[RISCV][LoopIdiomVectorize] Replace vp.icmp -> regular icmp. NFCish #157068
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
To match the general direction of the loop vectorizer and reduce our dependency on trivial VP intrinsics[1], replace the use of vp.icmp with a regular icmp. RISCVVLOptimizer can take care of propagating VL. [1]: https://discourse.llvm.org/t/rfc-remove-codegen-support-for-trivial-vp-intrinsics-in-the-risc-v-backend/87999
|
@llvm/pr-subscribers-backend-risc-v @llvm/pr-subscribers-vectorizers Author: Luke Lau (lukel97) ChangesTo match the general direction of the loop vectorizer and reduce our dependency on trivial VP intrinsics, replace the use of vp.icmp with a regular icmp. RISCVVLOptimizer can take care of propagating VL. Full diff: https://github.com/llvm/llvm-project/pull/157068.diff 3 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/LoopIdiomVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopIdiomVectorize.cpp
index 491f0b76f4ae0..8d1915a6a5ed4 100644
--- a/llvm/lib/Transforms/Vectorize/LoopIdiomVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopIdiomVectorize.cpp
@@ -574,13 +574,8 @@ Value *LoopIdiomVectorize::createPredicatedFindMismatch(
Intrinsic::vp_load, {VectorLoadType, VectorLhsGep->getType()},
{VectorRhsGep, AllTrueMask, VL}, nullptr, "rhs.load");
- StringRef PredicateStr = CmpInst::getPredicateName(CmpInst::ICMP_NE);
- auto *PredicateMDS = MDString::get(VectorLhsLoad->getContext(), PredicateStr);
- Value *Pred = MetadataAsValue::get(VectorLhsLoad->getContext(), PredicateMDS);
- Value *VectorMatchCmp = Builder.CreateIntrinsic(
- Intrinsic::vp_icmp, {VectorLhsLoad->getType()},
- {VectorLhsLoad, VectorRhsLoad, Pred, AllTrueMask, VL}, nullptr,
- "mismatch.cmp");
+ Value *VectorMatchCmp =
+ Builder.CreateICmpNE(VectorLhsLoad, VectorRhsLoad, "mismatch.cmp");
Value *CTZ = Builder.CreateIntrinsic(
Intrinsic::vp_cttz_elts, {ResType, VectorMatchCmp->getType()},
{VectorMatchCmp, /*ZeroIsPoison=*/Builder.getInt1(false), AllTrueMask,
diff --git a/llvm/test/CodeGen/RISCV/rvv/vfirst-byte-compare-index.ll b/llvm/test/CodeGen/RISCV/rvv/vfirst-byte-compare-index.ll
index abfb652f2206e..7c34475a11d04 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vfirst-byte-compare-index.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vfirst-byte-compare-index.ll
@@ -111,7 +111,7 @@ mismatch_vec_loop: ; preds = %mismatch_vec_loop_i
%lhs.load = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr %20, <vscale x 16 x i1> shufflevector (<vscale x 16 x i1> insertelement (<vscale x 16 x i1> poison, i1 true, i64 0), <vscale x 16 x i1> poison, <vscale x 16 x i32> zeroinitializer), i32 %19)
%21 = getelementptr inbounds i8, ptr %b, i64 %mismatch_vector_index
%rhs.load = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr %21, <vscale x 16 x i1> shufflevector (<vscale x 16 x i1> insertelement (<vscale x 16 x i1> poison, i1 true, i64 0), <vscale x 16 x i1> poison, <vscale x 16 x i32> zeroinitializer), i32 %19)
- %mismatch.cmp = call <vscale x 16 x i1> @llvm.vp.icmp.nxv16i8(<vscale x 16 x i8> %lhs.load, <vscale x 16 x i8> %rhs.load, metadata !"ne", <vscale x 16 x i1> shufflevector (<vscale x 16 x i1> insertelement (<vscale x 16 x i1> poison, i1 true, i64 0), <vscale x 16 x i1> poison, <vscale x 16 x i32> zeroinitializer), i32 %19)
+ %mismatch.cmp = icmp ne <vscale x 16 x i8> %lhs.load, %rhs.load
%22 = call i32 @llvm.vp.cttz.elts.i32.nxv16i1(<vscale x 16 x i1> %mismatch.cmp, i1 false, <vscale x 16 x i1> shufflevector (<vscale x 16 x i1> insertelement (<vscale x 16 x i1> poison, i1 true, i64 0), <vscale x 16 x i1> poison, <vscale x 16 x i32> zeroinitializer), i32 %19)
%23 = icmp ne i32 %22, %19
br i1 %23, label %mismatch_vec_loop_found, label %mismatch_vec_loop_inc
diff --git a/llvm/test/Transforms/LoopIdiom/RISCV/byte-compare-index.ll b/llvm/test/Transforms/LoopIdiom/RISCV/byte-compare-index.ll
index afc28cfda45aa..110b4a8b2e902 100644
--- a/llvm/test/Transforms/LoopIdiom/RISCV/byte-compare-index.ll
+++ b/llvm/test/Transforms/LoopIdiom/RISCV/byte-compare-index.ll
@@ -42,7 +42,7 @@ define i32 @compare_bytes_simple(ptr %a, ptr %b, i32 %len, i32 %n) {
; CHECK-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP20]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; CHECK-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP21]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
-; CHECK-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 16 x i1> @llvm.vp.icmp.nxv16i8(<vscale x 16 x i8> [[LHS_LOAD]], <vscale x 16 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
