cwpearson/astaroth
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Major improvement: uniforms can now be set to default values. The syntax is the same as for setting any other values, f.ex. 'uniform Scalar a = 1; uniform Scalar b = 0.5 * a;'. Undefined uniforms are still allowed, but in this case the user should load a proper value into it during runtime. Default uniform values can be overwritten by calling any of the uniform loader funcions (like acDeviceLoadScalarUniform). Improved also error checking. Now there are explicit warnings if the user tries to load an invalid value into a device constant.

Browse Source
This commit is contained in:
jpekkila
2020-01-28 18:17:31 +02:00
parent 6dfe3ed4d6
commit 0ccd4e3dbc
9 changed files with 197 additions and 115 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
acc/mhd_solver/stencil_kernel.ac
View File

@@ -56,7 +56,7 @@ uniform Scalar AC_cdt;
uniform Scalar AC_cdtv;
uniform Scalar AC_cdts;
uniform Scalar AC_nu_visc;
uniform Scalar AC_cs_sound;
uniform Scalar AC_cs_sound = 1.0;
uniform Scalar AC_eta;
uniform Scalar AC_mu0;
uniform Scalar AC_cp_sound;
@@ -96,7 +96,7 @@ uniform Scalar AC_G_const;
uniform Scalar AC_GM_star;
uniform Scalar AC_unit_mass;
uniform Scalar AC_sq2GM_star;
uniform Scalar AC_cs2_sound;
uniform Scalar AC_cs2_sound = AC_cs_sound * AC_cs_sound;
/*
* =============================================================================

