static
void
skc_raster_builder_cohort_grid_pfn_execute(skc_grid_t const grid)
{
//
// ALLOCATED RESOURCES
//
// path_ids i
// raster_ids i
// transforms i
// clips i
// fill_cmds i
// cq -
// cohort atomics -
// cmds -
// keys -
// meta -
//
// allocate the cohort
struct skc_raster_cohort * const cohort = skc_grid_get_data(grid);
// get impl
struct skc_raster_builder_impl * const impl = cohort->impl;
struct skc_runtime * const runtime = impl->runtime;
// acquire in-order cq
cohort->cq = skc_runtime_acquire_cq_in_order(runtime);
// alloc the snapshot -- could be zero-sized
skc_extent_phw1g_tdrNs_snap_alloc(runtime,
&impl->fill_cmds,
&cohort->fill_cmds,
cohort->cq,NULL);
// flush the cq to get the fill running
// cl(Flush(cohort->cq));
// create split atomics
skc_extent_thr_tdrw_alloc(runtime,&cohort->atomics,sizeof(struct skc_raster_cohort_atomic));
// zero the atomics
skc_extent_thr_tdrw_zero(&cohort->atomics,cohort->cq,NULL);
// get config
struct skc_config const * const config = runtime->config;
// acquire device-side extents
skc_extent_tdrw_alloc(runtime,
&cohort->cmds,
sizeof(union skc_cmd_rasterize) * config->raster_cohort.expand.cmds);
//
// FILLS EXPAND
//
// need result of cmd counts before launching RASTERIZE grids
//
// - OpenCL 1.2: copy atomic counters back to host and launch RASTERIZE grids from host
// - OpenCL 2.x: have a kernel size and launch RASTERIZE grids from device
// - or launch a device-wide grid that feeds itself but that's unsatisfying
//
// how many commands? could be zero
skc_uint const work_size = skc_extent_ring_snap_count(cohort->fill_cmds.snap);
if (work_size > 0)
{
cl(SetKernelArg(impl->kernels.fills_expand,0,SKC_CL_ARG(impl->runtime->block_pool.blocks.drw)));
cl(SetKernelArg(impl->kernels.fills_expand,1,SKC_CL_ARG(cohort->atomics.drw)));
cl(SetKernelArg(impl->kernels.fills_expand,2,SKC_CL_ARG(runtime->handle_pool.map.drw)));
cl(SetKernelArg(impl->kernels.fills_expand,3,SKC_CL_ARG(cohort->fill_cmds.drN)));
cl(SetKernelArg(impl->kernels.fills_expand,4,SKC_CL_ARG(cohort->cmds.drw)));
skc_device_enqueue_kernel(runtime->device,
SKC_DEVICE_KERNEL_ID_FILLS_EXPAND,
cohort->cq,
impl->kernels.fills_expand,
work_size,
0,NULL,NULL);
}
//
// copyback number of rasterization commands
//
cl_event complete;
skc_extent_thr_tdrw_read(&cohort->atomics,cohort->cq,&complete);
cl(SetEventCallback(complete,CL_COMPLETE,skc_raster_cohort_fills_expand_cb,grid));
cl(ReleaseEvent(complete));
// flush command queue
cl(Flush(cohort->cq));
//
// ALLOCATED RESOURCES
//
// path_ids i
// raster_ids i
// transforms i
// clips i
// fill_cmds s
// cq a
// cohort atomics a
// cmds a
// keys -
// meta -
//
}
static
void
skc_raster_cohort_rasterize(skc_grid_t const grid)
{
//
// ALLOCATED RESOURCES
//
// path_ids i
// raster_ids i
// transforms i
// clips i
// fill_cmds s
// cq a
// cohort atomics a
// cmds a
// cmds_quad a
// cmds_cubic a
// keys -
// meta -
// use the backpointers
struct skc_raster_cohort * const cohort = skc_grid_get_data(grid);
struct skc_raster_builder_impl * const impl = cohort->impl;
struct skc_runtime * const runtime = impl->runtime;
//
// RELEASED RESOURCES
//
// cmds snap
