[RISCV] Add cost for @llvm.vector.splice.{left,right} - #179219
Merged
lukel97 merged 4 commits intoFeb 9, 2026
Conversation
Currently vector splice intrinsics are costed through getShuffleCost when the offset is fixed. When the offset is variable though we can't use a shuffle mask so it currently returns invalid. This implements the cost in RISCVTTIImpl::getIntrinsicInstrCost as the cost of a slideup and a slidedown, which matches the codegen. It may be possible to reduce the cost in future when one of the vector operands is known to be poison, in which case we only generate a single slideup or slidedown.
lukel97
requested review from
ElvisWang123,
arcbbb,
preames,
topperc and
wangpc-pp
February 2, 2026 12:20
llvmbot
added
backend:RISC-V
llvm:analysis
Member
|
@llvm/pr-subscribers-backend-risc-v @llvm/pr-subscribers-llvm-analysis Author: Luke Lau (lukel97) ChangesCurrently vector splice intrinsics are costed through getShuffleCost when the offset is fixed. When the offset is variable though we can't use a shuffle mask so it currently returns invalid. This implements the cost in RISCVTTIImpl::getIntrinsicInstrCost as the cost of a slideup and a slidedown, which matches the codegen. It may be possible to reduce the cost in future when one of the vector operands is known to be poison, in which case we only generate a single slideup or slidedown. Patch is 76.79 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/179219.diff 2 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index dc4bf0784e2d1..77604e6457400 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -1628,6 +1628,16 @@ RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
getRISCVInstructionCost(RISCV::VADD_VX, LT.second, CostKind);
return 1 + (LT.first - 1);
}
+ case Intrinsic::vector_splice_left:
+ case Intrinsic::vector_splice_right: {
+ auto LT = getTypeLegalizationCost(RetTy);
+ if (ST->hasVInstructions() && LT.second.isVector()) {
+ return LT.first *
+ getRISCVInstructionCost({RISCV::VSLIDEDOWN_VX, RISCV::VSLIDEUP_VX},
+ LT.second, CostKind);
+ }
+ break;
+ }
case Intrinsic::experimental_cttz_elts: {
Type *ArgTy = ICA.getArgTypes()[0];
EVT ArgType = TLI->getValueType(DL, ArgTy, true);
diff --git a/llvm/test/Analysis/CostModel/RISCV/splice.ll b/llvm/test/Analysis/CostModel/RISCV/splice.ll
index 5250c3dc1171a..0dad42c945809 100644
--- a/llvm/test/Analysis/CostModel/RISCV/splice.ll
+++ b/llvm/test/Analysis/CostModel/RISCV/splice.ll
@@ -4,7 +4,7 @@
; RUN: opt < %s -passes="print<cost-model>" -cost-kind=code-size 2>&1 -disable-output -S -mtriple=riscv64 -mattr=+v,+f,+d,+zfh,+zvfh | FileCheck %s --check-prefix=SIZE
; RUN: opt < %s -passes="print<cost-model>" -cost-kind=code-size 2>&1 -disable-output -S -mtriple=riscv64 -mattr=+v,+f,+d,+zfh,+zvfhmin | FileCheck %s --check-prefix=SIZE
-define void @vector_splice(i32 zeroext %offset) {
+define void @vector_splice() {
; CHECK-LABEL: 'vector_splice'
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %1 = call <vscale x 1 x i8> @llvm.vector.splice.right.nxv1i8(<vscale x 1 x i8> zeroinitializer, <vscale x 1 x i8> zeroinitializer, i32 1)
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %2 = call <vscale x 2 x i8> @llvm.vector.splice.right.nxv2i8(<vscale x 2 x i8> zeroinitializer, <vscale x 2 x i8> zeroinitializer, i32 1)
@@ -62,8 +62,6 @@ define void @vector_splice(i32 zeroext %offset) {
; CHECK-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %54 = call <vscale x 16 x double> @llvm.vector.splice.right.nxv16f64(<vscale x 16 x double> zeroinitializer, <vscale x 16 x double> zeroinitializer, i32 1)
; CHECK-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %55 = call <vscale x 32 x double> @llvm.vector.splice.right.nxv32f64(<vscale x 32 x double> zeroinitializer, <vscale x 32 x double> zeroinitializer, i32 1)
; CHECK-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %56 = call <vscale x 64 x double> @llvm.vector.splice.right.nxv64f64(<vscale x 64 x double> zeroinitializer, <vscale x 64 x double> zeroinitializer, i32 1)