81
acc/src/code_generator.c
View File

@@ -224,6 +224,7 @@ print_symbol_table(void)
*/
static bool inside_declaration = false;
static bool inside_kernel = false;
static bool inside_function = false;
/*
* =============================================================================
@@ -246,6 +247,8 @@ traverse(const ASTNode* node)
}
if (node->type == NODE_DECLARATION)
inside_declaration = true;
if (node->type == NODE_FUNCTION_DEFINITION)
inside_function = true;
if (node->token == KERNEL)
inside_kernel = true;
@@ -335,7 +338,12 @@ traverse(const ASTNode* node)
// Translate the new symbol
if (tqualifier == UNIFORM) {
// Do nothing
if (tspecifier == SCALAR || tspecifier == VECTOR || tspecifier == INT ||
tspecifier == INT3) {
fprintf(CUDAHEADER, "static %s %s_DEFAULT_VALUE", translate(tspecifier),
identifier);
}
// else do nothing
}
else if (tqualifier == KERNEL) {
fprintf(CUDAHEADER, "%s %s\n%s", //
@@ -404,7 +412,13 @@ traverse(const ASTNode* node)
if (symbol->type_qualifier == UNIFORM) {
if (inside_kernel && symbol->type_specifier == SCALARARRAY)
fprintf(CUDAHEADER, "buffer.profiles[%s] ", symbol->identifier);
else
else if (!inside_function &&
((symbol->type_specifier == SCALAR) ||
(symbol->type_specifier == VECTOR) || (symbol->type_specifier == INT) ||
(symbol->type_specifier == INT3))) // Global scope and an uniform which
// can be set to a default value
fprintf(CUDAHEADER, "%s_DEFAULT_VALUE ", symbol->identifier);
else // Use device constants inside device functions
fprintf(CUDAHEADER, "DCONST(%s) ", symbol->identifier);
}
else if (node->parent->type != NODE_DECLARATION) {
@@ -462,12 +476,27 @@ traverse(const ASTNode* node)
}
if (node->type == NODE_DECLARATION)
inside_declaration = false;
if (node->type == NODE_FUNCTION_DEFINITION)
inside_function = false;
// Postfix translation
if (!inside_declaration && translate(node->postfix))
fprintf(CUDAHEADER, "%s", translate(node->postfix));
}
#define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
static const char* builtin_int_params[] = {
"AC_nx", "AC_ny", "AC_nz", "AC_mx", "AC_my",
"AC_mz", "AC_nx_min", "AC_ny_min", "AC_nz_min", "AC_nx_max",
"AC_ny_max", "AC_nz_max", "AC_mxy", "AC_nxy", "AC_nxyz",
};
static const char* builtin_int3_params[] = {
"AC_global_grid_n",
"AC_multigpu_offset",
};
static void
generate_preprocessed_structures(void)
{
@@ -528,6 +557,16 @@ generate_preprocessed_structures(void)
CUDAHEADER = fopen(cudaheader_filename, "w+");
fprintf(CUDAHEADER, "#pragma once\n");
// Add built-in params (the best way would be to inject these to user src with AC syntax)
for (size_t i = 0; i < ARRAY_SIZE(builtin_int_params); ++i) {
fprintf(CUDAHEADER, "static const int %s_DEFAULT_VALUE = 0;\n", builtin_int_params[i]);
}
for (size_t i = 0; i < ARRAY_SIZE(builtin_int3_params); ++i) {
fprintf(CUDAHEADER, "static const int3 %s_DEFAULT_VALUE = make_int3(0, 0, 0);\n",
builtin_int3_params[i]);
}
fprintf(CUDAHEADER, "typedef struct {\n");
for (size_t i = 0; i < num_symbols[current_nest]; ++i) {
if (symbol_table[i].type_qualifier == PREPROCESSED)
@@ -566,7 +605,7 @@ generate_header(void)
fprintf(DSLHEADER, "#define AC_FOR_USER_INT_PARAM_TYPES(FUNC)");
for (size_t i = 0; i < num_symbols[current_nest]; ++i) {
if (symbol_table[i].type_specifier == INT && symbol_table[i].type_qualifier == UNIFORM) {
fprintf(DSLHEADER, "\\\nFUNC(%s),", symbol_table[i].identifier);
fprintf(DSLHEADER, "\\\nFUNC(%s)", symbol_table[i].identifier);
}
}
fprintf(DSLHEADER, "\n\n");
@@ -575,7 +614,7 @@ generate_header(void)
fprintf(DSLHEADER, "#define AC_FOR_USER_INT3_PARAM_TYPES(FUNC)");
for (size_t i = 0; i < num_symbols[current_nest]; ++i) {
if (symbol_table[i].type_specifier == INT3 && symbol_table[i].type_qualifier == UNIFORM) {
fprintf(DSLHEADER, "\\\nFUNC(%s),", symbol_table[i].identifier);
fprintf(DSLHEADER, "\\\nFUNC(%s)", symbol_table[i].identifier);
}
}
fprintf(DSLHEADER, "\n\n");
@@ -584,7 +623,7 @@ generate_header(void)
fprintf(DSLHEADER, "#define AC_FOR_USER_REAL_PARAM_TYPES(FUNC)");
for (size_t i = 0; i < num_symbols[current_nest]; ++i) {
if (symbol_table[i].type_specifier == SCALAR && symbol_table[i].type_qualifier == UNIFORM) {
fprintf(DSLHEADER, "\\\nFUNC(%s),", symbol_table[i].identifier);
fprintf(DSLHEADER, "\\\nFUNC(%s)", symbol_table[i].identifier);
}
}
fprintf(DSLHEADER, "\n\n");
@@ -593,7 +632,7 @@ generate_header(void)
fprintf(DSLHEADER, "#define AC_FOR_USER_REAL3_PARAM_TYPES(FUNC)");
for (size_t i = 0; i < num_symbols[current_nest]; ++i) {
if (symbol_table[i].type_specifier == VECTOR && symbol_table[i].type_qualifier == UNIFORM) {
fprintf(DSLHEADER, "\\\nFUNC(%s),", symbol_table[i].identifier);
fprintf(DSLHEADER, "\\\nFUNC(%s)", symbol_table[i].identifier);
}
}
fprintf(DSLHEADER, "\n\n");
@@ -603,7 +642,7 @@ generate_header(void)
for (size_t i = 0; i < num_symbols[current_nest]; ++i) {
if (symbol_table[i].type_specifier == SCALARFIELD &&
symbol_table[i].type_qualifier == UNIFORM) {
fprintf(DSLHEADER, "\\\nFUNC(%s),", symbol_table[i].identifier);
fprintf(DSLHEADER, "\\\nFUNC(%s)", symbol_table[i].identifier);
}
}
fprintf(DSLHEADER, "\n\n");
@@ -613,7 +652,7 @@ generate_header(void)
for (size_t i = 0; i < num_symbols[current_nest]; ++i) {
if (symbol_table[i].type_specifier == SCALARARRAY &&
symbol_table[i].type_qualifier == UNIFORM) {
fprintf(DSLHEADER, "\\\nFUNC(%s),", symbol_table[i].identifier);
fprintf(DSLHEADER, "\\\nFUNC(%s)", symbol_table[i].identifier);
}
}
fprintf(DSLHEADER, "\n\n");
@@ -645,25 +684,13 @@ main(void)
assert(DSLHEADER);
assert(CUDAHEADER);
// Add built-in params
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_nx");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_ny");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_nz");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_mx");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_my");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_mz");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_nx_min");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_ny_min");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_nz_min");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_nx_max");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_ny_max");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_nz_max");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_mxy");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_nxy");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, "AC_nxyz");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT3, "AC_global_grid_n");
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT3, "AC_multigpu_offset");
// Add built-in param symbols
for (size_t i = 0; i < ARRAY_SIZE(builtin_int_params); ++i) {
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT, builtin_int_params[i]);
}
for (size_t i = 0; i < ARRAY_SIZE(builtin_int3_params); ++i) {
add_symbol(SYMBOLTYPE_OTHER, UNIFORM, INT3, builtin_int3_params[i]);
}
// Generate
traverse(root);

