[Mlir-commits] [mlir] [mlir][ArmSME] Verify ops on tile types post LLVM conversion (PR #92076)
llvmlistbot at llvm.org
llvmlistbot at llvm.org
Tue May 14 00:32:05 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-mlir
Author: Cullen Rhodes (c-rhodes)
<details>
<summary>Changes</summary>
Unsupported ops on tile types can become dead after
`-convert-arm-sme-to-llvm` resulting in incorrect results. Verify such
operations don't exist post-conversion and fail if they do.
Based on discussion from
https://discourse.llvm.org/t/on-improving-arm-sme-lowering-resilience-in-mlir/78543
---
Patch is 22.45 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/92076.diff
3 Files Affected:
- (modified) mlir/lib/Conversion/ArmSMEToLLVM/ArmSMEToLLVM.cpp (+27-2)
- (modified) mlir/test/Conversion/ArmSMEToLLVM/arm-sme-to-llvm.mlir (+99-52)
- (modified) mlir/test/Conversion/ArmSMEToLLVM/unsupported.mlir (+13)
``````````diff
diff --git a/mlir/lib/Conversion/ArmSMEToLLVM/ArmSMEToLLVM.cpp b/mlir/lib/Conversion/ArmSMEToLLVM/ArmSMEToLLVM.cpp
index 1ba1b88fc1234..607711abbcda4 100644
--- a/mlir/lib/Conversion/ArmSMEToLLVM/ArmSMEToLLVM.cpp
+++ b/mlir/lib/Conversion/ArmSMEToLLVM/ArmSMEToLLVM.cpp
@@ -17,6 +17,7 @@
#include "mlir/Dialect/Arith/IR/Arith.h"
#include "mlir/Dialect/ArmSME/IR/ArmSME.h"
#include "mlir/Dialect/ArmSME/Utils/Utils.h"
+#include "mlir/Dialect/ControlFlow/IR/ControlFlowOps.h"
#include "mlir/Dialect/Func/IR/FuncOps.h"
#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
#include "mlir/Dialect/MemRef/IR/MemRef.h"
@@ -872,9 +873,33 @@ struct ConvertArmSMEToLLVMPass
configureArmSMEToLLVMConversionLegality(target);
populateArmSMEToLLVMConversionPatterns(converter, patterns);
- if (failed(applyPartialConversion(getOperation(), target,
- std::move(patterns))))
+ Operation *moduleOp = getOperation();
+ if (failed(applyPartialConversion(moduleOp, target, std::move(patterns))))
signalPassFailure();
+
+ // Walk the function(s) and fail if there are unexpected operations on SME
+ // tile types after conversion.
+ moduleOp->walk([&](func::FuncOp funcOp) {
+ funcOp->walk([&](Operation *op) {
+ // These ops are legal post conversion, skip these.
+ // Note: 'test.some_use' is a dummy op we use extensively in the tests
+ // to prevent tile ops being DCE'd.
