[llvm] [InstCombine] Try to fold add into GEP x, C (PR #85090)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Mar 13 07:45:00 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-transforms
Author: David Green (davemgreen)
<details>
<summary>Changes</summary>
This attempts to GEP p, (x + C1), C2 with GEP p, x, C2+C1*S, removing the need
for the add by folding it into the following constant offset in the GEP. It
does not attempt to alter inbounds geps, to avoid the need to drop the
inbounds.
Example proof using a much reduced pointer address space:
https://alive2.llvm.org/ce/z/hG8xfF
---
Full diff: https://github.com/llvm/llvm-project/pull/85090.diff
2 Files Affected:
- (modified) llvm/lib/Transforms/InstCombine/InstructionCombining.cpp (+31)
- (added) llvm/test/Transforms/InstCombine/gepadd.ll (+108)
``````````diff
diff --git a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
index 1688005de2104d..7932822c0f7464 100644
--- a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -2893,6 +2893,37 @@ Instruction *InstCombinerImpl::visitGetElementPtrInst(GetElementPtrInst &GEP) {
}
}
}
+
+ // Try to replace GEP p, (x + C1), C2 with GEP p, x, C2+C1*S
+ if (GEP.getNumIndices() > 1) {
+ gep_type_iterator GTI = gep_type_begin(GEP);
+ for (User::op_iterator I = GEP.op_begin() + 1, E = GEP.op_end() - 1;
+ I != E; ++I, ++GTI) {
+ if (!GTI.isSequential())
+ break;
+ Value *X;
+ const APInt *C1, *C2;
+ User::op_iterator Next = std::next(I);
+ if (match(I->get(), m_Add(m_Value(X), m_APInt(C1))) &&
+ match(Next->get(), m_APInt(C2))) {
+ TypeSize Scale1 = GTI.getSequentialElementStride(DL);
+ if (Scale1.isScalable() || !(++GTI).isSequential())
+ break;
+ TypeSize Scale2 = GTI.getSequentialElementStride(DL);
+ if (Scale2.isScalable())
+ break;
+
+ // Update the GEP instruction indices, and add Add to the worklist
+ // so that it can be DCEd.
+ Instruction *Add = cast<Instruction>(*I);
+ *I = X;
+ *Next = ConstantInt::get((*Next)->getType(),
+ *C2 + *C1 * (Scale1 / Scale2));
+ addToWorklist(Add);
+ return &GEP;
+ }
+ }
+ }
}
if (Instruction *R = foldSelectGEP(GEP, Builder))
diff --git a/llvm/test/Transforms/InstCombine/gepadd.ll b/llvm/test/Transforms/InstCombine/gepadd.ll
new file mode 100644
index 00000000000000..895bb94ca99576
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/gepadd.ll
@@ -0,0 +1,108 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
+; RUN: opt < %s -S -passes=instcombine | FileCheck %s
+
+target datalayout = "e-p:64:64:64"
+
+define ptr @add1(ptr %p, i64 %o, i64 %n) {
+; CHECK-LABEL: define ptr @add1(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[O:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: [[G:%.*]] = getelementptr [10 x [10 x [10 x [10 x i32]]]], ptr [[P]], i64 0, i64 [[O]], i64 11, i64 -6, i64 [[N]]
+; CHECK-NEXT: ret ptr [[G]]
+;
+ %a = add i64 %o, 1
+ %g = getelementptr [10 x [10 x [10 x [10 x i32]]]], ptr %p, i64 0, i64 %a, i64 1, i64 -6, i64 %n
+ ret ptr %g
+}
+
+define ptr @sub1(ptr %p, i64 %o, i64 %n) {
+; CHECK-LABEL: define ptr @sub1(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[O:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: [[G:%.*]] = getelementptr [10 x [10 x [10 x [10 x i32]]]], ptr [[P]], i64 0, i64 [[O]], i64 -9, i64 -6, i64 [[N]]
+; CHECK-NEXT: ret ptr [[G]]
+;
+ %a = sub i64 %o, 1
+ %g = getelementptr [10 x [10 x [10 x [10 x i32]]]], ptr %p, i64 0, i64 %a, i64 1, i64 -6, i64 %n
+ ret ptr %g
+}
+