12
acc/stdlib/stdderiv.h
View File

@@ -3,12 +3,12 @@
#define STENCIL_ORDER (6)
#endif
uniform Scalar AC_dsx;
uniform Scalar AC_dsy;
uniform Scalar AC_dsz;
uniform Scalar AC_inv_dsx;
uniform Scalar AC_inv_dsy;
uniform Scalar AC_inv_dsz;
uniform Scalar AC_dsx = 0.04908738521;
uniform Scalar AC_dsy = 0.04908738521;
uniform Scalar AC_dsz = 0.04908738521;
uniform Scalar AC_inv_dsx = 1.0 / AC_dsx;
uniform Scalar AC_inv_dsy = 1.0 / AC_dsy;
uniform Scalar AC_inv_dsz = 1.0 / AC_dsz;
Scalar
first_derivative(Scalar pencil[], Scalar inv_ds)

10
include/astaroth.h
View File

@@ -82,7 +82,7 @@ typedef enum {
} Stream;
#define STREAM_ALL (NUM_STREAMS)
#define AC_GEN_ID(X) X
#define AC_GEN_ID(X) X,
typedef enum {
AC_FOR_USER_INT_PARAM_TYPES(AC_GEN_ID) //
NUM_INT_PARAMS
@@ -115,7 +115,7 @@ typedef enum {
#undef AC_GEN_ID
#define _UNUSED __attribute__((unused)) // Does not give a warning if unused
#define AC_GEN_STR(X) #X
#define AC_GEN_STR(X) #X,
static const char* intparam_names[] _UNUSED = {AC_FOR_USER_INT_PARAM_TYPES(AC_GEN_STR)};
static const char* int3param_names[] _UNUSED = {AC_FOR_USER_INT3_PARAM_TYPES(AC_GEN_STR)};
static const char* realparam_names[] _UNUSED = {AC_FOR_USER_REAL_PARAM_TYPES(AC_GEN_STR)};
@@ -451,8 +451,10 @@ AcResult acDeviceLoadScalarArray(const Device device, const Stream stream,
const AcReal* data, const size_t num);
/** */
AcResult acDeviceLoadMeshInfo(const Device device, const Stream stream,
const AcMeshInfo device_config);
AcResult acDeviceLoadMeshInfo(const Device device, const AcMeshInfo device_config);
/** */
AcResult acDeviceLoadDefaultUniforms(const Device device);
/** */
AcResult acDeviceLoadVertexBufferWithOffset(const Device device, const Stream stream,