//
// release the cmds extent and snap since it's only used by the expand stage
skc_extent_phw1g_tdrNs_snap_free(runtime,&cohort->fill_cmds);
//
// NEW ALLOCATED RESOURCES
//
// transforms snap
// clips snap
// ttrk keys
//
skc_extent_phw1g_tdrNs_snap_alloc(runtime,
&impl->transforms,
&cohort->transforms,
cohort->cq,NULL);
skc_extent_phw1g_tdrNs_snap_alloc(runtime,
&impl->clips,
&cohort->clips,
cohort->cq,NULL);
// acquire device-side extent
skc_extent_tdrw_alloc(runtime,
&cohort->keys,
sizeof(union skc_ttrk) * runtime->config->raster_cohort.rasterize.keys);
// skc_extent_thrw_tdrw_alloc(runtime,
// &cohort->keys,
// sizeof(union skc_ttrk) * runtime->config->raster_cohort.rasterize.keys);
//
// acquire out-of-order command queue
//
// and launch up to 3 kernels
//
// for each kernel:
//
// set runtime "global" kernel args:
//
// - block pool atomics
// - block pool extent
//
// set cohort "local" kernel args:
//
// - atomics
// - cmds
//
// enqueue barrier
// enqueue copy back of atomics on the command queue
// set callback on copy back event
// release command queue
//
struct skc_raster_cohort_atomic const * const atomics = cohort->atomics.hr;
if (atomics->cmds > 0)
{
cl(SetKernelArg(impl->kernels.rasterize_all,0,SKC_CL_ARG(runtime->block_pool.atomics.drw)));
cl(SetKernelArg(impl->kernels.rasterize_all,1,SKC_CL_ARG(runtime->block_pool.blocks.drw)));
cl(SetKernelArg(impl->kernels.rasterize_all,2,SKC_CL_ARG(runtime->block_pool.ids.drw)));
cl(SetKernelArg(impl->kernels.rasterize_all,3,SKC_CL_ARG(runtime->block_pool.size->ring_mask)));
cl(SetKernelArg(impl->kernels.rasterize_all,4,SKC_CL_ARG(cohort->atomics.drw)));
cl(SetKernelArg(impl->kernels.rasterize_all,5,SKC_CL_ARG(cohort->keys.drw)));
cl(SetKernelArg(impl->kernels.rasterize_all,6,SKC_CL_ARG(cohort->transforms.drN)));
cl(SetKernelArg(impl->kernels.rasterize_all,7,SKC_CL_ARG(cohort->clips.drN)));
cl(SetKernelArg(impl->kernels.rasterize_all,8,SKC_CL_ARG(cohort->cmds.drw)));
cl(SetKernelArg(impl->kernels.rasterize_all,9,SKC_CL_ARG(atomics->cmds)));
skc_device_enqueue_kernel(runtime->device,
SKC_DEVICE_KERNEL_ID_RASTERIZE_ALL,
cohort->cq,
impl->kernels.rasterize_all,
atomics->cmds,
0,NULL,NULL);
}
//
// copyback number of TTSK keys
//
cl_event complete;
skc_extent_thr_tdrw_read(&cohort->atomics,cohort->cq,&complete);
cl(SetEventCallback(complete,CL_COMPLETE,skc_raster_cohort_rasterize_cb,grid));
cl(ReleaseEvent(complete));
// flush command queue
cl(Flush(cohort->cq));
//
// ALLOCATED RESOURCES
//
// path_ids i
// raster_ids i
// transforms a
// clips a
// fill_cmds -
// cq a
// cohort atomics a
// cmds a
// keys a
// meta -
}
/* void setKernelArg (in long aIndex, in nsIVariant aValue, [optional] in long aType); */
NS_IMETHODIMP WebCLKernel::SetKernelArg(PRInt32 aIndex, nsIVariant *aValue, PRInt32 aType, JSContext *cx)
{
D_METHOD_START;
NS_ENSURE_ARG_POINTER (aValue);
NS_ENSURE_ARG_POINTER (cx);
nsresult rv = NS_OK;
if (aType == types::UNKNOWN)
{
PRUint16 variantType = 0;
rv = aValue->GetDataType (&variantType);
// If the type is unknown or user chose not to use explicit type, we'll try
// to guess it based on the type of the variant.