-; CHECK-NEXT: Cost Model: Invalid cost for instruction: %left.variable = call <vscale x 4 x i32> @llvm.vector.splice.left.nxv4i32(<vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> zeroinitializer, i32 %offset)
-; CHECK-NEXT: Cost Model: Invalid cost for instruction: %right.variable = call <vscale x 4 x i32> @llvm.vector.splice.right.nxv4i32(<vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> zeroinitializer, i32 %offset)
; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SIZE-LABEL: 'vector_splice'
@@ -123,8 +121,6 @@ define void @vector_splice(i32 zeroext %offset) {
; SIZE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %54 = call <vscale x 16 x double> @llvm.vector.splice.right.nxv16f64(<vscale x 16 x double> zeroinitializer, <vscale x 16 x double> zeroinitializer, i32 1)
; SIZE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %55 = call <vscale x 32 x double> @llvm.vector.splice.right.nxv32f64(<vscale x 32 x double> zeroinitializer, <vscale x 32 x double> zeroinitializer, i32 1)
; SIZE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %56 = call <vscale x 64 x double> @llvm.vector.splice.right.nxv64f64(<vscale x 64 x double> zeroinitializer, <vscale x 64 x double> zeroinitializer, i32 1)
-; SIZE-NEXT: Cost Model: Invalid cost for instruction: %left.variable = call <vscale x 4 x i32> @llvm.vector.splice.left.nxv4i32(<vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> zeroinitializer, i32 %offset)
-; SIZE-NEXT: Cost Model: Invalid cost for instruction: %right.variable = call <vscale x 4 x i32> @llvm.vector.splice.right.nxv4i32(<vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> zeroinitializer, i32 %offset)
; SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%splice.nxv1i8 = call <vscale x 1 x i8> @llvm.vector.splice.nxv1i8(<vscale x 1 x i8> zeroinitializer, <vscale x 1 x i8> zeroinitializer, i32 -1)
@@ -191,8 +187,377 @@ define void @vector_splice(i32 zeroext %offset) {
%splice.nxv32f64 = call <vscale x 32 x double> @llvm.vector.splice.nxv32f64(<vscale x 32 x double> zeroinitializer, <vscale x 32 x double> zeroinitializer, i32 -1)
%splice.nxv64f64 = call <vscale x 64 x double> @llvm.vector.splice.nxv64f64(<vscale x 64 x double> zeroinitializer, <vscale x 64 x double> zeroinitializer, i32 -1)
- %left.variable = call <vscale x 4 x i32> @llvm.vector.splice.left(<vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> zeroinitializer, i32 %offset)
- %right.variable = call <vscale x 4 x i32> @llvm.vector.splice.right(<vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> zeroinitializer, i32 %offset)
+ ret void
+}
+
+define void @vector_splice_left_variable(i32 zeroext %offset) {
+; CHECK-LABEL: 'vector_splice_left_variable'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv1i8 = call <vscale x 1 x i8> @llvm.vector.splice.left.nxv1i8(<vscale x 1 x i8> zeroinitializer, <vscale x 1 x i8> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv2i8 = call <vscale x 2 x i8> @llvm.vector.splice.left.nxv2i8(<vscale x 2 x i8> zeroinitializer, <vscale x 2 x i8> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv4i8 = call <vscale x 4 x i8> @llvm.vector.splice.left.nxv4i8(<vscale x 4 x i8> zeroinitializer, <vscale x 4 x i8> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv8i8 = call <vscale x 8 x i8> @llvm.vector.splice.left.nxv8i8(<vscale x 8 x i8> zeroinitializer, <vscale x 8 x i8> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %splice.nxv16i8 = call <vscale x 16 x i8> @llvm.vector.splice.left.nxv16i8(<vscale x 16 x i8> zeroinitializer, <vscale x 16 x i8> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %splice.nxv32i8 = call <vscale x 32 x i8> @llvm.vector.splice.left.nxv32i8(<vscale x 32 x i8> zeroinitializer, <vscale x 32 x i8> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %splice.nxv64i8 = call <vscale x 64 x