+; CHECK-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 16 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; CHECK-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv16i1(<vscale x 16 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; CHECK-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; CHECK-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -132,7 +132,7 @@ define i32 @compare_bytes_simple(ptr %a, ptr %b, i32 %len, i32 %n) {
; LMUL8-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr [[TMP20]], <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
; LMUL8-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; LMUL8-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr [[TMP21]], <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
-; LMUL8-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 64 x i1> @llvm.vp.icmp.nxv64i8(<vscale x 64 x i8> [[LHS_LOAD]], <vscale x 64 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
+; LMUL8-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 64 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; LMUL8-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv64i1(<vscale x 64 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
; LMUL8-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; LMUL8-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -218,7 +218,7 @@ define i32 @compare_bytes_simple(ptr %a, ptr %b, i32 %len, i32 %n) {
; LOOP-DEL-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP20]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; LOOP-DEL-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; LOOP-DEL-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP21]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
-; LOOP-DEL-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 16 x i1> @llvm.vp.icmp.nxv16i8(<vscale x 16 x i8> [[LHS_LOAD]], <vscale x 16 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
+; LOOP-DEL-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 16 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; LOOP-DEL-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv16i1(<vscale x 16 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; LOOP-DEL-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; LOOP-DEL-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -409,7 +409,7 @@ define i32 @compare_bytes_signed_wrap(ptr %a, ptr %b, i32 %len, i32 %n) {
; CHECK-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP20]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; CHECK-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP21]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
-; CHECK-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 16 x i1> @llvm.vp.icmp.nxv16i8(<vscale x 16 x i8> [[LHS_LOAD]], <vscale x 16 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
+; CHECK-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 16 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; CHECK-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv16i1(<vscale x 16 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; CHECK-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; CHECK-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -499,7 +499,7 @@ define i32 @compare_bytes_signed_wrap(ptr %a, ptr %b, i32 %len, i32 %n) {
; LMUL8-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr [[TMP20]], <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
; LMUL8-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; LMUL8-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr [[TMP21]], <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
-; LMUL8-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 64 x i1> @llvm.vp.icmp.nxv64i8(<vscale x 64 x i8> [[LHS_LOAD]], <vscale x 64 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
+; LMUL8-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 64 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; LMUL8-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv64i1(<vscale x 64 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
; LMUL8-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; LMUL8-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -585,7 +585,7 @@ define i32 @compare_bytes_signed_wrap(ptr %a, ptr %b, i32 %len, i32 %n) {
; LOOP-DEL-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP20]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; LOOP-DEL-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; LOOP-DEL-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP21]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
-; LOOP-DEL-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 16 x i1> @llvm.vp.icmp.nxv16i8(<vscale x 16 x i8> [[LHS_LOAD]], <vscale x 16 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