switch (variantType)
{
case nsIDataType::VTYPE_INT8: return SetKernelArg (aIndex, aValue, types::CHAR, cx);
case nsIDataType::VTYPE_INT16: return SetKernelArg (aIndex, aValue, types::SHORT, cx);
case nsIDataType::VTYPE_INT32: return SetKernelArg (aIndex, aValue, types::INT, cx);
case nsIDataType::VTYPE_INT64: return SetKernelArg (aIndex, aValue, types::LONG, cx);
case nsIDataType::VTYPE_UINT8: return SetKernelArg (aIndex, aValue, types::UCHAR, cx);
case nsIDataType::VTYPE_UINT16: return SetKernelArg (aIndex, aValue, types::USHORT, cx);
case nsIDataType::VTYPE_UINT32: return SetKernelArg (aIndex, aValue, types::UINT, cx);
case nsIDataType::VTYPE_UINT64: return SetKernelArg (aIndex, aValue, types::ULONG, cx);
case nsIDataType::VTYPE_FLOAT: return SetKernelArg (aIndex, aValue, types::FLOAT, cx);
case nsIDataType::VTYPE_DOUBLE: return SetKernelArg (aIndex, aValue, types::DOUBLE, cx);
case nsIDataType::VTYPE_BOOL: return SetKernelArg (aIndex, aValue, types::BOOL, cx);
case nsIDataType::VTYPE_CHAR:
case nsIDataType::VTYPE_WCHAR: return SetKernelArg (aIndex, aValue, types::CHAR, cx);
case nsIDataType::VTYPE_CHAR_STR:
case nsIDataType::VTYPE_WCHAR_STR:
case nsIDataType::VTYPE_UTF8STRING:
case nsIDataType::VTYPE_ASTRING:
case nsIDataType::VTYPE_CSTRING: return SetKernelArg (aIndex, aValue, types::STRING, cx);
case nsIDataType::VTYPE_INTERFACE:
case nsIDataType::VTYPE_INTERFACE_IS:
{
// Try conversions to supported WebCL interfaces.
nsCOMPtr<IWebCLMemoryObject> memObj;
rv = aValue->GetAsISupports (getter_AddRefs(memObj));
if (NS_SUCCEEDED (rv)) return SetKernelArg (aIndex, aValue, types::MEMORY_OBJECT, cx);
nsCOMPtr<IWebCLSampler> samplerObj;
rv = aValue->GetAsISupports (getter_AddRefs(samplerObj));
if (NS_SUCCEEDED (rv)) return SetKernelArg (aIndex, aValue, types::SAMPLER, cx);
// None found, intentional leak to default
}
case nsIDataType::VTYPE_ARRAY:
case nsIDataType::VTYPE_EMPTY_ARRAY:
D_LOG (LOG_LEVEL_ERROR, "Array support not implemented.");
WebCL_reportJSError (cx, "WebCLKernel::setKernelArg: Array support not implemented.");
return WEBCL_XPCOM_ERROR; //NS_ERROR_INVALID_ARG;
case nsIDataType::VTYPE_EMPTY:
case nsIDataType::VTYPE_VOID:
case nsIDataType::VTYPE_ID:
default:
D_LOG (LOG_LEVEL_ERROR,
"Unable to guess type from variant (type %u) and no type given by the user.",
variantType);
WebCL_reportJSError (cx, "WebCLKernel::setKernelArg: Unable to guess type from variant (type %u) and no type given by the user.",
variantType);
return WEBCL_XPCOM_ERROR; //NS_ERROR_INVALID_ARG;
}
}
size_t sze = 0;
void* value = 0;
switch (aType)
{
case types::BYTE:
case types::CHAR:
value = (void*)malloc (sze = sizeof (cl_char));
if (value)
rv = variantToCLChar (aValue, (cl_char*)value);
break;
case types::UCHAR:
value = (void*)malloc (sze = sizeof (cl_uchar));
if (value)
rv = variantToCLUChar (aValue, (cl_uchar*)value);
break;
case types::SHORT:
value = (void*)malloc (sze = sizeof (cl_short));
if (value)
rv = variantToCLShort (aValue, (cl_short*)value);
break;
case types::USHORT:
value = (void*)malloc (sze = sizeof (cl_ushort));
if (value)
rv = variantToCLUShort (aValue, (cl_ushort*)value);
break;
case types::BUILD_STATUS:
case types::INT:
value = (void*)malloc (sze = sizeof (cl_int));
if (value)
rv = variantToCLInt (aValue, (cl_int*)value);
break;
case types::ADRESSING_MODE:
case types::CHANNEL_ORDER:
case types::CHANNEL_TYPE:
case types::COMMAND_TYPE:
case types::DEVICE_LOCAL_MEM_TYPE:
case types::DEVICE_MEM_CACHE_TYPE:
case types::FILTER_MODE:
case types::GL_OBJECT_TYPE:
case types::MEM_OBJECT_TYPE:
case types::UINT:
value = (void*)malloc (sze = sizeof (cl_uint));
if (value)
rv = variantToCLUInt (aValue, (cl_uint*)value);
break;
case types::LONG:
value = (void*)malloc (sze = sizeof (cl_long));
if (value)
rv = variantToCLLong (aValue, (cl_long*)value);
break;
case types::COMMAND_QUEUE_PROPERTIES: // bitfield
case types::DEVICE_EXEC_CAPABILITIES: // bitfield
case types::DEVICE_FP_CONFIG: // bitfield
case types::DEVICE_TYPE: // bitfield
case types::MAP_FLAGS: // bitfield
case types::MEM_FLAGS: // bitfield
case types::ULONG:
value = (void*)malloc (sze = sizeof (cl_ulong));
if (value)
rv = variantToCLULong (aValue, (cl_ulong*)value);
break;
case types::BOOL:
value = (void*)malloc (sze = sizeof (cl_bool));
if (value)
rv = variantToCLBool (aValue, (cl_bool*)value);
break;
case types::SIZE_T:
value = (void*)malloc (sze = sizeof (size_t));
if (value)
rv = variantToCLSizeT (aValue, (size_t*)value);
break;
case types::HALF:
value = (void*)malloc (sze = sizeof (cl_half));
if (value)
rv = variantToCLHalf (aValue, (cl_half*)value);
break;
case types::FLOAT:
value = (void*)malloc (sze = sizeof (cl_float));
if (value)
rv = variantToCLFloat (aValue, (cl_float*)value);
break;
case types::DOUBLE:
value = (void*)malloc (sze = sizeof (cl_double));
if (value)
rv = variantToCLDouble (aValue, (cl_double*)value);
break;
case types::STRING:
{
nsCString str;
rv = aValue->GetAsACString (str);
if (NS_SUCCEEDED (rv))
{
cl_int err = mWrapper->setKernelArg (mInternal, aIndex, str.Length () + 1,
(void*)str.get ());
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (err);
}
break;
}
case types::BYTE_V:
case types::CHAR_V:
case types::UCHAR_V:
case types::SHORT_V:
case types::USHORT_V:
case types::CONTEXT_PROPERTIES:
case types::INT_V:
case types::UINT_V:
case types::LONG_V:
case types::ULONG_V:
case types::BOOL_V:
case types::SIZE_T_V:
case types::HALF_V:
case types::FLOAT_V:
case types::DOUBLE_V:
case types::STRING_V:
D_LOG (LOG_LEVEL_ERROR, "Array types are not supported.");
WebCL_reportJSError (cx, "WebCLKernel::setKernelArg: Array types are not supported.");
return WEBCL_XPCOM_ERROR; //NS_ERROR_NOT_IMPLEMENTED;
case types::MEMORY_OBJECT:
{
nsCOMPtr<nsISupports> isu;
rv = aValue->GetAsISupports (getter_AddRefs(isu));
if (NS_FAILED (rv)) break;
nsCOMPtr<WebCLMemoryObject> memObject = do_QueryInterface (isu, &rv);
if (NS_FAILED (rv)) break;
cl_mem wrapped = memObject->getInternal ();
cl_int wrErr = mWrapper->setKernelArg (mInternal, aIndex, sizeof(cl_mem),
(void*)&wrapped);
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (wrErr);
return NS_OK;
}
case types::SAMPLER:
{
nsCOMPtr<nsISupports> isu;
rv = aValue->GetAsISupports (getter_AddRefs(isu));
if (NS_FAILED (rv)) break;
nsCOMPtr<WebCLSampler> sampler = do_QueryInterface (isu, &rv);
if (NS_FAILED (rv)) break;
cl_sampler wrapped = sampler->getInternal ();
cl_int wrErr = mWrapper->setKernelArg (mInternal, aIndex, sizeof(cl_sampler),
(void*)&wrapped);
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (wrErr);
return NS_OK;
}
case types::PLATFORM:
{
nsCOMPtr<nsISupports> isu;
rv = aValue->GetAsISupports (getter_AddRefs(isu));
if (NS_FAILED (rv)) break;
nsCOMPtr<WebCLPlatform> platform = do_QueryInterface (isu, &rv);
if (NS_FAILED (rv)) break;