i8> @llvm.vector.splice.left.nxv64i8(<vscale x 64 x i8> zeroinitializer, <vscale x 64 x i8> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv1i16 = call <vscale x 1 x i16> @llvm.vector.splice.left.nxv1i16(<vscale x 1 x i16> zeroinitializer, <vscale x 1 x i16> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv2i16 = call <vscale x 2 x i16> @llvm.vector.splice.left.nxv2i16(<vscale x 2 x i16> zeroinitializer, <vscale x 2 x i16> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv4i16 = call <vscale x 4 x i16> @llvm.vector.splice.left.nxv4i16(<vscale x 4 x i16> zeroinitializer, <vscale x 4 x i16> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %splice.nxv8i16 = call <vscale x 8 x i16> @llvm.vector.splice.left.nxv8i16(<vscale x 8 x i16> zeroinitializer, <vscale x 8 x i16> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %splice.nxv16i16 = call <vscale x 16 x i16> @llvm.vector.splice.left.nxv16i16(<vscale x 16 x i16> zeroinitializer, <vscale x 16 x i16> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %splice.nxv32i16 = call <vscale x 32 x i16> @llvm.vector.splice.left.nxv32i16(<vscale x 32 x i16> zeroinitializer, <vscale x 32 x i16> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %splice.nxv64i16 = call <vscale x 64 x i16> @llvm.vector.splice.left.nxv64i16(<vscale x 64 x i16> zeroinitializer, <vscale x 64 x i16> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv1i32 = call <vscale x 1 x i32> @llvm.vector.splice.left.nxv1i32(<vscale x 1 x i32> zeroinitializer, <vscale x 1 x i32> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv2i32 = call <vscale x 2 x i32> @llvm.vector.splice.left.nxv2i32(<vscale x 2 x i32> zeroinitializer, <vscale x 2 x i32> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %splice.nxv4i32 = call <vscale x 4 x i32> @llvm.vector.splice.left.nxv4i32(<vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %splice.nxv8i32 = call <vscale x 8 x i32> @llvm.vector.splice.left.nxv8i32(<vscale x 8 x i32> zeroinitializer, <vscale x 8 x i32> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %splice.nxv16i32 = call <vscale x 16 x i32> @llvm.vector.splice.left.nxv16i32(<vscale x 16 x i32> zeroinitializer, <vscale x 16 x i32> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %splice.nxv32i32 = call <vscale x 32 x i32> @llvm.vector.splice.left.nxv32i32(<vscale x 32 x i32> zeroinitializer, <vscale x 32 x i32> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %splice.nxv64i32 = call <vscale x 64 x i32> @llvm.vector.splice.left.nxv64i32(<vscale x 64 x i32> zeroinitializer, <vscale x 64 x i32> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv1i64 = call <vscale x 1 x i64> @llvm.vector.splice.left.nxv1i64(<vscale x 1 x i64> zeroinitializer, <vscale x 1 x i64> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %splice.nxv2i64 = call <vscale x 2 x i64> @llvm.vector.splice.left.nxv2i64(<vscale x 2 x i64> zeroinitializer, <vscale x 2 x i64> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %splice.nxv4i64 = call <vscale x 4 x i64> @llvm.vector.splice.left.nxv4i64(<vscale x 4 x i64> zeroinitializer, <vscale x 4 x i64> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %splice.nxv8i64 = call <vscale x 8 x i64> @llvm.vector.splice.left.nxv8i64(<vscale x 8 x i64> zeroinitializer, <vscale x 8 x i64> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %splice.nxv16i64 = call <vscale x 16 x i64> @llvm.vector.splice.left.nxv16i64(<vscale x 16 x i64> zeroinitializer, <vscale x 16 x i64> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %splice.nxv32i64 = call <vscale x 32 x i64> @llvm.vector.splice.left.nxv32i64(<vscale x 32 x i64> zeroinitializer, <vscale x 32 x i64> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %splice.nxv64i64 = call <vscale x 64 x i64> @llvm.vector.splice.left.nxv64i64(<vscale x 64 x i64> zeroinitializer, <vscale x 64 x i64> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv1bf16 = call <vscale x 1 x