Doomhowl-Interactive/llvm-project
0
0
Fork 0
Code Issues Pull Requests Actions 5 Packages Projects Releases Wiki Activity
Files
main
llvm-project/llvm/test/Transforms/IndVarSimplify/pointer-loop-guards.ll
Florian Hahn a5d3522c13 [SCEV] Rewrite A - B = UMin(1, A - B) lazily for A != B loop guards. (#163787) 
Follow-up to 2d027260b0
(https://github.com/llvm/llvm-project/pull/160500)

Creating the SCEV subtraction eagerly is very expensive. To soften the
blow, just collect a map with inequalities and check if we can apply the
subtract rewrite when rewriting SCEVAddExpr.

Restores most of the regression:

http://llvm-compile-time-tracker.com/compare.php?from=0792478e4e133be96650444f3264e89d002fc058&to=7fca35db60fe6f423ea6051b45226046c067c252&stat=instructions:u
stage1-O3: -0.10%
stage1-ReleaseThinLTO: -0.09%
stage1-ReleaseLTO-g: -0.10%
stage1-O0-g: +0.02%
stage1-aarch64-O3: -0.09%
stage1-aarch64-O0-g: +0.00%
stage2-O3: -0.17%
stage2-O0-g: -0.05%
stage2-clang: -0.07%

There is still some negative impact compared to before 2d027260b0, but
there's probably not much we could do reduce this even more.

Compile-time improvement with 2d027260b0 reverted on top of the
current PR:
http://llvm-compile-time-tracker.com/compare.php?from=7fca35db60fe6f423ea6051b45226046c067c252&to=98dd152bdfc76b30d00190d3850d89406ca3c21f&stat=instructions:u

stage1-O3: 60628M (-0.03%)
stage1-ReleaseThinLTO: 76388M (-0.04%)
stage1-ReleaseLTO-g: 89228M (-0.02%)
stage1-O0-g: 18523M (-0.03%)
stage1-aarch64-O3: 67623M (-0.03%)
stage1-aarch64-O0-g: 22595M (+0.01%)
stage2-O3: 52336M (+0.01%)
stage2-O0-g: 16174M (+0.00%)
stage2-clang: 34890032M (-0.03%)

