-
Notifications
You must be signed in to change notification settings - Fork 14.3k
[CVP] Infer range return attribute #99620
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
23bdf84
to
b9128bc
Compare
We already infer this in IPSCCP, but as that pass runs very early, it cannot make use of simplifications (in particular post-inline simplifications). This fixes most cases from llvm#98946 (everything apart from f2 where the assume is dropped).
b9128bc
to
46ab5f8
Compare
@llvm/pr-subscribers-hlsl @llvm/pr-subscribers-llvm-transforms Author: Nikita Popov (nikic) ChangesWe already infer this in IPSCCP (which runs very early, so cannot benefit from inlining and simplifications) and SCCP (which runs without PredicateInfo, so does not use assumes). Do it in CVP as well, so it can handle cases that IPSCCP/SCCP can't. Fixes #98946 (everything apart from f2, where the assume is dropped by the frontend). (No compile-time impact: https://llvm-compile-time-tracker.com/compare.php?from=f1ff3a279f3320d8e0d4abbc1a8357bb51de25a2&to=46ab5f809414b3f06ab9956e0fc6910284417b6d&stat=instructions:u) Full diff: https://github.com/llvm/llvm-project/pull/99620.diff 7 Files Affected:
diff --git a/llvm/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp b/llvm/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp
index 95de8eceb6be59..ed2bddd0521967 100644
--- a/llvm/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp
+++ b/llvm/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp
@@ -1207,6 +1207,11 @@ static bool processAnd(BinaryOperator *BinOp, LazyValueInfo *LVI) {
static bool runImpl(Function &F, LazyValueInfo *LVI, DominatorTree *DT,
const SimplifyQuery &SQ) {
bool FnChanged = false;
+ std::optional<ConstantRange> RetRange;
+ if (F.hasExactDefinition() && F.getReturnType()->isIntOrIntVectorTy())
+ RetRange =
+ ConstantRange::getEmpty(F.getReturnType()->getScalarSizeInBits());
+
// Visiting in a pre-order depth-first traversal causes us to simplify early
// blocks before querying later blocks (which require us to analyze early
// blocks). Eagerly simplifying shallow blocks means there is strictly less
@@ -1277,6 +1282,11 @@ static bool runImpl(Function &F, LazyValueInfo *LVI, DominatorTree *DT,
// constant folding the return values of callees.
auto *RetVal = RI->getReturnValue();
if (!RetVal) break; // handle "ret void"
+ if (RetRange && !RetRange->isFullSet())
+ RetRange =
+ RetRange->unionWith(LVI->getConstantRange(RetVal, RI,
+ /*UndefAllowed=*/false));
+
if (isa<Constant>(RetVal)) break; // nothing to do
if (auto *C = getConstantAt(RetVal, RI, LVI)) {
++NumReturns;
@@ -1289,6 +1299,14 @@ static bool runImpl(Function &F, LazyValueInfo *LVI, DominatorTree *DT,
FnChanged |= BBChanged;
}
+ // Infer range attribute on return value.
+ if (RetRange && !RetRange->isFullSet()) {
+ Attribute RangeAttr = F.getRetAttribute(Attribute::Range);
+ if (RangeAttr.isValid())
+ RetRange = RetRange->intersectWith(RangeAttr.getRange());
+ if (!RetRange->isEmptySet() && !RetRange->isSingleElement())
+ F.addRangeRetAttr(*RetRange);
+ }
return FnChanged;
}
diff --git a/llvm/test/Transforms/CorrelatedValuePropagation/add.ll b/llvm/test/Transforms/CorrelatedValuePropagation/add.ll
index b29a496ab1deb9..b1151cdf26ffd5 100644
--- a/llvm/test/Transforms/CorrelatedValuePropagation/add.ll
+++ b/llvm/test/Transforms/CorrelatedValuePropagation/add.ll
@@ -311,7 +311,7 @@ exit:
@limit = external global i32
define i32 @test11(ptr %p, i32 %i) {
-; CHECK-LABEL: define i32 @test11(
+; CHECK-LABEL: define range(i32 0, 2147483645) i32 @test11(
; CHECK-SAME: ptr [[P:%.*]], i32 [[I:%.*]]) {
