[RISCV][TTI] Model the cost of insert/extractelt when the vector split into multiple register group and idx exceed single group. #118401
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
llvmbot
added
backend:RISC-V
llvm:analysis
Member
|
@llvm/pr-subscribers-backend-risc-v @llvm/pr-subscribers-llvm-analysis Author: Elvis Wang (ElvisWang123) ChangesThis patch implements the cost when the size of the vector need to split into multiple groups and the index exceed single vector group. Under this situation, we need the store the entire vector to stack then load the target element. After this patch, the cost of extractelement will close to the generated assembly. Patch is 36.77 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/118401.diff 2 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index 57f635ca6f42a8..20ca80aedab62c 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -1945,6 +1945,23 @@ InstructionCost RISCVTTIImpl::getVectorInstrCost(unsigned Opcode, Type *Val,
// TODO: should we count these special vsetvlis?
BaseCost = Opcode == Instruction::InsertElement ? 3 : 4;
}
+
+ // When the vector need to split into multiple register groups and the index
+ // exceed single vector resgister group, we need to extract the element via
+ // stack.
+ if (Opcode == Instruction::ExtractElement && LT.first > 1 &&
+ ((Index == -1U) || (Index > LT.second.getVectorMinNumElements() &&
+ LT.second.isScalableVector()))) {
+ Type *ScalarType = Val->getScalarType();
+ Align VecAlign = DL.getPrefTypeAlign(Val);
+ Align SclAlign = DL.getPrefTypeAlign(ScalarType);
+ // Store all split vectors into stack and load the target element.
+ return LT.first *
+ getMemoryOpCost(Instruction::Store, Val, VecAlign, 0, CostKind) +
+ getMemoryOpCost(Instruction::Load, ScalarType, SclAlign, 0,
+ CostKind);
+ }
+
return BaseCost + SlideCost;
}
diff --git a/llvm/test/Analysis/CostModel/RISCV/rvv-extractelement.ll b/llvm/test/Analysis/CostModel/RISCV/rvv-extractelement.ll
index 618b7bc8945a50..34a323066689ba 100644
--- a/llvm/test/Analysis/CostModel/RISCV/rvv-extractelement.ll
+++ b/llvm/test/Analysis/CostModel/RISCV/rvv-extractelement.ll
@@ -139,7 +139,7 @@ define void @extractelement_int(i32 %x) {
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i8_x = extractelement <vscale x 16 x i8> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i8_x = extractelement <vscale x 32 x i8> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv64i8_x = extractelement <vscale x 64 x i8> undef, i32 %x
-; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv128i8_x = extractelement <vscale x 128 x i8> undef, i32 %x
+; RV32V-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv128i8_x = extractelement <vscale x 128 x i8> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i16_x = extractelement <2 x i16> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i16_x = extractelement <4 x i16> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i16_x = extractelement <8 x i16> undef, i32 %x
@@ -151,7 +151,7 @@ define void @extractelement_int(i32 %x) {
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv8i16_x = extractelement <vscale x 8 x i16> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i16_x = extractelement <vscale x 16 x i16> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i16_x = extractelement <vscale x 32 x i16> undef, i32 %x
-; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv64i16_x = extractelement <vscale x 64 x i16> undef, i32 %x
+; RV32V-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv64i16_x = extractelement <vscale x 64 x i16> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i32_x = extractelement <2 x i32> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i32_x = extractelement <4 x i32> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i32_x = extractelement <8 x i32> undef, i32 %x
@@ -161,7 +161,7 @@ define void @extractelement_int(i32 %x) {
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv4i32_x = extractelement <vscale x 4 x i32> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv8i32_x = extractelement <vscale x 8 x i32> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i32_x = extractelement <vscale x 16 x i32> undef, i32 %x
-; RV32V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i32_x = extractelement <vscale x 32 x i32> undef, i32 %x
+; RV32V-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv32i32_x = extractelement <vscale x 32 x i32> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i64_x = extractelement <2 x i64> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i64_x = extractelement <4 x i64> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i64_x = extractelement <8 x i64> undef, i32 %x
@@ -169,7 +169,7 @@ define void @extractelement_int(i32 %x) {
; RV32V-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %nxv2i64_x = extractelement <vscale x 2 x i64> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %nxv4i64_x = extractelement <vscale x 4 x i64> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %nxv8i64_x = extractelement <vscale x 8 x i64> undef, i32 %x
-; RV32V-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %nxv16i64_x = extractelement <vscale x 16 x i64> undef, i32 %x
+; RV32V-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %nxv16i64_x = extractelement <vscale x 16 x i64> undef, i32 %x
; RV32V-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; RV64V-LABEL: 'extractelement_int'
@@ -304,7 +304,7 @@ define void @extractelement_int(i32 %x) {
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i8_x = extractelement <vscale x 16 x i8> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i8_x = extractelement <vscale x 32 x i8> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv64i8_x = extractelement <vscale x 64 x i8> undef, i32 %x
-; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv128i8_x = extractelement <vscale x 128 x i8> undef, i32 %x
+; RV64V-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv128i8_x = extractelement <vscale x 128 x i8> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i16_x = extractelement <2 x i16> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i16_x = extractelement <4 x i16> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i16_x = extractelement <8 x i16> undef, i32 %x
