[llvm] 8693ef1 - [SCEV] Add tests that benefit from rewriting SCEVAddExpr with guards.
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Sat Sep 20 11:24:55 PDT 2025
Author: Florian Hahn
Date: 2025-09-20T19:24:19+01:00
New Revision: 8693ef16f6f828275c6f9b65b855504019bea27a
URL: https://github.com/llvm/llvm-project/commit/8693ef16f6f828275c6f9b65b855504019bea27a
DIFF: https://github.com/llvm/llvm-project/commit/8693ef16f6f828275c6f9b65b855504019bea27a.diff
LOG: [SCEV] Add tests that benefit from rewriting SCEVAddExpr with guards.
Add additional tests benefiting from rewriting existing SCEVAddExprs with
guards.
Added:
Modified:
llvm/test/Analysis/ScalarEvolution/backedge-taken-count-guard-info-apply-to-adds.ll
llvm/test/Transforms/IndVarSimplify/canonicalize-cmp.ll
llvm/test/Transforms/LoopUnroll/peel-last-iteration-with-guards.ll
llvm/test/Transforms/LoopVectorize/runtime-checks-difference.ll
Removed:
################################################################################
diff --git a/llvm/test/Analysis/ScalarEvolution/backedge-taken-count-guard-info-apply-to-adds.ll b/llvm/test/Analysis/ScalarEvolution/backedge-taken-count-guard-info-apply-to-adds.ll
index 75014f3a58eb6..8df4b52757753 100644
--- a/llvm/test/Analysis/ScalarEvolution/backedge-taken-count-guard-info-apply-to-adds.ll
+++ b/llvm/test/Analysis/ScalarEvolution/backedge-taken-count-guard-info-apply-to-adds.ll
@@ -27,3 +27,32 @@ loop:
exit:
ret void
}
+
+declare void @clobber()
+
+define void @test_add_sub_1_guard(ptr %src, i32 %n) {
+; CHECK-LABEL: 'test_add_sub_1_guard'
+; CHECK-NEXT: Determining loop execution counts for: @test_add_sub_1_guard
+; CHECK-NEXT: Loop %loop: backedge-taken count is (zext i32 (-1 + (%n /u 2))<nsw> to i64)
+; CHECK-NEXT: Loop %loop: constant max backedge-taken count is i64 4294967295
+; CHECK-NEXT: Loop %loop: symbolic max backedge-taken count is (zext i32 (-1 + (%n /u 2))<nsw> to i64)
+; CHECK-NEXT: Loop %loop: Trip multiple is 1
+;
+entry:
+ %shr = lshr i32 %n, 1
+ %sub.1 = add i32 %shr, -1
+ %sub.ext = zext i32 %sub.1 to i64
+ %pre = icmp eq i32 %shr, 1
+ %end = getelementptr i8, ptr %src, i64 %sub.ext
+ br i1 %pre, label %loop, label %exit
+
+loop:
+ %iv = phi ptr [ %src, %entry ], [ %iv.next, %loop ]
+ call void @clobber()
+ %iv.next = getelementptr i8, ptr %iv, i64 1
+ %ec = icmp eq ptr %iv, %end
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
diff --git a/llvm/test/Transforms/IndVarSimplify/canonicalize-cmp.ll b/llvm/test/Transforms/IndVarSimplify/canonicalize-cmp.ll
index 99baa6105655d..4b52479fc6c4d 100644
--- a/llvm/test/Transforms/IndVarSimplify/canonicalize-cmp.ll
+++ b/llvm/test/Transforms/IndVarSimplify/canonicalize-cmp.ll
@@ -334,4 +334,88 @@ out_of_bounds:
ret i32 -1
}
+define void @slt_no_smax_needed(i64 %n, ptr %dst) {
+; CHECK-LABEL: @slt_no_smax_needed(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[N_TRUNC:%.*]] = trunc i64 [[N:%.*]] to i32
+; CHECK-NEXT: [[ADD_1:%.*]] = add i32 [[N_TRUNC]], 1
+; CHECK-NEXT: [[SHR:%.*]] = lshr i32 [[ADD_1]], 1
+; CHECK-NEXT: [[PRE:%.*]] = icmp ult i32 [[ADD_1]], 8
+; CHECK-NEXT: br i1 [[PRE]], label [[EXIT:%.*]], label [[LOOP_PREHEADER:%.*]]
+; CHECK: loop.preheader:
+; CHECK-NEXT: [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[SHR]], i32 1)
+; CHECK-NEXT: br label [[LOOP:%.*]]
+; CHECK: loop:
+; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i8, ptr [[DST:%.*]], i32 [[IV]]
+; CHECK-NEXT: store i8 0, ptr [[GEP]], align 1
+; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i32 [[IV_NEXT]], [[SMAX]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]]
+; CHECK: exit.loopexit:
+; CHECK-NEXT: br label [[EXIT]]
+; CHECK: exit:
+; CHECK-NEXT: ret void
+;
+entry:
+ %n.trunc = trunc i64 %n to i32
+ %add.1 = add i32 %n.trunc, 1
+ %shr = lshr i32 %add.1, 1
+ %pre = icmp ult i32 %add.1, 8
+ br i1 %pre, label %exit, label %loop
+
+loop:
+ %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
+ %gep = getelementptr inbounds i8, ptr %dst, i32 %iv
+ store i8 0, ptr %gep, align 1
+ %iv.next = add i32 %iv, 1
+ %ec = icmp slt i32 %iv.next, %shr
+ br i1 %ec, label %loop, label %exit
+
+exit:
+ ret void
+}
+
+define void @ult_no_umax_needed(i64 %n, ptr %dst) {
+; CHECK-LABEL: @ult_no_umax_needed(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[N_TRUNC:%.*]] = trunc i64 [[N:%.*]] to i32
+; CHECK-NEXT: [[ADD_1:%.*]] = add i32 [[N_TRUNC]], 1
+; CHECK-NEXT: [[SHR:%.*]] = lshr i32 [[ADD_1]], 1
+; CHECK-NEXT: [[PRE:%.*]] = icmp ult i32 [[ADD_1]], 8
+; CHECK-NEXT: br i1 [[PRE]], label [[EXIT:%.*]], label [[LOOP_PREHEADER:%.*]]
+; CHECK: loop.preheader:
+; CHECK-NEXT: [[UMAX:%.*]] = call i32 @llvm.umax.i32(i32 [[SHR]], i32 1)
+; CHECK-NEXT: br label [[LOOP:%.*]]
+; CHECK: loop:
+; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[LOOP_PREHEADER]] ]
+; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i8, ptr [[DST:%.*]], i32 [[IV]]
+; CHECK-NEXT: store i8 0, ptr [[GEP]], align 1
+; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i32 [[IV_NEXT]], [[UMAX]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]]
+; CHECK: exit.loopexit:
+; CHECK-NEXT: br label [[EXIT]]
+; CHECK: exit:
+; CHECK-NEXT: ret void
+;
+entry:
+ %n.trunc = trunc i64 %n to i32
+ %add.1 = add i32 %n.trunc, 1
+ %shr = lshr i32 %add.1, 1
+ %pre = icmp ult i32 %add.1, 8
+ br i1 %pre, label %exit, label %loop
+
+loop:
+ %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
+ %gep = getelementptr inbounds i8, ptr %dst, i32 %iv
+ store i8 0, ptr %gep, align 1
+ %iv.next = add i32 %iv, 1
+ %ec = icmp ult i32 %iv.next, %shr
+ br i1 %ec, label %loop, label %exit
+
+exit:
+ ret void
+}
+
!0 = !{i32 1, i32 2147483648}
diff --git a/llvm/test/Transforms/LoopUnroll/peel-last-iteration-with-guards.ll b/llvm/test/Transforms/LoopUnroll/peel-last-iteration-with-guards.ll
index 824e23fcf3e6e..c1d8395965577 100644
--- a/llvm/test/Transforms/LoopUnroll/peel-last-iteration-with-guards.ll
+++ b/llvm/test/Transforms/LoopUnroll/peel-last-iteration-with-guards.ll
@@ -201,6 +201,78 @@ loop.latch:
%ec = icmp eq i32 %iv.next, %n
br i1 %ec, label %exit, label %loop.header
+exit:
+ ret void
+}
+
+define void @test_peel_guard_sub_1_btc(i32 %n) {
+; CHECK-LABEL: define void @test_peel_guard_sub_1_btc(
+; CHECK-SAME: i32 [[N:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[SUB:%.*]] = add i32 [[N]], -1
+; CHECK-NEXT: [[PRE:%.*]] = icmp eq i32 [[SUB]], 0
+; CHECK-NEXT: br i1 [[PRE]], label %[[EXIT:.*]], label %[[LOOP_HEADER_PREHEADER:.*]]
