P4 to Git Change 1398097 by lmoriche@lmoriche_opencl_dev2 on 2017/04/13 13:01:56
SWDEV-102733 - [OCL-LC-ROCm] Cmake build Write CMakeLists.txt to enable building with and without the DK environment - Change the coding convention of the runtime files. Use Google's Style (https://google.github.io/styleguide/cppguide.html). Affected files ... ... //depot/stg/opencl/drivers/opencl/.clang-format#1 add ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_agent_amd.h#2 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_command.cpp#13 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_context.cpp#53 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_counter.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_d3d10.cpp#15 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_d3d11.cpp#22 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_d3d9.cpp#32 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_debugger_amd.cpp#8 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_debugger_amd.h#7 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_device.cpp#61 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_event.cpp#10 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_execute.cpp#23 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_gl.cpp#53 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_icd.cpp#27 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_icd_amd.h#18 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_kernel.h#24 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_kernel_info_amd.cpp#3 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_kernel_info_amd.h#4 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_lqdflash_amd.cpp#17 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_lqdflash_amd.h#6 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_memobj.cpp#81 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_object.cpp#3 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_pipe.cpp#6 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_platform_amd.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_platform_amd.h#2 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_profile_amd.cpp#3 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_profile_amd.h#2 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_program.cpp#41 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_sampler.cpp#6 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_sdi_amd.cpp#3 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_sdi_amd.h#2 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_semaphore_amd.h#3 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_svm.cpp#20 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_thread_trace_amd.cpp#8 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_thread_trace_amd.h#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/appprofile.cpp#17 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/appprofile.hpp#12 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/blit.cpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/blit.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/blitcl.cpp#11 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpubinary.cpp#11 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpubinary.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpubuiltins.cpp#13 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpubuiltins.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpucommand.cpp#66 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpucommand.hpp#40 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpudevice.cpp#280 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpudevice.hpp#96 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpufeat.hpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpukernel.hpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpumapping.cpp#6 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpumapping.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpuprogram.cpp#70 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpuprogram.hpp#14 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpusettings.cpp#33 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpusettings.hpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cputables.hpp#5 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpuvirtual.cpp#26 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpuvirtual.hpp#13 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/device.cpp#209 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#284 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuappprofile.cpp#12 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuappprofile.hpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpubinary.cpp#58 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpubinary.hpp#27 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpublit.cpp#126 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpublit.hpp#41 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpucompiler.cpp#156 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuconstbuf.cpp#10 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuconstbuf.hpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpucounters.cpp#12 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpucounters.hpp#9 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudebugger.hpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudebugmanager.cpp#10 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudebugmanager.hpp#6 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudefs.hpp#147 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#567 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.hpp#163 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#318 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.hpp#126 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpumemory.cpp#131 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpumemory.hpp#50 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprintf.cpp#44 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprintf.hpp#15 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprogram.cpp#232 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprogram.hpp#69 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuresource.cpp#238 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuresource.hpp#87 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusched.hpp#19 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuschedcl.cpp#35 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuscsi.cpp#37 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.cpp#350 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.hpp#98 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gputhreadtrace.cpp#9 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gputhreadtrace.hpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gputimestamp.cpp#27 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gputimestamp.hpp#16 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gputrap.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#410 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.hpp#140 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuwavelimiter.cpp#13 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuwavelimiter.hpp#9 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/hwdebug.cpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/hwdebug.hpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palappprofile.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palappprofile.hpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palbinary.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palbinary.hpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palblit.cpp#13 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palblit.hpp#5 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palcompiler.cpp#15 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palconstbuf.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palconstbuf.hpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palcounters.cpp#11 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palcounters.hpp#9 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldebugger.hpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldebugmanager.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldebugmanager.hpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldefs.hpp#16 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#45 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.hpp#16 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldeviced3d10.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldeviced3d11.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldeviced3d9.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevicegl.cpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.cpp#34 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.hpp#11 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palmemory.cpp#13 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palmemory.hpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprintf.cpp#5 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprintf.hpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.cpp#39 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.hpp#17 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.cpp#28 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.hpp#12 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsched.hpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palschedcl.cpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.cpp#24 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.hpp#10 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palthreadtrace.cpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palthreadtrace.hpp#5 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paltimestamp.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paltimestamp.hpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paltrap.hpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#48 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.hpp#21 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palwavelimiter.cpp#3 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palwavelimiter.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/mesa_glinterop.h#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocappprofile.cpp#6 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocappprofile.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocbinary.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.cpp#17 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.hpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompiler.cpp#32 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompilerlib.cpp#6 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompilerlib.hpp#5 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdefs.hpp#10 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#48 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.hpp#20 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocglinterop.cpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocglinterop.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#22 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.hpp#16 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.cpp#15 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.hpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprintf.cpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprintf.hpp#5 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#64 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.hpp#23 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocregisters.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#17 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.hpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#34 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.hpp#10 edit ... //depot/stg/opencl/drivers/opencl/runtime/os/alloc.cpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/os/alloc.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/os/os.cpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/os/os.hpp#30 edit ... //depot/stg/opencl/drivers/opencl/runtime/os/os_posix.cpp#42 edit ... //depot/stg/opencl/drivers/opencl/runtime/os/os_win32.cpp#47 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/agent.cpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/agent.hpp#6 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/command.cpp#78 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/command.hpp#83 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/commandqueue.cpp#23 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/commandqueue.hpp#18 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/context.cpp#42 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/context.hpp#26 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/counter.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/interop.hpp#12 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/kernel.cpp#23 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/kernel.hpp#18 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/memory.cpp#127 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/memory.hpp#100 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/ndrange.cpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/ndrange.hpp#9 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/object.cpp#2 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/object.hpp#17 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/perfctr.hpp#5 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/program.cpp#86 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/program.hpp#41 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/runtime.cpp#35 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/runtime.hpp#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/sampler.hpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/platform/threadtrace.hpp#6 edit ... //depot/stg/opencl/drivers/opencl/runtime/thread/atomic.hpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/thread/monitor.cpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/thread/monitor.hpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/thread/semaphore.cpp#10 edit ... //depot/stg/opencl/drivers/opencl/runtime/thread/semaphore.hpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/thread/thread.cpp#14 edit ... //depot/stg/opencl/drivers/opencl/runtime/thread/thread.hpp#15 edit ... //depot/stg/opencl/drivers/opencl/runtime/top.hpp#26 edit ... //depot/stg/opencl/drivers/opencl/runtime/utils/concurrent.hpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/utils/debug.cpp#5 edit ... //depot/stg/opencl/drivers/opencl/runtime/utils/debug.hpp#7 edit ... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.cpp#16 edit ... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#271 edit ... //depot/stg/opencl/drivers/opencl/runtime/utils/macros.hpp#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/utils/util.hpp#12 edit ... //depot/stg/opencl/drivers/opencl/runtime/utils/versions.hpp#2150 edit
This commit is contained in:
foreman
@@ -16,74 +16,69 @@
|
||||
|
||||
namespace cpu {
|
||||
|
||||
ClBinary::FeatureCheckResult
|
||||
ClBinary::checkFeatures()
|
||||
{
|
||||
/* Validate that all cpu features of loaded binary target (i.e. elf_target) exists in current target.
|
||||
* If some of elf_target features doesn't exist in current target we fail the build since we assume that elf LLVM-IR and binary are
|
||||
* target specific and can't be recompiled to current target*/
|
||||
uint16_t target = (uint16_t)dev().settings().cpuFeatures_;
|
||||
uint16_t elf_target;
|
||||
amd::OclElf::oclElfPlatform platform;
|
||||
if (!elfIn()->getTarget(elf_target, platform)){
|
||||
LogError("Loading OCL CPU binary: incorrect format");
|
||||
return ERROR;
|
||||
ClBinary::FeatureCheckResult ClBinary::checkFeatures() {
|
||||
/* Validate that all cpu features of loaded binary target (i.e. elf_target) exists in current
|
||||
* target.
|
||||
* If some of elf_target features doesn't exist in current target we fail the build since we
|
||||
* assume that elf LLVM-IR and binary are
|
||||
* target specific and can't be recompiled to current target*/
|
||||
uint16_t target = (uint16_t)dev().settings().cpuFeatures_;
|
||||
uint16_t elf_target;
|
||||
amd::OclElf::oclElfPlatform platform;
|
||||
if (!elfIn()->getTarget(elf_target, platform)) {
|
||||
LogError("Loading OCL CPU binary: incorrect format");
|
||||
return ERROR;
|
||||
}
|
||||
uint64_t chip_options = 0x0;
|
||||
if (platform == amd::OclElf::COMPLIB_PLATFORM) {
|
||||
// BIF 3.0
|
||||
uint32_t flag;
|
||||
if (!elfIn()->getFlags(flag)) {
|
||||
LogError("Loading OCL CPU binary: incorrect format");
|
||||
return ERROR;
|
||||
}
|
||||
uint64_t chip_options=0x0;
|
||||
if (platform == amd::OclElf::COMPLIB_PLATFORM) {
|
||||
// BIF 3.0
|
||||
uint32_t flag;
|
||||
if (!elfIn()->getFlags(flag)) {
|
||||
LogError("Loading OCL CPU binary: incorrect format");
|
||||
return ERROR;
|
||||
}
|
||||
aclTargetInfo tgtInfo = aclGetTargetInfoFromChipID(LP64_SWITCH("x86", "x86-64"), flag, NULL);
|
||||
chip_options = aclGetChipOptions(tgtInfo) ;
|
||||
if (((target & chip_options) != chip_options) ||
|
||||
((elf_target == EM_386) && (strcmp(LP64_SWITCH("x86", "x86-64"), "x86") != 0)) ||
|
||||
((elf_target == EM_X86_64) && (strcmp(LP64_SWITCH("x86", "x86-64"), "x86-64") != 0))){
|
||||
LogError("Loading OCL CPU binary: different target");
|
||||
return ERROR;
|
||||
}
|
||||
aclTargetInfo tgtInfo = aclGetTargetInfoFromChipID(LP64_SWITCH("x86", "x86-64"), flag, NULL);
|
||||
chip_options = aclGetChipOptions(tgtInfo);
|
||||
if (((target & chip_options) != chip_options) ||
|
||||
((elf_target == EM_386) && (strcmp(LP64_SWITCH("x86", "x86-64"), "x86") != 0)) ||
|
||||
((elf_target == EM_X86_64) && (strcmp(LP64_SWITCH("x86", "x86-64"), "x86-64") != 0))) {
|
||||
LogError("Loading OCL CPU binary: different target");
|
||||
return ERROR;
|
||||
}
|
||||
else {
|
||||
// BIF 2.0
|
||||
if ((platform != amd::OclElf::CPU_PLATFORM) ||
|
||||
((target & elf_target) != elf_target)) {
|
||||
LogError("Loading OCL CPU binary: different target");
|
||||
return ERROR;
|
||||
}
|
||||
} else {
|
||||
// BIF 2.0
|
||||
if ((platform != amd::OclElf::CPU_PLATFORM) || ((target & elf_target) != elf_target)) {
|
||||
LogError("Loading OCL CPU binary: different target");
|
||||
return ERROR;
|
||||
}
|
||||
char* section;
|
||||
size_t sz;
|
||||
}
|
||||
char* section;
|
||||
size_t sz;
|
||||
|
||||
/* If current target has more cpu features than the one for which the binary was (notice it must have all features as in elf_target
|
||||
* due to previous check), we can benefit from recompiling the LLVM-IR if exists in binary (if there are errors, ignore them !).*/
|
||||
if (((platform == amd::OclElf::CPU_PLATFORM) &&
|
||||
((target ^ elf_target) != 0)) ||
|
||||
((platform == amd::OclElf::COMPLIB_PLATFORM) &&
|
||||
((target ^ chip_options) != 0))) {
|
||||
if (elfIn_->getSection(amd::OclElf::LLVMIR, §ion, &sz)) {
|
||||
if ((section != NULL) && (sz > 0)) {
|
||||
// hasDLL being false to force recompiling
|
||||
RECOMPILE;
|
||||
}
|
||||
}
|
||||
/* If current target has more cpu features than the one for which the binary was (notice it must
|
||||
* have all features as in elf_target
|
||||
* due to previous check), we can benefit from recompiling the LLVM-IR if exists in binary (if
|
||||
* there are errors, ignore them !).*/
|
||||
if (((platform == amd::OclElf::CPU_PLATFORM) && ((target ^ elf_target) != 0)) ||
|
||||
((platform == amd::OclElf::COMPLIB_PLATFORM) && ((target ^ chip_options) != 0))) {
|
||||
if (elfIn_->getSection(amd::OclElf::LLVMIR, §ion, &sz)) {
|
||||
if ((section != NULL) && (sz > 0)) {
|
||||
// hasDLL being false to force recompiling
|
||||
RECOMPILE;
|
||||
}
|
||||
}
|
||||
return OK;
|
||||
}
|
||||
return OK;
|
||||
}
|
||||
|
||||
bool
|
||||
ClBinary::loadX86(Program& program, std::string& dllName, bool& hasDLL)
|
||||
{
|
||||
hasDLL = false;
|
||||