PR: https://github.com/llvm/llvm-project/pull/163787
2025-10-18 13:32:40 +01:00

224 lines
10 KiB
LLVM
Raw Permalink Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -p indvars -S %s | FileCheck %s
; RUN: opt -p indvars -data-layout='n32:64' -S %s | FileCheck --check-prefix=N32 %s
declare i1 @cond()
define i64 @test_ptr_compare_guard(ptr %start, ptr %end) {
; CHECK-LABEL: define i64 @test_ptr_compare_guard(
; CHECK-SAME: ptr [[START:%.*]], ptr [[END:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: [[C_0:%.*]] = icmp eq ptr [[START]], [[END]]
; CHECK-NEXT: br i1 [[C_0]], label %[[EXIT:.*]], label %[[LOOP_HEADER_PREHEADER:.*]]
; CHECK: [[LOOP_HEADER_PREHEADER]]:
; CHECK-NEXT: br label %[[LOOP_HEADER:.*]]
; CHECK: [[LOOP_HEADER]]:
; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ], [ [[START]], %[[LOOP_HEADER_PREHEADER]] ]
; CHECK-NEXT: [[I64_IV:%.*]] = phi i64 [ [[I64_IV_NEXT:%.*]], %[[LOOP_LATCH]] ], [ 0, %[[LOOP_HEADER_PREHEADER]] ]
; CHECK-NEXT: [[C_1:%.*]] = call i1 @cond()
; CHECK-NEXT: br i1 [[C_1]], label %[[LOOP_LATCH]], label %[[EXIT_LOOPEXIT:.*]]
; CHECK: [[LOOP_LATCH]]:
; CHECK-NEXT: [[PTR_IV_NEXT]] = getelementptr i8, ptr [[PTR_IV]], i64 1
; CHECK-NEXT: [[I64_IV_NEXT]] = add nuw i64 [[I64_IV]], 1
; CHECK-NEXT: [[C_2:%.*]] = icmp eq ptr [[PTR_IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[C_2]], label %[[EXIT_LOOPEXIT]], label %[[LOOP_HEADER]]
; CHECK: [[EXIT_LOOPEXIT]]:
; CHECK-NEXT: [[RES_PH:%.*]] = phi i64 [ 0, %[[LOOP_HEADER]] ], [ [[I64_IV]], %[[LOOP_LATCH]] ]
; CHECK-NEXT: br label %[[EXIT]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: [[RES:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[RES_PH]], %[[EXIT_LOOPEXIT]] ]
; CHECK-NEXT: ret i64 [[RES]]
;
; N32-LABEL: define i64 @test_ptr_compare_guard(
; N32-SAME: ptr [[START:%.*]], ptr [[END:%.*]]) {
; N32-NEXT: [[ENTRY:.*]]:
; N32-NEXT: [[START2:%.*]] = ptrtoint ptr [[START]] to i64
; N32-NEXT: [[END1:%.*]] = ptrtoint ptr [[END]] to i64
; N32-NEXT: [[C_0:%.*]] = icmp eq ptr [[START]], [[END]]
; N32-NEXT: br i1 [[C_0]], label %[[EXIT:.*]], label %[[LOOP_HEADER_PREHEADER:.*]]
; N32: [[LOOP_HEADER_PREHEADER]]:
; N32-NEXT: [[TMP0:%.*]] = add i64 [[END1]], -1
; N32-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[START2]]
; N32-NEXT: br label %[[LOOP_HEADER:.*]]
; N32: [[LOOP_HEADER]]:
; N32-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ], [ [[START]], %[[LOOP_HEADER_PREHEADER]] ]
; N32-NEXT: [[C_1:%.*]] = call i1 @cond()
; N32-NEXT: br i1 [[C_1]], label %[[LOOP_LATCH]], label %[[EXIT_LOOPEXIT:.*]]
; N32: [[LOOP_LATCH]]:
; N32-NEXT: [[PTR_IV_NEXT]] = getelementptr i8, ptr [[PTR_IV]], i64 1
; N32-NEXT: [[C_2:%.*]] = icmp eq ptr [[PTR_IV_NEXT]], [[END]]
; N32-NEXT: br i1 [[C_2]], label %[[EXIT_LOOPEXIT]], label %[[LOOP_HEADER]]
; N32: [[EXIT_LOOPEXIT]]:
; N32-NEXT: [[RES_PH:%.*]] = phi i64 [ 0, %[[LOOP_HEADER]] ], [ [[TMP1]], %[[LOOP_LATCH]] ]
; N32-NEXT: br label %[[EXIT]]
; N32: [[EXIT]]:
; N32-NEXT: [[RES:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[RES_PH]], %[[EXIT_LOOPEXIT]] ]
; N32-NEXT: ret i64 [[RES]]
;
entry:
%c.0 = icmp eq ptr %start, %end
br i1 %c.0, label %exit, label %loop.header
loop.header:
%ptr.iv = phi ptr [ %start, %entry ], [ %ptr.iv.next, %loop.latch ]
%i64.iv = phi i64 [ 0, %entry ], [ %i64.iv.next, %loop.latch ]
