mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-04-20 06:49:15 +00:00
[CK-Tile] Merge transpose examples (#2450)
* unify pipeline signature with existing example * iwyu * move stuff around in load-tile-transpose * cleanups in batched transpose pipeline * comments * use same inputs size * cleaner printf * print host args * use 64 block sides in the 37_transpose example * roll back grid dimension size adjustment for 37_transpose example * transpose grid for 37_transpose to unify with 35_batched_transpose * unify grid computation logic * make policy methods device only (since they are used only on device from the pipeline) * more host/device attribute cleanups * copy over problem * move over pipeline and policy * add switch to batched transpose api * make the lds problem more similar to original problem * factor out logic into traits * factor out conditional compilation into trait parameter * propagate pipeline to args * unhardcode pipeline dispatch parameter * refactor vector size * put warp tile out of dispatch * rename template parameter for trait * rewrite vector size in terms of problem * mark policy-internal struct variable as device * factor out input distribution and thread access pattern from policies * reword vector size * use datatype across batched transpose pipelines, problems and kernel * remove transpose traits from lds pipeline * add padding to the lds pipeline *interface* * add comment * remove ck_tile example #37 * update cmakelists * add test for new pipeline * update batched transpose test * roll back load_tile_transpose changes * remove comments * pack dispatch parameters into a config * padM can be enabled * adjust lds vector size to enable padding along N * update test * clean up logic * swap m/n input vector size * adjust perf test script * sweep over C/W in perf test * count both read and written bytes into bandwidth (x2 the number) * clang-format * widen size range for perf test * remove 64k x 64k case; it's too large for index * remove thread tile from dispatch * Solve merge conflict * fix compile * modify the transpose * solve the test error and clang format * Add v3 support for Groupd fwd conv+bias+clamp & ckProfiler (#2463) * Add logging to IsSupported. * Less casting in AddClamp * Conv+bias+clamp instances & profiler BF16 * Fix 3D instances & run just 1x for verification. * :Run just once for verification conv fwd. * ckProfiler conv fwd clampwq * Remove exec bit & formatting * Add support for MultiD for grouped conv fwd v3. * Enable 2Lds. * clean * align instances * align instances * profiler fixes * Fixes * fix * fix --------- Co-authored-by: Adam Osewski <root@quanta-ccs-aus-f01-19.cs-aus.dcgpu> Co-authored-by: Bartłomiej Kocot <barkocot@amd.com> * Fixing 0ms and inf GB/s issue in img2col (#2565) issue : ==== ``` sh $ bin/tile_example_img2col Perf: 0 ms, inf GB/s ``` solution : ====== Problem occured because config.time_kernel is false by default. if false, then no need to calculate perf, just print proper message `image_to_coloumn: pass, No Perf generated due to config.time_kernel=0` * merge with develop * solve clang format --------- Co-authored-by: ThomasNing <thomas.ning@amd.com> Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com> Co-authored-by: Adam Osewski <root@quanta-ccs-aus-f01-19.cs-aus.dcgpu> Co-authored-by: Bartłomiej Kocot <barkocot@amd.com> Co-authored-by: rahjain-amd <Rahul.Jain@amd.com>