+; CHECK: [[LOOP_HEADER_PREHEADER]]:
+; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[N]], -2
+; CHECK-NEXT: [[TMP1:%.*]] = icmp ne i32 [[TMP0]], 0
+; CHECK-NEXT: br i1 [[TMP1]], label %[[LOOP_HEADER_PREHEADER_SPLIT:.*]], label %[[EXIT_LOOPEXIT_PEEL_BEGIN:.*]]
+; CHECK: [[LOOP_HEADER_PREHEADER_SPLIT]]:
+; CHECK-NEXT: br label %[[LOOP_HEADER:.*]]
+; CHECK: [[LOOP_HEADER]]:
+; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ], [ 1, %[[LOOP_HEADER_PREHEADER_SPLIT]] ]
+; CHECK-NEXT: br i1 false, label %[[LOOP_LATCH]], label %[[THEN:.*]]
+; CHECK: [[THEN]]:
+; CHECK-NEXT: [[CALL136:%.*]] = load volatile ptr, ptr null, align 4294967296
+; CHECK-NEXT: br label %[[LOOP_LATCH]]
+; CHECK: [[LOOP_LATCH]]:
+; CHECK-NEXT: [[IV_NEXT]] = add nuw i32 [[IV]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = sub i32 [[N]], 1
+; CHECK-NEXT: [[EC:%.*]] = icmp eq i32 [[IV_NEXT]], [[TMP2]]
+; CHECK-NEXT: br i1 [[EC]], label %[[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT:.*]], label %[[LOOP_HEADER]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK: [[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT]]:
+; CHECK-NEXT: [[DOTPH:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP_LATCH]] ]
+; CHECK-NEXT: br label %[[EXIT_LOOPEXIT_PEEL_BEGIN]]
+; CHECK: [[EXIT_LOOPEXIT_PEEL_BEGIN]]:
+; CHECK-NEXT: [[TMP3:%.*]] = phi i32 [ 1, %[[LOOP_HEADER_PREHEADER]] ], [ [[DOTPH]], %[[EXIT_LOOPEXIT_PEEL_BEGIN_LOOPEXIT]] ]
+; CHECK-NEXT: br label %[[LOOP_HEADER_PEEL:.*]]
+; CHECK: [[LOOP_HEADER_PEEL]]:
+; CHECK-NEXT: [[CMP115_PEEL:%.*]] = icmp eq i32 [[TMP3]], [[SUB]]
+; CHECK-NEXT: br i1 [[CMP115_PEEL]], label %[[LOOP_LATCH_PEEL:.*]], label %[[THEN_PEEL:.*]]
+; CHECK: [[THEN_PEEL]]:
+; CHECK-NEXT: [[CALL136_PEEL:%.*]] = load volatile ptr, ptr null, align 4294967296
+; CHECK-NEXT: br label %[[LOOP_LATCH_PEEL]]
+; CHECK: [[LOOP_LATCH_PEEL]]:
+; CHECK-NEXT: [[IV_NEXT_PEEL:%.*]] = add nuw i32 [[TMP3]], 1
+; CHECK-NEXT: [[EC_PEEL:%.*]] = icmp eq i32 [[IV_NEXT_PEEL]], [[N]]
+; CHECK-NEXT: br i1 [[EC_PEEL]], label %[[EXIT_LOOPEXIT_PEEL_NEXT:.*]], label %[[EXIT_LOOPEXIT_PEEL_NEXT]]
+; CHECK: [[EXIT_LOOPEXIT_PEEL_NEXT]]:
+; CHECK-NEXT: br label %[[LOOP_HEADER_PEEL_NEXT:.*]]
+; CHECK: [[LOOP_HEADER_PEEL_NEXT]]:
+; CHECK-NEXT: br label %[[EXIT_LOOPEXIT:.*]]
+; CHECK: [[EXIT_LOOPEXIT]]:
+; CHECK-NEXT: br label %[[EXIT]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: ret void
+;
+entry:
+ %sub = add i32 %n, -1
+ %pre = icmp eq i32 %sub, 0
+ br i1 %pre, label %exit, label %loop.header
+
+loop.header: ; preds = %loop.latch, %entry
+ %iv = phi i32 [ %iv.next, %loop.latch ], [ 1, %entry ]
+ %cmp115 = icmp eq i32 %iv, %sub
+ br i1 %cmp115, label %loop.latch, label %then
+
+then:
+ %call136 = load volatile ptr, ptr null, align 4294967296
+ br label %loop.latch
+
+loop.latch:
+ %iv.next = add nuw i32 %iv, 1
+ %ec = icmp eq i32 %iv.next, %n
+ br i1 %ec, label %exit, label %loop.header
+
exit:
ret void
}
@@ -208,4 +280,5 @@ exit:
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]]}
; CHECK: [[META1]] = !{!"llvm.loop.peeled.count", i32 1}
; CHECK: [[LOOP2]] = distinct !{[[LOOP2]], [[META1]]}
+; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]}
;.