%c.1 = call i1 @cond()
br i1 %c.1, label %loop.latch, label %exit
loop.latch:
%ptr.iv.next = getelementptr i8, ptr %ptr.iv, i64 1
%i64.iv.next = add i64 %i64.iv, 1
%c.2 = icmp eq ptr %ptr.iv.next, %end
br i1 %c.2, label %exit, label %loop.header
exit:
%res = phi i64 [ 0, %entry ], [ %i64.iv, %loop.latch ], [ 0, %loop.header ]
ret i64 %res
}
define void @test_sub_cmp(ptr align 8 %start, ptr %end) {
; CHECK-LABEL: define void @test_sub_cmp(
; CHECK-SAME: ptr align 8 [[START:%.*]], ptr [[END:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[START_INT:%.*]] = ptrtoint ptr [[START]] to i64
; CHECK-NEXT: [[END_INT:%.*]] = ptrtoint ptr [[END]] to i64
; CHECK-NEXT: [[PTR_DIFF:%.*]] = sub i64 [[START_INT]], [[END_INT]]
; CHECK-NEXT: [[CMP_ENTRY:%.*]] = icmp eq ptr [[START]], [[END]]
; CHECK-NEXT: br i1 [[CMP_ENTRY]], label %[[EXIT:.*]], label %[[LOOP_HEADER_PREHEADER:.*]]
; CHECK: [[LOOP_HEADER_PREHEADER]]:
; CHECK-NEXT: br label %[[LOOP_HEADER:.*]]
; CHECK: [[LOOP_HEADER]]:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ], [ 0, %[[LOOP_HEADER_PREHEADER]] ]
; CHECK-NEXT: [[C_1:%.*]] = call i1 @cond()
; CHECK-NEXT: br i1 [[C_1]], label %[[EXIT_EARLY:.*]], label %[[LOOP_LATCH]]
; CHECK: [[LOOP_LATCH]]:
; CHECK-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1
; CHECK-NEXT: [[CMP_LATCH:%.*]] = icmp ult i64 [[IV_NEXT]], [[PTR_DIFF]]
; CHECK-NEXT: br i1 [[CMP_LATCH]], label %[[LOOP_HEADER]], label %[[EXIT_LOOPEXIT:.*]]
; CHECK: [[EXIT_EARLY]]:
; CHECK-NEXT: br label %[[EXIT]]
; CHECK: [[EXIT_LOOPEXIT]]:
; CHECK-NEXT: br label %[[EXIT]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
; N32-LABEL: define void @test_sub_cmp(
; N32-SAME: ptr align 8 [[START:%.*]], ptr [[END:%.*]]) {
; N32-NEXT: [[ENTRY:.*:]]
; N32-NEXT: [[START_INT:%.*]] = ptrtoint ptr [[START]] to i64
; N32-NEXT: [[END_INT:%.*]] = ptrtoint ptr [[END]] to i64
; N32-NEXT: [[PTR_DIFF:%.*]] = sub i64 [[START_INT]], [[END_INT]]
; N32-NEXT: [[CMP_ENTRY:%.*]] = icmp eq ptr [[START]], [[END]]
; N32-NEXT: br i1 [[CMP_ENTRY]], label %[[EXIT:.*]], label %[[LOOP_HEADER_PREHEADER:.*]]
; N32: [[LOOP_HEADER_PREHEADER]]:
; N32-NEXT: br label %[[LOOP_HEADER:.*]]
; N32: [[LOOP_HEADER]]:
; N32-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ], [ 0, %[[LOOP_HEADER_PREHEADER]] ]
; N32-NEXT: [[C_1:%.*]] = call i1 @cond()
; N32-NEXT: br i1 [[C_1]], label %[[EXIT_EARLY:.*]], label %[[LOOP_LATCH]]
; N32: [[LOOP_LATCH]]:
; N32-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], 1
; N32-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[IV_NEXT]], [[PTR_DIFF]]
; N32-NEXT: br i1 [[EXITCOND]], label %[[LOOP_HEADER]], label %[[EXIT_LOOPEXIT:.*]]
; N32: [[EXIT_EARLY]]:
; N32-NEXT: br label %[[EXIT]]
; N32: [[EXIT_LOOPEXIT]]:
; N32-NEXT: br label %[[EXIT]]
; N32: [[EXIT]]:
; N32-NEXT: ret void
;
entry:
%start.int = ptrtoint ptr %start to i64
%end.int = ptrtoint ptr %end to i64
%ptr.diff = sub i64 %start.int, %end.int
%cmp.entry = icmp eq ptr %start, %end
br i1 %cmp.entry, label %exit, label %loop.header
loop.header:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
%c.1 = call i1 @cond()
br i1 %c.1, label %exit.early, label %loop.latch
loop.latch:
%iv.next = add i64 %iv, 1
%cmp.latch = icmp ult i64 %iv.next, %ptr.diff
br i1 %cmp.latch, label %loop.header, label %exit
exit.early:
br label %exit
