- build:
- unify schema building in core and converter;
- add more build script for android;
- add linux build script for python;
- ops impl:
- add floor mod support in binary;
- use eltwise impl in add/max/sub/mul binary for optimization;
- remove fake double support in cast;
- fix 5d support for concat;
- add adjX and adjY support for batch matmul;
- optimize conv2d back prop filter;
- add pad mode support for conv3d;
- fix bug in conv2d & conv depthwise with very small feature map;
- optimize binary without broacast;
- add data types support for gather;
- add gather ND support;
- use uint8 data type in gather v2;
- add transpose support for matmul;
- add matrix band part;
- add dim != 4 support for padding, reshape & tensor convert;
- add pad type support for pool3d;
- make ops based on TensorFlow Lite quantization optional;
- add all & any support for reduction;
- use type in parameter as output type in reduction;
- add int support for unary;
- add variable weight support for conv2d;
- fix conv2d depthwise weights initialization;
- fix type support for transpose;
- fix grad outputs count for reduce grad and reshape grad;
- fix priorbox & detection output;
- fix metal softmax error;
- python:
- add runSessionWithCallBackInfo interface;
- add max nodes limit (1400) for visualization tool;
- fix save error in python3;
- align default dim;
- convert:
- add extra design for optimization;
- add more post converting optimizers;
- add caffe v1 weights blob support;
- add cast, unary, conv transpose support for onnx model;
- optimize batchnorm, conv with variable weights, prelu, reshape, slice, upsample for onnx model;
- add cos/sin/atan/tan support for unary for tensorflow model;
- add any/all support for reduction for tensorflow model;
- add elu, conv3d, pool3d support for tensorflow model;
- optimize argmax, batchnorm, concat, batch to space, conv with variable weights, prelu, slice for tensorflow model;
- others:
- fix size computer lock;
- fix thread pool deadlock;
- add express & parameters in express;
- rewrite blitter chooser without static map;
- add tests for expr;
2019-10-29 13:37:26 +08:00
|
|
|
//
|
|
|
|
// MatMulTest.cpp
|
|
|
|
// MNNTests
|
|
|
|
//
|
|
|
|
// Created by MNN on 2019/09/17.
|
|
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
|
|
//
|
|
|
|
|
|
|
|
#include <random>
|
|
|
|
#include <math.h>
|
2019-12-27 22:16:57 +08:00
|
|
|
#include <MNN/expr/ExprCreator.hpp>
|
- build:
- unify schema building in core and converter;
- add more build script for android;
- add linux build script for python;
- ops impl:
- add floor mod support in binary;
- use eltwise impl in add/max/sub/mul binary for optimization;
- remove fake double support in cast;
- fix 5d support for concat;
- add adjX and adjY support for batch matmul;
- optimize conv2d back prop filter;
- add pad mode support for conv3d;
- fix bug in conv2d & conv depthwise with very small feature map;
- optimize binary without broacast;
- add data types support for gather;
- add gather ND support;
- use uint8 data type in gather v2;
- add transpose support for matmul;
- add matrix band part;
- add dim != 4 support for padding, reshape & tensor convert;
- add pad type support for pool3d;
- make ops based on TensorFlow Lite quantization optional;
- add all & any support for reduction;
- use type in parameter as output type in reduction;
- add int support for unary;
- add variable weight support for conv2d;
- fix conv2d depthwise weights initialization;
- fix type support for transpose;
- fix grad outputs count for reduce grad and reshape grad;
- fix priorbox & detection output;
- fix metal softmax error;
- python:
- add runSessionWithCallBackInfo interface;
- add max nodes limit (1400) for visualization tool;
- fix save error in python3;
- align default dim;
- convert:
- add extra design for optimization;
- add more post converting optimizers;
- add caffe v1 weights blob support;
- add cast, unary, conv transpose support for onnx model;