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-checks-
diff erence.ll b/llvm/test/Transforms/LoopVectorize/runtime-checks-
diff erence.ll
index 23c6baede0f3b..648ebc7e6c3a5 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-checks-
diff erence.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-checks-
diff erence.ll
@@ -434,3 +434,62 @@ loop:
exit:
ret void
}
+
+
+define void @remove_
diff _checks_via_guards(i32 %x, i32 %y, ptr %A) {
+; CHECK-LABEL: define void @remove_
diff _checks_via_guards(
+; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], ptr [[A:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[OFFSET:%.*]] = sub i32 [[X]], [[Y]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp sge i32 [[OFFSET]], 0
+; CHECK-NEXT: br i1 [[CMP]], [[EXIT:label %.*]], label %[[LOOP_PREHEADER:.*]]
+; CHECK: [[LOOP_PREHEADER]]:
+; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[X]] to i64
+; CHECK-NEXT: [[TMP1:%.*]] = add nsw i64 [[TMP0]], 1
+; CHECK-NEXT: [[SMAX2:%.*]] = call i64 @llvm.smax.i64(i64 [[TMP1]], i64 0)
+; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[SMAX2]] to i32
+; CHECK-NEXT: [[TMP3:%.*]] = add nuw i32 [[TMP2]], 1
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP3]], 4
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], [[SCALAR_PH:label %.*]], label %[[VECTOR_SCEVCHECK:.*]]
+; CHECK: [[VECTOR_SCEVCHECK]]:
+; CHECK-NEXT: [[TMP4:%.*]] = sext i32 [[X]] to i64
+; CHECK-NEXT: [[TMP5:%.*]] = add nsw i64 [[TMP4]], 1
+; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[TMP5]], i64 0)
+; CHECK-NEXT: [[TMP6:%.*]] = trunc i64 [[SMAX]] to i32
+; CHECK-NEXT: [[TMP7:%.*]] = icmp slt i32 [[TMP6]], 0
+; CHECK-NEXT: [[TMP8:%.*]] = icmp ugt i64 [[SMAX]], 4294967295
+; CHECK-NEXT: [[TMP9:%.*]] = or i1 [[TMP7]], [[TMP8]]
+; CHECK-NEXT: [[TMP10:%.*]] = trunc i64 [[SMAX]] to i32
+; CHECK-NEXT: [[TMP11:%.*]] = add i32 [[OFFSET]], [[TMP10]]
+; CHECK-NEXT: [[TMP12:%.*]] = icmp slt i32 [[TMP11]], [[OFFSET]]
+; CHECK-NEXT: [[TMP13:%.*]] = icmp ugt i64 [[SMAX]], 4294967295
+; CHECK-NEXT: [[TMP14:%.*]] = or i1 [[TMP12]], [[TMP13]]
+; CHECK-NEXT: [[TMP15:%.*]] = or i1 [[TMP9]], [[TMP14]]
+; CHECK-NEXT: br i1 [[TMP15]], [[SCALAR_PH]], label %[[VECTOR_MEMCHECK:.*]]
+; CHECK: [[VECTOR_MEMCHECK]]:
+; CHECK-NEXT: [[TMP16:%.*]] = sext i32 [[OFFSET]] to i64
+; CHECK-NEXT: [[TMP17:%.*]] = shl nsw i64 [[TMP16]], 2
+; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP17]], 16
+; CHECK-NEXT: br i1 [[DIFF_CHECK]], [[SCALAR_PH]], [[VECTOR_PH1:label %.*]]
+;
+entry:
+ %offset = sub i32 %x, %y
+ %cmp = icmp sge i32 %offset, 0
+ br i1 %cmp, label %exit, label %loop
+
+loop:
+ %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
+ %iv.ext = sext i32 %iv to i64
+ %gep.A = getelementptr i32, ptr %A, i64 %iv.ext
+ %l = load i32, ptr %gep.A, align 1
+ %iv.offset = add i32 %iv, %offset
+ %iv.offset.ext = sext i32 %iv.offset to i64
+ %gep.A.offset = getelementptr i32, ptr %A, i64 %iv.offset.ext
+ store i32 %l, ptr %gep.A.offset, align 1
+ %iv.next = add i32 %iv, 1
+ %ec = icmp sgt i32 %iv, %x
+ br i1 %ec, label %exit, label %loop
+
+exit:
+ ret void
+}
More information about the llvm-commits
mailing list