bool ClBinary::loadX86(Program& program, std::string& dllName, bool& hasDLL) {
|
||||
hasDLL = false;
|
||||
|
||||
std::string tempName = amd::Os::getTempFileName();
|
||||
std::string tempName = amd::Os::getTempFileName();
|
||||
|
||||
dllName = tempName
|
||||
+ "." WINDOWS_SWITCH("dll",MACOS_SWITCH("dyld","so"));
|
||||
dllName = tempName + "." WINDOWS_SWITCH("dll", MACOS_SWITCH("dyld", "so"));
|
||||
|
||||
switch (checkFeatures()) {
|
||||
switch (checkFeatures()) {
|
||||
case ERROR:
|
||||
return false;
|
||||
case RECOMPILE:
|
||||
@@ -91,67 +86,63 @@ ClBinary::loadX86(Program& program, std::string& dllName, bool& hasDLL)
|
||||
case OK:
|
||||
// Fallthrough
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
char* section;
|
||||
size_t sz;
|
||||
char* section;
|
||||
size_t sz;
|
||||
|
||||
if (!elfIn_->getSection(amd::OclElf::DLL, §ion, &sz)) {
|
||||
LogError("Loading OCL CPU binary: error occured!");
|
||||
return false;
|
||||
}
|
||||
if (!elfIn_->getSection(amd::OclElf::DLL, §ion, &sz)) {
|
||||
LogError("Loading OCL CPU binary: error occured!");
|
||||
return false;
|
||||
}
|
||||
|
||||
if ((section == NULL) || (sz == 0)) {
|
||||
// hasDLL being false to force recompiling
|
||||
return true;
|
||||
}
|
||||
if ((section == NULL) || (sz == 0)) {
|
||||
// hasDLL being false to force recompiling
|
||||
return true;
|
||||
}
|
||||
|
||||
std::fstream f;
|
||||
f.open(dllName.c_str(), (std::fstream::out | std::fstream::binary));
|
||||
std::fstream f;
|
||||
f.open(dllName.c_str(), (std::fstream::out | std::fstream::binary));
|
||||
|
||||
if (!f.is_open()) {
|
||||
if (!f.is_open()) {
|
||||
#ifdef _WIN32
|
||||
amd::Os::unlink(tempName.c_str());
|
||||
#endif // _WIN32
|
||||
LogError("Loading OCL CPU binary: cannot open a file!");
|
||||
return false;
|
||||
}
|
||||
f.write(section, sz);
|
||||
f.close();
|
||||
amd::Os::unlink(tempName.c_str());
|
||||
#endif // _WIN32
|
||||
LogError("Loading OCL CPU binary: cannot open a file!");
|
||||
return false;
|
||||
}
|
||||
f.write(section, sz);
|
||||
f.close();
|
||||
|
||||
hasDLL = true;
|
||||
return true;
|
||||
hasDLL = true;
|
||||
return true;
|
||||
}
|
||||
|
||||
bool
|
||||
ClBinary::storeX86(Program& program, std::string& dllName)
|
||||
{
|
||||
std::fstream f;
|
||||
f.open(dllName.c_str(), (std::fstream::in | std::fstream::binary));
|
||||
if (!f.is_open()) {
|
||||
return false;
|
||||
}
|
||||
bool ClBinary::storeX86(Program& program, std::string& dllName) {
|
||||
std::fstream f;
|
||||
f.open(dllName.c_str(), (std::fstream::in | std::fstream::binary));
|
||||
if (!f.is_open()) {
|
||||
return false;
|
||||
}
|
||||
|
||||
f.seekg(0, std::fstream::end);
|
||||
size_t x86CodeSize = f.tellg();
|
||||
f.seekg(0, std::fstream::beg);
|
||||
f.seekg(0, std::fstream::end);
|
||||
size_t x86CodeSize = f.tellg();
|
||||
f.seekg(0, std::fstream::beg);
|
||||
|
||||
if (saveISA()) {
|
||||
char* x86Code = new char[x86CodeSize];
|
||||
f.read(x86Code, x86CodeSize);
|
||||
elfOut_->addSection(amd::OclElf::DLL, x86Code, x86CodeSize);
|
||||
delete [] x86Code;
|
||||
}
|
||||
f.close();
|
||||
return true;
|
||||
if (saveISA()) {
|
||||
char* x86Code = new char[x86CodeSize];
|
||||
f.read(x86Code, x86CodeSize);
|
||||
elfOut_->addSection(amd::OclElf::DLL, x86Code, x86CodeSize);
|
||||
delete[] x86Code;
|
||||
}
|
||||
f.close();
|
||||
return true;
|
||||
}
|
||||
|
||||
bool
|
||||
ClBinary::loadX86JIT(Program& program, bool& hasJITBinary)
|
||||
{
|
||||
bool ClBinary::loadX86JIT(Program& program, bool& hasJITBinary) {
|
||||
hasJITBinary = false;
|
||||
|
||||
switch (checkFeatures()) {
|
||||
switch (checkFeatures()) {
|
||||
case ERROR:
|
||||
return false;
|
||||
case RECOMPILE:
|
||||
@@ -159,66 +150,62 @@ ClBinary::loadX86JIT(Program& program, bool& hasJITBinary)
|
||||
case OK:
|
||||
// Fallthrough
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
char* section;
|
||||
size_t sz;
|
||||
char* section;
|
||||
size_t sz;
|
||||
|
||||
if (!elfIn_->getSection(amd::OclElf::JITBINARY, §ion, &sz)) {
|
||||
LogError("Loading OCL CPU JIT binary: error occured!");
|
||||
return false;
|
||||
}
|
||||
if (!elfIn_->getSection(amd::OclElf::JITBINARY, §ion, &sz)) {
|
||||
LogError("Loading OCL CPU JIT binary: error occured!");
|
||||
return false;
|
||||
}
|
||||
|
||||
if ((section == NULL) || (sz == 0)) {
|
||||
// force recompiling
|
||||
return true;
|
||||
}
|
||||
acl_error err = ACL_SUCCESS;
|
||||
program.setJITBinary(aclJITObjectImageCopy(program.compiler(), section, sz, &err));
|
||||
if (err != ACL_SUCCESS) {
|
||||
LogWarning("aclJITObjectImageCopy failed");
|
||||
return false;
|
||||
}
|
||||
hasJITBinary = true;
|
||||
if ((section == NULL) || (sz == 0)) {
|
||||
// force recompiling
|
||||
return true;
|
||||
}
|
||||
acl_error err = ACL_SUCCESS;
|
||||
program.setJITBinary(aclJITObjectImageCopy(program.compiler(), section, sz, &err));
|
||||
if (err != ACL_SUCCESS) {
|
||||
LogWarning("aclJITObjectImageCopy failed");
|
||||
return false;
|
||||
}
|
||||
hasJITBinary = true;
|
||||
return true;
|
||||
}
|
||||
|
||||
void checkDifference(const char* buf1, const char* buf2, size_t size) {
|
||||
for(size_t i = 0; i < size; ++i) {
|
||||
if(buf1[i] != buf2[i]) {
|
||||
printf("Index %d different",(int)i);
|
||||
for (size_t i = 0; i < size; ++i) {
|
||||
if (buf1[i] != buf2[i]) {
|
||||
printf("Index %d different", (int)i);
|
||||
return;
|
||||
}
|
||||
}
|
||||
}
|
||||
bool
|
||||
ClBinary::storeX86JIT(Program& program)
|
||||
{
|
||||
bool ClBinary::storeX86JIT(Program& program) {
|
||||
if (saveISA()) {
|
||||
acl_error err = ACL_SUCCESS;
|
||||
aclJITObjectImage objectImage = program.getJITBinary();
|
||||
size_t x86CodeSize = aclJITObjectImageSize(program.compiler(), objectImage, &err);
|
||||
if (err != ACL_SUCCESS) {
|
||||
LogWarning("aclJITObjectImageSize failed");
|
||||
return false;
|
||||
LogWarning("aclJITObjectImageSize failed");
|
||||
return false;
|
||||
}
|
||||
const char* x86CodePtr = aclJITObjectImageData(program.compiler(), objectImage, &err);
|
||||
if (err != ACL_SUCCESS) {
|
||||
LogWarning("aclJITObjectImageData failed");
|
||||
return false;
|
||||
LogWarning("aclJITObjectImageData failed");
|
||||
return false;
|
||||
}
|
||||
elfOut_->addSection(amd::OclElf::JITBINARY, x86CodePtr, x86CodeSize);
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
bool
|
||||
ClBinary::storeX86Asm(const char* buffer, size_t size)
|
||||
{
|
||||
bool ClBinary::storeX86Asm(const char* buffer, size_t size) {
|
||||
if (saveAS()) {
|
||||
elfOut_->addSection(amd::OclElf::ASTEXT, buffer, size);
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
@@ -17,69 +17,59 @@ class Device;
|
||||
class Program;
|
||||
|
||||
//! \class CPU binary
|
||||
class ClBinary : public device::ClBinary
|
||||
{
|
||||
public:
|
||||
//! Constructor
|
||||
ClBinary(const Device& dev) : device::ClBinary(dev) {}
|
||||
class ClBinary : public device::ClBinary {
|
||||
public:
|
||||
//! Constructor
|
||||
ClBinary(const Device& dev) : device::ClBinary(dev) {}
|
||||
|
||||
//! Destructor
|
||||
~ClBinary() {}
|
||||
//! Destructor
|
||||
~ClBinary() {}
|
||||
|
||||
//! Loads x86 executable code
|
||||
bool loadX86(
|
||||
Program& prorgam, //!< CPU Program object
|
||||
std::string& dllName, //!< Dll name of the CPU binary
|
||||
bool& hasDLL //!< indicate if the OCL binary has DLL
|
||||
);
|
||||
//! Loads x86 executable code
|
||||
bool loadX86(Program& prorgam, //!< CPU Program object
|
||||
std::string& dllName, //!< Dll name of the CPU binary
|
||||
bool& hasDLL //!< indicate if the OCL binary has DLL
|
||||
);
|
||||
|
||||
//! Stores x86 executable code
|
||||
bool storeX86(
|
||||
Program& program, //!< CPU Program object
|
||||
std::string& dllName //!< Dll name for the binary
|
||||
);
|
||||
//! Stores x86 executable code
|
||||
bool storeX86(Program& program, //!< CPU Program object
|
||||
std::string& dllName //!< Dll name for the binary
|
||||
);
|
||||
|
||||
//! Loads x86 executable in-memory code
|
||||
bool loadX86JIT(
|
||||
Program& prorgam, //!< CPU Program object
|
||||
bool& hasJITBin //!< indicate if the OCL binary has JIT binary
|
||||
);
|
||||
//! Loads x86 executable in-memory code
|
||||
bool loadX86JIT(Program& prorgam, //!< CPU Program object
|
||||
bool& hasJITBin //!< indicate if the OCL binary has JIT binary
|
||||
);
|
||||
|
||||
//! Stores x86 executable in-memory code
|
||||
bool storeX86JIT(
|
||||
Program& program //!< CPU Program object
|
||||
);
|
||||
//! Stores x86 executable in-memory code
|
||||
bool storeX86JIT(Program& program //!< CPU Program object
|
||||
);
|
||||
|
||||
//! Set elf header information for CPU target
|
||||
bool setElfTarget() {
|
||||
uint32_t target = dev().settings().cpuFeatures_;
|
||||
assert (((0xFFFF8000 & target) == 0) && "ASIC target ID >= 2^15");
|
||||
uint16_t elf_target = (uint16_t)(0x7FFF & target);
|
||||
return elfOut()->setTarget(elf_target, amd::OclElf::CPU_PLATFORM);
|
||||
}
|
||||
//! Set elf header information for CPU target
|
||||
bool setElfTarget() {
|
||||
uint32_t target = dev().settings().cpuFeatures_;
|
||||
assert(((0xFFFF8000 & target) == 0) && "ASIC target ID >= 2^15");
|
||||
uint16_t elf_target = (uint16_t)(0x7FFF & target);
|
||||
return elfOut()->setTarget(elf_target, amd::OclElf::CPU_PLATFORM);
|
||||
}
|
||||
|
||||
bool storeX86Asm(const char* buffer, size_t size);
|
||||
|
||||
private:
|
||||
private:
|
||||
enum FeatureCheckResult { ERROR, RECOMPILE, OK };
|
||||
|
||||
enum FeatureCheckResult {
|
||||
ERROR,
|
||||
RECOMPILE,
|
||||
OK
|
||||
};
|
||||
FeatureCheckResult checkFeatures();
|
||||
|
||||
FeatureCheckResult checkFeatures();
|
||||
//! Disable default copy constructor
|
||||
ClBinary(const ClBinary&);
|
||||
|
||||
//! Disable default copy constructor
|
||||
ClBinary(const ClBinary&);
|
||||
//! Disable default operator=
|
||||
ClBinary& operator=(const ClBinary&);
|
||||
|
||||
//! Disable default operator=
|
||||
ClBinary& operator=(const ClBinary&);
|
||||
|
||||
//! Returns the GPU device for this object
|
||||
const Device& dev() { return static_cast<const Device&>(dev_); }
|
||||
//! Returns the GPU device for this object
|
||||
const Device& dev() { return static_cast<const Device&>(dev_); }
|
||||
};
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
#endif // CPUBINARY_HPP_
|
||||
#endif // CPUBINARY_HPP_
|
||||
|
||||
@@ -6,52 +6,47 @@
|
||||
#include "device/cpu/cpucommand.hpp"
|
||||
|
||||
#include <amdocl/cl_kernel.h>
|
||||
#include <cstdio> // for printf
|
||||
#include <cstdio> // for printf
|
||||
#include <stdarg.h>
|
||||
|
||||
#define BUF_SIZE_PRINTF 4095
|
||||
//In the current implementation of printf in gcc 4.5.2 runtime libraries,inf/infinity and nan are not supported
|
||||
//The [-]infinity value is printed as [-]1.#INF00
|
||||
//The [-]nan value is printed as [-]1.#INF00
|
||||
//bufOutUpdate converts the all printed instanced of [-]1.#INF00 to inf,and
|
||||
// In the current implementation of printf in gcc 4.5.2 runtime libraries,inf/infinity and nan are
|
||||
// not supported
|
||||
// The [-]infinity value is printed as [-]1.#INF00
|
||||
// The [-]nan value is printed as [-]1.#INF00
|
||||
// bufOutUpdate converts the all printed instanced of [-]1.#INF00 to inf,and
|
||||
// all printed instanced of [-]1.#IND00 to nan
|
||||
void bufOutUpdate(std::string& sBufOut,const char* strToReplace,const char* strReplace)
|
||||
{
|
||||
size_t foundIdx = 0;
|
||||
while ((foundIdx = sBufOut.find(strToReplace,foundIdx)) != std::string::npos) {
|
||||
sBufOut.replace(foundIdx,strlen(strToReplace),strReplace,strlen(strReplace));
|
||||
foundIdx += 3;
|
||||
}
|
||||
void bufOutUpdate(std::string& sBufOut, const char* strToReplace, const char* strReplace) {
|
||||
size_t foundIdx = 0;
|
||||
while ((foundIdx = sBufOut.find(strToReplace, foundIdx)) != std::string::npos) {
|
||||
sBufOut.replace(foundIdx, strlen(strToReplace), strReplace, strlen(strReplace));
|
||||
foundIdx += 3;
|
||||
}
|
||||
}
|
||||
int cpuprintf(const char* format,...)
|
||||
{
|
||||
char cBufOut[BUF_SIZE_PRINTF];
|
||||
std::string sBufOut;
|
||||
va_list args;
|
||||
va_start(args, format);
|
||||
//write to the buffer
|
||||
vsprintf(cBufOut,format,args);
|
||||
sBufOut = cBufOut;
|
||||
int cpuprintf(const char* format, ...) {
|
||||
char cBufOut[BUF_SIZE_PRINTF];
|
||||
std::string sBufOut;
|
||||
va_list args;
|
||||
va_start(args, format);
|
||||
// write to the buffer
|
||||
vsprintf(cBufOut, format, args);
|
||||
sBufOut = cBufOut;
|
||||
|
||||
//convert to correct infinity/nan representation
|
||||
bufOutUpdate(sBufOut,"1.#INF00","inf");
|
||||
bufOutUpdate(sBufOut,"1.#IND00","nan");
|
||||
bufOutUpdate(sBufOut,"1.#QNAN0","nan");
|
||||
int ret = amd::Os::printf("%s",sBufOut.c_str());
|
||||
fflush(stdout);
|
||||
va_end (args);
|
||||
return ret;
|
||||
// convert to correct infinity/nan representation
|
||||
bufOutUpdate(sBufOut, "1.#INF00", "inf");
|
||||
bufOutUpdate(sBufOut, "1.#IND00", "nan");
|
||||
bufOutUpdate(sBufOut, "1.#QNAN0", "nan");
|
||||
int ret = amd::Os::printf("%s", sBufOut.c_str());
|
||||
fflush(stdout);
|
||||
va_end(args);
|
||||
return ret;
|
||||
}
|
||||
namespace cpu {
|
||||
|
||||
const clk_builtins_t
|
||||
Builtins::dispatchTable_ =
|
||||
{
|
||||
const clk_builtins_t Builtins::dispatchTable_ = {
|
||||
/* Synchronization functions */
|
||||
&WorkItem::barrier,
|
||||
/* AMD Only builtins: FIXME_lmoriche: remove or add an extension */
|
||||
NULL,
|
||||
cpuprintf
|
||||
};
|
||||
NULL, cpuprintf};
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
@@ -10,11 +10,10 @@
|
||||
|
||||
namespace cpu {
|
||||
|
||||
struct Builtins : public amd::AllStatic
|
||||
{
|
||||
static const clk_builtins_t dispatchTable_;
|
||||
struct Builtins : public amd::AllStatic {
|
||||
static const clk_builtins_t dispatchTable_;
|
||||
};
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
#endif /*BUILTINS_HPP_*/
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@@ -15,7 +15,7 @@
|
||||
|
||||
#if defined(ATI_ARCH_ARM)
|
||||
#include <setjmp.h>
|
||||
#endif // ATI_ARCH_ARM
|
||||
#endif // ATI_ARCH_ARM
|
||||
|
||||
namespace cpu {
|
||||
|
||||
@@ -27,94 +27,91 @@ namespace cpu {
|
||||
*/
|
||||
|
||||
//! A saved stack context
|
||||
class StackContext : public amd::StackObject
|
||||
{
|
||||
private:
|
||||
class StackContext : public amd::StackObject {
|
||||
private:
|
||||
#if defined(ATI_ARCH_ARM)
|
||||
jmp_buf env_;
|
||||
jmp_buf env_;
|
||||
#elif defined(_WIN64)
|
||||
intptr_t __declspec(align(16)) regs_[32];
|
||||
#else // !_WIN64
|
||||
intptr_t regs_[LP64_SWITCH(6,8)];
|
||||
#endif // !_WIN64
|
||||
intptr_t __declspec(align(16)) regs_[32];
|
||||
#else // !_WIN64
|
||||
intptr_t regs_[LP64_SWITCH(6, 8)];
|
||||
#endif // !_WIN64
|
||||
|
||||
public:
|
||||
//! Save the stack context. Return 0 if returning directly.
|
||||
inline intptr_t setjmp();
|
||||
public:
|
||||
//! Save the stack context. Return 0 if returning directly.
|
||||
inline intptr_t setjmp();
|
||||
|
||||
//! Restore the stack context
|
||||
inline void longjmp(intptr_t val) const;
|
||||
//! Restore the stack context
|
||||
inline void longjmp(intptr_t val) const;
|
||||
};
|
||||
|
||||
//! A thread fiber
|
||||
class Fiber : public amd::StackObject
|
||||
{
|
||||
private:
|
||||
//! Next fiber in the thread.
|
||||
Fiber* next_;
|
||||
class Fiber : public amd::StackObject {
|
||||
private:
|
||||
//! Next fiber in the thread.
|
||||
Fiber* next_;
|
||||
|
||||
//! This fiber's saved state.
|
||||
StackContext context_;
|
||||
//! This fiber's saved state.
|
||||
StackContext context_;
|
||||
|
||||
public:
|
||||
//! Construct a new Fiber
|
||||
Fiber() : next_(NULL) { }
|
||||
public:
|
||||
//! Construct a new Fiber
|
||||
Fiber() : next_(NULL) {}
|
||||
|
||||
//! Return the next fiber in the current thread.
|
||||
const Fiber* next() const { return next_; }
|
||||
//! Set the next fiber in the current thread.
|
||||
void setNext(Fiber* next) { next_ = next; }
|
||||
//! Return the next fiber in the current thread.
|
||||
const Fiber* next() const { return next_; }
|
||||
//! Set the next fiber in the current thread.
|
||||
void setNext(Fiber* next) { next_ = next; }
|
||||
|
||||
//! Save the state of this fiber. Return true if directly returning.
|
||||
ALWAYSINLINE bool save() { return context_.setjmp() == 0; }
|
||||
//! Restore this fiber from the saved context.
|
||||
void restore() const { context_.longjmp(1); }
|
||||
//! Save the state of this fiber. Return true if directly returning.
|
||||
ALWAYSINLINE bool save() { return context_.setjmp() == 0; }
|
||||
//! Restore this fiber from the saved context.
|
||||
void restore() const { context_.longjmp(1); }
|
||||
|
||||
//! Switch to the given fiber.
|
||||
void swap(const Fiber* fiber) { if (save()) { fiber->restore(); } }
|
||||
//! Switch to the given fiber.
|
||||
void swap(const Fiber* fiber) {
|
||||
if (save()) {
|
||||
fiber->restore();
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
|
||||
//! A CPU core operation (enqueued in the worker thread queue)
|
||||
class Operation : public amd::HeapObject
|
||||
{
|
||||
public:
|
||||
//! An atomic counter
|
||||
class Counter
|
||||
{
|
||||
// FIXME_lmoriche: recycle the counters, implement a thread local pool.
|
||||
private:
|
||||
amd::Event& event_;
|
||||
//! The atomic counter value.
|
||||
amd::Atomic<size_t> counter_;
|
||||
class Operation : public amd::HeapObject {
|
||||
public:
|
||||
//! An atomic counter
|
||||
class Counter {
|
||||
// FIXME_lmoriche: recycle the counters, implement a thread local pool.
|
||||
private:
|
||||
amd::Event& event_;
|
||||
//! The atomic counter value.
|
||||
amd::Atomic<size_t> counter_;
|
||||
|
||||
public:
|
||||
//! Initialize the counter with the given initial value.
|
||||
Counter(amd::Event& event, size_t initialValue) :
|
||||
event_(event), counter_(initialValue) { }
|
||||
//! Return the event associated with this counter.
|
||||
amd::Event& event() { return event_; }
|
||||
//! Decrement the counter and return the new value.
|
||||
size_t decrement() { return --counter_; }
|
||||
};
|
||||
public:
|
||||
//! Initialize the counter with the given initial value.
|
||||
Counter(amd::Event& event, size_t initialValue) : event_(event), counter_(initialValue) {}
|
||||
//! Return the event associated with this counter.
|
||||
amd::Event& event() { return event_; }
|
||||
//! Decrement the counter and return the new value.
|
||||
size_t decrement() { return --counter_; }
|
||||
};
|
||||
|
||||
protected:
|
||||
amd::Command& command_;
|
||||
protected:
|
||||
amd::Command& command_;
|
||||
|
||||
public:
|
||||
Operation(amd::Command& command) : command_(command)
|
||||
{ }
|
||||
public:
|
||||
Operation(amd::Command& command) : command_(command) {}
|
||||
|
||||
virtual ~Operation() {};
|
||||
virtual ~Operation(){};
|
||||
|
||||
virtual void clone(Operation* buf) = 0;
|
||||
virtual void clone(Operation* buf) = 0;
|
||||
|
||||
void cleanup();
|
||||
void cleanup();
|
||||
|
||||
amd::Command& command() { return command_;}
|
||||
amd::Command& command() { return command_; }
|
||||
|
||||
virtual void execute() = 0;
|
||||
virtual void execute() = 0;
|
||||
};
|
||||
|
||||
/*! @}
|
||||
@@ -123,243 +120,210 @@ public:
|
||||
*/
|
||||
|
||||
//! A work item instance
|
||||
class WorkItem : public Fiber
|
||||
{
|
||||
private:
|
||||
//! Thread info block (must be the last field).
|
||||
clk_thread_info_block_t tib_;
|
||||
class WorkItem : public Fiber {
|
||||
private:
|
||||
//! Thread info block (must be the last field).
|
||||
clk_thread_info_block_t tib_;
|
||||
|
||||
private:
|
||||
//! Cannot be deleted (allocated with placement new).
|
||||
void operator delete(void*) { ShouldNotCallThis(); }
|
||||
private:
|
||||
//! Cannot be deleted (allocated with placement new).
|
||||
void operator delete(void*) { ShouldNotCallThis(); }
|
||||
|
||||
public:
|
||||
//! Initialize this workgroup.
|
||||
WorkItem(
|
||||
const amd::NDRangeContainer& size,
|
||||
void* scratchMemPtr,
|
||||
void* localMemPtr);
|
||||
public:
|
||||
//! Initialize this workgroup.
