+; CHECK-NEXT:    store double [[L3]], ptr [[ARRAYIDX2_1]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_2:%.*]] = getelementptr inbounds double, ptr [[Y]], i64 2
+; CHECK-NEXT:    store double [[L5]], ptr [[ARRAYIDX2_2]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_3:%.*]] = getelementptr inbounds double, ptr [[Y]], i64 3
+; CHECK-NEXT:    store double [[L7]], ptr [[ARRAYIDX2_3]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %l0 = load double, ptr %x, align 4
+  %arrayidx.1 = getelementptr inbounds double, ptr %x, i64 1
+  %l2 = load double, ptr %arrayidx.1, align 4
+  %arrayidx.2 = getelementptr inbounds double, ptr %x, i64 2
+  %l4 = load double, ptr %arrayidx.2, align 4
+  %arrayidx.3 = getelementptr inbounds double, ptr %x, i64 3
+  %l6 = load double, ptr %arrayidx.3, align 4
+  %l1 = tail call double @llvm.ldexp.f64.i32(double %l0, i32 %exp)
+  %l3 = tail call double @llvm.ldexp.f64.i32(double %l2, i32 %exp)
+  %l5 = tail call double @llvm.ldexp.f64.i32(double %l4, i32 %exp)
+  %l7 = tail call double @llvm.ldexp.f64.i32(double %l6, i32 %exp)
+  store double %l1, ptr %y, align 4
+  %arrayidx2.1 = getelementptr inbounds double, ptr %y, i64 1
+  store double %l3, ptr %arrayidx2.1, align 4
+  %arrayidx2.2 = getelementptr inbounds double, ptr %y, i64 2
+  store double %l5, ptr %arrayidx2.2, align 4
+  %arrayidx2.3 = getelementptr inbounds double, ptr %y, i64 3
+  store double %l7, ptr %arrayidx2.3, align 4
+  ret void
+}
+
+define void @ldexp_f32i64(ptr %x, ptr %y, i64 %exp) {
+; CHECK-LABEL: @ldexp_f32i64(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[L0:%.*]] = load float, ptr [[X:%.*]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds float, ptr [[X]], i64 1
+; CHECK-NEXT:    [[L2:%.*]] = load float, ptr [[ARRAYIDX_1]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds float, ptr [[X]], i64 2
+; CHECK-NEXT:    [[L4:%.*]] = load float, ptr [[ARRAYIDX_2]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds float, ptr [[X]], i64 3
+; CHECK-NEXT:    [[L6:%.*]] = load float, ptr [[ARRAYIDX_3]], align 4
+; CHECK-NEXT:    [[L1:%.*]] = tail call float @llvm.ldexp.f32.i64(float [[L0]], i64 [[EXP:%.*]])
+; CHECK-NEXT:    [[L3:%.*]] = tail call float @llvm.ldexp.f32.i64(float [[L2]], i64 [[EXP]])
+; CHECK-NEXT:    [[L5:%.*]] = tail call float @llvm.ldexp.f32.i64(float [[L4]], i64 [[EXP]])
+; CHECK-NEXT:    [[L7:%.*]] = tail call float @llvm.ldexp.f32.i64(float [[L6]], i64 [[EXP]])
+; CHECK-NEXT:    store float [[L1]], ptr [[Y:%.*]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_1:%.*]] = getelementptr inbounds float, ptr [[Y]], i64 1
+; CHECK-NEXT:    store float [[L3]], ptr [[ARRAYIDX2_1]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_2:%.*]] = getelementptr inbounds float, ptr [[Y]], i64 2
+; CHECK-NEXT:    store float [[L5]], ptr [[ARRAYIDX2_2]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_3:%.*]] = getelementptr inbounds float, ptr [[Y]], i64 3
+; CHECK-NEXT:    store float [[L7]], ptr [[ARRAYIDX2_3]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %l0 = load float, ptr %x, align 4
+  %arrayidx.1 = getelementptr inbounds float, ptr %x, i64 1
+  %l2 = load float, ptr %arrayidx.1, align 4
+  %arrayidx.2 = getelementptr inbounds float, ptr %x, i64 2
+  %l4 = load float, ptr %arrayidx.2, align 4
+  %arrayidx.3 = getelementptr inbounds float, ptr %x, i64 3
+  %l6 = load float, ptr %arrayidx.3, align 4
+  %l1 = tail call float @llvm.ldexp.f32.i64(float %l0, i64 %exp)
+  %l3 = tail call float @llvm.ldexp.f32.i64(float %l2, i64 %exp)
+  %l5 = tail call float @llvm.ldexp.f32.i64(float %l4, i64 %exp)
+  %l7 = tail call float @llvm.ldexp.f32.i64(float %l6, i64 %exp)
+  store float %l1, ptr %y, align 4
+  %arrayidx2.1 = getelementptr inbounds float, ptr %y, i64 1
+  store float %l3, ptr %arrayidx2.1, align 4
+  %arrayidx2.2 = getelementptr inbounds float, ptr %y, i64 2
+  store float %l5, ptr %arrayidx2.2, align 4
+  %arrayidx2.3 = getelementptr inbounds float, ptr %y, i64 3
+  store float %l7, ptr %arrayidx2.3, align 4
+  ret void
+}
+