- optimize batchnorm, conv with variable weights, prelu, reshape, slice, upsample for onnx model;
- add cos/sin/atan/tan support for unary for tensorflow model;
- add any/all support for reduction for tensorflow model;
- add elu, conv3d, pool3d support for tensorflow model;
- optimize argmax, batchnorm, concat, batch to space, conv with variable weights, prelu, slice for tensorflow model;
- others:
- fix size computer lock;
- fix thread pool deadlock;
- add express & parameters in express;
- rewrite blitter chooser without static map;
- add tests for expr;
2019-10-29 13:37:26 +08:00
|
|
|
#include "MNNTestSuite.h"
|
|
|
|
#include "MNN_generated.h"
|
|
|
|
using namespace MNN::Express;
|
|
|
|
|
|
|
|
static void fillFloat(float* dst, int h, int w, float offset = 0.0f) {
|
|
|
|
for (int y=0; y<h; ++y) {
|
|
|
|
auto dstY = dst + w*y;
|
|
|
|
for (int x=0; x<w; ++x) {
|
|
|
|
dstY[x] = (float)x * 0.1f + (float)y + offset;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool checkMatMul(const float* C, const float* A, const float* B, int e, int l, int h) {
|
2020-03-07 23:01:34 +08:00
|
|
|
bool res = true;
|
- build:
- unify schema building in core and converter;
- add more build script for android;
- add linux build script for python;
- ops impl:
- add floor mod support in binary;
- use eltwise impl in add/max/sub/mul binary for optimization;
- remove fake double support in cast;
- fix 5d support for concat;
- add adjX and adjY support for batch matmul;
- optimize conv2d back prop filter;
- add pad mode support for conv3d;
- fix bug in conv2d & conv depthwise with very small feature map;
- optimize binary without broacast;
- add data types support for gather;
- add gather ND support;
- use uint8 data type in gather v2;
- add transpose support for matmul;
- add matrix band part;
- add dim != 4 support for padding, reshape & tensor convert;
- add pad type support for pool3d;
- make ops based on TensorFlow Lite quantization optional;
- add all & any support for reduction;
- use type in parameter as output type in reduction;
- add int support for unary;
- add variable weight support for conv2d;
- fix conv2d depthwise weights initialization;
- fix type support for transpose;
- fix grad outputs count for reduce grad and reshape grad;
- fix priorbox & detection output;
- fix metal softmax error;
- python:
- add runSessionWithCallBackInfo interface;
- add max nodes limit (1400) for visualization tool;
- fix save error in python3;
- align default dim;
- convert:
- add extra design for optimization;
- add more post converting optimizers;
- add caffe v1 weights blob support;
- add cast, unary, conv transpose support for onnx model;
- optimize batchnorm, conv with variable weights, prelu, reshape, slice, upsample for onnx model;
- add cos/sin/atan/tan support for unary for tensorflow model;
- add any/all support for reduction for tensorflow model;
- add elu, conv3d, pool3d support for tensorflow model;
- optimize argmax, batchnorm, concat, batch to space, conv with variable weights, prelu, slice for tensorflow model;
- others:
- fix size computer lock;
- fix thread pool deadlock;
- add express & parameters in express;
- rewrite blitter chooser without static map;
- add tests for expr;
2019-10-29 13:37:26 +08:00
|
|
|
for (int y=0; y<h; ++y) {
|
|
|
|
auto AY = A + l*y;
|
|
|
|
auto CY = C + e*y;
|
|
|
|
for (int x=0; x<e; ++x) {
|
|
|
|
auto BX = B + x;
|
|
|
|
float expected = 0.0f;
|
|
|
|
auto computed = CY[x];
|
|
|
|
for (int k=0; k<l; ++k) {
|
|
|
|
expected += AY[k] * BX[k*e];
|
|
|
|
}
|
|
|
|
auto diff = fabsf(expected-computed);
|
2020-03-07 23:01:34 +08:00
|
|
|
if (diff > 0.1f) {
|
2019-12-27 22:16:57 +08:00
|
|
|
MNN_PRINT("%f -> %f\n", expected, computed);
|
2020-03-07 23:01:34 +08:00
|
|
|
res = false;
|
- build:
- unify schema building in core and converter;
- add more build script for android;
- add linux build script for python;
- ops impl:
- add floor mod support in binary;
- use eltwise impl in add/max/sub/mul binary for optimization;
- remove fake double support in cast;
- fix 5d support for concat;