|
||||
WorkItem(const amd::NDRangeContainer& size, void* scratchMemPtr, void* localMemPtr);
|
||||
|
||||
//! Return the current WorkItem (based of the current stack pointer).
|
||||
static WorkItem* current() {
|
||||
return (WorkItem*)amd::alignUp((intptr_t) amd::Os::currentStackPtr(),
|
||||
CLK_PRIVATE_MEMORY_SIZE) - 1;
|
||||
}
|
||||
//! Return the current WorkItem (based of the current stack pointer).
|
||||
static WorkItem* current() {
|
||||
return (WorkItem*)amd::alignUp((intptr_t)amd::Os::currentStackPtr(), CLK_PRIVATE_MEMORY_SIZE) -
|
||||
1;
|
||||
}
|
||||
|
||||
clk_thread_info_block_t& infoBlock() { return tib_; }
|
||||
clk_thread_info_block_t& infoBlock() { return tib_; }
|
||||
|
||||
//! Return the native stack pointer base for this workitem.
|
||||
address nativeStackPtr() const {
|
||||
address newSp = amd::alignDown((address) this - CPUKERNEL_STACK_ALIGN,
|
||||
CPUKERNEL_STACK_ALIGN);
|
||||
WINDOWS_ONLY(NOT_WIN64(newSp += sizeof(void*)));
|
||||
return newSp;
|
||||
}
|
||||
//! Return the native stack pointer base for this workitem.
|
||||
address nativeStackPtr() const {
|
||||
address newSp = amd::alignDown((address) this - CPUKERNEL_STACK_ALIGN, CPUKERNEL_STACK_ALIGN);
|
||||
WINDOWS_ONLY(NOT_WIN64(newSp += sizeof(void*)));
|
||||
return newSp;
|
||||
}
|
||||
|
||||
//! These functions are mapping "n" from 1d index to the required dimension
|
||||
inline void setGroupId(
|
||||
const amd::NDRange& rangeLimits,
|
||||
const amd::NDRange& offset,
|
||||
size_t n);
|
||||
inline void incrementGroupId(
|
||||
const amd::NDRange& rangeLimits,
|
||||
const amd::NDRange& offset,
|
||||
size_t n);
|
||||
//! These functions are mapping "n" from 1d index to the required dimension
|
||||
inline void setGroupId(const amd::NDRange& rangeLimits, const amd::NDRange& offset, size_t n);
|
||||
inline void incrementGroupId(const amd::NDRange& rangeLimits, const amd::NDRange& offset,
|
||||
size_t n);
|
||||
|
||||
//! Execute a thread synchronization barrier.
|
||||
static void barrier(cl_mem_fence_flags flags);
|
||||
//! Execute a thread synchronization barrier.
|
||||
static void barrier(cl_mem_fence_flags flags);
|
||||
};
|
||||
|
||||
typedef void (*kernelentrypoint_t)(const void*);
|
||||
|
||||
//! Execute a workgroup (work-items).
|
||||
class WorkGroup
|
||||
{
|
||||
private:
|
||||
amd::NDRangeKernelCommand& command_;
|
||||
const cpu::Kernel& kernel_;
|
||||
WorkerThread& thread_;
|
||||
address params_;
|
||||
WorkItem* const workItem0_;
|
||||
const Fiber* workingFiber_;
|
||||
size_t numWorkItems_;
|
||||
class WorkGroup {
|
||||
private:
|
||||
amd::NDRangeKernelCommand& command_;
|
||||
const cpu::Kernel& kernel_;
|
||||
WorkerThread& thread_;
|
||||
address params_;
|
||||
WorkItem* const workItem0_;
|
||||
const Fiber* workingFiber_;
|
||||
size_t numWorkItems_;
|
||||
|
||||
public:
|
||||
WorkGroup(
|
||||
amd::NDRangeKernelCommand& parent,
|
||||
const cpu::Kernel& kernel,
|
||||
WorkerThread& thread,
|
||||
address params,
|
||||
WorkItem* workItem0,
|
||||
const size_t numWorkItems) :
|
||||
command_(parent),
|
||||
kernel_(kernel),
|
||||
thread_(thread),
|
||||
params_(params),
|
||||
workItem0_(workItem0),
|
||||
numWorkItems_(numWorkItems)
|
||||
{ }
|
||||
public:
|
||||
WorkGroup(amd::NDRangeKernelCommand& parent, const cpu::Kernel& kernel, WorkerThread& thread,
|
||||
address params, WorkItem* workItem0, const size_t numWorkItems)
|
||||
: command_(parent),
|
||||
kernel_(kernel),
|
||||
thread_(thread),
|
||||
params_(params),
|
||||
workItem0_(workItem0),
|
||||
numWorkItems_(numWorkItems) {}
|
||||
|
||||
WorkItem* getBaseWorkItem() { return workItem0_; }
|
||||
WorkerThread& getWorkerThread() { return thread_; }
|
||||
WorkItem* getBaseWorkItem() { return workItem0_; }
|
||||
WorkerThread& getWorkerThread() { return thread_; }
|
||||
|
||||
void executeWorkItem(); // In case of 1 WorkItem
|
||||
void executeWithBarrier();
|
||||
void executeWithoutBarrier();
|
||||
void executeWorkItem(); // In case of 1 WorkItem
|
||||
void executeWithBarrier();
|
||||
void executeWithoutBarrier();
|
||||
|
||||
void setNumWorkItems(size_t workItems) { numWorkItems_ = workItems; }
|
||||
size_t getNumWorkItems() { return numWorkItems_; }
|
||||
private:
|
||||
void callKernelRange(
|
||||
kernelentrypoint_t entryPoint,
|
||||
address stackPtr,
|
||||
clk_thread_info_block_t& tib);
|
||||
inline void callKernel(
|
||||
kernelentrypoint_t entryPoint,
|
||||
address stackPtr);
|
||||
inline void callKernelProtectedReturn(
|
||||
kernelentrypoint_t entryPoint,
|
||||
address stackPtr);
|
||||
void setNumWorkItems(size_t workItems) { numWorkItems_ = workItems; }
|
||||
size_t getNumWorkItems() { return numWorkItems_; }
|
||||
|
||||
private:
|
||||
void callKernelRange(kernelentrypoint_t entryPoint, address stackPtr,
|
||||
clk_thread_info_block_t& tib);
|
||||
inline void callKernel(kernelentrypoint_t entryPoint, address stackPtr);
|
||||
inline void callKernelProtectedReturn(kernelentrypoint_t entryPoint, address stackPtr);
|
||||
};
|
||||
|
||||
class NDRangeKernelBatch : public Operation
|
||||
{
|
||||
protected:
|
||||
size_t coreId_;
|
||||
const size_t numWorkGroups_;
|
||||
const size_t numCores_;
|
||||
volatile size_t currentOpId_;
|
||||
const amd::NDRange groupIds_; //!< Number of groups in each dimensions
|
||||
VirtualCPU& virtualDevice_;
|
||||
class NDRangeKernelBatch : public Operation {
|
||||
protected:
|
||||
size_t coreId_;
|
||||
const size_t numWorkGroups_;
|
||||
const size_t numCores_;
|
||||
volatile size_t currentOpId_;
|
||||
const amd::NDRange groupIds_; //!< Number of groups in each dimensions
|
||||
VirtualCPU& virtualDevice_;
|
||||
|
||||
public:
|
||||
enum ExecutionOrder {
|
||||
ORDER_DEFAULT,
|
||||
ORDER_ROUND_ROBIN = ORDER_DEFAULT,
|
||||
//ORDER_LINEAR
|
||||
};
|
||||
public:
|
||||
enum ExecutionOrder {
|
||||
ORDER_DEFAULT,
|
||||
ORDER_ROUND_ROBIN = ORDER_DEFAULT,
|
||||
// ORDER_LINEAR
|
||||
};
|
||||
|
||||
enum ExecutionNature {
|
||||
NATURE_WITH_BARRIER,
|
||||
NATURE_WITHOUT_BARRIER,
|
||||
NATURE_1_WORK_ITEM,
|
||||
NATURE_WG_LEVEL_EXEC
|
||||
};
|
||||
enum ExecutionNature {
|
||||
NATURE_WITH_BARRIER,
|
||||
NATURE_WITHOUT_BARRIER,
|
||||
NATURE_1_WORK_ITEM,
|
||||
NATURE_WG_LEVEL_EXEC
|
||||
};
|
||||
|
||||
NDRangeKernelBatch(
|
||||
amd::NDRangeKernelCommand& parent,
|
||||
VirtualCPU& virtualDevice,
|
||||
const amd::NDRange& groupIds, size_t numCores) :
|
||||
Operation(parent),
|
||||
coreId_(0),
|
||||
numWorkGroups_(groupIds.product()),
|
||||
numCores_(numCores),
|
||||
currentOpId_(0),
|
||||
groupIds_(groupIds),
|
||||
virtualDevice_(virtualDevice)
|
||||
{ }
|
||||
NDRangeKernelBatch(amd::NDRangeKernelCommand& parent, VirtualCPU& virtualDevice,
|
||||
const amd::NDRange& groupIds, size_t numCores)
|
||||
: Operation(parent),
|
||||
coreId_(0),
|
||||
numWorkGroups_(groupIds.product()),
|
||||
numCores_(numCores),
|
||||
currentOpId_(0),
|
||||
groupIds_(groupIds),
|
||||
virtualDevice_(virtualDevice) {}
|
||||
|
||||
virtual void clone(Operation* buf)
|
||||
{
|
||||
::new(buf) NDRangeKernelBatch(static_cast<amd::NDRangeKernelCommand&>(command_),
|
||||
virtualDevice_, groupIds_, numCores_);
|
||||
static_cast<NDRangeKernelBatch*>(buf)->setCoreId(coreId_);
|
||||
}
|
||||
virtual void clone(Operation* buf) {
|
||||
::new (buf) NDRangeKernelBatch(static_cast<amd::NDRangeKernelCommand&>(command_),
|
||||
virtualDevice_, groupIds_, numCores_);
|
||||
static_cast<NDRangeKernelBatch*>(buf)->setCoreId(coreId_);
|
||||
}
|
||||
|
||||
virtual void execute();
|
||||
virtual void execute();
|
||||
|
||||
void setCoreId(size_t coreId) { coreId_ = coreId; currentOpId_ = coreId; }
|
||||
void setCoreId(size_t coreId) {
|
||||
coreId_ = coreId;
|
||||
currentOpId_ = coreId;
|
||||
}
|
||||
|
||||
inline bool getNextOperationId(size_t& opId);
|
||||
inline size_t getNextOperationIds(size_t& opId, size_t count);
|
||||
inline bool getNextOperationId(size_t& opId);
|
||||
inline size_t getNextOperationIds(size_t& opId, size_t count);
|
||||
|
||||
private:
|
||||
bool patchParameters(
|
||||
const cpu::Kernel& kernel,
|
||||
address params,
|
||||
address& localMemPtr,
|
||||
const address localMemLimit,
|
||||
size_t localMemSize) const;
|
||||
private:
|
||||
bool patchParameters(const cpu::Kernel& kernel, address params, address& localMemPtr,
|
||||
const address localMemLimit, size_t localMemSize) const;
|
||||
};
|
||||
|
||||
class NativeFn : public Operation
|
||||
{
|
||||
public:
|
||||
NativeFn(amd::NativeFnCommand& parent) : Operation(parent)
|
||||
{ }
|
||||
class NativeFn : public Operation {
|
||||
public:
|
||||
NativeFn(amd::NativeFnCommand& parent) : Operation(parent) {}
|
||||
|
||||
virtual void clone(Operation* buf)
|
||||
{
|
||||
::new(buf) NativeFn(static_cast<amd::NativeFnCommand&>(command_));
|
||||
}
|
||||
virtual void clone(Operation* buf) {
|
||||
::new (buf) NativeFn(static_cast<amd::NativeFnCommand&>(command_));
|
||||
}
|
||||
|
||||
virtual void execute();
|
||||
virtual void execute();
|
||||
};
|
||||
#ifndef MAX
|
||||
#define MAX(x,y) ((x)>=(y) ?(x) : (y))
|
||||
#endif //MAX
|
||||
#define MAX(x, y) ((x) >= (y) ? (x) : (y))
|
||||
#endif // MAX
|
||||
|
||||
#define MAX_OPERATION_ALLOC_SIZE (MAX(sizeof(NDRangeKernelBatch), sizeof(NativeFn)))
|
||||
|
||||
//! A thread bound to a cpu core.
|
||||
class WorkerThread : public amd::Thread
|
||||
{
|
||||
private:
|
||||
Fiber mainFiber_; //!< main fiber for this worker thread.
|
||||
class WorkerThread : public amd::Thread {
|
||||
private:
|
||||
Fiber mainFiber_; //!< main fiber for this worker thread.
|
||||
|
||||
amd::Monitor queueLock_; //!< lock protecting the queue.
|
||||
volatile int waitingOp_;
|
||||
bool terminated_; //!< true if the thread is shutting down.
|
||||
|
||||
//! Local memory storage
|
||||
address localDataStorage_;
|
||||
//! Size of the local memory.
|
||||
size_t localDataSize_;
|
||||
amd::Monitor queueLock_; //!< lock protecting the queue.
|
||||
volatile int waitingOp_;
|
||||
bool terminated_; //!< true if the thread is shutting down.
|
||||
|
||||
char operation_[MAX_OPERATION_ALLOC_SIZE];
|
||||
//! Local memory storage
|
||||
address localDataStorage_;
|
||||
//! Size of the local memory.
|
||||
size_t localDataSize_;
|
||||
|
||||
address baseWorkItemsStack_;
|
||||
private:
|
||||
//! Awaits operations and execute them as they become ready.
|
||||
void loop();
|
||||
char operation_[MAX_OPERATION_ALLOC_SIZE];
|
||||
|
||||
public:
|
||||
//! Construct a new WorkerThread.
|
||||
WorkerThread(const cpu::Device& device);
|
||||
//! Destroy the worker thread.
|
||||
virtual ~WorkerThread();
|
||||
//! Cleanup the thread before termination.
|
||||
bool terminate();
|
||||
address baseWorkItemsStack_;
|
||||
|
||||
//! Return the main fiber for this thread.
|
||||
Fiber& mainFiber() { return mainFiber_; }
|
||||
//! Return the LDS for this thread
|
||||
address localDataStorage() const { return localDataStorage_; }
|
||||
//! Return the size of the local memory for this thread.
|
||||
size_t localDataSize() const { return localDataSize_; }
|
||||
private:
|
||||
//! Awaits operations and execute them as they become ready.
|
||||
void loop();
|
||||
|
||||
address baseWorkItemsStack() { return baseWorkItemsStack_; }
|
||||
public:
|
||||
//! Construct a new WorkerThread.
|
||||
WorkerThread(const cpu::Device& device);
|
||||
//! Destroy the worker thread.
|
||||
virtual ~WorkerThread();
|
||||
//! Cleanup the thread before termination.
|
||||
bool terminate();
|
||||
|
||||
Operation* operation() { return reinterpret_cast<Operation*>(operation_); }
|
||||
bool isOperationValid() { return waitingOp_ > 0; }
|
||||
//! Return the main fiber for this thread.
|
||||
Fiber& mainFiber() { return mainFiber_; }
|
||||
//! Return the LDS for this thread
|
||||
address localDataStorage() const { return localDataStorage_; }
|
||||
//! Return the size of the local memory for this thread.
|
||||
size_t localDataSize() const { return localDataSize_; }
|
||||
|
||||
//! Enqueue a new operation to execute in this thread.
|
||||
void enqueue(Operation& op);
|
||||
//! Signal to start processing the commands in the queue.
|
||||
void flush() { amd::ScopedLock sl(queueLock_); queueLock_.notify(); }
|
||||
address baseWorkItemsStack() { return baseWorkItemsStack_; }
|
||||
|
||||
//! This thread's execution engine.
|
||||
void run(void* data) {
|
||||
loop();
|
||||
}
|
||||
Operation* operation() { return reinterpret_cast<Operation*>(operation_); }
|
||||
bool isOperationValid() { return waitingOp_ > 0; }
|
||||
|
||||
bool isWorkerThread() const { return true; }
|
||||
//! Enqueue a new operation to execute in this thread.
|
||||
void enqueue(Operation& op);
|
||||
//! Signal to start processing the commands in the queue.
|
||||
void flush() {
|
||||
amd::ScopedLock sl(queueLock_);
|
||||
queueLock_.notify();
|
||||
}
|
||||
|
||||
//! Return the currently executing WorkerThread's instance.
|
||||
static WorkerThread* current()
|
||||
{
|
||||
return static_cast<WorkerThread*>(Thread::current());
|
||||
}
|
||||
//! This thread's execution engine.
|
||||
void run(void* data) { loop(); }
|
||||
|
||||
bool isWorkerThread() const { return true; }
|
||||
|
||||
//! Return the currently executing WorkerThread's instance.
|
||||
static WorkerThread* current() { return static_cast<WorkerThread*>(Thread::current()); }
|
||||
};
|
||||
|
||||
/*! @}
|
||||
@@ -371,59 +335,43 @@ extern "C" intptr_t _StackContext_setjmp(intptr_t* regs);
|
||||
#if !defined(ATI_ARCH_ARM)
|
||||
ALWAYSINLINE
|
||||
#endif
|
||||
intptr_t
|
||||
StackContext::setjmp()
|
||||
{
|
||||
intptr_t StackContext::setjmp() {
|
||||
#if defined(ATI_ARCH_ARM)
|
||||
return ::setjmp(env_);
|
||||
return ::setjmp(env_);
|
||||
#else
|
||||
return _StackContext_setjmp(regs_);
|
||||
return _StackContext_setjmp(regs_);
|
||||
#endif
|
||||
}
|
||||
|
||||
extern "C" void _StackContext_longjmp(const intptr_t* env, intptr_t val);
|
||||
|
||||
ALWAYSINLINE void
|
||||
StackContext::longjmp(intptr_t val) const
|
||||
{
|
||||
ALWAYSINLINE void StackContext::longjmp(intptr_t val) const {
|
||||
#if defined(ATI_ARCH_ARM)
|
||||
return ::longjmp(*const_cast<jmp_buf*>(&env_), val);
|
||||
return ::longjmp(*const_cast<jmp_buf*>(&env_), val);
|
||||
#else
|
||||
return _StackContext_longjmp(regs_, val);
|
||||
return _StackContext_longjmp(regs_, val);
|
||||
#endif
|
||||
}
|
||||
|
||||
|
||||
extern "C" void _WorkGroup_callKernel(address params, kernelentrypoint_t entryPoint,
|
||||
address stackPtr);
|
||||
|
||||
extern "C" void _WorkGroup_callKernel(
|
||||
address params,
|
||||
kernelentrypoint_t entryPoint,
|
||||
address stackPtr);
|
||||
|
||||
extern "C" void _WorkGroup_callKernelProtectedReturn(
|
||||
address params,
|
||||
kernelentrypoint_t entryPoint,
|
||||
address stackPtr);
|
||||
extern "C" void _WorkGroup_callKernelProtectedReturn(address params, kernelentrypoint_t entryPoint,
|
||||
address stackPtr);
|
||||
|
||||
|
||||
ALWAYSINLINE void
|
||||
WorkGroup::callKernel(
|
||||
kernelentrypoint_t entryPoint,
|
||||
address stackPtr)
|
||||
{
|
||||
_WorkGroup_callKernel(params_, entryPoint, stackPtr);
|
||||
ALWAYSINLINE void WorkGroup::callKernel(kernelentrypoint_t entryPoint, address stackPtr) {
|
||||
_WorkGroup_callKernel(params_, entryPoint, stackPtr);
|
||||
}
|
||||
|
||||
// This version support the case of changing the stack for fibers.