cl_platform_id wrapped = platform->getInternal ();
cl_int wrErr = mWrapper->setKernelArg (mInternal, aIndex, sizeof(cl_platform_id),
(void*)&wrapped);
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (wrErr);
return NS_OK;
}
case types::DEVICE:
{
nsCOMPtr<nsISupports> isu;
rv = aValue->GetAsISupports (getter_AddRefs(isu));
if (NS_FAILED (rv)) break;
nsCOMPtr<WebCLDevice> device = do_QueryInterface (isu, &rv);
if (NS_FAILED (rv)) break;
cl_device_id wrapped = device->getInternal ();
cl_int wrErr = mWrapper->setKernelArg (mInternal, aIndex, sizeof(cl_device_id),
(void*)&wrapped);
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (wrErr);
return NS_OK;
}
case types::CONTEXT:
{
nsCOMPtr<nsISupports> isu;
rv = aValue->GetAsISupports (getter_AddRefs(isu));
if (NS_FAILED (rv)) break;
nsCOMPtr<WebCLContext> context = do_QueryInterface (isu, &rv);
if (NS_FAILED (rv)) break;
cl_context wrapped = context->getInternal ();
cl_int wrErr = mWrapper->setKernelArg (mInternal, aIndex, sizeof(cl_context),
(void*)&wrapped);
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (wrErr);
return NS_OK;
}
case types::COMMAND_QUEUE:
{
nsCOMPtr<nsISupports> isu;
rv = aValue->GetAsISupports (getter_AddRefs(isu));
if (NS_FAILED (rv)) break;
nsCOMPtr<WebCLCommandQueue> cmdQueue = do_QueryInterface (isu, &rv);
if (NS_FAILED (rv)) break;
cl_command_queue wrapped = cmdQueue->getInternal ();
cl_int wrErr = mWrapper->setKernelArg (mInternal, aIndex, sizeof(cl_command_queue),
(void*)&wrapped);
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (wrErr);
return NS_OK;
}
case types::PROGRAM:
{
nsCOMPtr<nsISupports> isu;
rv = aValue->GetAsISupports (getter_AddRefs(isu));
if (NS_FAILED (rv)) break;
nsCOMPtr<WebCLProgram> program = do_QueryInterface (isu, &rv);
if (NS_FAILED (rv)) break;
cl_program wrapped = program->getInternal ();
cl_int wrErr = mWrapper->setKernelArg (mInternal, aIndex, sizeof(cl_program),
(void*)&wrapped);
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (wrErr);
return NS_OK;
}
case types::KERNEL:
{
nsCOMPtr<nsISupports> isu;
rv = aValue->GetAsISupports (getter_AddRefs(isu));
if (NS_FAILED (rv)) break;
nsCOMPtr<WebCLKernel> kernel = do_QueryInterface (isu, &rv);
if (NS_FAILED (rv)) break;
cl_kernel wrapped = kernel->getInternal ();
cl_int wrErr = mWrapper->setKernelArg (mInternal, aIndex, sizeof(cl_kernel),
(void*)&wrapped);
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (wrErr);
return NS_OK;
}
case types::EVENT:
{
nsCOMPtr<nsISupports> isu;
rv = aValue->GetAsISupports (getter_AddRefs(isu));
if (NS_FAILED (rv)) break;
nsCOMPtr<WebCLEvent> event = do_QueryInterface (isu, &rv);
if (NS_FAILED (rv)) break;
cl_event wrapped = event->getInternal ();
cl_int wrErr = mWrapper->setKernelArg (mInternal, aIndex, sizeof(cl_event),
(void*)&wrapped);
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (wrErr);
return NS_OK;
}
default:
D_LOG (LOG_LEVEL_ERROR, "Unsupported type %d at argument index %u", aType, aIndex);
WebCL_reportJSError (cx, "WebCLKernel::setKernelArg: Unsupported type %d at argument index %u.", aType, aIndex);
//rv = NS_ERROR_INVALID_ARG;
return WEBCL_XPCOM_ERROR;
}
if (NS_SUCCEEDED (rv))
{
if (value)
{
cl_int err = mWrapper->setKernelArg (mInternal, aIndex, sze, value);
ENSURE_LIB_WRAPPER_SUCCESS (mWrapper);
ENSURE_CL_OP_SUCCESS (err);
}
else
{
D_LOG (LOG_LEVEL_ERROR, "Memory allocation failed for kernel argument at index %d.", aIndex);