; CHECK-NEXT: [[LIMIT:%.*]] = load i32, ptr [[P]], align 4, !range [[RNG0:![0-9]+]]
; CHECK-NEXT: [[WITHIN_1:%.*]] = icmp ugt i32 [[LIMIT]], [[I]]
diff --git a/llvm/test/Transforms/CorrelatedValuePropagation/ashr.ll b/llvm/test/Transforms/CorrelatedValuePropagation/ashr.ll
index 25bec18a5dbf37..f719effac113e9 100644
--- a/llvm/test/Transforms/CorrelatedValuePropagation/ashr.ll
+++ b/llvm/test/Transforms/CorrelatedValuePropagation/ashr.ll
@@ -159,7 +159,7 @@ exit:
; check that ashr of -1 or 0 is optimized away
define i32 @test6(i32 %f, i32 %g) {
-; CHECK-LABEL: define i32 @test6(
+; CHECK-LABEL: define range(i32 -1, 1) i32 @test6(
; CHECK-SAME: i32 [[F:%.*]], i32 [[G:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[F]], 1
@@ -177,7 +177,7 @@ entry:
; same test as above with different numbers
define i32 @test7(i32 %f, i32 %g) {
-; CHECK-LABEL: define i32 @test7(
+; CHECK-LABEL: define range(i32 -1, 1) i32 @test7(
; CHECK-SAME: i32 [[F:%.*]], i32 [[G:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[TMP0:%.*]] = and i32 [[F]], -2
@@ -197,7 +197,7 @@ entry:
; check that ashr of -2 or 1 is not optimized away
define i32 @test8(i32 %f, i32 %g, i1 %s) {
-; CHECK-LABEL: define i32 @test8(
+; CHECK-LABEL: define range(i32 -2, 2) i32 @test8(
; CHECK-SAME: i32 [[F:%.*]], i32 [[G:%.*]], i1 [[S:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[TMP0:%.*]] = ashr i32 -2, [[F]]
@@ -213,7 +213,7 @@ entry:
}
define i32 @may_including_undef(i1 %c.1, i1 %c.2) {
-; CHECK-LABEL: define i32 @may_including_undef(
+; CHECK-LABEL: define range(i32 -1073741824, 1073741824) i32 @may_including_undef(
; CHECK-SAME: i1 [[C_1:%.*]], i1 [[C_2:%.*]]) {
; CHECK-NEXT: br i1 [[C_1]], label %[[TRUE_1:.*]], label %[[FALSE:.*]]
; CHECK: [[TRUE_1]]:
diff --git a/llvm/test/Transforms/CorrelatedValuePropagation/basic.ll b/llvm/test/Transforms/CorrelatedValuePropagation/basic.ll
index c529bab4ef4a72..3c3b1d4bef45bb 100644
--- a/llvm/test/Transforms/CorrelatedValuePropagation/basic.ll
+++ b/llvm/test/Transforms/CorrelatedValuePropagation/basic.ll
@@ -3,7 +3,7 @@
; PR2581
define i32 @test1(i1 %C) {
-; CHECK-LABEL: define i32 @test1
+; CHECK-LABEL: define range(i32 10, 12) i32 @test1
; CHECK-SAME: (i1 [[C:%.*]]) {
; CHECK-NEXT: br i1 [[C]], label [[EXIT:%.*]], label [[BODY:%.*]]
; CHECK: body:
@@ -82,7 +82,7 @@ bb2:
; PR1757
define i32 @test4(i32) {
-; CHECK-LABEL: define i32 @test4
+; CHECK-LABEL: define range(i32 0, 3) i32 @test4
; CHECK-SAME: (i32 [[TMP0:%.*]]) {
; CHECK-NEXT: EntryBlock:
; CHECK-NEXT: [[DOTDEMORGAN:%.*]] = icmp sgt i32 [[TMP0]], 2
@@ -210,7 +210,7 @@ return:
}
define i32 @switch1(i32 %s) {
-; CHECK-LABEL: define i32 @switch1
+; CHECK-LABEL: define range(i32 -1, 2) i32 @switch1
; CHECK-SAME: (i32 [[S:%.*]]) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[S]], 0
@@ -250,7 +250,7 @@ next:
}
define i32 @switch2(i32 %s) {
-; CHECK-LABEL: define i32 @switch2
+; CHECK-LABEL: define range(i32 -1, 2) i32 @switch2
; CHECK-SAME: (i32 [[S:%.*]]) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[S]], 0
@@ -284,7 +284,7 @@ next:
}
define i32 @switch3(i32 %s) {
-; CHECK-LABEL: define i32 @switch3
+; CHECK-LABEL: define range(i32 -1, 2) i32 @switch3
; CHECK-SAME: (i32 [[S:%.*]]) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[S]], 0
@@ -451,7 +451,7 @@ unreachable:
}
define i32 @switch_range(i32 %cond) {
-; CHECK-LABEL: define i32 @switch_range
+; CHECK-LABEL: define range(i32 1, 3) i32 @switch_range
; CHECK-SAME: (i32 [[COND:%.*]]) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[S:%.*]] = urem i32 [[COND]], 3
@@ -491,7 +491,7 @@ unreachable:
; switch condition, we should not change the default.