|
||||
ALWAYSINLINE void
|
||||
WorkGroup::callKernelProtectedReturn(
|
||||
kernelentrypoint_t entryPoint,
|
||||
address stackPtr)
|
||||
{
|
||||
_WorkGroup_callKernelProtectedReturn(params_, entryPoint, stackPtr);
|
||||
ALWAYSINLINE void WorkGroup::callKernelProtectedReturn(kernelentrypoint_t entryPoint,
|
||||
address stackPtr) {
|
||||
_WorkGroup_callKernelProtectedReturn(params_, entryPoint, stackPtr);
|
||||
}
|
||||
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
#endif /*OPERATION_HPP_*/
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@@ -21,9 +21,9 @@
|
||||
namespace cpu {
|
||||
|
||||
//! Maximum number of the supported samplers
|
||||
const static uint32_t MaxSamplers = 16;
|
||||
const static uint32_t MaxSamplers = 16;
|
||||
//! Maximum number of supported read images
|
||||
const static uint32_t MaxReadImage = 128;
|
||||
const static uint32_t MaxReadImage = 128;
|
||||
//! Maximum number of supported write images
|
||||
const static uint32_t MaxWriteImage = 64;
|
||||
//! Maximum number of supported read/write images
|
||||
@@ -40,203 +40,169 @@ const static uint32_t MaxReadWriteImage = 64;
|
||||
*/
|
||||
|
||||
//! A CPU device ordinal
|
||||
class Device : public amd::Device
|
||||
{
|
||||
protected:
|
||||
static aclCompiler* compiler_;
|
||||
public:
|
||||
aclCompiler* compiler() const { return compiler_; }
|
||||
class Device : public amd::Device {
|
||||
protected:
|
||||
static aclCompiler* compiler_;
|
||||
|
||||
public:
|
||||
static bool init(void);
|
||||
public:
|
||||
aclCompiler* compiler() const { return compiler_; }
|
||||
|
||||
//! Shutdown CPU device
|
||||
static void tearDown();
|
||||
public:
|
||||
static bool init(void);
|
||||
|
||||
//! Construct a new identifier
|
||||
Device(Device* parent = NULL) :
|
||||
amd::Device(parent),
|
||||
workerThreadsAffinity_(NULL)
|
||||
{}
|
||||
//! Shutdown CPU device
|
||||
static void tearDown();
|
||||
|
||||
virtual ~Device();
|
||||
//! Construct a new identifier
|
||||
Device(Device* parent = NULL) : amd::Device(parent), workerThreadsAffinity_(NULL) {}
|
||||
|
||||
bool create();
|
||||
virtual ~Device();
|
||||
|
||||
virtual cl_int createSubDevices(
|
||||
device::CreateSubDevicesInfo& create_info,
|
||||
cl_uint num_entries,
|
||||
cl_device_id* devices,
|
||||
cl_uint* num_devices);
|
||||
bool create();
|
||||
|
||||
//! Instantiate a new virtual device
|
||||
virtual device::VirtualDevice* createVirtualDevice(
|
||||
amd::CommandQueue* queue = NULL
|
||||
)
|
||||
{
|
||||
VirtualCPU* virtualCpu = new VirtualCPU(*this);
|
||||
if (virtualCpu != NULL && !virtualCpu->acceptingCommands()) {
|
||||
virtualCpu->terminate();
|
||||
delete virtualCpu;
|
||||
virtualCpu = NULL;
|
||||
}
|
||||
return virtualCpu;
|
||||
virtual cl_int createSubDevices(device::CreateSubDevicesInfo& create_info, cl_uint num_entries,
|
||||
cl_device_id* devices, cl_uint* num_devices);
|
||||
|
||||
//! Instantiate a new virtual device
|
||||
virtual device::VirtualDevice* createVirtualDevice(amd::CommandQueue* queue = NULL) {
|
||||
VirtualCPU* virtualCpu = new VirtualCPU(*this);
|
||||
if (virtualCpu != NULL && !virtualCpu->acceptingCommands()) {
|
||||
virtualCpu->terminate();
|
||||
delete virtualCpu;
|
||||
virtualCpu = NULL;
|
||||
}
|
||||
return virtualCpu;
|
||||
}
|
||||
|
||||
//! Compile the given source code.
|
||||
virtual device::Program* createProgram(amd::option::Options* options = NULL);
|
||||
//! Compile the given source code.
|
||||
virtual device::Program* createProgram(amd::option::Options* options = NULL);
|
||||
|
||||
//! Just returns NULL as CPU devices use the host memory
|
||||
virtual device::Memory* createMemory(amd::Memory& owner) const
|
||||
{
|
||||
return NULL;
|
||||
}
|
||||
//! Just returns NULL as CPU devices use the host memory
|
||||
virtual device::Memory* createMemory(amd::Memory& owner) const { return NULL; }
|
||||
|
||||
//! Sampler object allocation
|
||||
virtual bool createSampler(
|
||||
const amd::Sampler& owner, //!< abstraction layer sampler object
|
||||
device::Sampler** sampler //!< device sampler object
|
||||
) const
|
||||
{
|
||||
// Just return NULL on CPU device
|
||||
*sampler = NULL;
|
||||
return true;
|
||||
}
|
||||
//! Sampler object allocation
|
||||
virtual bool createSampler(const amd::Sampler& owner, //!< abstraction layer sampler object
|
||||
device::Sampler** sampler //!< device sampler object
|
||||
) const {
|
||||
// Just return NULL on CPU device
|
||||
*sampler = NULL;
|
||||
return true;
|
||||
}
|
||||
|
||||
//! Reallocates device memory obje
|
||||
virtual bool reallocMemory(amd::Memory& owner) const
|
||||
{
|
||||
return true;
|
||||
}
|
||||
//! Reallocates device memory obje
|
||||
virtual bool reallocMemory(amd::Memory& owner) const { return true; }
|
||||
|
||||
//! Just returns NULL as CPU devices use the host memory
|
||||
virtual device::Memory* createView(
|
||||
amd::Memory& owner, //!< Owner memory object
|
||||
const device::Memory& parent //!< Parent device memory object for the view
|
||||
) const
|
||||
{
|
||||
return NULL;
|
||||
}
|
||||
//! Just returns NULL as CPU devices use the host memory
|
||||
virtual device::Memory* createView(
|
||||
amd::Memory& owner, //!< Owner memory object
|
||||
const device::Memory& parent //!< Parent device memory object for the view
|
||||
) const {
|
||||
return NULL;
|
||||
}
|
||||
|
||||
//! Acquire external graphics API object in the host thread
|
||||
//! Needed for OpenGL objects on CPU device
|
||||
//! Acquire external graphics API object in the host thread
|
||||
//! Needed for OpenGL objects on CPU device
|
||||
|
||||
//! Return true if initialized interoperability, otherwise false
|
||||
virtual bool bindExternalDevice(uint flags, void* const pDevice[], void* pContext, bool validateOnly)
|
||||
{
|
||||
return true; // On CPU always avail if pD3DDevice is not NULL
|
||||
}
|
||||
//! Return true if initialized interoperability, otherwise false
|
||||
virtual bool bindExternalDevice(uint flags, void* const pDevice[], void* pContext,
|
||||
bool validateOnly) {
|
||||
return true; // On CPU always avail if pD3DDevice is not NULL
|
||||
}
|
||||
|
||||
virtual bool unbindExternalDevice(uint flags, void* const pDevice[], void* pContext, bool validateOnly)
|
||||
{
|
||||
return true;
|
||||
}
|
||||
virtual bool unbindExternalDevice(uint flags, void* const pDevice[], void* pContext,
|
||||
bool validateOnly) {
|
||||
return true;
|
||||
}
|
||||
|
||||
//! Gets a pointer to a region of host-visible memory for use as the target
|
||||
//! of a non-blocking map for a given memory object
|
||||
virtual void* allocMapTarget(
|
||||
amd::Memory& mem, //!< Abstraction layer memory object
|
||||
const amd::Coord3D& origin, //!< The map location in memory
|
||||
const amd::Coord3D& region, //!< The map region in memory
|
||||
uint mapFlags, //!< Map flags
|
||||
size_t* rowPitch = NULL, //!< Row pitch for the mapped memory
|
||||
size_t* slicePitch = NULL //!< Slice for the mapped memory
|
||||
);
|
||||
//! Gets a pointer to a region of host-visible memory for use as the target
|
||||
//! of a non-blocking map for a given memory object
|
||||
virtual void* allocMapTarget(amd::Memory& mem, //!< Abstraction layer memory object
|
||||
const amd::Coord3D& origin, //!< The map location in memory
|
||||
const amd::Coord3D& region, //!< The map region in memory
|
||||
uint mapFlags, //!< Map flags
|
||||
size_t* rowPitch = NULL, //!< Row pitch for the mapped memory
|
||||
size_t* slicePitch = NULL //!< Slice for the mapped memory
|
||||
);
|
||||
|
||||
//! Releases non-blocking map target memory
|
||||
virtual void freeMapTarget(amd::Memory& mem, void* target);
|
||||
//! Releases non-blocking map target memory
|
||||
virtual void freeMapTarget(amd::Memory& mem, void* target);
|
||||
|
||||
//! Empty implementation on a CPU device
|
||||
virtual bool globalFreeMemory(size_t* freeMemory) const { return false; }
|
||||
//! Empty implementation on a CPU device
|
||||
virtual bool globalFreeMemory(size_t* freeMemory) const { return false; }
|
||||
|
||||
//! Get CPU device settings
|
||||
const cpu::Settings& settings() const
|
||||
{ return reinterpret_cast<cpu::Settings&>(*settings_); }
|
||||
//! Get CPU device settings
|
||||
const cpu::Settings& settings() const { return reinterpret_cast<cpu::Settings&>(*settings_); }
|
||||
|
||||
bool hasAVXInstructions() const
|
||||
{ return (settings().cpuFeatures_ & Settings::AVXInstructions) ? true : false; }
|
||||
bool hasAVXInstructions() const {
|
||||
return (settings().cpuFeatures_ & Settings::AVXInstructions) ? true : false;
|
||||
}
|
||||
|
||||
bool hasFMA4Instructions() const
|
||||
{ return (settings().cpuFeatures_ & Settings::FMA4Instructions) ? true : false; }
|
||||
bool hasFMA4Instructions() const {
|
||||
return (settings().cpuFeatures_ & Settings::FMA4Instructions) ? true : false;
|
||||
}
|
||||
|
||||
static size_t getMaxWorkerThreadsNumber() { return maxWorkerThreads_; }
|
||||
static size_t getMaxWorkerThreadsNumber() { return maxWorkerThreads_; }
|
||||
|
||||
void setWorkerThreadsAffinity(
|
||||
cl_uint numWorkerThreads,
|
||||
const amd::Os::ThreadAffinityMask* threadsAffinityMask,
|
||||
uint& baseCoreId);
|
||||
void setWorkerThreadsAffinity(cl_uint numWorkerThreads,
|
||||
const amd::Os::ThreadAffinityMask* threadsAffinityMask,
|
||||
uint& baseCoreId);
|
||||
|
||||
const amd::Os::ThreadAffinityMask* getWorkerThreadsAffinity() const
|
||||
{
|
||||
return workerThreadsAffinity_;
|
||||
}
|
||||
//! host memory alloc
|
||||
virtual void* svmAlloc(amd::Context& context, size_t size, size_t alignment, cl_svm_mem_flags flags, void* svmPtr) const
|
||||
{
|
||||
return NULL;
|
||||
}
|
||||
const amd::Os::ThreadAffinityMask* getWorkerThreadsAffinity() const {
|
||||
return workerThreadsAffinity_;
|
||||
}
|
||||
//! host memory alloc
|
||||
virtual void* svmAlloc(amd::Context& context, size_t size, size_t alignment,
|
||||
cl_svm_mem_flags flags, void* svmPtr) const {
|
||||
return NULL;
|
||||
}
|
||||
|
||||
//! host memory deallocation
|
||||
virtual void svmFree(void* ptr) const
|
||||
{
|
||||
return;
|
||||
}
|
||||
private:
|
||||
bool initSubDevice(
|
||||
device::Info& info,
|
||||
cl_uint maxComputeUnits,
|
||||
const device::CreateSubDevicesInfo& create_info);
|
||||
//! host memory deallocation
|
||||
virtual void svmFree(void* ptr) const { return; }
|
||||
|
||||
cl_int partitionEqually(
|
||||
const device::CreateSubDevicesInfo& create_info,
|
||||
cl_uint num_entries,
|
||||
cl_device_id* devices,
|
||||
cl_uint* num_devices);
|
||||
private:
|
||||
bool initSubDevice(device::Info& info, cl_uint maxComputeUnits,
|
||||
const device::CreateSubDevicesInfo& create_info);
|
||||
|
||||
cl_int partitionByCounts(
|
||||
const device::CreateSubDevicesInfo& create_info,
|
||||
cl_uint num_entries,
|
||||
cl_device_id* devices,
|
||||
cl_uint* num_devices);
|
||||
cl_int partitionEqually(const device::CreateSubDevicesInfo& create_info, cl_uint num_entries,
|
||||
cl_device_id* devices, cl_uint* num_devices);
|
||||
|
||||
cl_int partitionByAffinityDomainNUMA(
|
||||
const device::CreateSubDevicesInfo& create_info,
|
||||
cl_uint num_entries,
|
||||
cl_device_id* devices,
|
||||
cl_uint* num_devices);
|
||||
cl_int partitionByCounts(const device::CreateSubDevicesInfo& create_info, cl_uint num_entries,
|
||||
cl_device_id* devices, cl_uint* num_devices);
|
||||
|
||||
cl_int partitionByAffinityDomainCacheLevel(
|
||||
const device::CreateSubDevicesInfo& create_info,
|
||||
cl_uint num_entries,
|
||||
cl_device_id* devices,
|
||||
cl_uint* num_devices);
|
||||
cl_int partitionByAffinityDomainNUMA(const device::CreateSubDevicesInfo& create_info,
|
||||
cl_uint num_entries, cl_device_id* devices,
|
||||
cl_uint* num_devices);
|
||||
|
||||
private:
|
||||
cl_int partitionByAffinityDomainCacheLevel(const device::CreateSubDevicesInfo& create_info,
|
||||
cl_uint num_entries, cl_device_id* devices,
|
||||
cl_uint* num_devices);
|
||||
|
||||
private:
|
||||
#if defined(__linux__) && defined(NUMA_SUPPORT)
|
||||
public:
|
||||
const nodemask_t* getNumaMask() const
|
||||
{
|
||||
return (info_.partitionCreateInfo_.type_ == device::PartitionType::BY_AFFINITY_DOMAIN &&
|
||||
info_.partitionCreateInfo_.byAffinityDomain_.numa_) ?
|
||||
numaMask_ : NULL;
|
||||
}
|
||||
public:
|
||||
const nodemask_t* getNumaMask() const {
|
||||
return (info_.partitionCreateInfo_.type_ == device::PartitionType::BY_AFFINITY_DOMAIN &&
|
||||
info_.partitionCreateInfo_.byAffinityDomain_.numa_)
|
||||
? numaMask_
|
||||
: NULL;
|
||||
}
|
||||
|
||||
private:
|
||||
union {
|
||||
nodemask_t* numaMask_;
|
||||
amd::Os::ThreadAffinityMask* workerThreadsAffinity_; //!< As the number of compute units.
|
||||
};
|
||||
private:
|
||||
union {
|
||||
nodemask_t* numaMask_;
|
||||
amd::Os::ThreadAffinityMask* workerThreadsAffinity_; //!< As the number of compute units.
|
||||
};
|
||||
#else
|
||||
amd::Os::ThreadAffinityMask* workerThreadsAffinity_; //!< As the number of compute units.
|
||||
amd::Os::ThreadAffinityMask* workerThreadsAffinity_; //!< As the number of compute units.