- add adjX and adjY support for batch matmul;
- optimize conv2d back prop filter;
- add pad mode support for conv3d;
- fix bug in conv2d & conv depthwise with very small feature map;
- optimize binary without broacast;
- add data types support for gather;
- add gather ND support;
- use uint8 data type in gather v2;
- add transpose support for matmul;
- add matrix band part;
- add dim != 4 support for padding, reshape & tensor convert;
- add pad type support for pool3d;
- make ops based on TensorFlow Lite quantization optional;
- add all & any support for reduction;
- use type in parameter as output type in reduction;
- add int support for unary;
- add variable weight support for conv2d;
- fix conv2d depthwise weights initialization;
- fix type support for transpose;
- fix grad outputs count for reduce grad and reshape grad;
- fix priorbox & detection output;
- fix metal softmax error;
- python:
- add runSessionWithCallBackInfo interface;
- add max nodes limit (1400) for visualization tool;
- fix save error in python3;
- align default dim;
- convert:
- add extra design for optimization;
- add more post converting optimizers;
- add caffe v1 weights blob support;
- add cast, unary, conv transpose support for onnx model;
- optimize batchnorm, conv with variable weights, prelu, reshape, slice, upsample for onnx model;
- add cos/sin/atan/tan support for unary for tensorflow model;
- add any/all support for reduction for tensorflow model;
- add elu, conv3d, pool3d support for tensorflow model;
- optimize argmax, batchnorm, concat, batch to space, conv with variable weights, prelu, slice for tensorflow model;
- others:
- fix size computer lock;
- fix thread pool deadlock;
- add express & parameters in express;
- rewrite blitter chooser without static map;
- add tests for expr;
2019-10-29 13:37:26 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2020-03-07 23:01:34 +08:00
|
|
|
return res;
|
- build:
- unify schema building in core and converter;
- add more build script for android;
- add linux build script for python;
- ops impl:
- add floor mod support in binary;
- use eltwise impl in add/max/sub/mul binary for optimization;
- remove fake double support in cast;
- fix 5d support for concat;
- add adjX and adjY support for batch matmul;
- optimize conv2d back prop filter;
- add pad mode support for conv3d;
- fix bug in conv2d & conv depthwise with very small feature map;
- optimize binary without broacast;
- add data types support for gather;
- add gather ND support;
- use uint8 data type in gather v2;
- add transpose support for matmul;
- add matrix band part;
- add dim != 4 support for padding, reshape & tensor convert;
- add pad type support for pool3d;
- make ops based on TensorFlow Lite quantization optional;
- add all & any support for reduction;
- use type in parameter as output type in reduction;
- add int support for unary;
- add variable weight support for conv2d;
- fix conv2d depthwise weights initialization;
- fix type support for transpose;
- fix grad outputs count for reduce grad and reshape grad;
- fix priorbox & detection output;
- fix metal softmax error;
- python:
- add runSessionWithCallBackInfo interface;
- add max nodes limit (1400) for visualization tool;
- fix save error in python3;
- align default dim;
- convert:
- add extra design for optimization;
- add more post converting optimizers;
- add caffe v1 weights blob support;
- add cast, unary, conv transpose support for onnx model;
- optimize batchnorm, conv with variable weights, prelu, reshape, slice, upsample for onnx model;
- add cos/sin/atan/tan support for unary for tensorflow model;
- add any/all support for reduction for tensorflow model;
- add elu, conv3d, pool3d support for tensorflow model;
- optimize argmax, batchnorm, concat, batch to space, conv with variable weights, prelu, slice for tensorflow model;
- others:
- fix size computer lock;
- fix thread pool deadlock;
- add express & parameters in express;
- rewrite blitter chooser without static map;
- add tests for expr;
