[llvm] [LoopDist] Add some runtime checks for cross partition loads (PR #145623)
Michael Berg via llvm-commits
llvm-commits at lists.llvm.org
Fri Jul 18 13:58:25 PDT 2025
================
@@ -0,0 +1,207 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -aa-pipeline=basic-aa -passes=loop-simplify,loop-distribute -enable-loop-distribute -verify-loop-info -verify-dom-info -S %s | FileCheck %s
+
+; Test emit safety guards for ptr access of %a and %c on a cross parition load which
+; has a corresponding store to the same address. This ensures that if %a and %c
+; overlap in some scenarios, that we execute the original loop for safety reasons.
+
+define dso_local void @_Z13distribution3PiS_S_S_i(ptr nocapture noundef %a, ptr nocapture noundef readonly %b, ptr nocapture noundef %c, ptr nocapture noundef writeonly %d, i32 noundef signext %len) {
+; CHECK-LABEL: define dso_local void @_Z13distribution3PiS_S_S_i(
+; CHECK-SAME: ptr noundef captures(none) [[A:%.*]], ptr noundef readonly captures(none) [[B:%.*]], ptr noundef captures(none) [[C:%.*]], ptr noundef writeonly captures(none) [[D:%.*]], i32 noundef signext [[LEN:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[LEN]], 0
+; CHECK-NEXT: br i1 [[CMP]], label %[[IF_THEN:.*]], label %[[IF_END:.*]]
+; CHECK: [[IF_THEN]]:
+; CHECK-NEXT: [[SUB_NEG:%.*]] = add nsw i32 [[LEN]], -2147483647
+; CHECK-NEXT: [[CMP1_NOT51:%.*]] = icmp eq i32 [[LEN]], 1
+; CHECK-NEXT: br i1 [[CMP1_NOT51]], label %[[IF_END]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: [[ADD:%.*]] = sub nuw i32 -2147483648, [[LEN]]
+; CHECK-NEXT: [[I:%.*]] = zext nneg i32 [[ADD]] to i64
+; CHECK-NEXT: br label %[[FOR_BODY_LVER_CHECK:.*]]
+; CHECK: [[FOR_BODY_LVER_CHECK]]:
+; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A]], i64 4
+; CHECK-NEXT: [[TMP0:%.*]] = sub i32 -2147483647, [[LEN]]
+; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
+; CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 2
+; CHECK-NEXT: [[TMP3:%.*]] = sub i64 8589934596, [[TMP2]]
+; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP3]]
+; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[C]], i64 [[TMP3]]
+; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[D]], i64 4
+; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr i8, ptr [[D]], i64 [[TMP3]]
+; CHECK-NEXT: [[SCEVGEP5:%.*]] = getelementptr i8, ptr [[B]], i64 4
+; CHECK-NEXT: [[SCEVGEP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP3]]
+; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP2]]
+; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[C]], [[SCEVGEP1]]
+; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
+; CHECK-NEXT: [[BOUND07:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP4]]