|
||||
#endif
|
||||
|
||||
static size_t maxWorkerThreads_; //!< Maximum number of Worker Threads
|
||||
static size_t maxWorkerThreads_; //!< Maximum number of Worker Threads
|
||||
};
|
||||
|
||||
/*! @}
|
||||
* @}
|
||||
*/
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
#endif // CPUDEVICE_HPP_
|
||||
#endif // CPUDEVICE_HPP_
|
||||
|
||||
@@ -24,4 +24,4 @@
|
||||
#define CPUFEAT_DX_SSE (1 < 25)
|
||||
#define CPUFEAT_DX_SSE2 (1 << 26)
|
||||
|
||||
#endif // CPUFEAT_HPP
|
||||
#endif // CPUFEAT_HPP
|
||||
|
||||
@@ -15,91 +15,79 @@
|
||||
namespace cpu {
|
||||
|
||||
//! \class CPU kernel
|
||||
class Kernel : public device::Kernel
|
||||
{
|
||||
private:
|
||||
const void* entryPoint_; //!< entry for the kernel
|
||||
class Kernel : public device::Kernel {
|
||||
private:
|
||||
const void* entryPoint_; //!< entry for the kernel
|
||||
|
||||
std::vector< std::pair<size_t, size_t> > args_;
|
||||
std::vector< std::pair < HCtoDCmap, size_t> > HCtoDCmaps_;
|
||||
std::vector< HCtoDCmap > internal_maps_;
|
||||
public:
|
||||
uint nature_; //!< kernel's nature
|
||||
uint privateSize_; //!< WorkItem's private memory size (in bytes)
|
||||
std::vector<std::pair<size_t, size_t> > args_;
|
||||
std::vector<std::pair<HCtoDCmap, size_t> > HCtoDCmaps_;
|
||||
std::vector<HCtoDCmap> internal_maps_;
|
||||
|
||||
private:
|
||||
//! Disable default copy constructor
|
||||
Kernel(const Kernel&);
|
||||
//! Disable operator=
|
||||
Kernel& operator=(const Kernel&);
|
||||
public:
|
||||
uint nature_; //!< kernel's nature
|
||||
uint privateSize_; //!< WorkItem's private memory size (in bytes)
|
||||
|
||||
public:
|
||||
void addArg(size_t size, size_t alignment) {
|
||||
args_.push_back(std::pair<size_t, size_t>(size, alignment));
|
||||
}
|
||||
private:
|
||||
//! Disable default copy constructor
|
||||
Kernel(const Kernel&);
|
||||
//! Disable operator=
|
||||
Kernel& operator=(const Kernel&);
|
||||
|
||||
size_t getArgSize(int argIndex) const {
|
||||
return args_[argIndex].first;
|
||||
}
|
||||
public:
|
||||
void addArg(size_t size, size_t alignment) {
|
||||
args_.push_back(std::pair<size_t, size_t>(size, alignment));
|
||||
}
|
||||
|
||||
size_t getArgAlignment(int argIndex) const {
|
||||
return args_[argIndex].second;
|
||||
}
|
||||
size_t getArgSize(int argIndex) const { return args_[argIndex].first; }
|
||||
|
||||
void addInternalMap(HCtoDCmap *new_map) {
|
||||
if (new_map != NULL) {
|
||||
internal_maps_.push_back(*new_map);
|
||||
this->addInternalMap(new_map->internal_field_map);
|
||||
this->addInternalMap(new_map->next_field_map);
|
||||
}
|
||||
else
|
||||
return;
|
||||
}
|
||||
size_t getArgAlignment(int argIndex) const { return args_[argIndex].second; }
|
||||
|
||||
void addHCtoDCmap(HCtoDCmap *new_map) {
|
||||
if (new_map != NULL) {
|
||||
if (HCtoDCmaps_.size() > 0)
|
||||
HCtoDCmaps_.push_back(std::pair< HCtoDCmap, size_t >(*new_map, HCtoDCmaps_.back().second));
|
||||
else
|
||||
HCtoDCmaps_.push_back(std::pair< HCtoDCmap, size_t >(*new_map, 0));
|
||||
}
|
||||
else
|
||||
return;
|
||||
}
|
||||
void addInternalMap(HCtoDCmap* new_map) {
|
||||
if (new_map != NULL) {
|
||||
internal_maps_.push_back(*new_map);
|
||||
this->addInternalMap(new_map->internal_field_map);
|
||||
this->addInternalMap(new_map->next_field_map);
|
||||
} else
|
||||
return;
|
||||
}
|
||||
|
||||
HCtoDCmap getHCtoDCmap(int mapIndex) const {
|
||||
return HCtoDCmaps_[mapIndex].first;
|
||||
}
|
||||
void addHCtoDCmap(HCtoDCmap* new_map) {
|
||||
if (new_map != NULL) {
|
||||
if (HCtoDCmaps_.size() > 0)
|
||||
HCtoDCmaps_.push_back(std::pair<HCtoDCmap, size_t>(*new_map, HCtoDCmaps_.back().second));
|
||||
else
|
||||
HCtoDCmaps_.push_back(std::pair<HCtoDCmap, size_t>(*new_map, 0));
|
||||
} else
|
||||
return;
|
||||
}
|
||||
|
||||
HCtoDCmap getHCtoDCmap(int mapIndex) const { return HCtoDCmaps_[mapIndex].first; }
|
||||
|
||||
|
||||
uint getArgNumber() {
|
||||
return HCtoDCmaps_.size();
|
||||
}
|
||||
uint getArgNumber() { return HCtoDCmaps_.size(); }
|
||||
|
||||
//! Default constructor
|
||||
Kernel(const std::string& name)
|
||||
: device::Kernel(name), entryPoint_(NULL), nature_(0),
|
||||
privateSize_(CLK_PRIVATE_MEMORY_SIZE)
|
||||
{
|
||||
workGroupInfo_.size_ = CPU_MAX_WORKGROUP_SIZE;
|
||||
}
|
||||
//! Default constructor
|
||||
Kernel(const std::string& name)
|
||||
: device::Kernel(name), entryPoint_(NULL), nature_(0), privateSize_(CLK_PRIVATE_MEMORY_SIZE) {
|
||||
workGroupInfo_.size_ = CPU_MAX_WORKGROUP_SIZE;
|
||||
}
|
||||
|
||||
//! Default destructor
|
||||
~Kernel() {}
|
||||
//! Default destructor
|
||||
~Kernel() {}
|
||||
|
||||
//! Returns the CPU kernel entry point
|
||||
const void* getEntryPoint() const { return entryPoint_; }
|
||||
//! Returns the CPU kernel entry point
|
||||
const void* getEntryPoint() const { return entryPoint_; }
|
||||
|
||||
//! Sets the CPU kernel entry point
|
||||
void setEntryPoint(const void* entryPoint) { entryPoint_ = entryPoint; }
|
||||
//! Sets the CPU kernel entry point
|
||||
void setEntryPoint(const void* entryPoint) { entryPoint_ = entryPoint; }
|
||||
|
||||
//! Returns true if the kernel has a call to barrier
|
||||
bool hasBarrier() const { return 0 != (nature_ & KN_HAS_BARRIER); }
|
||||
//! Returns true if the kernel has a call to barrier
|
||||
bool hasBarrier() const { return 0 != (nature_ & KN_HAS_BARRIER); }
|
||||
|
||||
//! Returns the private memory size of a single WorkItem
|
||||
uint getWorkItemPrivateMemSize() const { return privateSize_; }
|
||||
//! Returns the private memory size of a single WorkItem
|
||||
uint getWorkItemPrivateMemSize() const { return privateSize_; }
|
||||
};
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
#endif // CPUKERNEL_HPP_
|
||||
#endif // CPUKERNEL_HPP_
|
||||
|
||||
@@ -25,354 +25,403 @@ using std::min;
|
||||
using std::max;
|
||||
|
||||
namespace cpu {
|
||||
HCtoDCmap::HCtoDCmap(const clk_parameter_descriptor_t* desc, unsigned int level_alignment, unsigned int index, unsigned int init_offset)
|
||||
{
|
||||
level_alignment = std::max(level_alignment, 1u); // Minimal possible alignment is 1 and alignment is used as a divisor below.
|
||||
//Initialize fields
|
||||
hc_offset = 0;
|
||||
hc_size = 0;
|
||||
dc_offset = 0;
|
||||
dc_size = 0;
|
||||
hc_alignment = level_alignment;
|
||||
dc_alignment = level_alignment;
|
||||
internal_field_map = NULL;
|
||||
next_field_map = NULL;
|
||||
return;
|
||||
}
|
||||
HCtoDCmap::HCtoDCmap(const clk_parameter_descriptor_t* desc, unsigned int level_alignment,
|
||||
unsigned int index, unsigned int init_offset) {
|
||||
level_alignment =
|
||||
std::max(level_alignment,
|
||||
1u); // Minimal possible alignment is 1 and alignment is used as a divisor below.
|
||||
// Initialize fields
|
||||
hc_offset = 0;
|
||||
hc_size = 0;
|
||||
dc_offset = 0;
|
||||
dc_size = 0;
|
||||
hc_alignment = level_alignment;
|
||||
dc_alignment = level_alignment;
|
||||
internal_field_map = NULL;
|
||||
next_field_map = NULL;
|
||||
return;
|
||||
}
|
||||
|
||||
HCtoDCmap::~HCtoDCmap()
|
||||
{
|
||||
return;
|
||||
}
|
||||
HCtoDCmap::~HCtoDCmap() { return; }
|
||||
|
||||
//Helper to find sizes of each scalar type
|
||||
size_t HCtoDCmap::getHostScalarParamSize(const clk_value_type_t type) const
|
||||
{
|
||||
size_t size = 0;
|
||||
switch (type) {
|
||||
case T_CHAR:
|
||||
size = 1;
|
||||
break;
|
||||
case T_SHORT: case T_CHAR2:
|
||||
size = 2;
|
||||
break;
|
||||
case T_FLOAT: case T_INT: case T_CHAR4:
|
||||
case T_SHORT2: case T_CHAR3:
|
||||
size = 4;
|
||||
break;
|
||||
case T_SAMPLER:
|
||||
size = 4;
|
||||
break;
|
||||
case T_LONG: case T_DOUBLE: case T_CHAR8:
|
||||
case T_SHORT4: case T_INT2: case T_FLOAT2:
|
||||
case T_SHORT3:
|
||||
size = 8;
|
||||
break;
|
||||
case T_INT3: case T_FLOAT3:
|
||||
case T_CHAR16: case T_SHORT8: case T_INT4:
|
||||
case T_FLOAT4: case T_LONG2: case T_DOUBLE2:
|
||||
size = 16;
|
||||
break;
|
||||
case T_LONG3: case T_DOUBLE3:
|
||||
case T_SHORT16: case T_INT8: case T_FLOAT8:
|
||||
case T_LONG4: case T_DOUBLE4:
|
||||
size = 32;
|
||||
break;
|
||||
case T_INT16: case T_FLOAT16: case T_LONG8:
|
||||
case T_DOUBLE8:
|
||||
size = 64;
|
||||
break;
|
||||
case T_LONG16: case T_DOUBLE16:
|
||||
size = 128;
|
||||
break;
|
||||
case T_POINTER: case T_VOID:
|
||||
size = sizeof(void*);
|
||||
break;
|
||||
default:
|
||||
assert(0 && "unknown scalar parameter size");
|
||||
break;
|
||||
}
|
||||
return size;
|
||||
}
|
||||
// Helper to find sizes of each scalar type
|
||||
size_t HCtoDCmap::getHostScalarParamSize(const clk_value_type_t type) const {
|
||||
size_t size = 0;
|
||||
switch (type) {
|
||||
case T_CHAR:
|
||||
size = 1;
|
||||
break;
|
||||
case T_SHORT:
|
||||
case T_CHAR2:
|
||||
size = 2;
|
||||
break;
|
||||
case T_FLOAT:
|
||||
case T_INT:
|
||||
case T_CHAR4:
|
||||
case T_SHORT2:
|
||||
case T_CHAR3:
|
||||
size = 4;
|
||||
break;
|
||||
case T_SAMPLER:
|
||||
size = 4;
|
||||
break;
|
||||
case T_LONG:
|
||||
case T_DOUBLE:
|
||||
case T_CHAR8:
|
||||
case T_SHORT4:
|
||||
case T_INT2:
|
||||
case T_FLOAT2:
|
||||
case T_SHORT3:
|
||||
size = 8;
|
||||
break;
|
||||
case T_INT3:
|
||||
case T_FLOAT3:
|
||||
case T_CHAR16:
|
||||
case T_SHORT8:
|
||||
case T_INT4:
|
||||
case T_FLOAT4:
|
||||
case T_LONG2:
|
||||
case T_DOUBLE2:
|
||||
size = 16;
|
||||
break;
|
||||
case T_LONG3:
|
||||
case T_DOUBLE3:
|
||||
case T_SHORT16:
|
||||
case T_INT8:
|
||||
case T_FLOAT8:
|
||||
case T_LONG4:
|
||||
case T_DOUBLE4:
|
||||
size = 32;
|
||||
break;
|
||||
case T_INT16:
|
||||
case T_FLOAT16:
|
||||
case T_LONG8:
|
||||
case T_DOUBLE8:
|
||||
size = 64;
|
||||
break;
|
||||
case T_LONG16:
|
||||
case T_DOUBLE16:
|
||||
size = 128;
|
||||
break;
|
||||
case T_POINTER:
|
||||
case T_VOID:
|
||||
size = sizeof(void*);
|
||||
break;
|
||||
default:
|
||||
assert(0 && "unknown scalar parameter size");
|
||||
break;
|
||||
}
|
||||
return size;
|
||||
}
|
||||
|
||||
size_t HCtoDCmap::getScalarAlignment(const clk_value_type_t type, bool isHost) const
|
||||
{
|
||||
size_t align = 0;
|
||||
switch (type) {
|
||||
case T_CHAR:
|
||||
align = 1;
|
||||
break;
|
||||
case T_SHORT: case T_CHAR2:
|
||||
align = 2;
|
||||
break;
|
||||
case T_FLOAT: case T_INT: case T_CHAR4:
|
||||
case T_SHORT2: case T_CHAR3:
|
||||
align = 4;
|
||||
break;
|
||||
case T_SAMPLER:
|
||||
align = sizeof(uint32_t);
|
||||
break;
|
||||
case T_LONG:
|
||||
#if defined(_WIN32)
|
||||
align = 8;
|
||||
#else
|
||||
align = isHost? 8 : LP64_SWITCH(4, 8);
|
||||
#endif
|
||||
break;
|
||||
case T_DOUBLE:
|
||||
#if defined(_WIN32)
|
||||
align = 8;
|
||||
#else
|
||||
align = LP64_SWITCH(4, 8);
|
||||
#endif
|
||||
break;
|
||||
case T_CHAR8:
|
||||
case T_SHORT4: case T_INT2: case T_FLOAT2:
|
||||
case T_SHORT3:
|
||||
align = 4;
|
||||
break;
|
||||
case T_INT3: case T_FLOAT3:
|
||||
case T_CHAR16: case T_SHORT8: case T_INT4:
|
||||
case T_FLOAT4: case T_LONG2: case T_DOUBLE2:
|
||||
case T_LONG3: case T_DOUBLE3:
|
||||
case T_SHORT16: case T_INT8: case T_FLOAT8:
|
||||
case T_LONG4: case T_DOUBLE4:
|
||||
case T_INT16: case T_FLOAT16: case T_LONG8:
|
||||
case T_DOUBLE8:
|
||||
case T_LONG16: case T_DOUBLE16:
|
||||
align = LP64_SWITCH(4, 8);
|
||||
break;
|
||||
case T_POINTER: case T_VOID:
|
||||
align = sizeof(void*);
|
||||
break;
|
||||
default:
|
||||
assert(0 && "unknown scalar parameter alignment");
|
||||
break;
|
||||
}
|
||||
return align;
|
||||
}
|
||||
size_t HCtoDCmap::getScalarAlignment(const clk_value_type_t type, bool isHost) const {
|
||||
size_t align = 0;
|
||||
switch (type) {
|
||||
case T_CHAR:
|
||||
align = 1;
|
||||
break;
|
||||
case T_SHORT:
|
||||
case T_CHAR2:
|
||||
align = 2;
|
||||
break;
|
||||
case T_FLOAT:
|
||||
case T_INT:
|
||||
case T_CHAR4:
|
||||
case T_SHORT2:
|
||||
case T_CHAR3:
|
||||
align = 4;
|
||||
break;
|
||||
case T_SAMPLER:
|
||||
align = sizeof(uint32_t);
|
||||
break;
|
||||
case T_LONG:
|
||||
#if defined(_WIN32)
|
||||
align = 8;
|
||||
#else
|
||||
align = isHost ? 8 : LP64_SWITCH(4, 8);
|
||||
#endif
|
||||
break;
|
||||
case T_DOUBLE:
|
||||
#if defined(_WIN32)
|
||||
align = 8;
|
||||
#else
|
||||
align = LP64_SWITCH(4, 8);
|
||||
#endif
|
||||
break;
|
||||
case T_CHAR8:
|
||||
case T_SHORT4:
|
||||
case T_INT2:
|
||||
case T_FLOAT2:
|
||||
case T_SHORT3:
|
||||
align = 4;
|
||||
break;
|
||||
case T_INT3:
|
||||
case T_FLOAT3:
|
||||
case T_CHAR16:
|
||||
case T_SHORT8:
|
||||
case T_INT4:
|
||||
case T_FLOAT4:
|
||||
case T_LONG2:
|
||||
case T_DOUBLE2:
|
||||
case T_LONG3:
|
||||
case T_DOUBLE3:
|
||||
case T_SHORT16:
|
||||
case T_INT8:
|
||||
case T_FLOAT8:
|
||||
case T_LONG4:
|
||||
case T_DOUBLE4:
|
||||
case T_INT16:
|
||||
case T_FLOAT16:
|
||||
case T_LONG8:
|
||||
case T_DOUBLE8:
|
||||
case T_LONG16:
|
||||
case T_DOUBLE16:
|
||||
align = LP64_SWITCH(4, 8);
|
||||
break;
|
||||
case T_POINTER:
|
||||
case T_VOID:
|
||||
align = sizeof(void*);
|
||||
break;
|
||||
default:
|
||||
assert(0 && "unknown scalar parameter alignment");
|
||||
break;
|
||||
}
|
||||
return align;
|
||||
}
|
||||
|
||||
// Align up arguments within each map, return the size of current map parameter
|
||||
// Input current alignment of the parameter, size of outer struct if it exists
|
||||
void HCtoDCmap::align_map(unsigned outer_hc_alignment, unsigned outer_dc_alignment, unsigned &outer_hc_size, unsigned &outer_dc_size, int &inStruct)
|
||||
{
|
||||
unsigned map_param_size = 0;
|
||||
if (internal_field_map != NULL) {
|
||||
hc_size = 0; //Recalculate size to account for internal offsets
|
||||
inStruct++;
|
||||
internal_field_map->align_map(hc_alignment, dc_alignment, hc_size, dc_size, inStruct); // align internal struct, might alter size of this struct
|
||||
if (hc_alignment != 1 && hc_size%hc_alignment)
|
||||
hc_size = max(hc_size, hc_size - (hc_size%hc_alignment) + hc_alignment);
|
||||
if (dc_alignment != 1 && dc_size%dc_alignment)
|
||||
dc_size = max(dc_size, dc_size - (dc_size%dc_alignment) + dc_alignment);
|
||||
}
|
||||
// Use map_param_size to store current parameter size after adjusting alignment
|
||||
if (hc_alignment != 1 && hc_size % hc_alignment != 0) {
|
||||
map_param_size = max(hc_alignment, hc_size - (hc_size%hc_alignment) + hc_alignment);
|
||||
}
|
||||
else {
|
||||
map_param_size = max(hc_alignment, hc_size);
|
||||
}
|
||||
if (next_field_map != NULL) {
|
||||
next_field_map->hc_offset = this->next_offset(hc_offset, map_param_size, inStruct);
|
||||
next_field_map->align_map(outer_hc_alignment, outer_dc_alignment, outer_hc_size, outer_dc_size, inStruct);
|
||||
// Reset parameter size for char padding
|
||||
if (next_field_map->type == T_CHAR)
|
||||
map_param_size = 1;
|
||||
}
|
||||
else
|
||||
{
|
||||
// Moving out of struct
|
||||
if (inStruct > 0)
|
||||
inStruct--;
|
||||
if (type == T_CHAR)
|
||||
map_param_size = 1;
|
||||
}
|
||||
outer_hc_size = max(outer_hc_size, hc_offset+map_param_size);
|
||||
outer_dc_size = max(outer_dc_size, dc_offset+dc_size);
|
||||
return;
|
||||
}
|
||||
// Align up arguments within each map, return the size of current map parameter
|
||||
// Input current alignment of the parameter, size of outer struct if it exists
|
||||
void HCtoDCmap::align_map(unsigned outer_hc_alignment, unsigned outer_dc_alignment,
|
||||
unsigned& outer_hc_size, unsigned& outer_dc_size, int& inStruct) {
|
||||
unsigned map_param_size = 0;
|
||||
if (internal_field_map != NULL) {
|
||||
hc_size = 0; // Recalculate size to account for internal offsets
|
||||
inStruct++;
|
||||
internal_field_map->align_map(
|
||||
hc_alignment, dc_alignment, hc_size, dc_size,
|
||||
inStruct); // align internal struct, might alter size of this struct
|
||||
if (hc_alignment != 1 && hc_size % hc_alignment)
|
||||
hc_size = max(hc_size, hc_size - (hc_size % hc_alignment) + hc_alignment);
|
||||
if (dc_alignment != 1 && dc_size % dc_alignment)
|
||||
dc_size = max(dc_size, dc_size - (dc_size % dc_alignment) + dc_alignment);
|
||||
}
|
||||
// Use map_param_size to store current parameter size after adjusting alignment
|
||||
if (hc_alignment != 1 && hc_size % hc_alignment != 0) {
|
||||
map_param_size = max(hc_alignment, hc_size - (hc_size % hc_alignment) + hc_alignment);
|
||||
} else {
|
||||
map_param_size = max(hc_alignment, hc_size);
|
||||
}
|
||||
if (next_field_map != NULL) {
|
||||
next_field_map->hc_offset = this->next_offset(hc_offset, map_param_size, inStruct);
|
||||
next_field_map->align_map(outer_hc_alignment, outer_dc_alignment, outer_hc_size, outer_dc_size,
|
||||
inStruct);
|
||||
// Reset parameter size for char padding
|
||||
if (next_field_map->type == T_CHAR) map_param_size = 1;
|
||||
} else {
|
||||
// Moving out of struct
|
||||
if (inStruct > 0) inStruct--;
|
||||
if (type == T_CHAR) map_param_size = 1;
|
||||
}
|
||||
outer_hc_size = max(outer_hc_size, hc_offset + map_param_size);
|
||||
outer_dc_size = max(outer_dc_size, dc_offset + dc_size);
|
||||
return;
|
||||
}
|
||||
|
||||
// Return current size of map, calculate internal maps and process next args if in struct.