WebCL_reportJSError (cx, "WebCLKernel::setKernelArg: Memory allocation failed for kernel argument at index %d.", aIndex);
rv = WEBCL_XPCOM_ERROR; //NS_ERROR_OUT_OF_MEMORY;
}
}
else
{
D_LOG (LOG_LEVEL_ERROR, "Failed to convert kernel argument at index %d.", aIndex);
WebCL_reportJSError (cx, "WebCLKernel::setKernelArg: Failed to convert kernel argument at index %d.", aIndex);
rv = WEBCL_XPCOM_ERROR;
}
if (value)
free (value);
return rv;
}
static
void
skc_raster_cohort_sort_prefix(skc_grid_t const grid)
{
//
// ALLOCATED RESOURCES
//
// path_ids i
// raster_ids i
// transforms a
// clips a
// fill_cmds -
// cq a
// cohort atomics a
// cmds a
// keys a
// meta -
//
// use the backpointers
struct skc_raster_cohort * const cohort = skc_grid_get_data(grid);
struct skc_raster_builder_impl * const impl = cohort->impl;
struct skc_runtime * const runtime = impl->runtime;
// release transforms
skc_extent_phw1g_tdrNs_snap_free(runtime,&cohort->transforms);
// release clips
skc_extent_phw1g_tdrNs_snap_free(runtime,&cohort->clips);
// release expanded cmds
skc_extent_tdrw_free(runtime,&cohort->cmds);
// alloc the snapshost -- could be zero-sized
skc_extent_phrwg_tdrNs_snap_alloc(runtime,
&impl->raster_ids,
&cohort->raster_ids,
cohort->cq,NULL);
// will never be zero
skc_uint const rasters = skc_extent_ring_snap_count(cohort->raster_ids.snap);
// acquire fixed-size device-side extent
skc_extent_tdrw_alloc(runtime,
&cohort->metas,
sizeof(struct skc_raster_cohort_meta));
// skc_extent_thrw_tdrw_alloc(runtime,
// &cohort->metas,
// sizeof(struct skc_raster_cohort_meta));
// zero the metas
skc_extent_tdrw_zero(&cohort->metas,cohort->cq,NULL);
// get the read-only host copy of the device atomics
struct skc_raster_cohort_atomic const * const atomics = cohort->atomics.hr;
//
// SORT
//
if (atomics->keys > 0)
{
#ifndef NDEBUG
fprintf(stderr,"raster cohort sort: %u\n",atomics->keys);
#endif
//
//
//
uint32_t keys_padded_in, keys_padded_out;
hs_cl_pad(runtime->hs,atomics->keys,&keys_padded_in,&keys_padded_out);
hs_cl_sort(runtime->hs,
cohort->cq,
0,NULL,NULL,
cohort->keys.drw,
NULL,
atomics->keys,
keys_padded_in,
keys_padded_out,
false);
cl(SetKernelArg(impl->kernels.segment,0,SKC_CL_ARG(cohort->keys.drw)));
cl(SetKernelArg(impl->kernels.segment,1,SKC_CL_ARG(cohort->metas.drw)));
#ifndef NDEBUG
fprintf(stderr,"post-sort\n");
#endif
// find start of each tile
skc_device_enqueue_kernel(runtime->device,
SKC_DEVICE_KERNEL_ID_SEGMENT_TTRK,
cohort->cq,
impl->kernels.segment,
atomics->keys,
0,NULL,NULL);
#ifndef NDEBUG
fprintf(stderr,"post-segment\n");
#endif
//
// DELETE ALL THIS WHEN READY
//
#if 0
//
//
//
cl(Finish(cohort->cq));
// map keys to host
union skc_ttrk * const keys = skc_extent_thrw_tdrw_map(&cohort->keys,
cohort->cq,
NULL);
// map meta to host
struct skc_raster_cohort_meta * const metas = skc_extent_thrw_tdrw_map(&cohort->metas,
cohort->cq,
NULL);
// block until done
cl(Finish(cohort->cq));
// sort keys
qsort(keys,atomics->keys,sizeof(*keys),cmp64);
// mask to determine if rk id is a new block
skc_uint const subblock_mask = runtime->config->block.subblocks - 1;
//
// some counters
//
union skc_raster_cohort_meta_in meta_in = {
.blocks = 0,
.offset = 0,
.pk = 0,
.rk = 0
};
// get first key
union skc_ttrk curr = keys[0];
skc_uint ii=0, jj=0;
// for all TTRK keys
while (true)
{
// increment ttrk count
meta_in.rk += 1;
// was this a new block?