5
samples/ctest/main.c
View File

@@ -27,11 +27,6 @@ main(void)
{
AcMeshInfo info;
acLoadConfig(AC_DEFAULT_CONFIG, &info);
// Some real params must be calculated (for the MHD case) // TODO DANGEROUS
info.real_params[AC_inv_dsx] = (AcReal)(1.0) / info.real_params[AC_dsx];
info.real_params[AC_inv_dsy] = (AcReal)(1.0) / info.real_params[AC_dsy];
info.real_params[AC_inv_dsz] = (AcReal)(1.0) / info.real_params[AC_dsz];
info.real_params[AC_cs2_sound] = info.real_params[AC_cs_sound] * info.real_params[AC_cs_sound];
// Alloc
AcMesh model, candidate;

5
src/common/errchk.h
View File

@@ -66,6 +66,11 @@
if (!(retval)) \
WARNING(#retval " was false"); \
}
#define WARNCHK_ALWAYS(retval) \
{ \
if (!(retval)) \
WARNING(#retval " was false"); \
}
#define ERRCHK_ALWAYS(retval) \
{ \
if (!(retval)) \

3
src/core/device.cc
View File

@@ -150,7 +150,8 @@ acDeviceCreate(const int id, const AcMeshInfo device_config, Device* device_hand
ERRCHK_CUDA_ALWAYS(cudaMalloc((void**)&device->reduce_result, sizeof(AcReal)));
// Device constants
acDeviceLoadMeshInfo(device, STREAM_DEFAULT, device_config);
acDeviceLoadDefaultUniforms(device);
acDeviceLoadMeshInfo(device, device_config);
printf("Created device %d (%p)\n", device->id, device);
*device_handle = device;