|
||||
// Alignment: alignment flag for members in case of structs, alignment of scalar otherwise.
|
||||
int HCtoDCmap::compute_map(const clk_parameter_descriptor_t* desc, unsigned int &outer_hc_alignment, unsigned int &outer_dc_alignment, unsigned int init_offset, int& inStruct, int& index_out)
|
||||
{
|
||||
unsigned internal_index;
|
||||
// Return current size of map, calculate internal maps and process next args if in struct.
|
||||
// Alignment: alignment flag for members in case of structs, alignment of scalar otherwise.
|
||||
int HCtoDCmap::compute_map(const clk_parameter_descriptor_t* desc, unsigned int& outer_hc_alignment,
|
||||
unsigned int& outer_dc_alignment, unsigned int init_offset,
|
||||
int& inStruct, int& index_out) {
|
||||
unsigned internal_index;
|
||||
internal_index = index_out;
|
||||
unsigned int next_offset = init_offset;
|
||||
unsigned struct_size = 0;
|
||||
type = desc[internal_index].type;
|
||||
|
||||
if (desc[internal_index].type == T_STRUCT) {
|
||||
// Moving into struct, go to next index
|
||||
inStruct++;
|
||||
hc_offset = init_offset;
|
||||
if (desc[index_out + 1].type != T_VOID) {
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
internal_field_map = new HCtoDCmap(desc, 0, internal_index, init_offset);
|
||||
hc_size = internal_field_map->compute_map(desc, hc_alignment, dc_alignment, next_offset,
|
||||
inStruct, index_out);
|
||||
hc_alignment =
|
||||
max(hc_alignment,
|
||||
internal_field_map->hc_alignment); // Adjust alignment to biggest member alignment
|
||||
struct_size = hc_size;
|
||||
internal_index = index_out;
|
||||
outer_hc_alignment = max(outer_hc_alignment, hc_alignment);
|
||||
if (inStruct > 0) {
|
||||
if (desc[index_out + 1].type != T_VOID) {
|
||||
// Still inside struct and not done
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
next_field_map = new HCtoDCmap(desc, 0, internal_index, next_offset);
|
||||
struct_size = hc_size;
|
||||
struct_size += next_field_map->compute_map(desc, outer_hc_alignment, outer_dc_alignment,
|
||||
next_offset, inStruct, index_out);
|
||||
next_offset = max(next_field_map->hc_offset + next_field_map->hc_size,
|
||||
next_field_map->hc_offset + hc_alignment);
|
||||
// running count of strucdc_size = hc_size + size of next member
|
||||
return struct_size;
|
||||
} else {
|
||||
// Moving out of struct, go to next index
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
inStruct--;
|
||||
return hc_size; // return last struct member size
|
||||
}
|
||||
}
|
||||
}
|
||||
} else if (desc[internal_index].type == T_PAD) {
|
||||
// Struct has padding
|
||||
hc_offset = init_offset;
|
||||
if (desc[index_out + 1].type != T_VOID) {
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
internal_field_map = new HCtoDCmap(desc, 0, internal_index, init_offset);
|
||||
hc_size = internal_field_map->compute_map(desc, hc_alignment, dc_alignment, next_offset,
|
||||
inStruct, index_out);
|
||||
// Adjust alignment to biggest member alignment
|
||||
hc_alignment = 1;
|
||||
dc_alignment = 1;
|
||||
unsigned pad_size = hc_size;
|
||||
internal_index = index_out;
|
||||
if (desc[index_out + 1].type != T_VOID) {
|
||||
// Still inside padding and not done
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
unsigned int next_offset = init_offset;
|
||||
unsigned struct_size = 0;
|
||||
type = desc[internal_index].type;
|
||||
|
||||
if (desc[internal_index].type == T_STRUCT) {
|
||||
//Moving into struct, go to next index
|
||||
inStruct++;
|
||||
hc_offset = init_offset;
|
||||
if (desc[index_out+1].type != T_VOID) {
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
internal_field_map = new HCtoDCmap(desc, 0, internal_index, init_offset);
|
||||
hc_size = internal_field_map->compute_map(desc, hc_alignment, dc_alignment, next_offset, inStruct, index_out);
|
||||
hc_alignment = max(hc_alignment, internal_field_map->hc_alignment); // Adjust alignment to biggest member alignment
|
||||
struct_size = hc_size;
|
||||
internal_index = index_out;
|
||||
outer_hc_alignment = max(outer_hc_alignment, hc_alignment);
|
||||
if (inStruct > 0) {
|
||||
if (desc[index_out+1].type != T_VOID) {
|
||||
//Still inside struct and not done
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
next_field_map = new HCtoDCmap(desc, 0, internal_index, next_offset);
|
||||
struct_size = hc_size;
|
||||
struct_size += next_field_map->compute_map(desc, outer_hc_alignment, outer_dc_alignment, next_offset, inStruct, index_out);
|
||||
next_offset = max(next_field_map->hc_offset+next_field_map->hc_size, next_field_map->hc_offset+hc_alignment);
|
||||
// running count of strucdc_size = hc_size + size of next member
|
||||
return struct_size;
|
||||
}
|
||||
else {
|
||||
//Moving out of struct, go to next index
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
inStruct--;
|
||||
return hc_size; //return last struct member size
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
else if (desc[internal_index].type == T_PAD) {
|
||||
//Struct has padding
|
||||
hc_offset = init_offset;
|
||||
if (desc[index_out+1].type != T_VOID) {
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
internal_field_map = new HCtoDCmap(desc, 0, internal_index, init_offset);
|
||||
hc_size = internal_field_map->compute_map(desc, hc_alignment, dc_alignment, next_offset, inStruct, index_out);
|
||||
// Adjust alignment to biggest member alignment
|
||||
hc_alignment = 1;
|
||||
dc_alignment = 1;
|
||||
unsigned pad_size = hc_size;
|
||||
internal_index = index_out;
|
||||
if (desc[index_out+1].type != T_VOID) {
|
||||
//Still inside padding and not done
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
next_field_map = new HCtoDCmap(desc, 0, internal_index, next_offset);
|
||||
pad_size = hc_size;
|
||||
pad_size += next_field_map->compute_map(desc, outer_hc_alignment, outer_dc_alignment, next_offset, inStruct, index_out);
|
||||
next_offset = max(next_field_map->hc_offset+next_field_map->hc_size, next_field_map->hc_offset+hc_alignment);
|
||||
// running count of padding dc_size = hc_size + size of next member
|
||||
return pad_size;
|
||||
}
|
||||
else {
|
||||
//Moving out of struct, go to next index
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
return hc_size; //return last padding member size
|
||||
}
|
||||
}
|
||||
}
|
||||
else {
|
||||
//Scalar parameter
|
||||
hc_offset = init_offset;
|
||||
hc_size = getHostScalarParamSize(desc[internal_index].type);
|
||||
dc_size = hc_size;
|
||||
hc_alignment = getScalarAlignment(desc[internal_index].type, true);
|
||||
dc_alignment = getScalarAlignment(desc[internal_index].type, false);
|
||||
outer_hc_alignment = max(outer_hc_alignment, hc_alignment); //Adjust alignment of upper level struct if necessary, upper level alignment = max alignment of members
|
||||
outer_dc_alignment = max(outer_dc_alignment, dc_alignment); //Adjust alignment of upper level struct if necessary, upper level alignment = max alignment of members
|
||||
if (inStruct > 0) {
|
||||
if (desc[index_out+1].type != T_VOID) {
|
||||
//Still inside struct and not done
|
||||
index_out++;
|
||||
next_field_map = new HCtoDCmap(desc, outer_hc_alignment, internal_index, next_offset);
|
||||
struct_size = hc_size;
|
||||
struct_size += next_field_map->compute_map(desc, outer_hc_alignment, outer_dc_alignment, next_offset, inStruct, index_out);
|
||||
next_offset = hc_offset+hc_alignment;
|
||||
outer_hc_alignment = max(outer_hc_alignment, next_field_map->hc_alignment);
|
||||
outer_dc_alignment = max(outer_dc_alignment, next_field_map->dc_alignment);
|
||||
// running count of strucdc_size = hc_size + size of next member
|
||||
return struct_size;
|
||||
}
|
||||
else {
|
||||
//Moving out of struct, go to next index
|
||||
index_out++;
|
||||
inStruct--;
|
||||
return hc_size; //return last struct member size
|
||||
}
|
||||
}
|
||||
}
|
||||
return hc_size;
|
||||
next_field_map = new HCtoDCmap(desc, 0, internal_index, next_offset);
|
||||
pad_size = hc_size;
|
||||
pad_size += next_field_map->compute_map(desc, outer_hc_alignment, outer_dc_alignment,
|
||||
next_offset, inStruct, index_out);
|
||||
next_offset = max(next_field_map->hc_offset + next_field_map->hc_size,
|
||||
next_field_map->hc_offset + hc_alignment);
|
||||
// running count of padding dc_size = hc_size + size of next member
|
||||
return pad_size;
|
||||
} else {
|
||||
// Moving out of struct, go to next index
|
||||
index_out++;
|
||||
internal_index = index_out;
|
||||
return hc_size; // return last padding member size
|
||||
}
|
||||
}
|
||||
|
||||
// Adjust offset for source and target, return next source offset
|
||||
unsigned HCtoDCmap::next_offset(unsigned current_offset, unsigned &map_param_size, int& inStruct_flag)
|
||||
{
|
||||
unsigned next_offset = current_offset;
|
||||
if (next_field_map == NULL) {
|
||||
assert(0 && "invalid next struct field map");
|
||||
return next_offset;
|
||||
}
|
||||
else {
|
||||
// Ignore alignment when a char occurs to account for padding
|
||||
if (type == T_PAD) {
|
||||
next_field_map->dc_offset = dc_offset + dc_size;
|
||||
next_offset = current_offset + hc_size;
|
||||
}
|
||||
else {
|
||||
if ((dc_offset + dc_size) % next_field_map->dc_alignment != 0) {
|
||||
this->next_field_map->dc_offset = dc_offset + dc_size - (dc_size % next_field_map->dc_alignment) + next_field_map->dc_alignment;
|
||||
}
|
||||
else {
|
||||
this->next_field_map->dc_offset = dc_offset + max(dc_size, next_field_map->dc_alignment);
|
||||
}
|
||||
if ((hc_offset + hc_size) % next_field_map->hc_alignment != 0) {
|
||||
next_offset = hc_offset + hc_size - (hc_size % next_field_map->hc_alignment) + next_field_map->hc_alignment;
|
||||
}
|
||||
else {
|
||||
next_offset = hc_offset + max(next_field_map->hc_alignment, map_param_size);
|
||||
}
|
||||
}
|
||||
return next_offset;
|
||||
}
|
||||
} else {
|
||||
// Scalar parameter
|
||||
hc_offset = init_offset;
|
||||
hc_size = getHostScalarParamSize(desc[internal_index].type);
|
||||
dc_size = hc_size;
|
||||
hc_alignment = getScalarAlignment(desc[internal_index].type, true);
|
||||
dc_alignment = getScalarAlignment(desc[internal_index].type, false);
|
||||
outer_hc_alignment = max(outer_hc_alignment, hc_alignment); // Adjust alignment of upper level
|
||||
// struct if necessary, upper level
|
||||
// alignment = max alignment of
|
||||
// members
|
||||
outer_dc_alignment = max(outer_dc_alignment, dc_alignment); // Adjust alignment of upper level
|
||||
// struct if necessary, upper level
|
||||
// alignment = max alignment of
|
||||
// members
|
||||
if (inStruct > 0) {
|
||||
if (desc[index_out + 1].type != T_VOID) {
|
||||
// Still inside struct and not done
|
||||
index_out++;
|
||||
next_field_map = new HCtoDCmap(desc, outer_hc_alignment, internal_index, next_offset);
|
||||
struct_size = hc_size;
|
||||
struct_size += next_field_map->compute_map(desc, outer_hc_alignment, outer_dc_alignment,
|
||||
next_offset, inStruct, index_out);
|
||||
next_offset = hc_offset + hc_alignment;
|
||||
outer_hc_alignment = max(outer_hc_alignment, next_field_map->hc_alignment);
|
||||
outer_dc_alignment = max(outer_dc_alignment, next_field_map->dc_alignment);
|
||||
// running count of strucdc_size = hc_size + size of next member
|
||||
return struct_size;
|
||||
} else {
|
||||
// Moving out of struct, go to next index
|
||||
index_out++;
|
||||
inStruct--;
|
||||
return hc_size; // return last struct member size
|
||||
}
|
||||
}
|
||||
}
|
||||
return hc_size;
|
||||
}
|
||||
|
||||
// Copy memory according to mapping
|
||||
unsigned int HCtoDCmap::copy_params(void *dst, const void *src, unsigned int arg_offset, int& error_code, int &inStruct) const
|
||||
{
|
||||
unsigned int padding = 0;
|
||||
// Pad offset to be aligned by 8 if parameter is double, not as struct field
|
||||
if ((arg_offset) % 8 != 0 && (type == T_DOUBLE) && inStruct == 0)
|
||||
padding = hc_alignment-((arg_offset+dc_offset)%hc_alignment);
|
||||
#if defined(_WIN32)
|
||||
// In windows, double is aligned by 8, add padding to struct if it contains double
|
||||
if ((arg_offset+dc_offset) % 8 != 0 && hc_alignment == 8)
|
||||
padding = hc_alignment-((arg_offset+dc_offset)%hc_alignment);
|
||||
#endif
|
||||
::memcpy(reinterpret_cast<void *>(reinterpret_cast<unsigned char*>(dst)+padding), src, hc_size);
|
||||
#if defined(_WIN32)
|
||||
if (internal_field_map != NULL) {
|
||||
inStruct++;
|
||||
void *internal_dst = reinterpret_cast<void *>(reinterpret_cast<unsigned char*>(dst)+padding);
|
||||
internal_field_map->copy_params(internal_dst, src, arg_offset+padding, error_code, inStruct);
|
||||
inStruct--;
|
||||
}
|
||||
if (next_field_map != NULL) {
|
||||
void *next_dst = reinterpret_cast<void *>(reinterpret_cast<unsigned char*>(dst)+next_field_map->dc_offset); // Next field starts with padding
|
||||
const void *next_src = reinterpret_cast<const void *>(reinterpret_cast<const unsigned char*>(src)+next_field_map->hc_offset);
|
||||
next_field_map->copy_params(next_dst, next_src, arg_offset+next_field_map->dc_offset, error_code, inStruct);
|
||||
}
|
||||
#else
|
||||
if (internal_field_map != NULL) {
|
||||
inStruct++;
|
||||
internal_field_map->copy_params(dst, src, arg_offset, error_code, inStruct);
|
||||
inStruct--;
|
||||
}
|
||||
if (next_field_map != NULL) {
|
||||
void *next_dst = reinterpret_cast<void *>(reinterpret_cast<unsigned char*>(dst)+next_field_map->dc_offset);
|
||||
const void *next_src = reinterpret_cast<const void *>(reinterpret_cast<const unsigned char*>(src)+next_field_map->hc_offset);
|
||||
next_field_map->copy_params(next_dst, next_src, arg_offset, error_code, inStruct);
|
||||
}
|
||||
#endif
|
||||
return padding;
|
||||
// Adjust offset for source and target, return next source offset
|
||||
unsigned HCtoDCmap::next_offset(unsigned current_offset, unsigned& map_param_size,
|
||||
int& inStruct_flag) {
|
||||
unsigned next_offset = current_offset;
|
||||
if (next_field_map == NULL) {
|
||||
assert(0 && "invalid next struct field map");
|
||||
return next_offset;
|
||||
} else {
|
||||
// Ignore alignment when a char occurs to account for padding
|
||||
if (type == T_PAD) {
|
||||
next_field_map->dc_offset = dc_offset + dc_size;
|
||||
next_offset = current_offset + hc_size;
|
||||
} else {
|
||||
if ((dc_offset + dc_size) % next_field_map->dc_alignment != 0) {
|
||||
this->next_field_map->dc_offset = dc_offset + dc_size -
|
||||
(dc_size % next_field_map->dc_alignment) + next_field_map->dc_alignment;
|
||||
} else {
|
||||
this->next_field_map->dc_offset = dc_offset + max(dc_size, next_field_map->dc_alignment);
|
||||
}
|
||||
if ((hc_offset + hc_size) % next_field_map->hc_alignment != 0) {
|
||||
next_offset = hc_offset + hc_size - (hc_size % next_field_map->hc_alignment) +
|
||||
next_field_map->hc_alignment;
|
||||
} else {
|
||||
next_offset = hc_offset + max(next_field_map->hc_alignment, map_param_size);
|
||||
}
|
||||
}
|
||||
return next_offset;
|
||||
}
|
||||
}
|
||||
|
||||
} //namespace cpu
|
||||
// Copy memory according to mapping
|
||||
unsigned int HCtoDCmap::copy_params(void* dst, const void* src, unsigned int arg_offset,
|
||||
int& error_code, int& inStruct) const {
|
||||
unsigned int padding = 0;
|
||||
// Pad offset to be aligned by 8 if parameter is double, not as struct field
|
||||
if ((arg_offset) % 8 != 0 && (type == T_DOUBLE) && inStruct == 0)
|
||||
padding = hc_alignment - ((arg_offset + dc_offset) % hc_alignment);
|
||||
#if defined(_WIN32)
|
||||
// In windows, double is aligned by 8, add padding to struct if it contains double
|
||||
if ((arg_offset + dc_offset) % 8 != 0 && hc_alignment == 8)
|
||||
padding = hc_alignment - ((arg_offset + dc_offset) % hc_alignment);
|
||||
#endif
|
||||
::memcpy(reinterpret_cast<void*>(reinterpret_cast<unsigned char*>(dst) + padding), src, hc_size);
|
||||
#if defined(_WIN32)
|
||||
if (internal_field_map != NULL) {
|
||||
inStruct++;
|
||||
void* internal_dst = reinterpret_cast<void*>(reinterpret_cast<unsigned char*>(dst) + padding);
|
||||
internal_field_map->copy_params(internal_dst, src, arg_offset + padding, error_code, inStruct);
