AI/rocm-systems
2
0
Bifurcation 0
Code Tickets Demandes d'ajout Actions Paquets Projets Publications Wiki Activité

P4 to Git Change 1767752 by wchau@wchau_OCL_Linux on 2019/04/09 22:58:03

Parcourir la source 
SWDEV-165259 - Update OpenCL runtime to support MsgPack metadata
	- Add support for the V3 code objects

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/devkernel.cpp#19 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/devkernel.hpp#14 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/devprogram.cpp#39 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/devprogram.hpp#24 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#336 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.hpp#134 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palbe/inc/core/palCmdBuffer.h#63 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palbe/src/core/hw/gfxip/gfx6/gfx6ComputeCmdBuffer.cpp#63 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palbe/src/core/hw/gfxip/gfx9/gfx9ComputeCmdBuffer.cpp#69 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.cpp#77 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.hpp#27 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.cpp#90 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.cpp#76 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.hpp#21 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#130 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#52 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.hpp#27 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#103 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.hpp#47 edit
Cette révision appartient à :
foreman
2019-04-09 23:24:10 -04:00
Parent 0fb499bb71
révision 36a5f2a85f
15 fichiers modifiés avec 951 ajouts et 417 suppressions
Télécharger le Fichier Patch Télécharger le Fichier des Différences Développer tous les fichiers Réduire tous les fichiers
rocclr/runtime/device/devprogram.cpp
+19 -11
Voir le fichier
@@ -706,8 +706,6 @@ bool Program::compileImplLC(const std::string& sourceCode,
driverOptions.append(options->llvmOptions);
driverOptions.append(ProcessOptions(options));
// Force object code v2.
driverOptions.append(" -mno-code-object-v3");
// Set whole program mode
driverOptions.append(" -mllvm -amdgpu-early-inline-all -mllvm -amdgpu-prelink");
@@ -1532,8 +1530,6 @@ bool Program::linkImplLC(amd::option::Options* options) {
codegenOptions.append(" -mno-sram-ecc");
}
// Force object code v2.
codegenOptions.append(" -mno-code-object-v3");
// Set whole program mode
codegenOptions.append(" -mllvm -amdgpu-internalize-symbols -mllvm -amdgpu-early-inline-all");
@@ -1764,8 +1760,6 @@ bool Program::linkImplLC(amd::option::Options* options) {
std::ostream_iterator<std::string>(ostrstr, " "));
codegenOptions.append(" ").append(ostrstr.str());
// Force object code v2.
codegenOptions.append(" -mno-code-object-v3");
// Set whole program mode
codegenOptions.append(" -mllvm -amdgpu-internalize-symbols -mllvm -amdgpu-early-inline-all");
@@ -2880,8 +2874,19 @@ bool Program::createKernelMetadataMap() {
status = amd::Comgr::metadata_lookup(*metadata_, "Kernels", &kernelsMD);
if (status == AMD_COMGR_STATUS_SUCCESS) {
LogInfo("Using Code Object V2.");
hasKernelMD = true;
status = amd::Comgr::get_metadata_list_size(kernelsMD, &size);
codeObjectVer_ = 2;
}
else {
status = amd::Comgr::metadata_lookup(*metadata_, "amdhsa.kernels", &kernelsMD);
if (status == AMD_COMGR_STATUS_SUCCESS) {
LogInfo("Using Code Object V3.");
hasKernelMD = true;
codeObjectVer_ = 3;
status = amd::Comgr::get_metadata_list_size(kernelsMD, &size);
}
}
for (size_t i = 0; i < size && status == AMD_COMGR_STATUS_SUCCESS; i++) {
@@ -2896,7 +2901,9 @@ bool Program::createKernelMetadataMap() {
if (status == AMD_COMGR_STATUS_SUCCESS) {
hasKernelNode = true;
status = amd::Comgr::metadata_lookup(kernelNode, "Name", &nameMeta);
status = amd::Comgr::metadata_lookup(kernelNode,
(codeObjectVer() == 2) ? "Name" : ".name",
&nameMeta);
}
if (status == AMD_COMGR_STATUS_SUCCESS) {
@@ -2970,9 +2977,10 @@ bool Program::FindGlobalVarSize(void* binary, size_t binSize) {
buildLog_ += "Error: object code with old metadata is not supported\n";
return false;
}
else if (note->n_type == 10 /* NT_AMD_AMDGPU_HSA_METADATA V2 */ &&
note->n_namesz == sizeof "AMD" &&
!memcmp(name, "AMD", note->n_namesz)) {
else if ((note->n_type == 10 /* NT_AMD_AMDGPU_HSA_METADATA V2 */ &&
note->n_namesz == sizeof "AMD" && !memcmp(name, "AMD", note->n_namesz)) ||
(note->n_type == 32 /* NT_AMD_AMDGPU_HSA_METADATA V3 */ &&
note->n_namesz == sizeof "AMDGPU" && !memcmp(name, "AMDGPU", note->n_namesz))) {
#if defined(USE_COMGR_LIBRARY)
amd_comgr_status_t status;
amd_comgr_data_t binaryData;