+ if (isa<arm_sme::MaterializeSSATileOp, cf::BranchOp, scf::ForOp,
+ scf::YieldOp>(op) ||
+ op->getName().getStringRef() == "test.some_use")
+ return;
+ auto isSMETileType = [](Type type) -> bool {
+ auto vecType = dyn_cast<VectorType>(type);
+ return vecType && arm_sme::isValidSMETileVectorType(vecType);
+ };
+ if (llvm::any_of(op->getResultTypes(), isSMETileType) ||
+ llvm::any_of(op->getOperandTypes(), isSMETileType)) {
+ op->emitOpError("unexpected operation with SME tile type after "
+ "conversion to LLVM");
+ signalPassFailure();
+ }
+ });
+ });
}
};
diff --git a/mlir/test/Conversion/ArmSMEToLLVM/arm-sme-to-llvm.mlir b/mlir/test/Conversion/ArmSMEToLLVM/arm-sme-to-llvm.mlir
index f48046a8d7995..d6b7e47b2a765 100644
--- a/mlir/test/Conversion/ArmSMEToLLVM/arm-sme-to-llvm.mlir
+++ b/mlir/test/Conversion/ArmSMEToLLVM/arm-sme-to-llvm.mlir
@@ -630,18 +630,18 @@ func.func @arm_sme_streaming_vl_double_words() -> index {
// CHECK-LABEL: arm_sme_fmopa_2way_f16f16_to_f32
// CHECK: "arm_sme.intr.mopa.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xf16>, vector<[8]xf16>) -> ()
-func.func @arm_sme_fmopa_2way_f16f16_to_f32(%vecA: vector<[8]xf16>, %vecB: vector<[8]xf16>) -> vector<[4]x[4]xf32> {
+func.func @arm_sme_fmopa_2way_f16f16_to_f32(%vecA: vector<[8]xf16>, %vecB: vector<[8]xf16>) {
%result = arm_sme.fmopa_2way %vecA, %vecB : vector<[8]xf16>, vector<[8]xf16> into vector<[4]x[4]xf32>
- return %result : vector<[4]x[4]xf32>
+ "test.some_use"(%result) : (vector<[4]x[4]xf32>) -> ()
}
// -----
// CHECK-LABEL: arm_sme_fmopa_2way_bf16bf16_to_f32
// CHECK: "arm_sme.intr.mopa.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xbf16>, vector<[8]xbf16>) -> ()
-func.func @arm_sme_fmopa_2way_bf16bf16_to_f32(%vecA: vector<[8]xbf16>, %vecB: vector<[8]xbf16>) -> vector<[4]x[4]xf32> {
+func.func @arm_sme_fmopa_2way_bf16bf16_to_f32(%vecA: vector<[8]xbf16>, %vecB: vector<[8]xbf16>) {
%result = arm_sme.fmopa_2way %vecA, %vecB : vector<[8]xbf16>, vector<[8]xbf16> into vector<[4]x[4]xf32>
- return %result : vector<[4]x[4]xf32>
+ "test.some_use"(%result) : (vector<[4]x[4]xf32>) -> ()
}
//===----------------------------------------------------------------------===//
@@ -652,18 +652,18 @@ func.func @arm_sme_fmopa_2way_bf16bf16_to_f32(%vecA: vector<[8]xbf16>, %vecB: ve
// CHECK-LABEL: arm_sme_fmops_2way_f16f16_to_f32
// CHECK: "arm_sme.intr.mops.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xf16>, vector<[8]xf16>) -> ()
-func.func @arm_sme_fmops_2way_f16f16_to_f32(%vecA: vector<[8]xf16>, %vecB: vector<[8]xf16>) -> vector<[4]x[4]xf32> {
+func.func @arm_sme_fmops_2way_f16f16_to_f32(%vecA: vector<[8]xf16>, %vecB: vector<[8]xf16>) {
%result = arm_sme.fmops_2way %vecA, %vecB : vector<[8]xf16>, vector<[8]xf16> into vector<[4]x[4]xf32>
- return %result : vector<[4]x[4]xf32>
+ "test.some_use"(%result) : (vector<[4]x[4]xf32>) -> ()
}
// -----
// CHECK-LABEL: arm_sme_fmops_2way_bf16bf16_to_f32
// CHECK: "arm_sme.intr.mops.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xbf16>, vector<[8]xbf16>) -> ()
-func.func @arm_sme_fmops_2way_bf16bf16_to_f32(%vecA: vector<[8]xbf16>, %vecB: vector<[8]xbf16>) -> vector<[4]x[4]xf32> {
+func.func @arm_sme_fmops_2way_bf16bf16_to_f32(%vecA: vector<[8]xbf16>, %vecB: vector<[8]xbf16>) {
%result = arm_sme.fmops_2way %vecA, %vecB : vector<[8]xbf16>, vector<[8]xbf16> into vector<[4]x[4]xf32>
- return %result : vector<[4]x[4]xf32>
+ "test.some_use"(%result) : (vector<[4]x[4]xf32>) -> ()
}
//===----------------------------------------------------------------------===//
@@ -674,9 +674,9 @@ func.func @arm_sme_fmops_2way_bf16bf16_to_f32(%vecA: vector<[8]xbf16>, %vecB: ve
// CHECK-LABEL: arm_sme_smopa_2way_i16i16_to_i32
// CHECK: "arm_sme.intr.smopa.za32"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_smopa_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[4]x[4]xi32> {
+func.func @arm_sme_smopa_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
%result = arm_sme.smopa_2way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[4]x[4]xi32>
- return %result : vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
}
//===----------------------------------------------------------------------===//