+define ptr @add10(ptr %p, i64 %o, i64 %n) {
+; CHECK-LABEL: define ptr @add10(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[O:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: [[G:%.*]] = getelementptr [10 x [10 x [10 x [10 x i32]]]], ptr [[P]], i64 0, i64 [[O]], i64 101, i64 -6, i64 [[N]]
+; CHECK-NEXT: ret ptr [[G]]
+;
+ %a = add i64 %o, 10
+ %g = getelementptr [10 x [10 x [10 x [10 x i32]]]], ptr %p, i64 0, i64 %a, i64 1, i64 -6, i64 %n
+ ret ptr %g
+}
+
+define ptr @nooff(ptr %p, i64 %o, i64 %n) {
+; CHECK-LABEL: define ptr @nooff(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[O:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: [[A:%.*]] = add i64 [[O]], 10
+; CHECK-NEXT: [[G:%.*]] = getelementptr [10 x [10 x [10 x [10 x i32]]]], ptr [[P]], i64 0, i64 [[A]], i64 [[N]]
+; CHECK-NEXT: ret ptr [[G]]
+;
+ %a = add i64 %o, 10
+ %g = getelementptr [10 x [10 x [10 x [10 x i32]]]], ptr %p, i64 0, i64 %a, i64 %n
+ ret ptr %g
+}
+
+define ptr @inbounds(ptr %p, i64 %o, i64 %n) {
+; CHECK-LABEL: define ptr @inbounds(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[O:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: [[A:%.*]] = add i64 [[O]], 1
+; CHECK-NEXT: [[G:%.*]] = getelementptr inbounds [10 x [10 x [10 x [10 x i32]]]], ptr [[P]], i64 0, i64 [[A]], i64 1, i64 -6, i64 [[N]]
+; CHECK-NEXT: ret ptr [[G]]
+;
+ %a = add i64 %o, 1
+ %g = getelementptr inbounds [10 x [10 x [10 x [10 x i32]]]], ptr %p, i64 0, i64 %a, i64 1, i64 -6, i64 %n
+ ret ptr %g
+}
+
+%struct.Struct = type { i32, i32, ptr }
+define ptr @struct(ptr %p, i64 %o, i64 %n) {
+; CHECK-LABEL: define ptr @struct(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[O:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: [[A:%.*]] = add i64 [[O]], -1
+; CHECK-NEXT: [[G:%.*]] = getelementptr [[STRUCT_STRUCT:%.*]], ptr [[P]], i64 [[A]], i32 1
+; CHECK-NEXT: ret ptr [[G]]
+;
+ %a = add i64 %o, -1
+ %g = getelementptr %struct.Struct, ptr %p, i64 %a, i32 1
+ ret ptr %g
+}
+
+define ptr @lessargs(ptr %p, i64 %o, i64 %n) {
+; CHECK-LABEL: define ptr @lessargs(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[O:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: [[G:%.*]] = getelementptr [10 x [10 x [10 x i32]]], ptr [[P]], i64 0, i64 [[O]], i64 11
+; CHECK-NEXT: ret ptr [[G]]
+;
+ %a = add i64 %o, 1
+ %g = getelementptr [10 x [10 x [10 x i32]]], ptr %p, i64 0, i64 %a, i64 1
+ ret ptr %g
+}
+
+define ptr @twice(ptr %p, i64 %o, i64 %n) {
+; CHECK-LABEL: define ptr @twice(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[O:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: [[G:%.*]] = getelementptr [10 x [10 x [10 x [10 x [10 x i32]]]]], ptr [[P]], i64 0, i64 [[O]], i64 11, i64 [[O]], i64 -13
+; CHECK-NEXT: ret ptr [[G]]
+;
+ %a = add i64 %o, 1
+ %b = sub i64 %o, 1
+ %g = getelementptr [10 x [10 x [10 x [10 x [10 x i32]]]]], ptr %p, i64 0, i64 %a, i64 1, i64 %b, i64 -3
+ ret ptr %g
+}
+
+define ptr @simpler(ptr %p, i64 %o, i64 %n) {
+; CHECK-LABEL: define ptr @simpler(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[O:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: [[G:%.*]] = getelementptr [16 x [8 x i32]], ptr [[P]], i64 0, i64 [[O]], i64 65
+; CHECK-NEXT: ret ptr [[G]]
+;
+ %a = add i64 %o, 8
+ %g = getelementptr [16 x [8 x i32]], ptr %p, i64 0, i64 %a, i64 1
+ ret ptr %g
+}
``````````
</details>
https://github.com/llvm/llvm-project/pull/85090
More information about the llvm-commits
mailing list