Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

AMD-related Code Changes #1363

Merged
merged 5 commits into from
Oct 7, 2019
Merged

AMD-related Code Changes #1363

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
c46514c
Files from sid-hip
fthaler Oct 3, 2019
5925055
Cleanup
fthaler Oct 3, 2019
ea00272
Addressing review comment
fthaler Oct 4, 2019
7c39282
Addressing review comments
fthaler Oct 4, 2019
edaa482
Updated references
fthaler Oct 4, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
252 changes: 109 additions & 143 deletions include/gridtools/boundary_conditions/apply_gpu.hpp
View file
Open in desktop

Large diffs are not rendered by default.

19 changes: 11 additions & 8 deletions include/gridtools/interface/layout_transformation/layout_transformation_impl_cuda.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@
#include "../../common/cuda_util.hpp"
#include "../../common/defs.hpp"
#include "../../common/hypercube_iterator.hpp"
#include "../../common/integral_constant.hpp"
#include "../../common/layout_map_metafunctions.hpp"
#include "../../common/make_array.hpp"
#include "../../common/tuple_util.hpp"
Expand All @@ -24,6 +25,9 @@

namespace gridtools {
namespace impl {
// compile-time block size due to HIP-Clang bug https://github.com/ROCm-Developer-Tools/HIP/issues/1283
using block_size_1d_t = integral_constant<int_t, 8>;

template <typename DataType>
__global__ void transform_cuda_loop_kernel(DataType *dst,
DataType *src,
Expand All @@ -32,13 +36,13 @@ namespace gridtools {
gridtools::array<gridtools::uint_t, GT_TRANSFORM_MAX_DIM> src_strides,
gridtools::array<size_t, GT_TRANSFORM_MAX_DIM - 3> outer_dims) {

int i = blockIdx.x * blockDim.x + threadIdx.x;
int i = blockIdx.x * block_size_1d_t::value + threadIdx.x;
fthaler marked this conversation as resolved.
Show resolved Hide resolved
if (i >= dims[0])
return;
int j = blockIdx.y * blockDim.y + threadIdx.y;
int j = blockIdx.y * block_size_1d_t::value + threadIdx.y;
if (j >= dims[1])
return;
int k = blockIdx.z * blockDim.z + threadIdx.z;
int k = blockIdx.z * block_size_1d_t::value + threadIdx.z;
if (k >= dims[2])
return;

Expand Down Expand Up @@ -67,16 +71,15 @@ namespace gridtools {
const std::vector<uint_t> &dims,
const std::vector<uint_t> &dst_strides,
const std::vector<uint_t> &src_strides) {
int block_size_1d = 8;

auto a_dims = impl::vector_to_dims_array<GT_TRANSFORM_MAX_DIM>(dims);
gridtools::array<size_t, GT_TRANSFORM_MAX_DIM - 3> outer_dims;
std::copy(a_dims.begin() + 3, a_dims.end(), outer_dims.begin());

dim3 grid_size((a_dims[0] + block_size_1d - 1) / block_size_1d,
(a_dims[1] + block_size_1d - 1) / block_size_1d,
(a_dims[2] + block_size_1d - 1) / block_size_1d);
dim3 block_size(block_size_1d, block_size_1d, block_size_1d);
dim3 grid_size((a_dims[0] + block_size_1d_t::value - 1) / block_size_1d_t::value,
(a_dims[1] + block_size_1d_t::value - 1) / block_size_1d_t::value,
(a_dims[2] + block_size_1d_t::value - 1) / block_size_1d_t::value);
dim3 block_size(block_size_1d_t::value, block_size_1d_t::value, block_size_1d_t::value);