This commit is contained in:
Max Podkorytov
@@ -2,41 +2,93 @@
|
||||
// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
|
||||
#include "batched_transpose_example.hpp"
|
||||
|
||||
template <typename ts_type,
|
||||
ck_tile::index_t block_x,
|
||||
ck_tile::index_t block_y,
|
||||
ck_tile::index_t warp_x,
|
||||
ck_tile::index_t warp_y,
|
||||
ck_tile::index_t thread_x,
|
||||
ck_tile::index_t thread_y,
|
||||
bool kPadM,
|
||||
bool kPadN>
|
||||
namespace {
|
||||
|
||||
template <int32_t pipeline_id>
|
||||
struct kernel_traits;
|
||||
|
||||
template <>
|
||||
struct kernel_traits<0>
|
||||
{
|
||||
template <typename ts_type, typename block_tile, typename warp_layout, bool kPadM, bool kPadN>
|
||||
using Problem =
|
||||
ck_tile::BatchedTransposeProblem<ts_type, block_tile, warp_layout, kPadM, kPadN>;
|
||||
using Policy = ck_tile::BatchedTransposePolicy;
|
||||
template <typename ts_type, typename block_tile, typename warp_layout, bool kPadM, bool kPadN>
|
||||
using Pipeline =
|
||||
ck_tile::BatchedTransposePipeline<Problem<ts_type, block_tile, warp_layout, kPadM, kPadN>,
|
||||
Policy>;
|
||||
};
|
||||
|
||||
template <>
|
||||
struct kernel_traits<1>
|
||||
{
|
||||
template <typename ts_type, typename block_tile, typename warp_layout, bool kPadM, bool kPadN>
|
||||
using Problem =
|
||||
ck_tile::BatchedTransposeLdsProblem<ts_type, block_tile, warp_layout, kPadM, kPadN>;
|
||||
using Policy = ck_tile::BatchedTransposeLdsPolicy;
|
||||
template <typename ts_type, typename block_tile, typename warp_layout, bool kPadM, bool kPadN>
|
||||
using Pipeline = ck_tile::BatchedTransposeLdsPipeline<
|
||||
Problem<ts_type, block_tile, warp_layout, kPadM, kPadN>,
|
||||
Policy>;
|
||||
};
|
||||
} // namespace
|
||||
|
||||
template <typename InputType_,
|
||||
ck_tile::index_t BlockX_,
|
||||
ck_tile::index_t BlockY_,
|
||||
ck_tile::index_t NumWarpsX_,
|
||||
ck_tile::index_t NumWarpsY_,
|
||||
bool PadM_,
|
||||
bool PadN_,
|
||||
ck_tile::index_t PipelineId_>
|
||||
struct BatchedTransposeConfig
|
||||
{
|
||||
using InputType = InputType_;
|
||||
static constexpr ck_tile::index_t kBlockX = BlockX_;
|
||||
static constexpr ck_tile::index_t kBlockY = BlockY_;
|
||||
static constexpr ck_tile::index_t kNumWarpsX = NumWarpsX_;
|
||||
static constexpr ck_tile::index_t kNumWarpsY = NumWarpsY_;
|
||||
static constexpr bool kPadM = PadM_;
|
||||
static constexpr bool kPadN = PadN_;
|
||||
static constexpr ck_tile::index_t kPipelineId = PipelineId_;
|
||||
};
|
||||
|
||||
template <typename Config>
|
||||
float batched_transpose_dispatch(batched_transpose_kargs& a, ck_tile::stream_config& s)
|
||||
{
|
||||
uint32_t dim_stride = a.height * a.width;
|
||||
|
||||
a.dim_stride = dim_stride;
|
||||
a.dim_block_h = block_y;
|
||||
a.dim_block_w = block_x;
|
||||
a.dim_block_h = Config::kBlockY;
|
||||
a.dim_block_w = Config::kBlockX;
|
||||
|
||||
using block_tile = ck_tile::sequence<block_x, block_y>;
|
||||
using warp_tile = ck_tile::sequence<warp_x, warp_y>;
|
||||
using thread_tile = ck_tile::sequence<thread_x, thread_y>;
|
||||
|
||||
using ts_problem =
|
||||
ck_tile::BatchedTransposeProblem<ts_type, block_tile, warp_tile, thread_tile, kPadM, kPadN>;
|
||||