define i32 @switch_range_not_full(i32 %cond) {
-; CHECK-LABEL: define i32 @switch_range_not_full
+; CHECK-LABEL: define range(i32 0, 3) i32 @switch_range_not_full
; CHECK-SAME: (i32 [[COND:%.*]]) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[S:%.*]] = urem i32 [[COND]], 3
@@ -2052,7 +2052,7 @@ define i1 @binop_eval_order(i32 %x) {
}
define range(i32 0, 1024) i32 @range_larger(i8 %x) {
-; CHECK-LABEL: define range(i32 0, 1024) i32 @range_larger
+; CHECK-LABEL: define range(i32 0, 256) i32 @range_larger
; CHECK-SAME: (i8 [[X:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext i8 [[X]] to i32
; CHECK-NEXT: ret i32 [[ZEXT]]
@@ -2072,7 +2072,7 @@ define range(i32 0, 128) i32 @range_smaller(i8 %x) {
}
define range(i32 128, 512) i32 @range_intersect(i8 %x) {
-; CHECK-LABEL: define range(i32 128, 512) i32 @range_intersect
+; CHECK-LABEL: define range(i32 128, 256) i32 @range_intersect
; CHECK-SAME: (i8 [[X:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext i8 [[X]] to i32
; CHECK-NEXT: ret i32 [[ZEXT]]
diff --git a/llvm/test/Transforms/CorrelatedValuePropagation/cond-using-block-value.ll b/llvm/test/Transforms/CorrelatedValuePropagation/cond-using-block-value.ll
index 252f6596cedc5e..a7a1803bccc263 100644
--- a/llvm/test/Transforms/CorrelatedValuePropagation/cond-using-block-value.ll
+++ b/llvm/test/Transforms/CorrelatedValuePropagation/cond-using-block-value.ll
@@ -38,7 +38,7 @@ end:
}
define i64 @test_sext_from_implied_cond(i32 %a, i32 %b) {
-; CHECK-LABEL: define i64 @test_sext_from_implied_cond(
+; CHECK-LABEL: define range(i64 0, 2147483647) i64 @test_sext_from_implied_cond(
; CHECK-SAME: i32 [[A:%.*]], i32 [[B:%.*]]) {
; CHECK-NEXT: [[A_CMP:%.*]] = icmp slt i32 [[A]], 0
; CHECK-NEXT: br i1 [[A_CMP]], label [[END:%.*]], label [[L1:%.*]]
diff --git a/llvm/test/Transforms/CorrelatedValuePropagation/select.ll b/llvm/test/Transforms/CorrelatedValuePropagation/select.ll
index 2054b0fc99d499..be027fbc698744 100644
--- a/llvm/test/Transforms/CorrelatedValuePropagation/select.ll
+++ b/llvm/test/Transforms/CorrelatedValuePropagation/select.ll
@@ -141,7 +141,7 @@ else:
}
define i8 @not_correlated(i1, i1) {
-; CHECK-LABEL: define i8 @not_correlated
+; CHECK-LABEL: define range(i8 0, 2) i8 @not_correlated
; CHECK-SAME: (i1 [[TMP0:%.*]], i1 [[TMP1:%.*]]) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[S:%.*]] = select i1 [[TMP0]], i8 0, i8 1
@@ -361,7 +361,7 @@ exit:
}
define i64 @select_cond_may_undef(i32 %a) {
-; CHECK-LABEL: define i64 @select_cond_may_undef
+; CHECK-LABEL: define range(i64 -2147483648, 2147483648) i64 @select_cond_may_undef
; CHECK-SAME: (i32 [[A:%.*]]) {
; CHECK-NEXT: [[IS_A_NONNEGATIVE:%.*]] = icmp sgt i32 [[A]], 1
; CHECK-NEXT: [[NARROW:%.*]] = select i1 [[IS_A_NONNEGATIVE]], i32 [[A]], i32 0
diff --git a/llvm/test/Transforms/CorrelatedValuePropagation/vectors.ll b/llvm/test/Transforms/CorrelatedValuePropagation/vectors.ll
index 351a2c79cdff42..bda270e99bc2e6 100644
--- a/llvm/test/Transforms/CorrelatedValuePropagation/vectors.ll
+++ b/llvm/test/Transforms/CorrelatedValuePropagation/vectors.ll
@@ -60,7 +60,7 @@ define <2 x i1> @cmp_signedness(<2 x i8> %a) {
}
define <2 x i16> @infer_nowrap(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @infer_nowrap(
+; CHECK-LABEL: define range(i16 1, 257) <2 x i16> @infer_nowrap(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: [[RES:%.*]] = add nuw nsw <2 x i16> [[ZEXT]], <i16 1, i16 1>
@@ -72,7 +72,7 @@ define <2 x i16> @infer_nowrap(<2 x i8> %a) {
}