+; CHECK-NEXT: [[BOUND18:%.*]] = icmp ult ptr [[SCEVGEP3]], [[SCEVGEP1]]
+; CHECK-NEXT: [[FOUND_CONFLICT9:%.*]] = and i1 [[BOUND07]], [[BOUND18]]
+; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT9]]
+; CHECK-NEXT: [[BOUND010:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP6]]
+; CHECK-NEXT: [[BOUND111:%.*]] = icmp ult ptr [[SCEVGEP5]], [[SCEVGEP1]]
+; CHECK-NEXT: [[FOUND_CONFLICT12:%.*]] = and i1 [[BOUND010]], [[BOUND111]]
+; CHECK-NEXT: [[CONFLICT_RDX13:%.*]] = or i1 [[CONFLICT_RDX]], [[FOUND_CONFLICT12]]
+; CHECK-NEXT: [[BOUND014:%.*]] = icmp ult ptr [[C]], [[SCEVGEP4]]
+; CHECK-NEXT: [[BOUND115:%.*]] = icmp ult ptr [[SCEVGEP3]], [[SCEVGEP2]]
+; CHECK-NEXT: [[FOUND_CONFLICT16:%.*]] = and i1 [[BOUND014]], [[BOUND115]]
+; CHECK-NEXT: [[CONFLICT_RDX17:%.*]] = or i1 [[CONFLICT_RDX13]], [[FOUND_CONFLICT16]]
+; CHECK-NEXT: [[BOUND018:%.*]] = icmp ult ptr [[SCEVGEP3]], [[SCEVGEP6]]
+; CHECK-NEXT: [[BOUND119:%.*]] = icmp ult ptr [[SCEVGEP5]], [[SCEVGEP4]]
+; CHECK-NEXT: [[FOUND_CONFLICT20:%.*]] = and i1 [[BOUND018]], [[BOUND119]]
+; CHECK-NEXT: [[CONFLICT_RDX21:%.*]] = or i1 [[CONFLICT_RDX17]], [[FOUND_CONFLICT20]]
+; CHECK-NEXT: [[TMP4:%.*]] = sub i32 -2147483647, [[LEN]]
+; CHECK-NEXT: [[TMP5:%.*]] = zext i32 [[TMP4]] to i64
+; CHECK-NEXT: [[TMP6:%.*]] = sub i64 2147483647, [[TMP5]]
+; CHECK-NEXT: [[TMP7:%.*]] = trunc i64 [[TMP6]] to i32
+; CHECK-NEXT: [[TMP8:%.*]] = add i32 [[TMP4]], [[TMP7]]
+; CHECK-NEXT: [[TMP9:%.*]] = icmp ult i32 [[TMP8]], [[TMP4]]
+; CHECK-NEXT: [[TMP10:%.*]] = icmp ugt i64 [[TMP6]], 4294967295
+; CHECK-NEXT: [[TMP11:%.*]] = or i1 [[TMP9]], [[TMP10]]
+; CHECK-NEXT: [[TMP12:%.*]] = trunc i64 [[TMP6]] to i32
+; CHECK-NEXT: [[TMP13:%.*]] = add i32 1, [[TMP12]]
+; CHECK-NEXT: [[TMP14:%.*]] = icmp slt i32 [[TMP13]], 1
+; CHECK-NEXT: [[TMP15:%.*]] = icmp ugt i64 [[TMP6]], 4294967295
+; CHECK-NEXT: [[TMP16:%.*]] = or i1 [[TMP14]], [[TMP15]]
+; CHECK-NEXT: [[TMP17:%.*]] = or i1 [[TMP11]], [[TMP16]]
+; CHECK-NEXT: [[LVER_SAFE:%.*]] = or i1 [[CONFLICT_RDX21]], [[TMP17]]
+; CHECK-NEXT: br i1 [[LVER_SAFE]], label %[[FOR_BODY_PH_LVER_ORIG:.*]], label %[[FOR_BODY_PH_LDIST1:.*]]
+; CHECK: [[FOR_BODY_PH_LVER_ORIG]]:
+; CHECK-NEXT: br label %[[FOR_BODY_LVER_ORIG:.*]]
+; CHECK: [[FOR_BODY_LVER_ORIG]]:
+; CHECK-NEXT: [[INDVARS_IV_LVER_ORIG:%.*]] = phi i64 [ [[I]], %[[FOR_BODY_PH_LVER_ORIG]] ], [ [[INDVARS_IV_NEXT_LVER_ORIG:%.*]], %[[FOR_BODY_LVER_ORIG]] ]
+; CHECK-NEXT: [[I1_LVER_ORIG:%.*]] = trunc i64 [[INDVARS_IV_LVER_ORIG]] to i32
+; CHECK-NEXT: [[SUB3_LVER_ORIG:%.*]] = add i32 [[SUB_NEG]], [[I1_LVER_ORIG]]
+; CHECK-NEXT: [[IDXPROM_LVER_ORIG:%.*]] = sext i32 [[SUB3_LVER_ORIG]] to i64
+; CHECK-NEXT: [[ARRAYIDX_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IDXPROM_LVER_ORIG]]