2019-10-29 13:37:26 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
class MatMulTest : public MNNTestCase {
|
|
|
|
public:
|
|
|
|
virtual bool run() {
|
|
|
|
int e=5, h=4, l=6;
|
|
|
|
{
|
|
|
|
//Test MatMul
|
|
|
|
std::unique_ptr<MNN::OpT> op(new MNN::OpT);
|
|
|
|
op->type = MNN::OpType_MatMul;
|
|
|
|
op->main.type = MNN::OpParameter_MatMul;
|
|
|
|
op->main.value = new MNN::MatMulT;
|
|
|
|
auto matmulParam = op->main.AsMatMul();
|
|
|
|
matmulParam->transposeA = false;
|
|
|
|
matmulParam->transposeB = false;
|
2019-12-27 22:16:57 +08:00
|
|
|
|
- build:
- unify schema building in core and converter;
- add more build script for android;
- add linux build script for python;
- ops impl:
- add floor mod support in binary;
- use eltwise impl in add/max/sub/mul binary for optimization;
- remove fake double support in cast;
- fix 5d support for concat;
- add adjX and adjY support for batch matmul;
- optimize conv2d back prop filter;
- add pad mode support for conv3d;
- fix bug in conv2d & conv depthwise with very small feature map;
- optimize binary without broacast;
- add data types support for gather;
- add gather ND support;
- use uint8 data type in gather v2;
- add transpose support for matmul;
- add matrix band part;
- add dim != 4 support for padding, reshape & tensor convert;
- add pad type support for pool3d;
- make ops based on TensorFlow Lite quantization optional;
- add all & any support for reduction;
- use type in parameter as output type in reduction;
- add int support for unary;
- add variable weight support for conv2d;
- fix conv2d depthwise weights initialization;
- fix type support for transpose;
- fix grad outputs count for reduce grad and reshape grad;
- fix priorbox & detection output;
- fix metal softmax error;
- python:
- add runSessionWithCallBackInfo interface;
- add max nodes limit (1400) for visualization tool;
- fix save error in python3;
- align default dim;
- convert:
- add extra design for optimization;
- add more post converting optimizers;
- add caffe v1 weights blob support;
- add cast, unary, conv transpose support for onnx model;
- optimize batchnorm, conv with variable weights, prelu, reshape, slice, upsample for onnx model;
- add cos/sin/atan/tan support for unary for tensorflow model;
- add any/all support for reduction for tensorflow model;
- add elu, conv3d, pool3d support for tensorflow model;
- optimize argmax, batchnorm, concat, batch to space, conv with variable weights, prelu, slice for tensorflow model;
- others:
- fix size computer lock;
- fix thread pool deadlock;
- add express & parameters in express;
- rewrite blitter chooser without static map;
- add tests for expr;
2019-10-29 13:37:26 +08:00
|
|
|
auto x0 = _Input({}, NHWC, halide_type_of<float>());
|
|
|
|
auto x1 = _Input({}, NHWC, halide_type_of<float>());
|
|
|
|
auto y = Variable::create(Expr::create(op.get(), {x0, x1}));
|
|
|
|
x0->resize({h, l});
|
|
|
|
x1->resize({l, e});
|
|
|
|
fillFloat(x0->writeMap<float>(), h, l);
|
|
|
|
fillFloat(x1->writeMap<float>(), l, e);
|
2019-12-27 22:16:57 +08:00
|
|
|
|
- build:
- unify schema building in core and converter;
- add more build script for android;
- add linux build script for python;
- ops impl:
- add floor mod support in binary;
- use eltwise impl in add/max/sub/mul binary for optimization;
- remove fake double support in cast;
- fix 5d support for concat;
- add adjX and adjY support for batch matmul;
- optimize conv2d back prop filter;
- add pad mode support for conv3d;
- fix bug in conv2d & conv depthwise with very small feature map;
- optimize binary without broacast;
- add data types support for gather;
- add gather ND support;
- use uint8 data type in gather v2;
- add transpose support for matmul;
- add matrix band part;
- add dim != 4 support for padding, reshape & tensor convert;
- add pad type support for pool3d;
- make ops based on TensorFlow Lite quantization optional;
- add all & any support for reduction;
- use type in parameter as output type in reduction;