using ts_pipeline = ck_tile::BatchedTransposePipeline<ts_problem>;
|
||||
|
||||
using kernel = ck_tile::BatchedTransposeKernel<ts_pipeline>;
|
||||
// TODO: this is fragile and slow to compile
|
||||
using kernel = ck_tile::BatchedTransposeKernel<
|
||||
typename kernel_traits<Config::kPipelineId>::template Pipeline<
|
||||
typename Config::InputType,
|
||||
ck_tile::sequence<Config::kBlockX, Config::kBlockY>,
|
||||
ck_tile::sequence<Config::kNumWarpsX, Config::kNumWarpsY>,
|
||||
Config::kPadM,
|
||||
Config::kPadN>>;
|
||||
|
||||
auto kargs = kernel::MakeKargs(a);
|
||||
|
||||
const dim3 grids = kernel::GridSize(a);
|
||||
constexpr dim3 blocks = kernel::BlockSize();
|
||||
|
||||
printf("Grid: %u %u %u\n", grids.x, grids.y, grids.z);
|
||||
printf("Block: %u %u %u\n", blocks.x, blocks.y, blocks.z);
|
||||
printf("kargs: kargs.batch %d kargs.height %d kargs.width %d kargs.dim_strid %d\n",
|
||||
printf("Pipeline: %d\n", Config::kPipelineId);
|
||||
printf("Grid: x=%u y=%u z=%u\n", grids.x, grids.y, grids.z);
|
||||
printf("Block: x=%u y=%u z=%u\n", blocks.x, blocks.y, blocks.z);
|
||||
printf(
|
||||
"Host args: batch=%d, height=%d, width=%d, dim_stride=%d, dim_block_h=%d, dim_block_w=%d\n",
|
||||
a.batch,
|
||||
a.height,
|
||||
a.width,
|
||||
a.dim_stride,
|
||||
a.dim_block_h,
|
||||
a.dim_block_w);
|
||||
printf("kargs: kargs.batch=%d kargs.height=%d kargs.width=%d kargs.dim_stride=%d\n",
|
||||
kargs.batch,
|
||||
kargs.height,
|
||||
kargs.width,
|
||||
@@ -52,22 +104,29 @@ float batched_transpose_dispatch(batched_transpose_kargs& a, ck_tile::stream_con
|
||||
return ave_time;
|
||||
}
|
||||
|
||||
// Param Comb: type_size, block_x & y, warp_x & y, thread_x & y
|
||||
#define FOREACH_TRANSPOSE_PARAM(F) \
|
||||
F(fp8, ck_tile::fp8_t, 64, 64, 64, 64, 8, 8, true, true) \
|
||||
F(fp8, ck_tile::fp8_t, 64, 64, 64, 64, 8, 8, false, false) \
|
||||
F(fp16, ck_tile::fp16_t, 64, 64, 64, 64, 8, 8, true, true) \
|
||||
F(fp16, ck_tile::fp16_t, 64, 64, 64, 64, 8, 8, false, false) \
|
||||
F(bf16, ck_tile::bf16_t, 64, 64, 64, 64, 8, 8, true, true) \
|
||||
F(bf16, ck_tile::bf16_t, 64, 64, 64, 64, 8, 8, false, false)
|
||||
// Param Comb: type_size, block_x & y, WarpNum_x & y
|
||||
#define FOREACH_TRANSPOSE_PARAM(F) \
|
||||
F(fp8, ck_tile::fp8_t, 64, 64, 1, 1, true, true, 0) \
|
||||
F(fp8, ck_tile::fp8_t, 64, 64, 1, 1, false, false, 0) \
|
||||
F(fp16, ck_tile::fp16_t, 64, 64, 1, 1, true, true, 0) \
|
||||
F(fp16, ck_tile::fp16_t, 64, 64, 1, 1, false, false, 0) \
|
||||
F(bf16, ck_tile::bf16_t, 64, 64, 1, 1, true, true, 0) \
|
||||
F(bf16, ck_tile::bf16_t, 64, 64, 1, 1, false, false, 0) \
|
||||
F(fp8, ck_tile::fp8_t, 64, 64, 1, 1, true, true, 1) \
|
||||
F(fp8, ck_tile::fp8_t, 64, 64, 1, 1, false, false, 1) \
|
||||
F(fp16, ck_tile::fp16_t, 64, 64, 1, 1, true, true, 1) \
|
||||
F(fp16, ck_tile::fp16_t, 64, 64, 1, 1, false, false, 1) \
|
||||
F(bf16, ck_tile::bf16_t, 64, 64, 1, 1, true, true, 1) \
|
||||
F(bf16, ck_tile::bf16_t, 64, 64, 1, 1, false, false, 1)
|
||||
|
||||
// Macro that defines one static function per line
|
||||
#define GEN_TRANSPOSE_FN(SHORT_NAME, REAL_TYPE, BX, BY, WX, WY, TX, TY, PADM, PADN) \
|
||||
static float \
|
||||
transpose_fn_##SHORT_NAME##_##BX##_##BY##_##WX##_##WY##_##TX##_##TY##_##PADM##_##PADN( \
|
||||
batched_transpose_kargs& a, ck_tile::stream_config& s) \