define <2 x i16> @infer_nowrap_nonsplat(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @infer_nowrap_nonsplat(
+; CHECK-LABEL: define range(i16 1, 258) <2 x i16> @infer_nowrap_nonsplat(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: [[RES:%.*]] = add nuw nsw <2 x i16> [[ZEXT]], <i16 1, i16 2>
@@ -84,7 +84,7 @@ define <2 x i16> @infer_nowrap_nonsplat(<2 x i8> %a) {
}
define <vscale x 2 x i16> @infer_nowrap_scalable(<vscale x 2 x i8> %a) {
-; CHECK-LABEL: define <vscale x 2 x i16> @infer_nowrap_scalable(
+; CHECK-LABEL: define range(i16 1, 257) <vscale x 2 x i16> @infer_nowrap_scalable(
; CHECK-SAME: <vscale x 2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <vscale x 2 x i8> [[A]] to <vscale x 2 x i16>
; CHECK-NEXT: [[RES:%.*]] = add nuw nsw <vscale x 2 x i16> [[ZEXT]], shufflevector (<vscale x 2 x i16> insertelement (<vscale x 2 x i16> poison, i16 1, i64 0), <vscale x 2 x i16> poison, <vscale x 2 x i32> zeroinitializer)
@@ -96,7 +96,7 @@ define <vscale x 2 x i16> @infer_nowrap_scalable(<vscale x 2 x i8> %a) {
}
define <2 x i16> @infer_nowrap_poison(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @infer_nowrap_poison(
+; CHECK-LABEL: define range(i16 1, 257) <2 x i16> @infer_nowrap_poison(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: [[RES:%.*]] = add nuw nsw <2 x i16> [[ZEXT]], <i16 1, i16 poison>
@@ -108,7 +108,7 @@ define <2 x i16> @infer_nowrap_poison(<2 x i8> %a) {
}
define <2 x i16> @infer_nowrap_nonsplat_nsw_only(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @infer_nowrap_nonsplat_nsw_only(
+; CHECK-LABEL: define range(i16 -1, 257) <2 x i16> @infer_nowrap_nonsplat_nsw_only(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: [[RES:%.*]] = add nsw <2 x i16> [[ZEXT]], <i16 1, i16 -1>
@@ -120,7 +120,7 @@ define <2 x i16> @infer_nowrap_nonsplat_nsw_only(<2 x i8> %a) {
}
define <2 x i16> @abs(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @abs(
+; CHECK-LABEL: define range(i16 0, 256) <2 x i16> @abs(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: ret <2 x i16> [[ZEXT]]
@@ -131,7 +131,7 @@ define <2 x i16> @abs(<2 x i8> %a) {
}
define <2 x i16> @saturating(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @saturating(
+; CHECK-LABEL: define range(i16 1, 257) <2 x i16> @saturating(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: [[RES:%.*]] = add nuw nsw <2 x i16> [[ZEXT]], <i16 1, i16 1>
@@ -156,7 +156,7 @@ define {<2 x i16>, <2 x i1>} @with_overflow(<2 x i8> %a) {
}
define <2 x i16> @srem1(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @srem1(
+; CHECK-LABEL: define range(i16 0, 42) <2 x i16> @srem1(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: [[RES1_LHS_TRUNC:%.*]] = trunc <2 x i16> [[ZEXT]] to <2 x i8>
@@ -170,7 +170,7 @@ define <2 x i16> @srem1(<2 x i8> %a) {
}
define <2 x i16> @srem2(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @srem2(
+; CHECK-LABEL: define range(i16 -41, 42) <2 x i16> @srem2(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = sext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: [[RES_LHS_TRUNC:%.*]] = trunc <2 x i16> [[ZEXT]] to <2 x i8>
@@ -184,7 +184,7 @@ define <2 x i16> @srem2(<2 x i8> %a) {
}
define <2 x i16> @ashr(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @ashr(
+; CHECK-LABEL: define range(i16 0, 128) <2 x i16> @ashr(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: [[RES:%.*]] = lshr <2 x i16> [[ZEXT]], <i16 1, i16 1>
@@ -196,7 +196,7 @@ define <2 x i16> @ashr(<2 x i8> %a) {
}
define <2 x i32> @sext(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i32> @sext(