|
||||
inStruct--;
|
||||
}
|
||||
if (next_field_map != NULL) {
|
||||
void* next_dst =
|
||||
reinterpret_cast<void*>(reinterpret_cast<unsigned char*>(dst) +
|
||||
next_field_map->dc_offset); // Next field starts with padding
|
||||
const void* next_src = reinterpret_cast<const void*>(
|
||||
reinterpret_cast<const unsigned char*>(src) + next_field_map->hc_offset);
|
||||
next_field_map->copy_params(next_dst, next_src, arg_offset + next_field_map->dc_offset,
|
||||
error_code, inStruct);
|
||||
}
|
||||
#else
|
||||
if (internal_field_map != NULL) {
|
||||
inStruct++;
|
||||
internal_field_map->copy_params(dst, src, arg_offset, error_code, inStruct);
|
||||
inStruct--;
|
||||
}
|
||||
if (next_field_map != NULL) {
|
||||
void* next_dst =
|
||||
reinterpret_cast<void*>(reinterpret_cast<unsigned char*>(dst) + next_field_map->dc_offset);
|
||||
const void* next_src = reinterpret_cast<const void*>(
|
||||
reinterpret_cast<const unsigned char*>(src) + next_field_map->hc_offset);
|
||||
next_field_map->copy_params(next_dst, next_src, arg_offset, error_code, inStruct);
|
||||
}
|
||||
#endif
|
||||
return padding;
|
||||
}
|
||||
|
||||
} // namespace cpu
|
||||
@@ -9,36 +9,37 @@
|
||||
|
||||
namespace cpu {
|
||||
|
||||
class HCtoDCmap
|
||||
{
|
||||
class HCtoDCmap {
|
||||
public:
|
||||
unsigned int hc_offset, hc_size; // Offset and size of this parameter in host compiler
|
||||
unsigned int dc_offset, dc_size; // Offset and size of this parameter in device compiler
|
||||
unsigned int hc_alignment; // Alignment of parameter in host compiler
|
||||
unsigned int dc_alignment; // Alignment of parameter in device compiler
|
||||
clk_value_type_t type; // Type of parameter
|
||||
HCtoDCmap*
|
||||
internal_field_map; // Pointer to internal mapping when current parameter is of type T_STRUCT
|
||||
HCtoDCmap*
|
||||
next_field_map; // Pointer to next struct field when current parameter is a struct member
|
||||
|
||||
public:
|
||||
unsigned int hc_offset, hc_size; // Offset and size of this parameter in host compiler
|
||||
unsigned int dc_offset, dc_size; // Offset and size of this parameter in device compiler
|
||||
unsigned int hc_alignment; // Alignment of parameter in host compiler
|
||||
unsigned int dc_alignment; // Alignment of parameter in device compiler
|
||||
clk_value_type_t type; // Type of parameter
|
||||
HCtoDCmap *internal_field_map; // Pointer to internal mapping when current parameter is of type T_STRUCT
|
||||
HCtoDCmap *next_field_map; // Pointer to next struct field when current parameter is a struct member
|
||||
HCtoDCmap(const clk_parameter_descriptor_t*, unsigned int, unsigned int, unsigned int);
|
||||
virtual ~HCtoDCmap();
|
||||
int compute_map(const clk_parameter_descriptor_t*, unsigned int&, unsigned int&, unsigned int,
|
||||
int&, int&);
|
||||
unsigned next_offset(unsigned, unsigned&, int&);
|
||||
size_t getHostScalarParamSize(const clk_value_type_t) const;
|
||||
size_t getScalarAlignment(const clk_value_type_t, bool) const;
|
||||
void align_map(unsigned, unsigned, unsigned&, unsigned&, int&);
|
||||
unsigned int copy_params(void*, const void*, unsigned int, int&, int&) const;
|
||||
|
||||
HCtoDCmap(const clk_parameter_descriptor_t*, unsigned int, unsigned int, unsigned int);
|
||||
virtual ~HCtoDCmap();
|
||||
int compute_map(const clk_parameter_descriptor_t*, unsigned int &, unsigned int &, unsigned int, int&, int&);
|
||||
unsigned next_offset(unsigned, unsigned &, int &);
|
||||
size_t getHostScalarParamSize(const clk_value_type_t) const;
|
||||
size_t getScalarAlignment(const clk_value_type_t, bool) const;
|
||||
void align_map(unsigned, unsigned, unsigned&, unsigned&, int&);
|
||||
unsigned int copy_params(void *, const void *, unsigned int, int&, int&) const;
|
||||
|
||||
private:
|
||||
private:
|
||||
};
|
||||
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
#endif // CPUMAPPING_HPP_
|
||||
// Mapping rule
|
||||
// Long types are treated with 8 byte alignment in runtime when passed in as arguments
|
||||
// but they are treated with 4 byte alignment in compiler
|
||||
// Double members have 8 byte alignment when passed as scalar argument
|
||||
// but have 4 byte alignment as a field inside a struct
|
||||
#endif // CPUMAPPING_HPP_
|
||||
// Mapping rule
|
||||
// Long types are treated with 8 byte alignment in runtime when passed in as arguments
|
||||
// but they are treated with 4 byte alignment in compiler
|
||||
// Double members have 8 byte alignment when passed as scalar argument
|
||||
// but have 4 byte alignment as a field inside a struct
|
||||
+1068
-1129
File diff suppressed because it is too large
Load Diff
@@ -14,106 +14,94 @@
|
||||
namespace amd {
|
||||
namespace option {
|
||||
class Options;
|
||||
} // option
|
||||
} // amd
|
||||
} // option
|
||||
} // amd
|
||||
|
||||
//! \namespace cpu CPU Device Implementation
|
||||
namespace cpu {
|
||||
|
||||
//! \class CPU program
|
||||
class Program : public device::Program
|
||||
{
|
||||
private:
|
||||
aclJITObjectImage JITBinary;
|
||||
std::string sourceFileName_; //!< The source image.
|
||||
void* handle_; // @todo: remove me
|
||||
class Program : public device::Program {
|
||||
private:
|
||||
aclJITObjectImage JITBinary;
|
||||
std::string sourceFileName_; //!< The source image.
|
||||
void* handle_; // @todo: remove me
|
||||
|
||||
public:
|
||||
//! Default constructor
|
||||
Program(Device& cpuDev)
|
||||
: device::Program(cpuDev), JITBinary(NULL), handle_(NULL) {}
|
||||
public:
|
||||
//! Default constructor
|
||||
Program(Device& cpuDev) : device::Program(cpuDev), JITBinary(NULL), handle_(NULL) {}
|
||||
|
||||
//! Default destructor
|
||||
~Program();
|
||||
//! Default destructor
|
||||
~Program();
|
||||
|
||||
//! pre-compile setup for CPU
|
||||
virtual bool initBuild(amd::option::Options* options);
|
||||
//! pre-compile setup for CPU
|
||||
virtual bool initBuild(amd::option::Options* options);
|
||||
|
||||
//! post-compile setup for CPU
|
||||
virtual bool finiBuild(bool isBuildGood);
|
||||
//! post-compile setup for CPU
|
||||
virtual bool finiBuild(bool isBuildGood);
|
||||
|
||||
//! Compiles CPU program
|
||||
virtual bool compileImpl(
|
||||
const std::string& sourceCode,
|
||||
const std::vector<const std::string*>& headers,
|
||||
const char** headerIncludeNames,
|
||||
amd::option::Options* options );
|
||||
//! Compiles CPU program
|
||||
virtual bool compileImpl(const std::string& sourceCode,
|
||||
const std::vector<const std::string*>& headers,
|
||||
const char** headerIncludeNames, amd::option::Options* options);
|
||||
|
||||
//! Links CPU program
|
||||
virtual bool linkImpl(amd::option::Options* options = NULL);
|
||||
//! Links CPU program
|
||||
virtual bool linkImpl(amd::option::Options* options = NULL);
|
||||
|
||||
//! Links CPU programs
|
||||
virtual bool linkImpl(
|
||||
const std::vector<device::Program*>& inputPrograms,
|
||||
amd::option::Options* options = NULL,
|
||||
bool createLibrary = false);
|
||||
//! Links CPU programs
|
||||
virtual bool linkImpl(const std::vector<device::Program*>& inputPrograms,
|
||||
amd::option::Options* options = NULL, bool createLibrary = false);
|
||||
|
||||
virtual bool createBinary(amd::option::Options* options);
|
||||
virtual bool createBinary(amd::option::Options* options);
|
||||
|
||||
//! Returns the device object, associated with this program.
|
||||
const Device& device() {
|
||||
return static_cast<const Device&>(device::Program::device());
|
||||
}
|
||||
//! Returns the device object, associated with this program.
|
||||
const Device& device() { return static_cast<const Device&>(device::Program::device()); }
|
||||
|
||||
/*! \brief Invokes the LLC compiler for the LLVM binary compilation
|
||||
* to x86 ASM text source code and ISA binary
|
||||
*
|
||||
* \return True if we successefully compiled a CPU program
|
||||
*/
|
||||
bool compileBinaryToISA(
|
||||
amd::option::Options* options //!< options for compilation
|
||||
);
|
||||
/*! \brief Invokes the LLC compiler for the LLVM binary compilation
|
||||
* to x86 ASM text source code and ISA binary
|
||||
*
|
||||
* \return True if we successefully compiled a CPU program
|
||||
*/
|
||||
bool compileBinaryToISA(amd::option::Options* options //!< options for compilation
|
||||
);
|
||||
|
||||
//! Load the library into memory
|
||||
bool loadDllCode(amd::option::Options* options, bool addElfSymbols=false);
|
||||
//! Load the library into memory
|
||||
bool loadDllCode(amd::option::Options* options, bool addElfSymbols = false);
|
||||
|
||||
//! Initialize binary for CPU
|
||||
virtual bool initClBinary();
|
||||
//! Initialize binary for CPU
|
||||
virtual bool initClBinary();
|
||||
|
||||
//! Release binary for CPU
|
||||
virtual void releaseClBinary();
|
||||
//! Release binary for CPU
|
||||
virtual void releaseClBinary();
|
||||
|
||||
ClBinary* clBinary() {
|
||||
return static_cast<ClBinary*>(device::Program::clBinary());
|
||||
}
|
||||
const ClBinary* clBinary() const {
|
||||
return static_cast<const ClBinary*>(device::Program::clBinary());
|
||||
}
|
||||
ClBinary* clBinary() { return static_cast<ClBinary*>(device::Program::clBinary()); }
|
||||
const ClBinary* clBinary() const {
|
||||
return static_cast<const ClBinary*>(device::Program::clBinary());
|
||||
}
|
||||
|
||||
aclJITObjectImage getJITBinary() { return this->JITBinary; }
|
||||
void setJITBinary(aclJITObjectImage JITBinary) { this->JITBinary = JITBinary; }
|
||||
aclJITObjectImage getJITBinary() { return this->JITBinary; }
|
||||
void setJITBinary(aclJITObjectImage JITBinary) { this->JITBinary = JITBinary; }
|
||||
|
||||
//! Returns the pointer to the Compiler struct
|
||||
//! Became public (prev. private) due to use in cpubinary for aclJIT functionality
|
||||
aclCompiler* compiler() { return static_cast<const Device&>(device()).compiler(); }
|
||||
//! Returns the pointer to the Compiler struct
|
||||
//! Became public (prev. private) due to use in cpubinary for aclJIT functionality
|
||||
aclCompiler* compiler() { return static_cast<const Device&>(device()).compiler(); }
|
||||
|
||||
private:
|
||||
private:
|
||||
//! Disable default copy constructor
|
||||
Program(const Program&);
|
||||
|
||||
//! Disable default copy constructor
|
||||
Program(const Program&);
|
||||
//! Disable operator=
|
||||
Program& operator=(const Program&);
|
||||
|
||||
//! Disable operator=
|
||||
Program& operator=(const Program&);
|
||||
std::string dllFileName_; //!< File name of the dll with kernels
|
||||
protected:
|
||||
virtual bool isElf(const char* bin) const {
|
||||
return amd::isElfHeader(bin, LP64_SWITCH(ELFCLASS32, ELFCLASS64));
|
||||
}
|
||||
|
||||
std::string dllFileName_; //!< File name of the dll with kernels
|
||||
protected:
|
||||
virtual bool isElf(const char* bin) const {
|
||||
return amd::isElfHeader(bin, LP64_SWITCH(ELFCLASS32, ELFCLASS64));
|
||||
}
|
||||
|
||||
virtual const aclTargetInfo & info(const char * str = "");
|
||||
virtual const aclTargetInfo& info(const char* str = "");
|
||||
};
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
#endif // CPUPROGRAM_HPP_
|
||||
#endif // CPUPROGRAM_HPP_
|
||||
|
||||
@@ -7,99 +7,92 @@
|
||||
|
||||
namespace cpu {
|
||||
|
||||
bool
|
||||
Settings::create()
|
||||
{
|
||||
// This code is temporary until cl_khr_fp64 is unconditional
|
||||
if (flagIsDefault(CL_KHR_FP64) || CL_KHR_FP64) {
|
||||
enableExtension(ClKhrFp64);
|
||||
}
|
||||
bool Settings::create() {
|
||||
// This code is temporary until cl_khr_fp64 is unconditional
|
||||
if (flagIsDefault(CL_KHR_FP64) || CL_KHR_FP64) {
|
||||
enableExtension(ClKhrFp64);
|
||||
}
|
||||
|
||||
enableExtension(ClAmdFp64);
|
||||
enableExtension(ClKhrGlobalInt32BaseAtomics);
|
||||
enableExtension(ClKhrGlobalInt32ExtendedAtomics);
|
||||
enableExtension(ClKhrLocalInt32BaseAtomics);
|
||||
enableExtension(ClKhrLocalInt32ExtendedAtomics);
|
||||
enableExtension(ClAmdFp64);
|
||||
enableExtension(ClKhrGlobalInt32BaseAtomics);
|
||||
enableExtension(ClKhrGlobalInt32ExtendedAtomics);
|
||||
enableExtension(ClKhrLocalInt32BaseAtomics);
|
||||
enableExtension(ClKhrLocalInt32ExtendedAtomics);
|
||||
|
||||
#ifdef _LP64
|
||||
enableExtension(ClKhrInt64BaseAtomics);
|
||||
enableExtension(ClKhrInt64ExtendedAtomics);
|
||||
#endif // _LP64
|
||||
enableExtension(ClKhrByteAddressableStore);
|
||||
enableExtension(ClKhrGlSharing);
|
||||
enableExtension(ClKhrGlEvent);
|
||||
enableExtension(ClExtDeviceFission);
|
||||
enableExtension(ClAmdDeviceAttributeQuery);
|
||||
enableExtension(ClAmdVec3);
|
||||
enableExtension(ClAmdMediaOps);
|
||||
enableExtension(ClAmdMediaOps2);
|
||||
enableExtension(ClAmdPopcnt);
|
||||
enableExtension(ClAmdPrintf);
|
||||
enableExtension(ClKhrInt64BaseAtomics);
|
||||
enableExtension(ClKhrInt64ExtendedAtomics);
|
||||
#endif // _LP64
|
||||
enableExtension(ClKhrByteAddressableStore);
|
||||
enableExtension(ClKhrGlSharing);
|
||||
enableExtension(ClKhrGlEvent);
|
||||
enableExtension(ClExtDeviceFission);
|
||||
enableExtension(ClAmdDeviceAttributeQuery);
|
||||
enableExtension(ClAmdVec3);
|
||||
enableExtension(ClAmdMediaOps);
|
||||
enableExtension(ClAmdMediaOps2);
|
||||
enableExtension(ClAmdPopcnt);
|
||||
enableExtension(ClAmdPrintf);
|
||||
|
||||
// enableExtension(ClKhrSelectFpRoundingMode);
|
||||
enableExtension(ClKhr3DImageWrites);
|
||||
// enableExtension(ClKhrSelectFpRoundingMode);
|
||||
enableExtension(ClKhr3DImageWrites);
|
||||
|
||||
// enableExtension(ClKhrFp16);
|
||||
// enableExtension(ClKhrFp16);
|
||||
|
||||
#if defined(_WIN32)
|
||||
enableExtension(ClKhrD3d10Sharing);
|
||||
#endif // _WIN32
|
||||
enableExtension(ClKhrSpir);
|
||||
// ToDo: enable this after conformance test is updated to accept it
|
||||
// enableExtension(ClKhrIlProgram);
|
||||
enableExtension(ClKhrD3d10Sharing);
|
||||
#endif // _WIN32
|
||||
enableExtension(ClKhrSpir);
|
||||
// ToDo: enable this after conformance test is updated to accept it
|
||||
// enableExtension(ClKhrIlProgram);
|
||||
|
||||
// Enable some OpenCL 2.0 extensions
|
||||
if ((OPENCL_MAJOR >= 2) && (CPU_OPENCL_VERSION >= 200)) {
|
||||
partialDispatch_ = true;
|
||||
enableExtension(ClKhrSubGroups);
|
||||
supportDepthsRGB_ = true;
|
||||
enableExtension(ClKhrDepthImages);
|
||||
}
|
||||
// Enable some OpenCL 2.0 extensions
|
||||
if ((OPENCL_MAJOR >= 2) && (CPU_OPENCL_VERSION >= 200)) {
|
||||
partialDispatch_ = true;
|
||||
enableExtension(ClKhrSubGroups);
|
||||
supportDepthsRGB_ = true;
|
||||
enableExtension(ClKhrDepthImages);
|
||||
}
|
||||
|
||||
// Map CPUID feature bits to our own feature bits
|
||||
const int sse2_features = CPUFEAT_DX_SSE | CPUFEAT_DX_SSE2;
|
||||
const int avx_features = CPUFEAT_CX_SSE3 | CPUFEAT_CX_SSSE3 |
|
||||
CPUFEAT_CX_SSE4_1 | CPUFEAT_CX_SSE4_2 |
|
||||
CPUFEAT_CX_POPCNT | CPUFEAT_CX_AVX |
|
||||
CPUFEAT_CX_OSXSAVE;
|
||||
const int fma3_features = INTEL_CPUFEAT_CX_FMA3;
|
||||
const int fma4_features = AMD_CPUFEAT_CX_FMA4 | AMD_CPUFEAT_CX_XOP;
|
||||
int regs[4];
|
||||
// Map CPUID feature bits to our own feature bits
|
||||
const int sse2_features = CPUFEAT_DX_SSE | CPUFEAT_DX_SSE2;
|
||||
const int avx_features = CPUFEAT_CX_SSE3 | CPUFEAT_CX_SSSE3 | CPUFEAT_CX_SSE4_1 |
|
||||
CPUFEAT_CX_SSE4_2 | CPUFEAT_CX_POPCNT | CPUFEAT_CX_AVX | CPUFEAT_CX_OSXSAVE;
|
||||
const int fma3_features = INTEL_CPUFEAT_CX_FMA3;
|
||||
const int fma4_features = AMD_CPUFEAT_CX_FMA4 | AMD_CPUFEAT_CX_XOP;
|
||||
int regs[4];
|
||||
|
||||
#if defined(ATI_ARCH_X86)
|
||||
amd::Os::cpuid(regs, 0x0);
|
||||
bool isAmd = regs[1] == ('A' | ('u' << 8) | ('t' << 16) | ('h' << 24));
|
||||
bool isIntel = regs[1] == ('G' | ('e' << 8) | ('n' << 16) | ('u' << 24));
|
||||
amd::Os::cpuid(regs, 0x0);
|
||||
bool isAmd = regs[1] == ('A' | ('u' << 8) | ('t' << 16) | ('h' << 24));
|
||||
bool isIntel = regs[1] == ('G' | ('e' << 8) | ('n' << 16) | ('u' << 24));
|
||||
|
||||
amd::Os::cpuid(regs, 0x1);
|
||||
amd::Os::cpuid(regs, 0x1);
|
||||
|
||||
cpuFeatures_ = (regs[3] & sse2_features) == sse2_features ?