107
src/core/kernels/kernels.cu
View File

@@ -96,33 +96,22 @@ static __device__ inline acComplex operator*(const acComplex& a, const acComplex
#include "packing.cuh"
#include "reductions.cuh"
AcResult
acDeviceLoadMeshInfo(const Device device, const Stream stream, const AcMeshInfo device_config)
{
cudaSetDevice(device->id);
ERRCHK_ALWAYS(device_config.int_params[AC_nx] == device->local_config.int_params[AC_nx]);
ERRCHK_ALWAYS(device_config.int_params[AC_ny] == device->local_config.int_params[AC_ny]);
ERRCHK_ALWAYS(device_config.int_params[AC_nz] == device->local_config.int_params[AC_nz]);
ERRCHK_ALWAYS(device_config.int_params[AC_multigpu_offset] ==
device->local_config.int_params[AC_multigpu_offset]);
ERRCHK_CUDA_ALWAYS(cudaMemcpyToSymbolAsync(d_mesh_info, &device_config, sizeof(device_config),
0, cudaMemcpyHostToDevice, device->streams[stream]));
return AC_SUCCESS;
}
AcResult
acDeviceLoadScalarUniform(const Device device, const Stream stream, const AcRealParam param,
const AcReal value)
{
cudaSetDevice(device->id);
if (!is_valid(value)) {
fprintf(stderr, "WARNING: Passed an invalid value %g to device constant %s. Skipping.\n",
(double)value, realparam_names[param]);
return AC_FAILURE;
}
if (param >= NUM_REAL_PARAMS)
return AC_FAILURE;
const size_t offset = (size_t)&d_mesh_info.real_params[param] - (size_t)&d_mesh_info;
ERRCHK_CUDA(cudaMemcpyToSymbolAsync(d_mesh_info, &value, sizeof(value), offset,
ERRCHK_CUDA_ALWAYS(cudaMemcpyToSymbolAsync(d_mesh_info, &value, sizeof(value), offset,
cudaMemcpyHostToDevice, device->streams[stream]));
return AC_SUCCESS;
}
@@ -132,12 +121,18 @@ acDeviceLoadVectorUniform(const Device device, const Stream stream, const AcReal
const AcReal3 value)
{
cudaSetDevice(device->id);
if (!is_valid(value)) {
fprintf(stderr,
"WARNING: Passed an invalid value (%g, %g, %g) to device constant %s. Skipping.\n",
(double)value.x, (double)value.y, (double)value.z, real3param_names[param]);
return AC_FAILURE;
}
if (param >= NUM_REAL3_PARAMS || !NUM_REAL3_PARAMS)
return AC_FAILURE;
const size_t offset = (size_t)&d_mesh_info.real3_params[param] - (size_t)&d_mesh_info;
ERRCHK_CUDA(cudaMemcpyToSymbolAsync(d_mesh_info, &value, sizeof(value), offset,
ERRCHK_CUDA_ALWAYS(cudaMemcpyToSymbolAsync(d_mesh_info, &value, sizeof(value), offset,
cudaMemcpyHostToDevice, device->streams[stream]));
return AC_SUCCESS;
}
@@ -147,12 +142,17 @@ acDeviceLoadIntUniform(const Device device, const Stream stream, const AcIntPara
const int value)
{
cudaSetDevice(device->id);
if (!is_valid(value)) {
fprintf(stderr, "WARNING: Passed an invalid value %d to device constant %s. Skipping.\n",
value, intparam_names[param]);
return AC_FAILURE;
}
if (param >= NUM_INT_PARAMS)
return AC_FAILURE;
const size_t offset = (size_t)&d_mesh_info.int_params[param] - (size_t)&d_mesh_info;
ERRCHK_CUDA(cudaMemcpyToSymbolAsync(d_mesh_info, &value, sizeof(value), offset,
ERRCHK_CUDA_ALWAYS(cudaMemcpyToSymbolAsync(d_mesh_info, &value, sizeof(value), offset,
cudaMemcpyHostToDevice, device->streams[stream]));
return AC_SUCCESS;
}
@@ -162,12 +162,79 @@ acDeviceLoadInt3Uniform(const Device device, const Stream stream, const AcInt3Pa
const int3 value)
{
cudaSetDevice(device->id);
if (!is_valid(value.x) || !is_valid(value.y) || !is_valid(value.z)) {
fprintf(
stderr,
"WARNING: Passed an invalid value (%d, %d, %def) to device constant %s. Skipping.\n",