@@ -316,7 +316,7 @@ define void @extractelement_int(i32 %x) {
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv8i16_x = extractelement <vscale x 8 x i16> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i16_x = extractelement <vscale x 16 x i16> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i16_x = extractelement <vscale x 32 x i16> undef, i32 %x
-; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv64i16_x = extractelement <vscale x 64 x i16> undef, i32 %x
+; RV64V-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv64i16_x = extractelement <vscale x 64 x i16> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i32_x = extractelement <2 x i32> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i32_x = extractelement <4 x i32> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i32_x = extractelement <8 x i32> undef, i32 %x
@@ -326,7 +326,7 @@ define void @extractelement_int(i32 %x) {
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv4i32_x = extractelement <vscale x 4 x i32> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv8i32_x = extractelement <vscale x 8 x i32> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i32_x = extractelement <vscale x 16 x i32> undef, i32 %x
-; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i32_x = extractelement <vscale x 32 x i32> undef, i32 %x
+; RV64V-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv32i32_x = extractelement <vscale x 32 x i32> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i64_x = extractelement <2 x i64> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i64_x = extractelement <4 x i64> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i64_x = extractelement <8 x i64> undef, i32 %x
@@ -334,7 +334,7 @@ define void @extractelement_int(i32 %x) {
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv2i64_x = extractelement <vscale x 2 x i64> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv4i64_x = extractelement <vscale x 4 x i64> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv8i64_x = extractelement <vscale x 8 x i64> undef, i32 %x
-; RV64V-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i64_x = extractelement <vscale x 16 x i64> undef, i32 %x
+; RV64V-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv16i64_x = extractelement <vscale x 16 x i64> undef, i32 %x
; RV64V-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; RV32ZVE64X-LABEL: 'extractelement_int'
@@ -462,44 +462,44 @@ define void @extractelement_int(i32 %x) {
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i8_x = extractelement <16 x i8> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v32i8_x = extractelement <32 x i8> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v64i8_x = extractelement <64 x i8> undef, i32 %x
-; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v128i8_x = extractelement <128 x i8> undef, i32 %x
+; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v128i8_x = extractelement <128 x i8> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv2i8_x = extractelement <vscale x 2 x i8> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv4i8_x = extractelement <vscale x 4 x i8> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv8i8_x = extractelement <vscale x 8 x i8> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i8_x = extractelement <vscale x 16 x i8> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i8_x = extractelement <vscale x 32 x i8> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv64i8_x = extractelement <vscale x 64 x i8> undef, i32 %x
-; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv128i8_x = extractelement <vscale x 128 x i8> undef, i32 %x
+; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv128i8_x = extractelement <vscale x 128 x i8> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i16_x = extractelement <2 x i16> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i16_x = extractelement <4 x i16> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i16_x = extractelement <8 x i16> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i16_x = extractelement <16 x i16> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v32i16_x = extractelement <32 x i16> undef, i32 %x
-; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v64i16_x = extractelement <64 x i16> undef, i32 %x
+; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v64i16_x = extractelement <64 x i16> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv2i16_x = extractelement <vscale x 2 x i16> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv4i16_x = extractelement <vscale x 4 x i16> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv8i16_x = extractelement <vscale x 8 x i16> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i16_x = extractelement <vscale x 16 x i16> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i16_x = extractelement <vscale x 32 x i16> undef, i32 %x
-; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv64i16_x = extractelement <vscale x 64 x i16> undef, i32 %x
+; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv64i16_x = extractelement <vscale x 64 x i16> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i32_x = extractelement <2 x i32> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i32_x = extractelement <4 x i32> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i32_x = extractelement <8 x i32> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i32_x = extractelement <16 x i32> undef, i32 %x
-; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v32i32_x = extractelement <32 x i32> undef, i32 %x
+; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v32i32_x = extractelement <32 x i32> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv2i32_x = extractelement <vscale x 2 x i32> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv4i32_x = extractelement <vscale x 4 x i32> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv8i32_x = extractelement <vscale x 8 x i32> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i32_x = extractelement <vscale x 16 x i32> undef, i32 %x
-; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i32_x = extractelement <vscale x 32 x i32> undef, i32 %x
+; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv32i32_x = extractelement <vscale x 32 x i32> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i64_x = extractelement <2 x i64> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i64_x = extractelement <4 x i64> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i64_x = extractelement <8 x i64> undef, i32 %x
-; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i64_x = extractelement <16 x i64> undef, i32 %x
+; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %v16i64_x = extractelement <16 x i64> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %nxv2i64_x = extractelement <vscale x 2 x i64> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %nxv4i64_x = extractelement <vscale x 4 x i64> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %nxv8i64_x = extractelement <vscale x 8 x i64> undef, i32 %x
-; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %nxv16i64_x = extractelement <vscale x 16 x i64> undef, i32 %x
+; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %nxv16i64_x = extractelement <vscale x 16 x i64> undef, i32 %x
; RV32ZVE64X-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; RV64ZVE64X-LABEL: 'extractelement_int'
@@ -627,44 +627,44 @@ define void @extractelement_int(i32 %x) {
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i8_x = extractelement <16 x i8> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v32i8_x = extractelement <32 x i8> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v64i8_x = extractelement <64 x i8> undef, i32 %x
-; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v128i8_x = extractelement <128 x i8> undef, i32 %x
+; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v128i8_x = extractelement <128 x i8> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv2i8_x = extractelement <vscale x 2 x i8> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv4i8_x = extractelement <vscale x 4 x i8> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv8i8_x = extractelement <vscale x 8 x i8> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i8_x = extractelement <vscale x 16 x i8> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i8_x = extractelement <vscale x 32 x i8> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv64i8_x = extractelement <vscale x 64 x i8> undef, i32 %x
-; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv128i8_x = extractelement <vscale x 128 x i8> undef, i32 %x
+; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %nxv128i8_x = extractelement <vscale x 128 x i8> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i16_x = extractelement <2 x i16> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i16_x = extractelement <4 x i16> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i16_x = extractelement <8 x i16> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i16_x = extractelement <16 x i16> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v32i16_x = extractelement <32 x i16> undef, i32 %x
-; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v64i16_x = extractelement <64 x i16> undef, i32 %x
+; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v64i16_x = extractelement <64 x i16> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv2i16_x = extractelement <vscale x 2 x i16> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv4i16_x = extractelement <vscale x 4 x i16> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv8i16_x = extractelement <vscale x 8 x i16> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv16i16_x = extractelement <vscale x 16 x i16> undef, i32 %x
; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv32i16_x = extractelement <vscale x 32 x i16> undef, i32 %x
-; RV64ZVE64X-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %nxv64i16_x = extracte...
[truncated]
|
topperc
reviewed
Dec 2, 2024
ElvisWang123
changed the title
Contributor
Author
|
CodeGen for extract/insert element exceed single register group. https://godbolt.org/z/r1zKEoK1P |
preames
requested changes
Dec 10, 2024
Collaborator
There was a problem hiding this comment.
You shouldn't need to multiply by LT.first here. getMemoryOpCost does that internally. Or, at least, it should. If it's not doing so in the split case, we should fix that in getMemoryOpCost instead.
Contributor
Author
There was a problem hiding this comment.
Removed, thanks!
Contributor
Author
There was a problem hiding this comment.
Removed, thanks!
Collaborator
There was a problem hiding this comment.
You're missing the addressing cost in both cases here. For the vector, that should be handled inside getMemoryOpCost, but you need to include the ADDI for the non-constant index case on the scalar load or store.
Contributor
Author
There was a problem hiding this comment.
Added, thanks!
Contributor
Author
There was a problem hiding this comment.
Removed, thanks.
Collaborator
There was a problem hiding this comment.
This cost seems way too big.
256 x i8 on V is 16 registers worth of data. Storing that should be ~16 in cost. A single scalar load and addressing is ~2-3. So, I'd expect something in the order of ~20, not ~129 here.
Contributor
Author
There was a problem hiding this comment.
I think 256 x i8 need 32 register under zve64x. 256 / 64 * 8 = 32
The cost of 128 x i8 is 18 and that is close to your expectation.
Collaborator
Thank you for including this. It made spot checking the codegen much simpler. |
…vslide. This patch implement the cost when the size of the vector need to split into multiple groups and the index exceed single vector group. Under this situation, we need the store the vector to stack and load the target element. After this patch, the cost of extract element will close to the generated assembly.
ElvisWang123
force-pushed
the
upstream-getVectorInstrCost-new
branch
from
c41b47e to
fe662e0
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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.
This patch implements the cost when the size of the vector need to split into multiple groups and the index exceed single vector group.
For extract element, we need to store split vectors to stack and load the target element.
For insert element, we need to store split vectors to stack and store the target element and load vectors back.
After this patch, the cost of insert/extract element will close to the generated assembly.