- add int support for unary;
- add variable weight support for conv2d;
- fix conv2d depthwise weights initialization;
- fix type support for transpose;
- fix grad outputs count for reduce grad and reshape grad;
- fix priorbox & detection output;
- fix metal softmax error;
- python:
- add runSessionWithCallBackInfo interface;
- add max nodes limit (1400) for visualization tool;
- fix save error in python3;
- align default dim;
- convert:
- add extra design for optimization;
- add more post converting optimizers;
- add caffe v1 weights blob support;
- add cast, unary, conv transpose support for onnx model;
- optimize batchnorm, conv with variable weights, prelu, reshape, slice, upsample for onnx model;
- add cos/sin/atan/tan support for unary for tensorflow model;
- add any/all support for reduction for tensorflow model;
- add elu, conv3d, pool3d support for tensorflow model;
- optimize argmax, batchnorm, concat, batch to space, conv with variable weights, prelu, slice for tensorflow model;
- others:
- fix size computer lock;
- fix thread pool deadlock;
- add express & parameters in express;
- rewrite blitter chooser without static map;
- add tests for expr;
2019-10-29 13:37:26 +08:00
|
|
|
auto res = checkMatMul(y->readMap<float>(), x0->readMap<float>(), x1->readMap<float>(), e, l, h);
|
|
|
|
if (!res) {
|
|
|
|
FUNC_PRINT(1);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
auto tranposeA = _Transpose(x0, {1, 0});
|
|
|
|
matmulParam->transposeA = true;
|
|
|
|
matmulParam->transposeB = false;
|
|
|
|
y = Variable::create(Expr::create(op.get(), {tranposeA, x1}));
|
|
|
|
res = checkMatMul(y->readMap<float>(), x0->readMap<float>(), x1->readMap<float>(), e, l, h);
|
|
|
|
if (!res) {
|
|
|
|
FUNC_PRINT(1);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
auto tranposeB = _Transpose(x1, {1, 0});
|
|
|
|
matmulParam->transposeA = true;
|
|
|
|
matmulParam->transposeB = true;
|
|
|
|
y = Variable::create(Expr::create(op.get(), {tranposeA, tranposeB}));
|
|
|
|
res = checkMatMul(y->readMap<float>(), x0->readMap<float>(), x1->readMap<float>(), e, l, h);
|
|
|
|
if (!res) {
|
|
|
|
FUNC_PRINT(1);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
matmulParam->transposeA = false;
|
|
|
|
matmulParam->transposeB = true;
|
|
|
|
y = Variable::create(Expr::create(op.get(), {x0, tranposeB}));
|
|
|
|
res = checkMatMul(y->readMap<float>(), x0->readMap<float>(), x1->readMap<float>(), e, l, h);
|
|
|
|
if (!res) {
|
|
|
|
FUNC_PRINT(1);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
{
|
|
|
|
std::unique_ptr<MNN::OpT> op(new MNN::OpT);
|
|
|
|
op->type = MNN::OpType_BatchMatMul;
|
|
|
|
op->main.type = MNN::OpParameter_BatchMatMulParam;
|
|
|
|
op->main.value = new MNN::BatchMatMulParamT;
|
|
|
|
auto param = op->main.AsBatchMatMulParam();
|
|
|
|
param->adjX = false;
|
|
|
|
param->adjY = false;
|
2019-12-27 22:16:57 +08:00
|
|
|
|
- build:
- unify schema building in core and converter;
- add more build script for android;
- add linux build script for python;
- ops impl:
- add floor mod support in binary;
- use eltwise impl in add/max/sub/mul binary for optimization;
- remove fake double support in cast;
- fix 5d support for concat;
- add adjX and adjY support for batch matmul;
- optimize conv2d back prop filter;
- add pad mode support for conv3d;
- fix bug in conv2d & conv depthwise with very small feature map;
- optimize binary without broacast;
- add data types support for gather;
- add gather ND support;
- use uint8 data type in gather v2;
- add transpose support for matmul;
- add matrix band part;
- add dim != 4 support for padding, reshape & tensor convert;
- add pad type support for pool3d;
- make ops based on TensorFlow Lite quantization optional;
- add all & any support for reduction;
- use type in parameter as output type in reduction;
- add int support for unary;
- add variable weight support for conv2d;
- fix conv2d depthwise weights initialization;
- fix type support for transpose;
- fix grad outputs count for reduce grad and reshape grad;
