New subgroups - full changes set (#1074)

* Extended subgroups - extended types types

* Extended subgroups - non uniform vote tests

* Extended subgroups - non uniform arithmetic tests

* Extended subgroups - ballot tests

* Extended subgroups - clustered reduce tests

* Extended subgroups - shuffle tests

* Extended subgroups - formating issues

* Extended subgroups - review fixes

* Extended subgroups - review fixes

Fixed: removed additional brakes, kernel_sstr

* Extended subgroups - fix macos build error

* Extended subgroups - review fixes

Fixed: mac os build error

* Extended subgroups - data type verification example

* Extended subgroups - error unification

* Extended subgroups - fix header years

* Extended subgroups - use is_half_nan

* Extended subgroups - compare half as float

* Review fixes mostly for ballot functions.

- Modify kernels for better handling active/inactive workitems
- Modify gen/chk functions for handling non uniform workgroup sizes
- Introduce new variables naming convention
- minor fixes

* Extended subgroups - simplification data generation for ballot lsb/msb functions

* Extended subgroups - minor fixes

* Extended subgroups - move common code to function

* Extended subgroups - formatting errors fix

* Extended subgroups - fix build error

* Extended subgroups - sub_group_elect more sophisticated

Define mask which is 4bytes pattern where bit 1 means work item is active.
If workitem in subgroup matches pattern then run sub_group_elect()

* Extended subgroups - fix Ubuntu build error

* Extended subgroups - voting function review fixes

* adjust all function for using masks
* remove calculate templates
* merge code to one common template
* check results only in active workitems
* normalize values on host side
* minor fixes

* Extended subgroups - fix typos

* Set of fixes and improvements after review

* define WorkGroupParams to stop extended parameters list in function
* better workitems mask handing (WorkGroupParams)
* narrow values of data input generation to avoid overflows (arithmetic func)
* implement work item masks for arithmetic functions
* enable half type testing for reduction/scan/broadcast
* minor fixes

* Extended subgroups - fix Linux issues

* Extended subgroups - fix sub_group_local_id data type

* Extended subgroups - use vector instead of array.

* Extended subgroups - change names to subgroup

* Extended subgroups - uncomment code, fix build

* Extended subgroups - build fix, use cl_half_from_float func

* Extended subgroups - remove is_half_nan

* Extended subgroups - do no use undef min/max

* Extended subgroups - use parenthesis, fix formatting
This commit is contained in:
Grzegorz Wawiorko
2021-04-06 18:25:48 +02:00
committed by GitHub
parent c5e4ca6c91
commit 71bef8563e
18 changed files with 5075 additions and 957 deletions