exit:
ret void
}
define void @test_ptr_diff_with_assume(ptr align 8 %start, ptr align 8 %end, ptr %P) {
; CHECK-LABEL: define void @test_ptr_diff_with_assume(
; CHECK-SAME: ptr align 8 [[START:%.*]], ptr align 8 [[END:%.*]], ptr [[P:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[START_INT:%.*]] = ptrtoint ptr [[START]] to i64
; CHECK-NEXT: [[END_INT:%.*]] = ptrtoint ptr [[END]] to i64
; CHECK-NEXT: [[PTR_DIFF:%.*]] = sub i64 [[START_INT]], [[END_INT]]
; CHECK-NEXT: [[DIFF_CMP:%.*]] = icmp ult i64 [[PTR_DIFF]], 2
; CHECK-NEXT: call void @llvm.assume(i1 [[DIFF_CMP]])
; CHECK-NEXT: [[COMPUTED_END:%.*]] = getelementptr i8, ptr [[START]], i64 [[PTR_DIFF]]
; CHECK-NEXT: [[ENTRY_CMP:%.*]] = icmp eq ptr [[START]], [[END]]
; CHECK-NEXT: br i1 [[ENTRY_CMP]], label %[[EXIT:.*]], label %[[LOOP_BODY_PREHEADER:.*]]
; CHECK: [[LOOP_BODY_PREHEADER]]:
; CHECK-NEXT: br label %[[LOOP_BODY:.*]]
; CHECK: [[LOOP_BODY]]:
; CHECK-NEXT: [[IV:%.*]] = phi ptr [ [[IV_NEXT:%.*]], %[[LOOP_BODY]] ], [ [[START]], %[[LOOP_BODY_PREHEADER]] ]
; CHECK-NEXT: [[TMP0:%.*]] = call i1 @cond()
; CHECK-NEXT: [[IV_NEXT]] = getelementptr i8, ptr [[IV]], i64 1
; CHECK-NEXT: [[LOOP_CMP:%.*]] = icmp eq ptr [[IV_NEXT]], [[COMPUTED_END]]
; CHECK-NEXT: br i1 [[LOOP_CMP]], label %[[EXIT_LOOPEXIT:.*]], label %[[LOOP_BODY]]
; CHECK: [[EXIT_LOOPEXIT]]:
; CHECK-NEXT: br label %[[EXIT]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
; N32-LABEL: define void @test_ptr_diff_with_assume(
; N32-SAME: ptr align 8 [[START:%.*]], ptr align 8 [[END:%.*]], ptr [[P:%.*]]) {
; N32-NEXT: [[ENTRY:.*:]]
; N32-NEXT: [[START_INT:%.*]] = ptrtoint ptr [[START]] to i64
; N32-NEXT: [[END_INT:%.*]] = ptrtoint ptr [[END]] to i64
; N32-NEXT: [[PTR_DIFF:%.*]] = sub i64 [[START_INT]], [[END_INT]]
; N32-NEXT: [[DIFF_CMP:%.*]] = icmp ult i64 [[PTR_DIFF]], 2
; N32-NEXT: call void @llvm.assume(i1 [[DIFF_CMP]])
; N32-NEXT: [[COMPUTED_END:%.*]] = getelementptr i8, ptr [[START]], i64 [[PTR_DIFF]]
; N32-NEXT: [[ENTRY_CMP:%.*]] = icmp eq ptr [[START]], [[END]]
; N32-NEXT: br i1 [[ENTRY_CMP]], label %[[EXIT:.*]], label %[[LOOP_BODY_PREHEADER:.*]]
; N32: [[LOOP_BODY_PREHEADER]]:
; N32-NEXT: br label %[[LOOP_BODY:.*]]
; N32: [[LOOP_BODY]]:
; N32-NEXT: [[IV:%.*]] = phi ptr [ [[IV_NEXT:%.*]], %[[LOOP_BODY]] ], [ [[START]], %[[LOOP_BODY_PREHEADER]] ]
; N32-NEXT: [[TMP0:%.*]] = call i1 @cond()
; N32-NEXT: [[IV_NEXT]] = getelementptr i8, ptr [[IV]], i64 1
; N32-NEXT: [[LOOP_CMP:%.*]] = icmp eq ptr [[IV_NEXT]], [[COMPUTED_END]]
; N32-NEXT: br i1 [[LOOP_CMP]], label %[[EXIT_LOOPEXIT:.*]], label %[[LOOP_BODY]]
; N32: [[EXIT_LOOPEXIT]]:
; N32-NEXT: br label %[[EXIT]]
; N32: [[EXIT]]:
; N32-NEXT: ret void
;
entry:
%start.int = ptrtoint ptr %start to i64
%end.int = ptrtoint ptr %end to i64
%ptr.diff = sub i64 %start.int, %end.int
%diff.cmp = icmp ult i64 %ptr.diff, 2
call void @llvm.assume(i1 %diff.cmp)
%computed.end = getelementptr i8, ptr %start, i64 %ptr.diff
%entry.cmp = icmp eq ptr %start, %end
br i1 %entry.cmp, label %exit, label %loop.body
loop.body:
%iv = phi ptr [ %start, %entry ], [ %iv.next, %loop.body ]
call i1 @cond()
%iv.next = getelementptr i8, ptr %iv, i64 1
%loop.cmp = icmp eq ptr %iv.next, %computed.end
br i1 %loop.cmp, label %exit, label %loop.body
exit:
ret void
}
Reference in New Issue View Git Blame Copy Permalink