- fix priorbox & detection output;
- fix metal softmax error;
- python:
- add runSessionWithCallBackInfo interface;
- add max nodes limit (1400) for visualization tool;
- fix save error in python3;
- align default dim;
- convert:
- add extra design for optimization;
- add more post converting optimizers;
- add caffe v1 weights blob support;
- add cast, unary, conv transpose support for onnx model;
- optimize batchnorm, conv with variable weights, prelu, reshape, slice, upsample for onnx model;
- add cos/sin/atan/tan support for unary for tensorflow model;
- add any/all support for reduction for tensorflow model;
- add elu, conv3d, pool3d support for tensorflow model;
- optimize argmax, batchnorm, concat, batch to space, conv with variable weights, prelu, slice for tensorflow model;
- others:
- fix size computer lock;
- fix thread pool deadlock;
- add express & parameters in express;
- rewrite blitter chooser without static map;
- add tests for expr;
2019-10-29 13:37:26 +08:00
|
|
|
int batch = 5;
|
|
|
|
auto x0 = _Input({}, NHWC, halide_type_of<float>());
|
|
|
|
auto x1 = _Input({}, NHWC, halide_type_of<float>());
|
|
|
|
x0->resize({5, h, l});
|
|
|
|
x1->resize({5, l, e});
|
|
|
|
auto x0Ptr = x0->writeMap<float>();
|
|
|
|
auto x1Ptr = x1->writeMap<float>();
|
|
|
|
for (int b=0; b<batch; ++b) {
|
|
|
|
fillFloat(x0Ptr + b*h*l, h, l, (float)b*10);
|
|
|
|
fillFloat(x1Ptr + b*e*l, l, e, (float)b*10);
|
|
|
|
}
|
|
|
|
auto y = Variable::create(Expr::create(op.get(), {x0, x1}));
|
|
|
|
auto yPtr = y->readMap<float>();
|
|
|
|
for (int b=0; b<batch; ++b) {
|
|
|
|
auto res = checkMatMul(yPtr+b*e*h, x0Ptr + b*h*l, x1Ptr + b*e*l, e, l, h);
|
|
|
|
if (!res) {
|
|
|
|
FUNC_PRINT(1);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2019-12-27 22:16:57 +08:00
|
|
|
|
- build:
- unify schema building in core and converter;
- add more build script for android;
- add linux build script for python;
- ops impl:
- add floor mod support in binary;
- use eltwise impl in add/max/sub/mul binary for optimization;
- remove fake double support in cast;
- fix 5d support for concat;
- add adjX and adjY support for batch matmul;
- optimize conv2d back prop filter;
- add pad mode support for conv3d;
- fix bug in conv2d & conv depthwise with very small feature map;
- optimize binary without broacast;
- add data types support for gather;
- add gather ND support;
- use uint8 data type in gather v2;
- add transpose support for matmul;
- add matrix band part;
- add dim != 4 support for padding, reshape & tensor convert;
- add pad type support for pool3d;
- make ops based on TensorFlow Lite quantization optional;
- add all & any support for reduction;
- use type in parameter as output type in reduction;
- add int support for unary;
- add variable weight support for conv2d;
- fix conv2d depthwise weights initialization;
- fix type support for transpose;
- fix grad outputs count for reduce grad and reshape grad;
- fix priorbox & detection output;
- fix metal softmax error;
- python:
- add runSessionWithCallBackInfo interface;
- add max nodes limit (1400) for visualization tool;
- fix save error in python3;
- align default dim;
- convert:
- add extra design for optimization;
- add more post converting optimizers;
- add caffe v1 weights blob support;
- add cast, unary, conv transpose support for onnx model;
- optimize batchnorm, conv with variable weights, prelu, reshape, slice, upsample for onnx model;
- add cos/sin/atan/tan support for unary for tensorflow model;
- add any/all support for reduction for tensorflow model;
- add elu, conv3d, pool3d support for tensorflow model;
- optimize argmax, batchnorm, concat, batch to space, conv with variable weights, prelu, slice for tensorflow model;
- others:
- fix size computer lock;
- fix thread pool deadlock;
- add express & parameters in express;
- rewrite blitter chooser without static map;
- add tests for expr;
2019-10-29 13:37:26 +08:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
};
|
|
|
|
MNNTestSuiteRegister(MatMulTest, "expr/MatMul");
|