+; CHECK-LABEL: define range(i32 0, 256) <2 x i32> @sext(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: [[RES:%.*]] = zext nneg <2 x i16> [[ZEXT]] to <2 x i32>
@@ -220,7 +220,7 @@ define <2 x float> @sitofp(<2 x i8> %a) {
}
define <2 x i16> @and(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @and(
+; CHECK-LABEL: define range(i16 0, 256) <2 x i16> @and(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: ret <2 x i16> [[ZEXT]]
@@ -231,7 +231,7 @@ define <2 x i16> @and(<2 x i8> %a) {
}
define <2 x i16> @and_with_poison(<2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @and_with_poison(
+; CHECK-LABEL: define range(i16 0, 256) <2 x i16> @and_with_poison(
; CHECK-SAME: <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
; CHECK-NEXT: ret <2 x i16> [[ZEXT]]
@@ -242,7 +242,7 @@ define <2 x i16> @and_with_poison(<2 x i8> %a) {
}
define <4 x i64> @issue_97674_getConstantOnEdge(i1 %cond) {
-; CHECK-LABEL: define <4 x i64> @issue_97674_getConstantOnEdge(
+; CHECK-LABEL: define range(i64 0, 2) <4 x i64> @issue_97674_getConstantOnEdge(
; CHECK-SAME: i1 [[COND:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: br i1 [[COND]], label %[[IF_THEN:.*]], label %[[IF_END:.*]]
@@ -277,7 +277,7 @@ entry:
}
define <2 x i16> @phi_merge1(i1 %c, <2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @phi_merge1(
+; CHECK-LABEL: define range(i16 2, 259) <2 x i16> @phi_merge1(
; CHECK-SAME: i1 [[C:%.*]], <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
@@ -303,7 +303,7 @@ join:
}
define <2 x i16> @phi_merge2(i1 %c, <2 x i8> %a) {
-; CHECK-LABEL: define <2 x i16> @phi_merge2(
+; CHECK-LABEL: define range(i16 2, 259) <2 x i16> @phi_merge2(
; CHECK-SAME: i1 [[C:%.*]], <2 x i8> [[A:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: [[ZEXT:%.*]] = zext <2 x i8> [[A]] to <2 x i16>
|
Attribute RangeAttr = F.getRetAttribute(Attribute::Range); | ||
if (RangeAttr.isValid()) | ||
RetRange = RetRange->intersectWith(RangeAttr.getRange()); | ||
if (!RetRange->isEmptySet() && !RetRange->isSingleElement()) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've excluded the single element case to make this less spammy. I didn't want to have range attributes on every single function returning a constant (in which case IPSCCP would already propagate the cosntant).
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe leave that as a comment?
Missing fold: https://alive2.llvm.org/ce/z/2rmc3h |
if (RangeAttr.isValid()) | ||
RetRange = RetRange->intersectWith(RangeAttr.getRange()); | ||
if (!RetRange->isEmptySet() && !RetRange->isSingleElement()) | ||
F.addRangeRetAttr(*RetRange); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Missing FnChanged = true
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Good point, fixed.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Could we cause an inf loop if existing range attr is same as new one?
In principle, I think this could be supported by making decomposeBitTestICmp detect that |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Thank you!
We already infer this in IPSCCP (which runs very early, so cannot benefit from inlining and simplifications) and SCCP (which runs without PredicateInfo, so does not use assumes). Do it in CVP as well, so it can handle cases that IPSCCP/SCCP can't.
Fixes #98946 (everything apart from f2, where the assume is dropped by the frontend).
(No compile-time impact: https://llvm-compile-time-tracker.com/compare.php?from=f1ff3a279f3320d8e0d4abbc1a8357bb51de25a2&to=46ab5f809414b3f06ab9956e0fc6910284417b6d&stat=instructions:u)