|
||||
{ \
|
||||
return batched_transpose_dispatch<REAL_TYPE, BX, BY, WX, WY, TX, TY, PADM, PADN>(a, s); \
|
||||
#define GEN_TRANSPOSE_FN(SHORT_NAME, REAL_TYPE, BX, BY, WX, WY, PADM, PADN, PIPE) \
|
||||
static float \
|
||||
transpose_fn_##SHORT_NAME##_##BX##_##BY##_##WX##_##WY##_##PADM##_##PADN##_v##PIPE( \
|
||||
batched_transpose_kargs& a, ck_tile::stream_config& s) \
|
||||
{ \
|
||||
return batched_transpose_dispatch< \
|
||||
BatchedTransposeConfig<REAL_TYPE, BX, BY, WX, WY, PADM, PADN, PIPE>>(a, s); \
|
||||
}
|
||||
|
||||
FOREACH_TRANSPOSE_PARAM(GEN_TRANSPOSE_FN)
|
||||
@@ -76,38 +135,78 @@ float batched_transpose(batched_transpose_trait t,
|
||||
batched_transpose_kargs a,
|
||||
ck_tile::stream_config s)
|
||||
{
|
||||
if(t.type == "fp8")
|
||||
if(t.pipeline == "0")
|
||||
{
|
||||
if(a.height % 64 == 0 && a.width % 64 == 0)
|
||||
if(t.type == "fp8")
|
||||
{
|
||||
return transpose_fn_fp8_64_64_64_64_8_8_false_false(a, s);
|
||||
if(a.height % 64 == 0 && a.width % 64 == 0)
|
||||
{
|
||||
return transpose_fn_fp8_64_64_1_1_false_false_v0(a, s);
|
||||
}
|
||||
else
|
||||
{
|
||||
return transpose_fn_fp8_64_64_1_1_true_true_v0(a, s);
|
||||
}
|
||||
}
|
||||
else
|
||||
else if(t.type == "fp16")
|
||||
{
|
||||
return transpose_fn_fp8_64_64_64_64_8_8_true_true(a, s);
|
||||
if(a.height % 64 == 0 && a.width % 64 == 0)
|
||||
{
|
||||
return transpose_fn_fp16_64_64_1_1_false_false_v0(a, s);
|
||||
}
|
||||
else
|
||||
{
|
||||
return transpose_fn_fp16_64_64_1_1_true_true_v0(a, s);
|
||||
}
|
||||
}
|
||||
else if(t.type == "bf16")
|
||||
{
|
||||
if(a.height % 64 == 0 && a.width % 64 == 0)
|
||||
{
|
||||
return transpose_fn_bf16_64_64_1_1_false_false_v0(a, s);
|
||||
}
|
||||
else
|
||||
{
|
||||
return transpose_fn_bf16_64_64_1_1_true_true_v0(a, s);
|
||||
}
|
||||
}
|
||||
}
|
||||
else if(t.type == "fp16")
|
||||
else if(t.pipeline == "1")
|
||||
{
|
||||
if(a.height % 64 == 0 && a.width % 64 == 0)
|
||||
if(t.type == "fp8")
|
||||
{
|
||||
return transpose_fn_fp16_64_64_64_64_8_8_false_false(a, s);
|
||||
if(a.height % 64 == 0 && a.width % 64 == 0)
|
||||
{
|
||||
return transpose_fn_fp8_64_64_1_1_false_false_v1(a, s);
|
||||
}
|
||||
else
|
||||
{
|
||||
return transpose_fn_fp8_64_64_1_1_true_true_v1(a, s);
|
||||
}
|
||||
}
|
||||
else
|
||||
else if(t.type == "fp16")
|
||||
{
|
||||
return transpose_fn_fp16_64_64_64_64_8_8_true_true(a, s);
|
||||
}
|
||||
}
|
||||
else if(t.type == "bf16")
|
||||
{
|
||||
if(a.height % 64 == 0 && a.width % 64 == 0)
|
||||
{
|
||||
return transpose_fn_bf16_64_64_64_64_8_8_false_false(a, s);
|
||||
}
|
||||
else
|
||||
{
|
||||
return transpose_fn_bf16_64_64_64_64_8_8_true_true(a, s);
|
||||
if(a.height % 64 == 0 && a.width % 64 == 0)
|
||||
{
|
||||
return transpose_fn_fp16_64_64_1_1_false_false_v1(a, s);
|
||||
}
|
||||
else
|
||||
{
|
||||
return transpose_fn_fp16_64_64_1_1_true_true_v1(a, s);
|
||||
}
|
||||
}
|
||||
else if(t.type == "bf16")
|
||||
{
|
||||
if(a.height % 64 == 0 && a.width % 64 == 0)
|
||||
{
|
||||
return transpose_fn_bf16_64_64_1_1_false_false_v1(a, s);
|
||||
}
|
||||
else
|
||||
{
|
||||
return transpose_fn_bf16_64_64_1_1_true_true_v1(a, s);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return -1;
|
||||
}
|
||||
|
||||
@@ -102,7 +102,8 @@ auto create_args(int argc, char* argv[])