if ((curr.u32v2.lo & subblock_mask) == 0)
meta_in.blocks += 1;
// break if we're out of keys
if (++ii >= atomics->keys)
break;
// otherwise, process next key
union skc_ttrk const next = keys[ii];
// if new cohort then save curr meta and init next meta
if (next.cohort != curr.cohort)
{
fprintf(stderr,"[ %u, %u, %u, %u ]\n",
meta_in.blocks,
meta_in.offset,
meta_in.pk,
meta_in.rk);
// store back to buffer
metas->inout[curr.cohort].in = meta_in;
// update meta_in
meta_in.blocks = 0;
meta_in.offset = ii;
meta_in.pk = 0;
meta_in.rk = 0;
}
// otherwise, if same y but new x then increment TTPK count
else if ((next.y == curr.y) && (next.x != curr.x))
{
meta_in.pk += 1;
#if 0
fprintf(stderr,"%3u : %3u : ( %3u, %3u ) -> ( %3u )\n",
jj++,curr.cohort,curr.y,curr.x,next.x);
#endif
}
#if 0
fprintf(stderr,"( %3u, %3u )\n",next.y,next.x);
#endif
curr = next;
}
fprintf(stderr,"[ %u, %u, %u, %u ]\n",
meta_in.blocks,
meta_in.offset,
meta_in.pk,
meta_in.rk);
// store back to buffer
metas->inout[curr.cohort].in = meta_in;
// unmap
skc_extent_thrw_tdrw_unmap(&cohort->keys,
keys,
cohort->cq,
NULL);
// unmap
skc_extent_thrw_tdrw_unmap(&cohort->metas,
metas,
cohort->cq,
NULL);
#endif
}
#ifndef NDEBUG
fprintf(stderr,"rasters_alloc: %u\n",rasters);
#endif
//
// RASTER ALLOC/INIT
//
cl(SetKernelArg(impl->kernels.rasters_alloc,0,SKC_CL_ARG(runtime->block_pool.atomics.drw)));
cl(SetKernelArg(impl->kernels.rasters_alloc,1,SKC_CL_ARG(runtime->block_pool.ids.drw)));
cl(SetKernelArg(impl->kernels.rasters_alloc,2,SKC_CL_ARG(runtime->block_pool.size->ring_mask)));
cl(SetKernelArg(impl->kernels.rasters_alloc,3,SKC_CL_ARG(runtime->handle_pool.map.drw)));
cl(SetKernelArg(impl->kernels.rasters_alloc,4,SKC_CL_ARG(cohort->metas.drw)));
cl(SetKernelArg(impl->kernels.rasters_alloc,5,SKC_CL_ARG(cohort->raster_ids.drN)));
cl(SetKernelArg(impl->kernels.rasters_alloc,6,SKC_CL_ARG(rasters)));
skc_device_enqueue_kernel(runtime->device,
SKC_DEVICE_KERNEL_ID_RASTERS_ALLOC,
cohort->cq,
impl->kernels.rasters_alloc,
rasters,
0,NULL,NULL);
#ifndef NDEBUG
fprintf(stderr,"post-alloc\n");
#endif
//
// PREFIX
//
cl(SetKernelArg(impl->kernels.prefix,0,SKC_CL_ARG(runtime->block_pool.atomics.drw)));
cl(SetKernelArg(impl->kernels.prefix,1,SKC_CL_ARG(runtime->block_pool.ids.drw)));
cl(SetKernelArg(impl->kernels.prefix,2,SKC_CL_ARG(runtime->block_pool.blocks.drw)));
cl(SetKernelArg(impl->kernels.prefix,3,SKC_CL_ARG(runtime->block_pool.size->ring_mask)));
cl(SetKernelArg(impl->kernels.prefix,4,SKC_CL_ARG(cohort->keys.drw)));
cl(SetKernelArg(impl->kernels.prefix,5,SKC_CL_ARG(runtime->handle_pool.map.drw)));
cl(SetKernelArg(impl->kernels.prefix,6,SKC_CL_ARG(cohort->metas.drw)));
cl(SetKernelArg(impl->kernels.prefix,7,SKC_CL_ARG(rasters)));
cl_event complete;
skc_device_enqueue_kernel(runtime->device,
SKC_DEVICE_KERNEL_ID_PREFIX,
cohort->cq,
impl->kernels.prefix,
rasters,
0,NULL,
&complete);