+; CHECK-NEXT: [[I2_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDX_LVER_ORIG]], align 4, !tbaa [[TBAA0:![0-9]+]]
+; CHECK-NEXT: [[ADD4_LVER_ORIG:%.*]] = add nsw i32 [[I2_LVER_ORIG]], 1
+; CHECK-NEXT: [[ARRAYIDX8_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IDXPROM_LVER_ORIG]]
+; CHECK-NEXT: store i32 [[ADD4_LVER_ORIG]], ptr [[ARRAYIDX8_LVER_ORIG]], align 4, !tbaa [[TBAA0]]
+; CHECK-NEXT: [[I3_LVER_ORIG:%.*]] = getelementptr i32, ptr [[C]], i64 [[IDXPROM_LVER_ORIG]]
+; CHECK-NEXT: [[ARRAYIDX17_LVER_ORIG:%.*]] = getelementptr i8, ptr [[I3_LVER_ORIG]], i64 -4
+; CHECK-NEXT: [[I4_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDX17_LVER_ORIG]], align 4, !tbaa [[TBAA0]]
+; CHECK-NEXT: [[SUB18_LVER_ORIG:%.*]] = sub nsw i32 [[ADD4_LVER_ORIG]], [[I4_LVER_ORIG]]
+; CHECK-NEXT: store i32 [[SUB18_LVER_ORIG]], ptr [[I3_LVER_ORIG]], align 4, !tbaa [[TBAA0]]
+; CHECK-NEXT: [[I5_LVER_ORIG:%.*]] = load i32, ptr [[ARRAYIDX8_LVER_ORIG]], align 4, !tbaa [[TBAA0]]
+; CHECK-NEXT: [[ADD27_LVER_ORIG:%.*]] = add nsw i32 [[I5_LVER_ORIG]], 2
+; CHECK-NEXT: [[ARRAYIDX31_LVER_ORIG:%.*]] = getelementptr inbounds i32, ptr [[D]], i64 [[IDXPROM_LVER_ORIG]]
+; CHECK-NEXT: store i32 [[ADD27_LVER_ORIG]], ptr [[ARRAYIDX31_LVER_ORIG]], align 4, !tbaa [[TBAA0]]
+; CHECK-NEXT: [[INDVARS_IV_NEXT_LVER_ORIG]] = add i64 [[INDVARS_IV_LVER_ORIG]], 1
+; CHECK-NEXT: [[I6_LVER_ORIG:%.*]] = and i64 [[INDVARS_IV_NEXT_LVER_ORIG]], 4294967295
+; CHECK-NEXT: [[CMP1_NOT_LVER_ORIG:%.*]] = icmp eq i64 [[I6_LVER_ORIG]], 2147483647
+; CHECK-NEXT: br i1 [[CMP1_NOT_LVER_ORIG]], label %[[IF_END_LOOPEXIT_LOOPEXIT:.*]], label %[[FOR_BODY_LVER_ORIG]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK: [[FOR_BODY_PH_LDIST1]]:
+; CHECK-NEXT: br label %[[FOR_BODY_LDIST1:.*]]
+; CHECK: [[FOR_BODY_LDIST1]]:
+; CHECK-NEXT: [[INDVARS_IV_LDIST1:%.*]] = phi i64 [ [[I]], %[[FOR_BODY_PH_LDIST1]] ], [ [[INDVARS_IV_NEXT_LDIST1:%.*]], %[[FOR_BODY_LDIST1]] ]
+; CHECK-NEXT: [[I1_LDIST1:%.*]] = trunc i64 [[INDVARS_IV_LDIST1]] to i32
+; CHECK-NEXT: [[SUB3_LDIST1:%.*]] = add i32 [[SUB_NEG]], [[I1_LDIST1]]
+; CHECK-NEXT: [[IDXPROM_LDIST1:%.*]] = sext i32 [[SUB3_LDIST1]] to i64
+; CHECK-NEXT: [[ARRAYIDX_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IDXPROM_LDIST1]]
+; CHECK-NEXT: [[I2_LDIST1:%.*]] = load i32, ptr [[ARRAYIDX_LDIST1]], align 4, !tbaa [[TBAA0]], !alias.scope [[META6:![0-9]+]]
+; CHECK-NEXT: [[ADD4_LDIST1:%.*]] = add nsw i32 [[I2_LDIST1]], 1
+; CHECK-NEXT: [[ARRAYIDX8_LDIST1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IDXPROM_LDIST1]]
+; CHECK-NEXT: store i32 [[ADD4_LDIST1]], ptr [[ARRAYIDX8_LDIST1]], align 4, !tbaa [[TBAA0]], !alias.scope [[META9:![0-9]+]], !noalias [[META11:![0-9]+]]
+; CHECK-NEXT: [[I3_LDIST1:%.*]] = getelementptr i32, ptr [[C]], i64 [[IDXPROM_LDIST1]]