View File

@@ -0,0 +1,217 @@
//
// Copyright (c) 2017 The Khronos Group Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//
#include "procs.h"
#include "subhelpers.h"
#include "subgroup_common_kernels.h"
#include "subgroup_common_templates.h"
#include "harness/conversions.h"
#include "harness/typeWrappers.h"
namespace {
// Any/All test functions
template <NonUniformVoteOp operation> struct AA
{
static void gen(cl_int *x, cl_int *t, cl_int *m,
const WorkGroupParams &test_params)
{
int i, ii, j, k, n;
int ng = test_params.global_workgroup_size;
int nw = test_params.local_workgroup_size;
int ns = test_params.subgroup_size;
int nj = (nw + ns - 1) / ns;
int e;
ng = ng / nw;
ii = 0;
log_info(" sub_group_%s...\n", operation_names(operation));
for (k = 0; k < ng; ++k)
{
for (j = 0; j < nj; ++j)
{
ii = j * ns;
n = ii + ns > nw ? nw - ii : ns;
e = (int)(genrand_int32(gMTdata) % 3);
// Initialize data matrix indexed by local id and sub group id
switch (e)
{
case 0: memset(&t[ii], 0, n * sizeof(cl_int)); break;
case 1:
memset(&t[ii], 0, n * sizeof(cl_int));
i = (int)(genrand_int32(gMTdata) % (cl_uint)n);
t[ii + i] = 41;
break;
case 2: memset(&t[ii], 0xff, n * sizeof(cl_int)); break;
}
}
// Now map into work group using map from device
for (j = 0; j < nw; ++j)
{
x[j] = t[j];
}
x += nw;
m += 4 * nw;
}
}
static int chk(cl_int *x, cl_int *y, cl_int *mx, cl_int *my, cl_int *m,
const WorkGroupParams &test_params)
{
int ii, i, j, k, n;
int ng = test_params.global_workgroup_size;
int nw = test_params.local_workgroup_size;
int ns = test_params.subgroup_size;
int nj = (nw + ns - 1) / ns;
cl_int taa, raa;
ng = ng / nw;
for (k = 0; k < ng; ++k)
{
// Map to array indexed to array indexed by local ID and sub group
for (j = 0; j < nw; ++j)
{
mx[j] = x[j];
my[j] = y[j];
}
for (j = 0; j < nj; ++j)
{
ii = j * ns;
n = ii + ns > nw ? nw - ii : ns;
// Compute target
if (operation == NonUniformVoteOp::any)
{
taa = 0;
for (i = 0; i < n; ++i) taa |= mx[ii + i] != 0;
}
if (operation == NonUniformVoteOp::all)
{
taa = 1;
for (i = 0; i < n; ++i) taa &= mx[ii + i] != 0;
}
// Check result
for (i = 0; i < n; ++i)
{
raa = my[ii + i] != 0;
if (raa != taa)
{
log_error("ERROR: sub_group_%s mismatch for local id "
"%d in sub group %d in group %d\n",
operation_names(operation), i, j, k);
return TEST_FAIL;
}
}
}
x += nw;
y += nw;
m += 4 * nw;
}
log_info(" sub_group_%s... passed\n", operation_names(operation));
return TEST_PASS;
}
};
static const char *any_source = "__kernel void test_any(const __global Type "
"*in, __global int4 *xy, __global Type *out)\n"
"{\n"
" int gid = get_global_id(0);\n"
" XY(xy,gid);\n"
" out[gid] = sub_group_any(in[gid]);\n"
"}\n";
static const char *all_source = "__kernel void test_all(const __global Type "
"*in, __global int4 *xy, __global Type *out)\n"
"{\n"
" int gid = get_global_id(0);\n"
" XY(xy,gid);\n"
" out[gid] = sub_group_all(in[gid]);\n"
"}\n";
template <typename T>
int run_broadcast_scan_reduction_for_type(RunTestForType rft)
{
int error = rft.run_impl<T, BC<T, SubgroupsBroadcastOp::broadcast>>(
"test_bcast", bcast_source);
error |= rft.run_impl<T, RED_NU<T, ArithmeticOp::add_>>("test_redadd",
redadd_source);
error |= rft.run_impl<T, RED_NU<T, ArithmeticOp::max_>>("test_redmax",
redmax_source);
error |= rft.run_impl<T, RED_NU<T, ArithmeticOp::min_>>("test_redmin",
redmin_source);
error |= rft.run_impl<T, SCIN_NU<T, ArithmeticOp::add_>>("test_scinadd",
scinadd_source);
error |= rft.run_impl<T, SCIN_NU<T, ArithmeticOp::max_>>("test_scinmax",
scinmax_source);
error |= rft.run_impl<T, SCIN_NU<T, ArithmeticOp::min_>>("test_scinmin",
scinmin_source);
error |= rft.run_impl<T, SCEX_NU<T, ArithmeticOp::add_>>("test_scexadd",
scexadd_source);
error |= rft.run_impl<T, SCEX_NU<T, ArithmeticOp::max_>>("test_scexmax",
scexmax_source);
error |= rft.run_impl<T, SCEX_NU<T, ArithmeticOp::min_>>("test_scexmin",
scexmin_source);
return error;
}
}
// Entry point from main
int test_subgroup_functions(cl_device_id device, cl_context context,
cl_command_queue queue, int num_elements,
bool useCoreSubgroups)
{
constexpr size_t global_work_size = 2000;
constexpr size_t local_work_size = 200;
WorkGroupParams test_params(global_work_size, local_work_size);
RunTestForType rft(device, context, queue, num_elements, test_params);
int error =
rft.run_impl<cl_int, AA<NonUniformVoteOp::any>>("test_any", any_source);
error |=
rft.run_impl<cl_int, AA<NonUniformVoteOp::all>>("test_all", all_source);
error |= run_broadcast_scan_reduction_for_type<cl_int>(rft);
error |= run_broadcast_scan_reduction_for_type<cl_uint>(rft);
error |= run_broadcast_scan_reduction_for_type<cl_long>(rft);
error |= run_broadcast_scan_reduction_for_type<cl_ulong>(rft);
error |= run_broadcast_scan_reduction_for_type<cl_float>(rft);
error |= run_broadcast_scan_reduction_for_type<cl_double>(rft);
error |= run_broadcast_scan_reduction_for_type<subgroups::cl_half>(rft);
return error;
}
int test_subgroup_functions_core(cl_device_id device, cl_context context,
cl_command_queue queue, int num_elements)
{
return test_subgroup_functions(device, context, queue, num_elements, true);
}
int test_subgroup_functions_ext(cl_device_id device, cl_context context,
cl_command_queue queue, int num_elements)
{
bool hasExtension = is_extension_available(device, "cl_khr_subgroups");
if (!hasExtension)
{
log_info(
"Device does not support 'cl_khr_subgroups'. Skipping the test.\n");
return TEST_SKIPPED_ITSELF;
}
return test_subgroup_functions(device, context, queue, num_elements, false);
}