@@ -687,9 +687,9 @@ func.func @arm_sme_smopa_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vecto
// CHECK-LABEL: arm_sme_smops_2way_i16i16_to_i32
// CHECK: "arm_sme.intr.smops.za32"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_smops_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[4]x[4]xi32> {
+func.func @arm_sme_smops_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
%result = arm_sme.smops_2way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[4]x[4]xi32>
- return %result : vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
}
//===----------------------------------------------------------------------===//
@@ -700,9 +700,10 @@ func.func @arm_sme_smops_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vecto
// CHECK-LABEL: arm_sme_umopa_2way_i16i16_to_i32
// CHECK: "arm_sme.intr.umopa.za32"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_umopa_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[4]x[4]xi32> {
+func.func @arm_sme_umopa_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
%result = arm_sme.umopa_2way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[4]x[4]xi32>
- return %result : vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
+ return
}
//===----------------------------------------------------------------------===//
@@ -713,9 +714,10 @@ func.func @arm_sme_umopa_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vecto
// CHECK-LABEL: arm_sme_umops_2way_i16i16_to_i32
// CHECK: "arm_sme.intr.umops.za32"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_umops_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[4]x[4]xi32> {
+func.func @arm_sme_umops_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
%result = arm_sme.umops_2way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[4]x[4]xi32>
- return %result : vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
+ return
}
//===----------------------------------------------------------------------===//
@@ -726,18 +728,20 @@ func.func @arm_sme_umops_2way_i16i16_to_i32(%vecA: vector<[8]xi16>, %vecB: vecto
// CHECK-LABEL: arm_sme_smopa_4way_i8i8_to_i32
// CHECK: "arm_sme.intr.smopa.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[16]xi1>, vector<[16]xi1>, vector<[16]xi8>, vector<[16]xi8>) -> ()
-func.func @arm_sme_smopa_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) -> vector<[4]x[4]xi32> {
+func.func @arm_sme_smopa_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) {
%result = arm_sme.smopa_4way %vecA, %vecB : vector<[16]xi8>, vector<[16]xi8> into vector<[4]x[4]xi32>
- return %result : vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
+ return
}
// -----
// CHECK-LABEL: arm_sme_smopa_4way_i16i16_to_i64
// CHECK: "arm_sme.intr.smopa.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_smopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[2]x[2]xi64> {
+func.func @arm_sme_smopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
%result = arm_sme.smopa_4way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[2]x[2]xi64>
- return %result : vector<[2]x[2]xi64>
+ "test.some_use"(%result) : (vector<[2]x[2]xi64>) -> ()
+ return
}
//===----------------------------------------------------------------------===//
@@ -748,18 +752,20 @@ func.func @arm_sme_smopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vecto
// CHECK-LABEL: arm_sme_smops_4way_i8i8_to_i32
// CHECK: "arm_sme.intr.smops.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[16]xi1>, vector<[16]xi1>, vector<[16]xi8>, vector<[16]xi8>) -> ()
-func.func @arm_sme_smops_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) -> vector<[4]x[4]xi32> {
+func.func @arm_sme_smops_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) {
%result = arm_sme.smops_4way %vecA, %vecB : vector<[16]xi8>, vector<[16]xi8> into vector<[4]x[4]xi32>
- return %result : vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
+ return
}
// -----
// CHECK-LABEL: arm_sme_smops_4way_i16i16_to_i64
// CHECK: "arm_sme.intr.smops.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_smops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[2]x[2]xi64> {
+func.func @arm_sme_smops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
%result = arm_sme.smops_4way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[2]x[2]xi64>