+; CHECK-NEXT: [[ARRAYIDX17_LDIST1:%.*]] = getelementptr i8, ptr [[I3_LDIST1]], i64 -4
+; CHECK-NEXT: [[I4_LDIST1:%.*]] = load i32, ptr [[ARRAYIDX17_LDIST1]], align 4, !tbaa [[TBAA0]], !alias.scope [[META14:![0-9]+]], !noalias [[META15:![0-9]+]]
+; CHECK-NEXT: [[SUB18_LDIST1:%.*]] = sub nsw i32 [[ADD4_LDIST1]], [[I4_LDIST1]]
+; CHECK-NEXT: store i32 [[SUB18_LDIST1]], ptr [[I3_LDIST1]], align 4, !tbaa [[TBAA0]], !alias.scope [[META14]], !noalias [[META15]]
+; CHECK-NEXT: [[INDVARS_IV_NEXT_LDIST1]] = add i64 [[INDVARS_IV_LDIST1]], 1
+; CHECK-NEXT: [[I6_LDIST1:%.*]] = and i64 [[INDVARS_IV_NEXT_LDIST1]], 4294967295
+; CHECK-NEXT: [[CMP1_NOT_LDIST1:%.*]] = icmp eq i64 [[I6_LDIST1]], 2147483647
+; CHECK-NEXT: br i1 [[CMP1_NOT_LDIST1]], label %[[FOR_BODY_PH:.*]], label %[[FOR_BODY_LDIST1]], !llvm.loop [[LOOP16:![0-9]+]]
+; CHECK: [[FOR_BODY_PH]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[I]], %[[FOR_BODY_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY]] ]
+; CHECK-NEXT: [[I1:%.*]] = trunc i64 [[INDVARS_IV]] to i32
+; CHECK-NEXT: [[SUB3:%.*]] = add i32 [[SUB_NEG]], [[I1]]
+; CHECK-NEXT: [[IDXPROM:%.*]] = sext i32 [[SUB3]] to i64
+; CHECK-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IDXPROM]]
+; CHECK-NEXT: [[I5:%.*]] = load i32, ptr [[ARRAYIDX8]], align 4, !tbaa [[TBAA0]], !alias.scope [[META9]], !noalias [[META11]]
+; CHECK-NEXT: [[ADD27:%.*]] = add nsw i32 [[I5]], 2
+; CHECK-NEXT: [[ARRAYIDX31:%.*]] = getelementptr inbounds i32, ptr [[D]], i64 [[IDXPROM]]
+; CHECK-NEXT: store i32 [[ADD27]], ptr [[ARRAYIDX31]], align 4, !tbaa [[TBAA0]], !alias.scope [[META15]], !noalias [[META6]]
+; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
+; CHECK-NEXT: [[I6:%.*]] = and i64 [[INDVARS_IV_NEXT]], 4294967295
+; CHECK-NEXT: [[CMP1_NOT:%.*]] = icmp eq i64 [[I6]], 2147483647
+; CHECK-NEXT: br i1 [[CMP1_NOT]], label %[[IF_END_LOOPEXIT_LOOPEXIT22:.*]], label %[[FOR_BODY]], !llvm.loop [[LOOP16]]
+; CHECK: [[IF_END_LOOPEXIT_LOOPEXIT]]:
+; CHECK-NEXT: br label %[[IF_END_LOOPEXIT:.*]]
+; CHECK: [[IF_END_LOOPEXIT_LOOPEXIT22]]:
+; CHECK-NEXT: br label %[[IF_END_LOOPEXIT]]
+; CHECK: [[IF_END_LOOPEXIT]]:
+; CHECK-NEXT: br label %[[IF_END]]
+; CHECK: [[IF_END]]:
+; CHECK-NEXT: ret void
+;
+entry:
+ %cmp = icmp sgt i32 %len, 0
+ br i1 %cmp, label %if.then, label %if.end
+
+if.then: ; preds = %entry
+ %sub.neg = add nsw i32 %len, -2147483647
+ %cmp1.not51 = icmp eq i32 %len, 1
+ br i1 %cmp1.not51, label %if.end, label %for.body.preheader
+
+for.body.preheader: ; preds = %if.then
+ %add = sub nuw i32 -2147483648, %len
+ %i = zext nneg i32 %add to i64
+ br label %for.body
----------------
mcberg2021 wrote:
I'll look into simplifying it.
https://github.com/llvm/llvm-project/pull/145623
More information about the llvm-commits
mailing list