bfloat> @llvm.vector.splice.left.nxv1bf16(<vscale x 1 x bfloat> zeroinitializer, <vscale x 1 x bfloat> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv2bf16 = call <vscale x 2 x bfloat> @llvm.vector.splice.left.nxv2bf16(<vscale x 2 x bfloat> zeroinitializer, <vscale x 2 x bfloat> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv4bf16 = call <vscale x 4 x bfloat> @llvm.vector.splice.left.nxv4bf16(<vscale x 4 x bfloat> zeroinitializer, <vscale x 4 x bfloat> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %splice.nxv8bf16 = call <vscale x 8 x bfloat> @llvm.vector.splice.left.nxv8bf16(<vscale x 8 x bfloat> zeroinitializer, <vscale x 8 x bfloat> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %splice.nxv16bf16 = call <vscale x 16 x bfloat> @llvm.vector.splice.left.nxv16bf16(<vscale x 16 x bfloat> zeroinitializer, <vscale x 16 x bfloat> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %splice.nxv32bf16 = call <vscale x 32 x bfloat> @llvm.vector.splice.left.nxv32bf16(<vscale x 32 x bfloat> zeroinitializer, <vscale x 32 x bfloat> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %splice.nxv64bf16 = call <vscale x 64 x bfloat> @llvm.vector.splice.left.nxv64bf16(<vscale x 64 x bfloat> zeroinitializer, <vscale x 64 x bfloat> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv1f16 = call <vscale x 1 x half> @llvm.vector.splice.left.nxv1f16(<vscale x 1 x half> zeroinitializer, <vscale x 1 x half> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv2f16 = call <vscale x 2 x half> @llvm.vector.splice.left.nxv2f16(<vscale x 2 x half> zeroinitializer, <vscale x 2 x half> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv4f16 = call <vscale x 4 x half> @llvm.vector.splice.left.nxv4f16(<vscale x 4 x half> zeroinitializer, <vscale x 4 x half> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %splice.nxv8f16 = call <vscale x 8 x half> @llvm.vector.splice.left.nxv8f16(<vscale x 8 x half> zeroinitializer, <vscale x 8 x half> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %splice.nxv16f16 = call <vscale x 16 x half> @llvm.vector.splice.left.nxv16f16(<vscale x 16 x half> zeroinitializer, <vscale x 16 x half> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %splice.nxv32f16 = call <vscale x 32 x half> @llvm.vector.splice.left.nxv32f16(<vscale x 32 x half> zeroinitializer, <vscale x 32 x half> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %splice.nxv64f16 = call <vscale x 64 x half> @llvm.vector.splice.left.nxv64f16(<vscale x 64 x half> zeroinitializer, <vscale x 64 x half> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv1f32 = call <vscale x 1 x float> @llvm.vector.splice.left.nxv1f32(<vscale x 1 x float> zeroinitializer, <vscale x 1 x float> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv2f32 = call <vscale x 2 x float> @llvm.vector.splice.left.nxv2f32(<vscale x 2 x float> zeroinitializer, <vscale x 2 x float> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %splice.nxv4f32 = call <vscale x 4 x float> @llvm.vector.splice.left.nxv4f32(<vscale x 4 x float> zeroinitializer, <vscale x 4 x float> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %splice.nxv8f32 = call <vscale x 8 x float> @llvm.vector.splice.left.nxv8f32(<vscale x 8 x float> zeroinitializer, <vscale x 8 x float> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %splice.nxv16f32 = call <vscale x 16 x float> @llvm.vector.splice.left.nxv16f32(<vscale x 16 x float> zeroinitializer, <vscale x 16 x float> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %splice.nxv32f32 = call <vscale x 32 x float> @llvm.vector.splice.left.nxv32f32(<vscale x 32 x float> zeroinitializer, <vscale x 32 x float> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %splice.nxv64f32 = call <vscale x 64 x float> @llvm.vector.splice.left.nxv64f32(<vscale x 64 x float> zeroinitializer, <vscale x 64 x float> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv1f64 = call <vscale x 1 x double> @llvm.vector.splice.left.nxv1f64(<vscale