- return %result : vector<[2]x[2]xi64>
+ "test.some_use"(%result) : (vector<[2]x[2]xi64>) -> ()
+ return
}
//===----------------------------------------------------------------------===//
@@ -770,18 +776,20 @@ func.func @arm_sme_smops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vecto
// CHECK-LABEL: arm_sme_umopa_4way_i8i8_to_i32
// CHECK: "arm_sme.intr.umopa.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[16]xi1>, vector<[16]xi1>, vector<[16]xi8>, vector<[16]xi8>) -> ()
-func.func @arm_sme_umopa_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) -> vector<[4]x[4]xi32> {
+func.func @arm_sme_umopa_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) {
%result = arm_sme.umopa_4way %vecA, %vecB : vector<[16]xi8>, vector<[16]xi8> into vector<[4]x[4]xi32>
- return %result : vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
+ return
}
// -----
// CHECK-LABEL: arm_sme_umopa_4way_i16i16_to_i64
// CHECK: "arm_sme.intr.umopa.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_umopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[2]x[2]xi64> {
+func.func @arm_sme_umopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
%result = arm_sme.umopa_4way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[2]x[2]xi64>
- return %result : vector<[2]x[2]xi64>
+ "test.some_use"(%result) : (vector<[2]x[2]xi64>) -> ()
+ return
}
//===----------------------------------------------------------------------===//
@@ -792,18 +800,20 @@ func.func @arm_sme_umopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vecto
// CHECK-LABEL: arm_sme_umops_4way_i8i8_to_i32
// CHECK: "arm_sme.intr.umops.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[16]xi1>, vector<[16]xi1>, vector<[16]xi8>, vector<[16]xi8>) -> ()
-func.func @arm_sme_umops_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) -> vector<[4]x[4]xi32> {
+func.func @arm_sme_umops_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) {
%result = arm_sme.umops_4way %vecA, %vecB : vector<[16]xi8>, vector<[16]xi8> into vector<[4]x[4]xi32>
- return %result : vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
+ return
}
// -----
// CHECK-LABEL: arm_sme_umops_4way_i16i16_to_i64
// CHECK: "arm_sme.intr.umops.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_umops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[2]x[2]xi64> {
+func.func @arm_sme_umops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
%result = arm_sme.umops_4way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[2]x[2]xi64>
- return %result : vector<[2]x[2]xi64>
+ "test.some_use"(%result) : (vector<[2]x[2]xi64>) -> ()
+ return
}
//===----------------------------------------------------------------------===//
@@ -814,18 +824,20 @@ func.func @arm_sme_umops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vecto
// CHECK-LABEL: arm_sme_sumopa_4way_i8i8_to_i32
// CHECK: "arm_sme.intr.sumopa.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[16]xi1>, vector<[16]xi1>, vector<[16]xi8>, vector<[16]xi8>) -> ()
-func.func @arm_sme_sumopa_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) -> vector<[4]x[4]xi32> {
+func.func @arm_sme_sumopa_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) {
%result = arm_sme.sumopa_4way %vecA, %vecB : vector<[16]xi8>, vector<[16]xi8> into vector<[4]x[4]xi32>
- return %result : vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
+ return
}
// -----
// CHECK-LABEL: arm_sme_sumopa_4way_i16i16_to_i64
// CHECK: "arm_sme.intr.sumopa.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_sumopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[2]x[2]xi64> {
+func.func @arm_sme_sumopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
%result = arm_sme.sumopa_4way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[2]x[2]xi64>
- return %result : vector<[2]x[2]xi64>
+ "test.some_use"(%result) : (vector<[2]x[2]xi64>) -> ()
+ return
}
//===----------------------------------------------------------------------===//
@@ -836,18 +848,20 @@ func.func @arm_sme_sumopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vect
// CHECK-LABEL: arm_sme_sumops_4way_i8i8_to_i32