|
||||
.insert("warmup", "50", "number of iterations before benchmark the kernel")
|
||||
.insert("repeat", "100", "number of iterations to benchmark the kernel")
|
||||
.insert("seed", "-1", "seed to be used, -1 means random every time")
|
||||
.insert("kname", "0", "t to 1 will print kernel name");
|
||||
.insert("kname", "0", "t to 1 will print kernel name")
|
||||
.insert("pipeline", "0", "0: no LDS usage, 1: LDS-accelerated (gfx950)");
|
||||
|
||||
bool result = arg_parser.parse(argc, argv);
|
||||
return std::make_tuple(result, arg_parser);
|
||||
@@ -121,6 +122,7 @@ bool run_batched_transpose(ck_tile::ArgParser args)
|
||||
int n_repeat = args.get_int("repeat");
|
||||
std::string layout_in = args.get_str("layout_in");
|
||||
std::string layout_out = args.get_str("layout_out");
|
||||
std::string pipeline = args.get_str("pipeline");
|
||||
int seed = args.get_int("seed");
|
||||
|
||||
int dim_in[4], dim_out[4];
|
||||
@@ -166,7 +168,7 @@ bool run_batched_transpose(ck_tile::ArgParser args)
|
||||
|
||||
x_dev.ToDevice(x_host.data());
|
||||
|
||||
auto trait = batched_transpose_trait{prec, layout_in};
|
||||
auto trait = batched_transpose_trait{prec, layout_in, pipeline};
|
||||
|
||||
uint32_t height = nchw2nhwc ? C : H * W;
|
||||
uint32_t width = nchw2nhwc ? H * W : C;
|
||||
@@ -185,17 +187,15 @@ bool run_batched_transpose(ck_tile::ArgParser args)
|
||||
|
||||
auto ms = batched_transpose(trait, karg, sc);
|
||||
|
||||
std::size_t num_operations = N * C * H * (W - 1);
|
||||
std::size_t num_bytes = N * C * H * W * sizeof(Type);
|
||||
std::size_t num_bytes = N * C * H * W * sizeof(Type) * 2; // read + written
|
||||
|
||||
float ave_time = ms * 1E-3;
|
||||
float gb_per_sec = num_bytes / ms * 1.E-6;
|
||||
float tflops = static_cast<float>(num_operations) / ms * 1.E-6;
|
||||
|
||||
std::cout << "Run Batched Transpose kernel with N=" << N << ", C=" << C << ", H=" << H
|
||||
<< ", W=" << W << ", layout_in=" << layout_in << ", layout_out=" << layout_out
|
||||
<< " : " << ms << " ms (" << ave_time << " ave_time), " << tflops << " TFlops"
|
||||
<< gb_per_sec << " GB/s, " << std::endl;
|
||||
<< " : " << std::endl
|
||||
<< ms << " ms " << std::endl
|
||||
<< gb_per_sec << " GB/s " << std::endl;
|
||||
|
||||
printf("[%s]N:%d, C:%d, H:%d, W:%d, layout_in:%s, %f\n",
|
||||
prec.c_str(),
|
||||
|
||||
@@ -14,6 +14,7 @@ struct batched_transpose_trait
|
||||
{
|
||||
std::string type;
|
||||
std::string layout;
|
||||
std::string pipeline;
|
||||
};
|
||||
|
||||
struct batched_transpose_kargs : public ck_tile::BatchedTransposeHostArgs
|
||||
|
||||
@@ -5,10 +5,14 @@
|
||||
|
||||
EXE=./build/bin/tile_example_batched_transpose
|
||||
|
||||
for C in "64" "256" "1024" "4096" "16384"; do
|
||||
for W in "64" "256" "1024" "4096" "16384"; do
|
||||
for pr in "fp8" "fp16" "bf16"; do
|
||||
$EXE -pr=$pr -N=1 -C=64 -H=1 -W=64 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=1 -C=1024 -H=1 -W=1024 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=1 -C=1024 -H=1 -W=2048 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=1 -C=4096 -H=1 -W=2048 -layout_in='NCHW' -layout_out='NHWC'
|
||||
for pipeline in "0" "1"; do
|
||||
|
||||
$EXE -pipeline=$pipeline -pr=$pr -N=1 -C=$C -H=1 -W=$W -layout_in='NCHW' -layout_out='NHWC'
|
||||
|
||||
done
|
||||
done
|
||||