x 1 x double> zeroinitializer, <vscale x 1 x double> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %splice.nxv2f64 = call <vscale x 2 x double> @llvm.vector.splice.left.nxv2f64(<vscale x 2 x double> zeroinitializer, <vscale x 2 x double> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %splice.nxv4f64 = call <vscale x 4 x double> @llvm.vector.splice.left.nxv4f64(<vscale x 4 x double> zeroinitializer, <vscale x 4 x double> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %splice.nxv8f64 = call <vscale x 8 x double> @llvm.vector.splice.left.nxv8f64(<vscale x 8 x double> zeroinitializer, <vscale x 8 x double> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %splice.nxv16f64 = call <vscale x 16 x double> @llvm.vector.splice.left.nxv16f64(<vscale x 16 x double> zeroinitializer, <vscale x 16 x double> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %splice.nxv32f64 = call <vscale x 32 x double> @llvm.vector.splice.left.nxv32f64(<vscale x 32 x double> zeroinitializer, <vscale x 32 x double> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %splice.nxv64f64 = call <vscale x 64 x double> @llvm.vector.splice.left.nxv64f64(<vscale x 64 x double> zeroinitializer, <vscale x 64 x double> zeroinitializer, i32 %offset)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SIZE-LABEL: 'vector_splice_left_variable'
+; SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv1i8 = call <vscale x 1 x i8> @llvm.vector.splice.left.nxv1i8(<vscale x 1 x i8> zeroinitializer, <vscale x 1 x i8> zeroinitializer, i32 %offset)
+; SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv2i8 = call <vscale x 2 x i8> @llvm.vector.splice.left.nxv2i8(<vscale x 2 x i8> zeroinitializer, <vscale x 2 x i8> zeroinitializer, i32 %offset)
+; SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv4i8 = call <vscale x 4 x i8> @llvm.vector.splice.left.nxv4i8(<vscale x 4 x i8> zeroinitializer, <vscale x 4 x i8> zeroinitializer, i32 %offset)
+; SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv8i8 = call <vscale x 8 x i8> @llvm.vector.splice.left.nxv8i8(<vscale x 8 x i8> zeroinitializer, <vscale x 8 x i8> zeroinitializer, i32 %offset)
+; SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splice.nxv16i8 = call <vscale x 16 x i8> @llvm.vector.splice.left.nxv16i8(<vscale x 16 x i8> zeroinitializer, <vscale x 16 x i8> zeroinitializer, i32 %offset)
+; SIZE-NEXT: Cost Model: Found an estimated cost o...
[truncated]
|
🐧 Linux x64 Test Results
✅ The build succeeded and all tests passed. |
🪟 Windows x64 Test Results
✅ The build succeeded and all tests passed. |
Contributor
Author
|
Ping |
rishabhmadan19
pushed a commit
to rishabhmadan19/llvm-project
that referenced
this pull request
Feb 9, 2026
Currently vector splice intrinsics are costed through getShuffleCost when the offset is fixed. When the offset is variable though we can't use a shuffle mask so it currently returns invalid. This implements the cost in RISCVTTIImpl::getIntrinsicInstrCost as the cost of a slideup and a slidedown, which matches the codegen. It also implements the type based cost whenever the offset argument isn't available. It may be possible to reduce the cost in future when one of the vector operands is known to be poison, in which case we only generate a single slideup or slidedown.
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Currently vector splice intrinsics are costed through getShuffleCost when the offset is fixed. When the offset is variable though we can't use a shuffle mask so it currently returns invalid.
This implements the cost in RISCVTTIImpl::getIntrinsicInstrCost as the cost of a slideup and a slidedown, which matches the codegen.
It also implements the type based cost whenever the offset argument isn't available.
It may be possible to reduce the cost in future when one of the vector operands is known to be poison, in which case we only generate a single slideup or slidedown.