+define void @ldexp_f64i64(ptr %x, ptr %y, i64 %exp) {
+; CHECK-LABEL: @ldexp_f64i64(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[L0:%.*]] = load double, ptr [[X:%.*]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds double, ptr [[X]], i64 1
+; CHECK-NEXT:    [[L2:%.*]] = load double, ptr [[ARRAYIDX_1]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds double, ptr [[X]], i64 2
+; CHECK-NEXT:    [[L4:%.*]] = load double, ptr [[ARRAYIDX_2]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds double, ptr [[X]], i64 3
+; CHECK-NEXT:    [[L6:%.*]] = load double, ptr [[ARRAYIDX_3]], align 4
+; CHECK-NEXT:    [[L1:%.*]] = tail call double @llvm.ldexp.f64.i64(double [[L0]], i64 [[EXP:%.*]])
+; CHECK-NEXT:    [[L3:%.*]] = tail call double @llvm.ldexp.f64.i64(double [[L2]], i64 [[EXP]])
+; CHECK-NEXT:    [[L5:%.*]] = tail call double @llvm.ldexp.f64.i64(double [[L4]], i64 [[EXP]])
+; CHECK-NEXT:    [[L7:%.*]] = tail call double @llvm.ldexp.f64.i64(double [[L6]], i64 [[EXP]])
+; CHECK-NEXT:    store double [[L1]], ptr [[Y:%.*]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_1:%.*]] = getelementptr inbounds double, ptr [[Y]], i64 1
+; CHECK-NEXT:    store double [[L3]], ptr [[ARRAYIDX2_1]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_2:%.*]] = getelementptr inbounds double, ptr [[Y]], i64 2
+; CHECK-NEXT:    store double [[L5]], ptr [[ARRAYIDX2_2]], align 4
+; CHECK-NEXT:    [[ARRAYIDX2_3:%.*]] = getelementptr inbounds double, ptr [[Y]], i64 3
+; CHECK-NEXT:    store double [[L7]], ptr [[ARRAYIDX2_3]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %l0 = load double, ptr %x, align 4
+  %arrayidx.1 = getelementptr inbounds double, ptr %x, i64 1
+  %l2 = load double, ptr %arrayidx.1, align 4
+  %arrayidx.2 = getelementptr inbounds double, ptr %x, i64 2
+  %l4 = load double, ptr %arrayidx.2, align 4
+  %arrayidx.3 = getelementptr inbounds double, ptr %x, i64 3
+  %l6 = load double, ptr %arrayidx.3, align 4
+  %l1 = tail call double @llvm.ldexp.f64.i64(double %l0, i64 %exp)
+  %l3 = tail call double @llvm.ldexp.f64.i64(double %l2, i64 %exp)
+  %l5 = tail call double @llvm.ldexp.f64.i64(double %l4, i64 %exp)
+  %l7 = tail call double @llvm.ldexp.f64.i64(double %l6, i64 %exp)
+  store double %l1, ptr %y, align 4
+  %arrayidx2.1 = getelementptr inbounds double, ptr %y, i64 1
+  store double %l3, ptr %arrayidx2.1, align 4
+  %arrayidx2.2 = getelementptr inbounds double, ptr %y, i64 2
+  store double %l5, ptr %arrayidx2.2, align 4
+  %arrayidx2.3 = getelementptr inbounds double, ptr %y, i64 3
+  store double %l7, ptr %arrayidx2.3, align 4
+  ret void
+}
+
+define void @ldexp_f32i32_i64(ptr %x, ptr %y, i32 %exp32, i64 %exp64) {
+; CHECK-LABEL: @ldexp_f32i32_i64(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[L0:%.*]] = load float, ptr [[X:%.*]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds float, ptr [[X]], i64 1
+; CHECK-NEXT:    [[L2:%.*]] = load float, ptr [[ARRAYIDX_1]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds float, ptr [[X]], i64 2
+; CHECK-NEXT:    [[L4:%.*]] = load float, ptr [[ARRAYIDX_2]], align 4
+; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds float, ptr [[X]], i64 3
+; CHECK-NEXT:    [[L6:%.*]] = load float, ptr [[ARRAYIDX_3]], align 4
+; CHECK-NEXT:    [[L1:%.*]] = tail call float @llvm.l...
[truncated]

fhahn
fhahn approved these changes Aug 7, 2025
View reviewed changes
Copy link
Contributor

@fhahn fhahn left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks. The issue that caused the revert has been fixed, should be good to go now

@artagnon artagnon merged commit edeee82 into llvm:main Aug 7, 2025
12 checks passed
@artagnon artagnon deleted the triviallyvec-missing-reland branch August 7, 2025 11:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@fhahn fhahn fhahn approved these changes

@davemgreen davemgreen Awaiting requested review from davemgreen

Assignees
No one assigned
Labels
llvm:analysis Includes value tracking, cost tables and constant folding llvm:transforms
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@artagnon @llvmbot @fhahn