done
|
||||
done
|
||||
@@ -6,25 +6,27 @@
|
||||
EXE=./build/bin/tile_example_batched_transpose
|
||||
|
||||
for pr in "fp8" "fp16" "bf16"; do
|
||||
$EXE -pr=$pr -N=1 -C=32 -H=1 -W=32 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=1 -C=64 -H=1 -W=64 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=2 -C=12 -H=1 -W=32 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -N=3 -C=1334 -H=1 -W=37 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -N=4 -C=27 -H=1 -W=32 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=5 -C=1234 -H=1 -W=12 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=1 -C=1 -H=1 -W=1 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=1 -C=1 -H=1 -W=1 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -N=128 -C=1024 -H=64 -W=64 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=128 -C=1024 -H=64 -W=64 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -N=16 -C=64 -H=32 -W=128 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=16 -C=64 -H=128 -W=32 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -N=1 -C=2048 -H=1 -W=1 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=1 -C=2048 -H=1 -W=1 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -N=1 -C=1 -H=1024 -W=1024 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=1 -C=1 -H=1024 -W=1024 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -N=8 -C=16 -H=8 -W=16 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=8 -C=16 -H=8 -W=16 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -N=1 -C=64 -H=1 -W=1024 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -N=1 -C=64 -H=1024 -W=1 -layout_in='NHWC' -layout_out='NCHW'
|
||||
for pipeline in "0" "1"; do
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=32 -H=1 -W=32 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=64 -H=1 -W=64 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=2 -C=12 -H=1 -W=32 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=3 -C=1334 -H=1 -W=37 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=4 -C=27 -H=1 -W=32 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=5 -C=1234 -H=1 -W=12 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=1 -H=1 -W=1 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=1 -H=1 -W=1 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=128 -C=1024 -H=64 -W=64 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=128 -C=1024 -H=64 -W=64 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=16 -C=64 -H=32 -W=128 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=16 -C=64 -H=128 -W=32 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=2048 -H=1 -W=1 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=2048 -H=1 -W=1 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=1 -H=1024 -W=1024 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=1 -H=1024 -W=1024 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=8 -C=16 -H=8 -W=16 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=8 -C=16 -H=8 -W=16 -layout_in='NHWC' -layout_out='NCHW'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=64 -H=1 -W=1024 -layout_in='NCHW' -layout_out='NHWC'
|
||||
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=64 -H=1024 -W=1 -layout_in='NHWC' -layout_out='NCHW'
|
||||
|
||||
done
|
||||
done
|
||||
|
||||
Reference in New Issue
Block a user