+; LOOP-DEL-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 16 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; LOOP-DEL-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv16i1(<vscale x 16 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; LOOP-DEL-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; LOOP-DEL-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -797,7 +797,7 @@ define i32 @compare_bytes_simple_end_ne_found(ptr %a, ptr %b, ptr %c, ptr %d, i3
; CHECK-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP20]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; CHECK-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP21]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
-; CHECK-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 16 x i1> @llvm.vp.icmp.nxv16i8(<vscale x 16 x i8> [[LHS_LOAD]], <vscale x 16 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
+; CHECK-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 16 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; CHECK-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv16i1(<vscale x 16 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; CHECK-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; CHECK-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -898,7 +898,7 @@ define i32 @compare_bytes_simple_end_ne_found(ptr %a, ptr %b, ptr %c, ptr %d, i3
; LMUL8-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr [[TMP20]], <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
; LMUL8-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; LMUL8-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr [[TMP21]], <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
-; LMUL8-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 64 x i1> @llvm.vp.icmp.nxv64i8(<vscale x 64 x i8> [[LHS_LOAD]], <vscale x 64 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
+; LMUL8-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 64 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; LMUL8-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv64i1(<vscale x 64 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
; LMUL8-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; LMUL8-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -995,7 +995,7 @@ define i32 @compare_bytes_simple_end_ne_found(ptr %a, ptr %b, ptr %c, ptr %d, i3
; LOOP-DEL-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP20]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; LOOP-DEL-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; LOOP-DEL-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP21]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
-; LOOP-DEL-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 16 x i1> @llvm.vp.icmp.nxv16i8(<vscale x 16 x i8> [[LHS_LOAD]], <vscale x 16 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
+; LOOP-DEL-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 16 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; LOOP-DEL-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv16i1(<vscale x 16 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; LOOP-DEL-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; LOOP-DEL-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -1248,7 +1248,7 @@ define i32 @compare_bytes_extra_cmp(ptr %a, ptr %b, i32 %len, i32 %n, i32 %x) {
; CHECK-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP20]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; CHECK-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP21]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
-; CHECK-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 16 x i1> @llvm.vp.icmp.nxv16i8(<vscale x 16 x i8> [[LHS_LOAD]], <vscale x 16 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
+; CHECK-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 16 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; CHECK-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv16i1(<vscale x 16 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; CHECK-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; CHECK-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -1344,7 +1344,7 @@ define i32 @compare_bytes_extra_cmp(ptr %a, ptr %b, i32 %len, i32 %n, i32 %x) {
; LMUL8-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr [[TMP20]], <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
; LMUL8-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; LMUL8-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr [[TMP21]], <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