transform_cuda_loop_kernel<<<grid_size, block_size>>>(dst,
src,
Expand Down
164 changes: 82 additions & 82 deletions pyutils/perftest/references/structured/double/128/daint_nvcc_gcc/result.cuda.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
{
"datetime": "2019-09-27T13:21:05.866185+0000",
"datetime": "2019-10-04T11:43:23.453322+0000",
"domain": [
128,
128,
Expand All @@ -9,131 +9,131 @@
"backend": "cuda",
"clustername": "daint",
"compiler": "/opt/nvidia/cudatoolkit9.2/9.2.148_3.19-6.0.7.1_2.1__g3d9acc8/bin/nvcc 9.2.148 (/opt/cray/pe/craype/2.5.15/bin/CC 6.2.0)",
"datetime": "2019-09-27T13:07:16.000000+0000",
"datetime": "2019-10-04T10:48:55.000000+0000",
"grid": "structured",
"hostname": "daint102",
"hostname": "daint103",
"name": "gridtools",
"precision": "double",
"version": "28825da5472abbe21262c1e9129fd5b6c2048f47"
"version": "201c1c2f2593747cd3415771d3c35946e40e7055"
},
"times": [
{
"measurements": [
0.000433504,
0.000432544,
0.000432,
0.000430784,
0.000430848,
0.000430848,
0.000430464,
0.000430784,
0.000434144,
0.000430272
0.000430528,
0.000431552,
0.0004336,
0.000431072,
0.000431264,
0.000430432,
0.000431712,
0.00043344,
0.000432032,
0.000432032
],
"stencil": "copy"
},
{
"measurements": [
0.00125597,
0.00125155,
0.00126035,
0.00125395,
0.00125923,
0.00125386,
0.00126048,
0.00125405,
0.00125814,
0.00125542
0.00125914,
0.0012567,
0.00125594,
0.00125827,
0.00125648,
0.0012577,
0.00125856,
0.00125898,
0.00126195,
0.00125888
],
"stencil": "horizontal diffusion"
},
{
"measurements": [
0.000982208,
0.000979008,
0.000981312,
0.000982816,
0.000982688,
0.00097936,
0.000981184,
0.000980736,
0.00098,
0.00098272
0.000982528,
0.000983296,
0.000981824,
0.000981248,
0.000981632,
0.000984928,
0.000982944,
0.000981792,
0.000981824,
0.000981696
],
"stencil": "horizontal diffusion fused"
},
{
"measurements": [
0.00243578,
0.00243782,
0.00244026,
0.00244189,
0.0024305,
0.00244522,
0.00244659,
0.00243091,
0.0024401,
0.0024392,
0.00243846,
0.0024321
0.00243872,
0.00243949,
0.00243859,
0.0024393,
0.00244128,
0.00244131,
0.00243984
],
"stencil": "simple horizontal diffusion"
},
{
"measurements": [
0.00239014,
0.00238928,
0.00236496,
0.00236294,
0.00236957,
0.00237619,
0.00237904,
0.00238253,
0.00238272,
0.00237405
0.00237965,
0.00238557,
0.00236499,
0.00236755,
0.00237571,
0.00236826,
0.00237562,
0.00237699,
0.00236566,
0.00237792
],
"stencil": "vertical advection"
},
{
"measurements": [
0.00552717,
0.00553178,
0.00553981,
0.00553309,
0.00553046,
0.00553018,
0.00553389,
0.00553254,
0.00553488,
0.00553142
0.00553043,
0.00553568,
0.00553043,
0.00553594,
0.00553533,
0.00552864,
0.00553149,
0.00553261,
0.00552995,
0.00553114
],
"stencil": "advection pd bott"
},
{
"measurements": [
0.000645248,
0.000650144,
0.00065088,
0.000648064,
0.000649952,
0.000646432,
0.000650592,
0.000647872,
0.000643872,
0.00064672,
0.000649472,
0.000647104
0.000646848,
0.000649504,
0.000649568,
0.000649408,
0.000647616,
0.000650016,
0.000648928
],
"stencil": "layout transformation"
},
{
"measurements": [
0.000405024,
0.00040352,
0.000403072,
0.00040128,
0.000400192,
0.000406784,
0.000403424,
0.00040304,
0.000407936,
0.000401696
0.000386912,
0.000388736,
0.000384768,
0.000388928,
0.000388096,
0.000389248,
0.000385984,
0.00038512,
0.000386208,
0.000385312
],
"stencil": "boundary conditions"
}
Expand Down
Loading