
After pr96656.ll were added to LAA and LoopVersioning, it was decided that the bug is in a caller of LoopVersioning, not in LAA or LoopVersioning itself. The new candidate was LoopLoadElim, but #96656 has since been marked invalid. Hence, re-organize the added tests to avoid confusion, and the testcase from the investigation to LoopLoadElim.
86 lines
4.2 KiB
LLVM
86 lines
4.2 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: opt -passes=loop-versioning -S %s | FileCheck %s
|
|
|
|
; Callers should not call LoopVersioning on single-iteration loops, as it
|
|
; is very likely not profitable.
|
|
; LoopVersioning faithfully versions single-iteration loops when the stride
|
|
; is unknown.
|
|
|
|
define double @single_iteration_unknown_stride(i32 %x, ptr %y, i1 %cond) {
|
|
; CHECK-LABEL: define double @single_iteration_unknown_stride(
|
|
; CHECK-SAME: i32 [[X:%.*]], ptr [[Y:%.*]], i1 [[COND:%.*]]) {
|
|
; CHECK-NEXT: [[ENTRY:.*:]]
|
|
; CHECK-NEXT: br i1 [[COND]], label %[[NOLOOP_EXIT:.*]], label %[[LOOP_BODY_LVER_CHECK:.*]]
|
|
; CHECK: [[LOOP_BODY_LVER_CHECK]]:
|
|
; CHECK-NEXT: [[SEXT7:%.*]] = sext i32 [[X]] to i64
|
|
; CHECK-NEXT: [[GEP8:%.*]] = getelementptr i8, ptr [[Y]], i64 8
|
|
; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[X]], 1
|
|
; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[LOOP_BODY_PH_LVER_ORIG:.*]], label %[[LOOP_BODY_PH:.*]]
|
|
; CHECK: [[LOOP_BODY_PH_LVER_ORIG]]:
|
|
; CHECK-NEXT: br label %[[LOOP_BODY_LVER_ORIG:.*]]
|
|
; CHECK: [[LOOP_BODY_LVER_ORIG]]:
|
|
; CHECK-NEXT: [[PHI_LVER_ORIG:%.*]] = phi i64 [ 0, %[[LOOP_BODY_PH_LVER_ORIG]] ], [ [[ADD_LVER_ORIG:%.*]], %[[LOOP_BODY_LVER_ORIG]] ]
|
|
; CHECK-NEXT: [[MUL_LVER_ORIG:%.*]] = mul i64 [[PHI_LVER_ORIG]], [[SEXT7]]
|
|
; CHECK-NEXT: [[GEP10_LVER_ORIG:%.*]] = getelementptr double, ptr [[GEP8]], i64 [[MUL_LVER_ORIG]]
|
|
; CHECK-NEXT: [[LOAD11_LVER_ORIG:%.*]] = load double, ptr [[GEP10_LVER_ORIG]], align 8
|
|
; CHECK-NEXT: store double [[LOAD11_LVER_ORIG]], ptr [[Y]], align 8
|
|
; CHECK-NEXT: [[ADD_LVER_ORIG]] = add i64 [[PHI_LVER_ORIG]], 1
|
|
; CHECK-NEXT: [[ICMP_LVER_ORIG:%.*]] = icmp eq i64 [[PHI_LVER_ORIG]], 0
|
|
; CHECK-NEXT: br i1 [[ICMP_LVER_ORIG]], label %[[LOOP_EXIT_LOOPEXIT:.*]], label %[[LOOP_BODY_LVER_ORIG]]
|
|
; CHECK: [[LOOP_BODY_PH]]:
|
|
; CHECK-NEXT: br label %[[LOOP_BODY:.*]]
|
|
; CHECK: [[LOOP_BODY]]:
|
|
; CHECK-NEXT: [[PHI:%.*]] = phi i64 [ 0, %[[LOOP_BODY_PH]] ], [ [[ADD:%.*]], %[[LOOP_BODY]] ]
|
|
; CHECK-NEXT: [[MUL:%.*]] = mul i64 [[PHI]], [[SEXT7]]
|
|
; CHECK-NEXT: [[GEP10:%.*]] = getelementptr double, ptr [[GEP8]], i64 [[MUL]]
|
|
; CHECK-NEXT: [[LOAD11:%.*]] = load double, ptr [[GEP10]], align 8
|
|
; CHECK-NEXT: store double [[LOAD11]], ptr [[Y]], align 8
|
|
; CHECK-NEXT: [[ADD]] = add i64 [[PHI]], 1
|
|
; CHECK-NEXT: [[ICMP:%.*]] = icmp eq i64 [[PHI]], 0
|
|
; CHECK-NEXT: br i1 [[ICMP]], label %[[LOOP_EXIT_LOOPEXIT1:.*]], label %[[LOOP_BODY]]
|
|
; CHECK: [[NOLOOP_EXIT]]:
|
|
; CHECK-NEXT: [[SEXT:%.*]] = sext i32 [[X]] to i64
|
|
; CHECK-NEXT: [[GEP:%.*]] = getelementptr double, ptr [[Y]], i64 [[SEXT]]
|
|
; CHECK-NEXT: [[LOAD5:%.*]] = load double, ptr [[GEP]], align 8
|
|
; CHECK-NEXT: ret double [[LOAD5]]
|
|
; CHECK: [[LOOP_EXIT_LOOPEXIT]]:
|
|
; CHECK-NEXT: br label %[[LOOP_EXIT:.*]]
|
|
; CHECK: [[LOOP_EXIT_LOOPEXIT1]]:
|
|
; CHECK-NEXT: br label %[[LOOP_EXIT]]
|
|
; CHECK: [[LOOP_EXIT]]:
|
|
; CHECK-NEXT: [[SEXT2:%.*]] = sext i32 [[X]] to i64
|
|
; CHECK-NEXT: [[GEP2:%.*]] = getelementptr double, ptr [[Y]], i64 [[SEXT2]]
|
|
; CHECK-NEXT: [[LOAD6:%.*]] = load double, ptr [[GEP2]], align 8
|
|
; CHECK-NEXT: ret double [[LOAD6]]
|
|
;
|
|
entry:
|
|
br i1 %cond, label %noloop.exit, label %loop.ph
|
|
|
|
loop.ph: ; preds = %entry
|
|
%sext7 = sext i32 %x to i64
|
|
%gep8 = getelementptr i8, ptr %y, i64 8
|
|
br label %loop.body
|
|
|
|
loop.body: ; preds = %loop.body, %loop.ph
|
|
%iv = phi i64 [ 0, %loop.ph ], [ %iv.next, %loop.body ]
|
|
%mul = mul i64 %iv, %sext7
|
|
%gep10 = getelementptr double, ptr %gep8, i64 %mul
|
|
%load11 = load double, ptr %gep10, align 8
|
|
store double %load11, ptr %y, align 8
|
|
%iv.next = add i64 %iv, 1
|
|
%icmp = icmp eq i64 %iv, 0
|
|
br i1 %icmp, label %loop.exit, label %loop.body
|
|
|
|
noloop.exit: ; preds = %entry
|
|
%sext = sext i32 %x to i64
|
|
%gep = getelementptr double, ptr %y, i64 %sext
|
|
%load5 = load double, ptr %gep, align 8
|
|
ret double %load5
|
|
|
|
loop.exit: ; preds = %loop.body
|
|
%sext2 = sext i32 %x to i64
|
|
%gep2 = getelementptr double, ptr %y, i64 %sext2
|
|
%load6 = load double, ptr %gep2, align 8
|
|
ret double %load6
|
|
}
|