// CHECK: "arm_sme.intr.sumops.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[16]xi1>, vector<[16]xi1>, vector<[16]xi8>, vector<[16]xi8>) -> ()
-func.func @arm_sme_sumops_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) -> vector<[4]x[4]xi32> {
+func.func @arm_sme_sumops_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) {
%result = arm_sme.sumops_4way %vecA, %vecB : vector<[16]xi8>, vector<[16]xi8> into vector<[4]x[4]xi32>
- return %result : vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
+ return
}
// -----
// CHECK-LABEL: arm_sme_sumops_4way_i16i16_to_i64
// CHECK: "arm_sme.intr.sumops.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_sumops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[2]x[2]xi64> {
+func.func @arm_sme_sumops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
%result = arm_sme.sumops_4way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[2]x[2]xi64>
- return %result : vector<[2]x[2]xi64>
+ "test.some_use"(%result) : (vector<[2]x[2]xi64>) -> ()
+ return
}
//===----------------------------------------------------------------------===//
@@ -858,18 +872,20 @@ func.func @arm_sme_sumops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vect
// CHECK-LABEL: arm_sme_usmopa_4way_i8i8_to_i32
// CHECK: "arm_sme.intr.usmopa.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[16]xi1>, vector<[16]xi1>, vector<[16]xi8>, vector<[16]xi8>) -> ()
-func.func @arm_sme_usmopa_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) -> vector<[4]x[4]xi32> {
- %reuslt = arm_sme.usmopa_4way %vecA, %vecB : vector<[16]xi8>, vector<[16]xi8> into vector<[4]x[4]xi32>
- return %reuslt : vector<[4]x[4]xi32>
+func.func @arm_sme_usmopa_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) {
+ %result = arm_sme.usmopa_4way %vecA, %vecB : vector<[16]xi8>, vector<[16]xi8> into vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
+ return
}
// -----
// CHECK-LABEL: arm_sme_usmopa_4way_i16i16_to_i64
// CHECK: "arm_sme.intr.usmopa.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_usmopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[2]x[2]xi64> {
- %reuslt = arm_sme.usmopa_4way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[2]x[2]xi64>
- return %reuslt : vector<[2]x[2]xi64>
+func.func @arm_sme_usmopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
+ %result = arm_sme.usmopa_4way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[2]x[2]xi64>
+ "test.some_use"(%result) : (vector<[2]x[2]xi64>) -> ()
+ return
}
//===----------------------------------------------------------------------===//
@@ -880,16 +896,47 @@ func.func @arm_sme_usmopa_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vect
// CHECK-LABEL: arm_sme_usmops_4way_i8i8_to_i32
// CHECK: "arm_sme.intr.usmops.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[16]xi1>, vector<[16]xi1>, vector<[16]xi8>, vector<[16]xi8>) -> ()
-func.func @arm_sme_usmops_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) -> vector<[4]x[4]xi32> {
- %reuslt = arm_sme.usmops_4way %vecA, %vecB : vector<[16]xi8>, vector<[16]xi8> into vector<[4]x[4]xi32>
- return %reuslt : vector<[4]x[4]xi32>
+func.func @arm_sme_usmops_4way_i8i8_to_i32(%vecA: vector<[16]xi8>, %vecB: vector<[16]xi8>) {
+ %result = arm_sme.usmops_4way %vecA, %vecB : vector<[16]xi8>, vector<[16]xi8> into vector<[4]x[4]xi32>
+ "test.some_use"(%result) : (vector<[4]x[4]xi32>) -> ()
+ return
}
// -----
// CHECK-LABEL: arm_sme_usmops_4way_i16i16_to_i64
// CHECK: "arm_sme.intr.usmops.wide"({{.*}}) <{tile_id = 0 : i32}> : (vector<[8]xi1>, vector<[8]xi1>, vector<[8]xi16>, vector<[8]xi16>) -> ()
-func.func @arm_sme_usmops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) -> vector<[2]x[2]xi64> {
- %reuslt = arm_sme.usmops_4way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[2]x[2]xi64>
- return %reuslt : vector<[2]x[2]xi64>
+func.func @arm_sme_usmops_4way_i16i16_to_i64(%vecA: vector<[8]xi16>, %vecB: vector<[8]xi16>) {
+ %result = arm_sme.usmops_4way %vecA, %vecB : vector<[8]xi16>, vector<[8]xi16> into vector<[2]x[2]xi64>
+ "test.some_use"(%result) : (vector<[2]x[2]xi64>) -> ()
+ return
+}
+
+//===---------------------------------------...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/92076
More information about the Mlir-commits
mailing list