-; LMUL8-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 64 x i1> @llvm.vp.icmp.nxv64i8(<vscale x 64 x i8> [[LHS_LOAD]], <vscale x 64 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
+; LMUL8-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 64 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; LMUL8-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv64i1(<vscale x 64 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 64 x i1> splat (i1 true), i32 [[TMP19]])
; LMUL8-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; LMUL8-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -1436,7 +1436,7 @@ define i32 @compare_bytes_extra_cmp(ptr %a, ptr %b, i32 %len, i32 %n, i32 %x) {
; LOOP-DEL-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP20]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; LOOP-DEL-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[MISMATCH_VECTOR_INDEX]]
; LOOP-DEL-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP21]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
-; LOOP-DEL-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 16 x i1> @llvm.vp.icmp.nxv16i8(<vscale x 16 x i8> [[LHS_LOAD]], <vscale x 16 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
+; LOOP-DEL-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 16 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; LOOP-DEL-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv16i1(<vscale x 16 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 16 x i1> splat (i1 true), i32 [[TMP19]])
; LOOP-DEL-NEXT: [[TMP22:%.*]] = icmp ne i32 [[FIRST]], [[TMP19]]
; LOOP-DEL-NEXT: br i1 [[TMP22]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -1656,7 +1656,7 @@ define void @compare_bytes_cleanup_block(ptr %src1, ptr %src2) {
; CHECK-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP16]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP15]])
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[SRC2]], i64 [[MISMATCH_VECTOR_INDEX]]
; CHECK-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.vp.load.nxv16i8.p0(ptr [[TMP17]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP15]])
-; CHECK-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 16 x i1> @llvm.vp.icmp.nxv16i8(<vscale x 16 x i8> [[LHS_LOAD]], <vscale x 16 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 16 x i1> splat (i1 true), i32 [[TMP15]])
+; CHECK-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 16 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; CHECK-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv16i1(<vscale x 16 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 16 x i1> splat (i1 true), i32 [[TMP15]])
; CHECK-NEXT: [[TMP18:%.*]] = icmp ne i32 [[FIRST]], [[TMP15]]
; CHECK-NEXT: br i1 [[TMP18]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
@@ -1745,7 +1745,7 @@ define void @compare_bytes_cleanup_block(ptr %src1, ptr %src2) {
; LMUL8-NEXT: [[LHS_LOAD:%.*]] = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr [[TMP16]], <vscale x 64 x i1> splat (i1 true), i32 [[TMP15]])
; LMUL8-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[SRC2]], i64 [[MISMATCH_VECTOR_INDEX]]
; LMUL8-NEXT: [[RHS_LOAD:%.*]] = call <vscale x 64 x i8> @llvm.vp.load.nxv64i8.p0(ptr [[TMP17]], <vscale x 64 x i1> splat (i1 true), i32 [[TMP15]])
-; LMUL8-NEXT: [[MISMATCH_CMP:%.*]] = call <vscale x 64 x i1> @llvm.vp.icmp.nxv64i8(<vscale x 64 x i8> [[LHS_LOAD]], <vscale x 64 x i8> [[RHS_LOAD]], metadata !"ne", <vscale x 64 x i1> splat (i1 true), i32 [[TMP15]])
+; LMUL8-NEXT: [[MISMATCH_CMP:%.*]] = icmp ne <vscale x 64 x i8> [[LHS_LOAD]], [[RHS_LOAD]]
; LMUL8-NEXT: [[FIRST:%.*]] = call i32 @llvm.vp.cttz.elts.i32.nxv64i1(<vscale x 64 x i1> [[MISMATCH_CMP]], i1 false, <vscale x 64 x i1> splat (i1 true), i32 [[TMP15]])
; LMUL8-NEXT: [[TMP18:%.*]] = icmp ne i32 [[FIRST]], [[TMP15]]
; LMUL8-NEXT: br i1 [[TMP18]], label [[MISMATCH_VECTOR_LOOP_FOUND:%.*]], label [[MISMATCH_VECTOR_LOOP_INC]]
|
lukel97
added a commit
to lukel97/llvm-project
that referenced
this pull request
Sep 5, 2025
Similar to llvm#157068, remove our reliance on trivial VP intrinsics by using a regular select instruction and letting RISCVVLOptimizer take care of VL.
lukel97
added a commit
to lukel97/llvm-project
that referenced
this pull request
Sep 5, 2025
Similar to llvm#157070 and llvm#157068, let RISCVVLOptimizer do the work.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
To match the general direction of the loop vectorizer and reduce our dependency on trivial VP intrinsics, replace the use of vp.icmp with a regular icmp. RISCVVLOptimizer can take care of propagating VL.