value.x, value.y, value.z, int3param_names[param]);
return AC_FAILURE;
}
if (param >= NUM_INT3_PARAMS)
return AC_FAILURE;
const size_t offset = (size_t)&d_mesh_info.int3_params[param] - (size_t)&d_mesh_info;
ERRCHK_CUDA(cudaMemcpyToSymbolAsync(d_mesh_info, &value, sizeof(value), offset,
ERRCHK_CUDA_ALWAYS(cudaMemcpyToSymbolAsync(d_mesh_info, &value, sizeof(value), offset,
cudaMemcpyHostToDevice, device->streams[stream]));
return AC_SUCCESS;
}
AcResult
acDeviceLoadMeshInfo(const Device device, const AcMeshInfo device_config)
{
cudaSetDevice(device->id);
ERRCHK_ALWAYS(device_config.int_params[AC_nx] == device->local_config.int_params[AC_nx]);
ERRCHK_ALWAYS(device_config.int_params[AC_ny] == device->local_config.int_params[AC_ny]);
ERRCHK_ALWAYS(device_config.int_params[AC_nz] == device->local_config.int_params[AC_nz]);
ERRCHK_ALWAYS(device_config.int_params[AC_multigpu_offset] ==
device->local_config.int_params[AC_multigpu_offset]);
for (int i = 0; i < NUM_INT_PARAMS; ++i)
acDeviceLoadIntUniform(device, STREAM_DEFAULT, (AcIntParam)i, device_config.int_params[i]);
for (int i = 0; i < NUM_INT3_PARAMS; ++i)
acDeviceLoadInt3Uniform(device, STREAM_DEFAULT, (AcInt3Param)i,
device_config.int3_params[i]);
for (int i = 0; i < NUM_REAL_PARAMS; ++i)
acDeviceLoadScalarUniform(device, STREAM_DEFAULT, (AcRealParam)i,
device_config.real_params[i]);
for (int i = 0; i < NUM_REAL3_PARAMS; ++i)
acDeviceLoadVectorUniform(device, STREAM_DEFAULT, (AcReal3Param)i,
device_config.real3_params[i]);
return AC_SUCCESS;
}
AcResult
acDeviceLoadDefaultUniforms(const Device device)
{
cudaSetDevice(device->id);
// clang-format off
// Scalar
#define LOAD_DEFAULT_UNIFORM(X) acDeviceLoadScalarUniform(device, STREAM_DEFAULT, X, X##_DEFAULT_VALUE);
AC_FOR_USER_REAL_PARAM_TYPES(LOAD_DEFAULT_UNIFORM)
#undef LOAD_DEFAULT_UNIFORM
// Vector
#define LOAD_DEFAULT_UNIFORM(X) acDeviceLoadVectorUniform(device, STREAM_DEFAULT, X, X##_DEFAULT_VALUE);
AC_FOR_USER_REAL3_PARAM_TYPES(LOAD_DEFAULT_UNIFORM)
#undef LOAD_DEFAULT_UNIFORM
// Int
#define LOAD_DEFAULT_UNIFORM(X) acDeviceLoadIntUniform(device, STREAM_DEFAULT, X, X##_DEFAULT_VALUE);
AC_FOR_USER_INT_PARAM_TYPES(LOAD_DEFAULT_UNIFORM)
#undef LOAD_DEFAULT_UNIFORM
// Int3
#define LOAD_DEFAULT_UNIFORM(X) acDeviceLoadInt3Uniform(device, STREAM_DEFAULT, X, X##_DEFAULT_VALUE);
AC_FOR_USER_INT3_PARAM_TYPES(LOAD_DEFAULT_UNIFORM)
#undef LOAD_DEFAULT_UNIFORM
// clang-format on
ERRCHK_CUDA_KERNEL_ALWAYS();
return AC_SUCCESS;
}

15
src/core/kernels/kernels.h
View File

@@ -62,21 +62,6 @@ AcReal acKernelReduceVec(const cudaStream_t stream, const ReductionType rtype, c
const int3 end, const AcReal* vtxbuf0, const AcReal* vtxbuf1,
const AcReal* vtxbuf2, AcReal* scratchpad, AcReal* reduce_result);
AcResult acDeviceLoadMeshInfo(const Device device, const Stream stream,
const AcMeshInfo device_config);
AcResult acDeviceLoadScalarUniform(const Device device, const Stream stream,
const AcRealParam param, const AcReal value);
AcResult acDeviceLoadVectorUniform(const Device device, const Stream stream,
const AcReal3Param param, const AcReal3 value);
AcResult acDeviceLoadIntUniform(const Device device, const Stream stream, const AcIntParam param,
const int value);
AcResult acDeviceLoadInt3Uniform(const Device device, const Stream stream, const AcInt3Param param,
const int3 value);
#ifdef __cplusplus
} // extern "C"
#endif