cl(SetEventCallback(complete,CL_COMPLETE,skc_raster_cohort_prefix_cb,grid));
cl(ReleaseEvent(complete));
#ifndef NDEBUG
fprintf(stderr,"post-prefix\n");
#endif
// flush command queue
cl(Flush(cohort->cq));
//
// ALLOCATED RESOURCES
//
// path_ids a
// raster_ids a
// transforms -
// clips -
// fill_cmds -
// cq a
// cohort atomics a
// cmds -
// keys a
// meta a
//
}
static
void
skc_surface_grid_pfn_execute(skc_grid_t const grid)
{
struct skc_surface_render * const render = skc_grid_get_data(grid);
struct skc_surface_impl * const impl = render->impl;
struct skc_runtime * const runtime = impl->runtime;
// get the composition args
struct skc_composition_impl * const ci = render->composition->impl;
struct skc_place_atomics * const atomics = ci->atomics.hr;
if (atomics->offsets > 0)
{
// acquire the rbo/tex
if (render->fb->type != SKC_FRAMEBUFFER_CL_IMAGE2D)
cl(EnqueueAcquireGLObjects(impl->cq,1,&render->fb->mem,0,NULL,NULL));
// get the styling args
struct skc_styling_impl * const si = render->styling->impl;
cl(SetKernelArg(impl->kernels.render,0,SKC_CL_ARG(si->layers.drN)));
cl(SetKernelArg(impl->kernels.render,1,SKC_CL_ARG(si->groups.drN)));
cl(SetKernelArg(impl->kernels.render,2,SKC_CL_ARG(si->extras.drN)));
cl(SetKernelArg(impl->kernels.render,3,SKC_CL_ARG(ci->keys.drw)));
cl(SetKernelArg(impl->kernels.render,4,SKC_CL_ARG(atomics->keys)));
cl(SetKernelArg(impl->kernels.render,5,SKC_CL_ARG(ci->offsets.drw)));
cl(SetKernelArg(impl->kernels.render,6,SKC_CL_ARG(atomics->offsets)));
// block pool
cl(SetKernelArg(impl->kernels.render,7,SKC_CL_ARG(impl->runtime->block_pool.blocks.drw)));
// surface
cl(SetKernelArg(impl->kernels.render,8,SKC_CL_ARG(render->fb->mem)));
#if 1
// tile clip
cl(SetKernelArg(impl->kernels.render,9,sizeof(skc_uint4),render->clip));
#else
// surface pitch (height)
skc_uint const surface_pitch = SKC_SURFACE_HEIGHT;
cl(SetKernelArg(impl->kernels.render,9,SKC_CL_ARG(surface_pitch)));
// tile clip
cl(SetKernelArg(impl->kernels.render,10,sizeof(skc_uint4),render->clip));
#endif
// launch render kernel
skc_device_enqueue_kernel(runtime->device,
SKC_DEVICE_KERNEL_ID_RENDER,
impl->cq,
impl->kernels.render,
atomics->offsets,
0,NULL,NULL);
cl_event complete;
// give the rbo back
if (render->fb->type != SKC_FRAMEBUFFER_CL_IMAGE2D)
{
cl(EnqueueReleaseGLObjects(impl->cq,1,&render->fb->mem,0,NULL,&complete));
//
// blit the rbo to fbo0
//
render->fb->post_render(render->fb->interop);
//
// clear the rbo -- FIXME -- we shouldn't have to do this here
//
float const rgba[4] = { 1.0f, 1.0f, 1.0f, 1.0f };
uint32_t rect[4] = { 0 };
skc_interop_get_size(render->fb->interop,rect+2,rect+3);
skc_surface_debug_clear(impl,render->fb,rgba,rect);
}
// notify anyone listening...
cl(SetEventCallback(complete,CL_COMPLETE,skc_surface_render_cb,render));
cl(ReleaseEvent(complete));
// flush it
cl(Flush(impl->cq));
}
else
{
skc_surface_render_complete(render);
}
}