|
||||
SSE2Instructions : 0;
|
||||
cpuFeatures_ = (regs[3] & sse2_features) == sse2_features ? SSE2Instructions : 0;
|
||||
|
||||
if ((regs[2] & avx_features) == avx_features) {
|
||||
// Check for state support
|
||||
uint64_t xcr0 = amd::Os::xgetbv(0);
|
||||
if ((regs[2] & avx_features) == avx_features) {
|
||||
// Check for state support
|
||||
uint64_t xcr0 = amd::Os::xgetbv(0);
|
||||
|
||||
// Check for SSE and YMM bits (1 and 2)
|
||||
if (((uint32_t)xcr0 & 0x6U) == 0x6U) {
|
||||
cpuFeatures_ |= AVXInstructions;
|
||||
// Check for SSE and YMM bits (1 and 2)
|
||||
if (((uint32_t)xcr0 & 0x6U) == 0x6U) {
|
||||
cpuFeatures_ |= AVXInstructions;
|
||||
|
||||
// Now check for FMA and XOP
|
||||
if (isIntel) {
|
||||
cpuFeatures_ |= (regs[2] & fma3_features) == fma3_features ?
|
||||
FMA3Instructions : 0;
|
||||
}
|
||||
// Now check for FMA and XOP
|
||||
if (isIntel) {
|
||||
cpuFeatures_ |= (regs[2] & fma3_features) == fma3_features ? FMA3Instructions : 0;
|
||||
}
|
||||
|
||||
if (isAmd) {
|
||||
amd::Os::cpuid(regs, 0x80000001);
|
||||
cpuFeatures_ |= (regs[2] & fma4_features) == fma4_features ?
|
||||
FMA4Instructions : 0;
|
||||
}
|
||||
}
|
||||
if (isAmd) {
|
||||
amd::Os::cpuid(regs, 0x80000001);
|
||||
cpuFeatures_ |= (regs[2] & fma4_features) == fma4_features ? FMA4Instructions : 0;
|
||||
}
|
||||
}
|
||||
#endif // ATI_ARCH_X86
|
||||
}
|
||||
#endif // ATI_ARCH_X86
|
||||
|
||||
return true;
|
||||
return true;
|
||||
}
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
@@ -13,32 +13,31 @@
|
||||
namespace cpu {
|
||||
|
||||
//! Device settings
|
||||
class Settings : public device::Settings
|
||||
{
|
||||
public:
|
||||
enum CpuFeatures {
|
||||
SSE2Instructions = 0x01,
|
||||
AVXInstructions = 0x02, // Processor reports SSSE3, SSE4_1, SSE4_2
|
||||
// POPCNT and AVX
|
||||
FMA3Instructions = 0x04, // Intel processor reports FMA3
|
||||
FMA4Instructions = 0x08 // AMD processor reports FMA4 and XOP
|
||||
};
|
||||
uint32_t cpuFeatures_; //!< CPU features
|
||||
class Settings : public device::Settings {
|
||||
public:
|
||||
enum CpuFeatures {
|
||||
SSE2Instructions = 0x01,
|
||||
AVXInstructions = 0x02, // Processor reports SSSE3, SSE4_1, SSE4_2
|
||||
// POPCNT and AVX
|
||||
FMA3Instructions = 0x04, // Intel processor reports FMA3
|
||||
FMA4Instructions = 0x08 // AMD processor reports FMA4 and XOP
|
||||
};
|
||||
uint32_t cpuFeatures_; //!< CPU features
|
||||
|
||||
//! Default constructor
|
||||
Settings() { cpuFeatures_ = 0; }
|
||||
//! Default constructor
|
||||
Settings() { cpuFeatures_ = 0; }
|
||||
|
||||
//! Creates settings
|
||||
bool create();
|
||||
//! Creates settings
|
||||
bool create();
|
||||
|
||||
private:
|
||||
//! Disable copy constructor
|
||||
Settings(const Settings&);
|
||||
private:
|
||||
//! Disable copy constructor
|
||||
Settings(const Settings&);
|
||||
|
||||
//! Disable assignment
|
||||
Settings& operator=(const Settings&);
|
||||
//! Disable assignment
|
||||
Settings& operator=(const Settings&);
|
||||
};
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
#endif // CPUSETTINGS_HPP_
|
||||
#endif // CPUSETTINGS_HPP_
|
||||
|
||||
+1004
-2009
File diff suppressed because it is too large
Load Diff
@@ -17,367 +17,328 @@ namespace cpu {
|
||||
|
||||
amd::Atomic<size_t> VirtualCPU::numWorkerThreads_(0);
|
||||
|
||||
VirtualCPU::VirtualCPU(Device& device)
|
||||
: device::VirtualDevice(device), acceptingCommands_(false)
|
||||
{
|
||||
const size_t numCores = device.info().maxComputeUnits_;
|
||||
VirtualCPU::VirtualCPU(Device& device) : device::VirtualDevice(device), acceptingCommands_(false) {
|
||||
const size_t numCores = device.info().maxComputeUnits_;
|
||||
|
||||
if ((numWorkerThreads_ += numCores) >= Device::getMaxWorkerThreadsNumber()) {
|
||||
numWorkerThreads_ -= numCores;
|
||||
cores_ = NULL;
|
||||
return;
|
||||
}
|
||||
|
||||
cores_ = new(std::nothrow) WorkerThread*[numCores];
|
||||
if (cores_ == NULL) {
|
||||
return;
|
||||
}
|
||||
|
||||
// Clear memory for the worker threads
|
||||
memset(cores_, 0, numCores * sizeof(WorkerThread*));
|
||||
|
||||
#if defined(__linux__)
|
||||
const bool isNuma =
|
||||
#if defined(NUMA_SUPPORT)
|
||||
device.getNumaMask() == NULL;
|
||||
#else
|
||||
false;
|
||||
#endif // NUMA_SUPPORT
|
||||
const amd::Os::ThreadAffinityMask* affinityMask = isNuma ? NULL :
|
||||
#else
|
||||
const amd::Os::ThreadAffinityMask* affinityMask =
|
||||
#endif
|
||||
device.getWorkerThreadsAffinity();
|
||||
|
||||
uint coreId = affinityMask != NULL ? affinityMask->getFirstSet() : (uint)-1;
|
||||
|
||||
for (size_t i = 0; i < numCores; ++i) {
|
||||
WorkerThread* thread = cores_[i] = new WorkerThread(device);
|
||||
if (thread == NULL) {
|
||||
for (size_t j = 0; j < i; ++j) {
|
||||
cores_[j]->resume();
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
if (thread->state() != amd::Thread::INITIALIZED) {
|
||||
return;
|
||||
}
|
||||
|
||||
#if defined(__linux__)
|
||||
if (!isNuma) {
|
||||
if (coreId == (uint)-1) {
|
||||
thread->setAffinity((uint) i);
|
||||
}
|
||||
else {
|
||||
thread->setAffinity(coreId);
|
||||
coreId = affinityMask->getNextSet(coreId);
|
||||
}
|
||||
}
|
||||
#else // On Windows we set an affinity mask and not a specific ID.
|
||||
if (coreId != (uint)-1) {
|
||||
thread->setAffinity(*affinityMask);
|
||||
}
|
||||
#endif
|
||||
thread->start();
|
||||
}
|
||||
|
||||
blitMgr_ = new device::HostBlitManager(*this);
|
||||
if ((NULL == blitMgr_) || !blitMgr_->create(device)) {
|
||||
LogError("Could not create BlitManager!");
|
||||
return;
|
||||
}
|
||||
|
||||
acceptingCommands_ = true;
|
||||
}
|
||||
|
||||
VirtualCPU::~VirtualCPU()
|
||||
{
|
||||
if (cores_ == NULL) {
|
||||
return;
|
||||
}
|
||||
|
||||
delete blitMgr_;
|
||||
|
||||
const size_t numCores = device().info().maxComputeUnits_;
|
||||
for (size_t i = 0; i < numCores; ++i) {
|
||||
delete cores_[i];
|
||||
}
|
||||
if ((numWorkerThreads_ += numCores) >= Device::getMaxWorkerThreadsNumber()) {
|
||||
numWorkerThreads_ -= numCores;
|
||||
delete[] cores_;
|
||||
cores_ = NULL;
|
||||
return;
|
||||
}
|
||||
|
||||
cores_ = new (std::nothrow) WorkerThread*[numCores];
|
||||
if (cores_ == NULL) {
|
||||
return;
|
||||
}
|
||||
|
||||
// Clear memory for the worker threads
|
||||
memset(cores_, 0, numCores * sizeof(WorkerThread*));
|
||||
|
||||
#if defined(__linux__)
|
||||
const bool isNuma =
|
||||
#if defined(NUMA_SUPPORT)
|
||||
device.getNumaMask() == NULL;
|
||||
#else
|
||||
false;
|
||||
#endif // NUMA_SUPPORT
|
||||
const amd::Os::ThreadAffinityMask* affinityMask = isNuma ? NULL :
|
||||
#else
|
||||
const amd::Os::ThreadAffinityMask* affinityMask =
|
||||
#endif
|
||||
device.getWorkerThreadsAffinity();
|
||||
|
||||
uint coreId = affinityMask != NULL ? affinityMask->getFirstSet() : (uint)-1;
|
||||
|
||||
for (size_t i = 0; i < numCores; ++i) {
|
||||
WorkerThread* thread = cores_[i] = new WorkerThread(device);
|
||||
if (thread == NULL) {
|
||||
for (size_t j = 0; j < i; ++j) {
|
||||
cores_[j]->resume();
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
if (thread->state() != amd::Thread::INITIALIZED) {
|
||||
return;
|
||||
}
|
||||
|
||||
#if defined(__linux__)
|
||||
if (!isNuma) {
|
||||
if (coreId == (uint)-1) {
|
||||
thread->setAffinity((uint)i);
|
||||
} else {
|
||||
thread->setAffinity(coreId);
|
||||
coreId = affinityMask->getNextSet(coreId);
|
||||
}
|
||||
}
|
||||
#else // On Windows we set an affinity mask and not a specific ID.
|
||||
if (coreId != (uint)-1) {
|
||||
thread->setAffinity(*affinityMask);
|
||||
}
|
||||
#endif
|
||||
thread->start();
|
||||
}
|
||||
|
||||
blitMgr_ = new device::HostBlitManager(*this);
|
||||
if ((NULL == blitMgr_) || !blitMgr_->create(device)) {
|
||||
LogError("Could not create BlitManager!");
|
||||
return;
|
||||
}
|
||||
|
||||
acceptingCommands_ = true;
|
||||
}
|
||||
|
||||
bool
|
||||
VirtualCPU::terminate()
|
||||
{
|
||||
if (cores_ == NULL) {
|
||||
return true;
|
||||
}
|
||||
VirtualCPU::~VirtualCPU() {
|
||||
if (cores_ == NULL) {
|
||||
return;
|
||||
}
|
||||
|
||||
const size_t numCores = device().info().maxComputeUnits_;
|
||||
for (size_t i = 0; i < numCores; ++i) {
|
||||
if (cores_[i]) {
|
||||
cores_[i]->terminate();
|
||||
}
|
||||
}
|
||||
delete blitMgr_;
|
||||
|
||||
const size_t numCores = device().info().maxComputeUnits_;
|
||||
for (size_t i = 0; i < numCores; ++i) {
|
||||
delete cores_[i];
|
||||
}
|
||||
numWorkerThreads_ -= numCores;
|
||||
delete[] cores_;
|
||||
}
|
||||
|
||||
bool VirtualCPU::terminate() {
|
||||
if (cores_ == NULL) {
|
||||
return true;
|
||||
}
|
||||
|
||||
const size_t numCores = device().info().maxComputeUnits_;
|
||||
for (size_t i = 0; i < numCores; ++i) {
|
||||
if (cores_[i]) {
|
||||
cores_[i]->terminate();
|
||||
}
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitReadMemory(amd::ReadMemoryCommand& vcmd)
|
||||
{
|
||||
vcmd.setStatus(CL_RUNNING);
|
||||
void VirtualCPU::submitReadMemory(amd::ReadMemoryCommand& vcmd) {
|
||||
vcmd.setStatus(CL_RUNNING);
|
||||
|
||||
bool result = false;
|
||||
device::Memory memory(vcmd.source());
|
||||
bool result = false;
|
||||
device::Memory memory(vcmd.source());
|
||||
|
||||
// Ensure memory up-to-date
|
||||
vcmd.source().cacheWriteBack();
|
||||
// Ensure memory up-to-date
|
||||
vcmd.source().cacheWriteBack();
|
||||
|
||||
switch (vcmd.type()) {
|
||||
switch (vcmd.type()) {
|
||||
case CL_COMMAND_READ_BUFFER:
|
||||
result = blitMgr().readBuffer(memory, vcmd.destination(),
|
||||
vcmd.origin(), vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().readBuffer(memory, vcmd.destination(), vcmd.origin(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
case CL_COMMAND_READ_BUFFER_RECT:
|
||||
result = blitMgr().readBufferRect(memory,
|
||||
vcmd.destination(), vcmd.bufRect(), vcmd.hostRect(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().readBufferRect(memory, vcmd.destination(), vcmd.bufRect(), vcmd.hostRect(),
|
||||
vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
case CL_COMMAND_READ_IMAGE:
|
||||
result = blitMgr().readImage(memory, vcmd.destination(),
|
||||
vcmd.origin(), vcmd.size(), vcmd.rowPitch(), vcmd.slicePitch(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().readImage(memory, vcmd.destination(), vcmd.origin(), vcmd.size(),
|
||||
vcmd.rowPitch(), vcmd.slicePitch(), vcmd.isEntireMemory());
|
||||
break;
|
||||
default:
|
||||
LogError("Unsupported type for the read command");
|
||||
break;
|
||||
}
|
||||
LogError("Unsupported type for the read command");
|
||||
break;
|
||||
}
|
||||
|
||||
if (!result) {
|
||||
LogError("submitReadMemory failed!");
|
||||
vcmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
else {
|
||||
vcmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
if (!result) {
|
||||
LogError("submitReadMemory failed!");
|
||||
vcmd.setStatus(CL_INVALID_OPERATION);
|
||||
} else {
|
||||
vcmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitWriteMemory(amd::WriteMemoryCommand& vcmd)
|
||||
{
|
||||
vcmd.setStatus(CL_RUNNING);
|
||||
void VirtualCPU::submitWriteMemory(amd::WriteMemoryCommand& vcmd) {
|
||||
vcmd.setStatus(CL_RUNNING);
|
||||
|
||||
bool result = false;
|
||||
device::Memory memory(vcmd.destination());
|
||||
bool result = false;
|
||||
device::Memory memory(vcmd.destination());
|
||||
|
||||
// Ensure memory up-to-date
|
||||
vcmd.destination().cacheWriteBack();
|
||||
// Ensure memory up-to-date
|
||||
vcmd.destination().cacheWriteBack();
|
||||
|
||||
// Process different write commands
|
||||
switch (vcmd.type()) {
|
||||
// Process different write commands
|
||||
switch (vcmd.type()) {
|
||||
case CL_COMMAND_WRITE_BUFFER:
|
||||
result = blitMgr().writeBuffer(vcmd.source(), memory,
|
||||
vcmd.origin(), vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().writeBuffer(vcmd.source(), memory, vcmd.origin(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
case CL_COMMAND_WRITE_BUFFER_RECT:
|
||||
result = blitMgr().writeBufferRect(vcmd.source(), memory,
|
||||
vcmd.hostRect(), vcmd.bufRect(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().writeBufferRect(vcmd.source(), memory, vcmd.hostRect(), vcmd.bufRect(),
|
||||
vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
case CL_COMMAND_WRITE_IMAGE:
|
||||
result = blitMgr().writeImage(vcmd.source(), memory,
|
||||
vcmd.origin(), vcmd.size(), vcmd.rowPitch(), vcmd.slicePitch(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().writeImage(vcmd.source(), memory, vcmd.origin(), vcmd.size(),
|
||||
vcmd.rowPitch(), vcmd.slicePitch(), vcmd.isEntireMemory());
|
||||
break;
|
||||
default:
|
||||
LogError("Unsupported type for the write command");
|
||||
break;
|
||||
}
|
||||
LogError("Unsupported type for the write command");
|
||||
break;
|
||||
}
|
||||
|
||||
// Mark cache as clean (CPU works directly on backing store)
|
||||
vcmd.destination().signalWrite(NULL);
|
||||
// Mark cache as clean (CPU works directly on backing store)
|
||||
vcmd.destination().signalWrite(NULL);
|
||||
|
||||
if (!result) {
|
||||
LogError("submitWriteMemory failed!");
|
||||
vcmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
else {
|
||||
vcmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
if (!result) {
|
||||
LogError("submitWriteMemory failed!");
|
||||
vcmd.setStatus(CL_INVALID_OPERATION);
|
||||
} else {
|
||||
vcmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
void
|
||||
VirtualCPU::submitCopyMemory(amd::CopyMemoryCommand& vcmd)
|
||||
{
|
||||
vcmd.setStatus(CL_RUNNING);
|
||||
void VirtualCPU::submitCopyMemory(amd::CopyMemoryCommand& vcmd) {
|
||||
vcmd.setStatus(CL_RUNNING);
|
||||
|
||||
// Ensure memory up-to-date
|
||||
vcmd.source().cacheWriteBack();
|
||||
vcmd.destination().cacheWriteBack();
|
||||
// Ensure memory up-to-date
|
||||
vcmd.source().cacheWriteBack();
|
||||
vcmd.destination().cacheWriteBack();
|
||||
|
||||
// Translate memory references and ensure cache up-to-date
|
||||
device::Memory dstMemory(vcmd.destination());
|
||||
device::Memory srcMemory(vcmd.source());
|
||||
// Translate memory references and ensure cache up-to-date
|
||||
device::Memory dstMemory(vcmd.destination());
|
||||
device::Memory srcMemory(vcmd.source());
|
||||
|
||||
bool result = false;
|
||||
bool result = false;
|
||||
|
||||
// Check if HW can be used for memory copy
|
||||
switch (vcmd.type()) {
|
||||
// Check if HW can be used for memory copy
|
||||
switch (vcmd.type()) {
|
||||
case CL_COMMAND_COPY_BUFFER:
|
||||
result = blitMgr().copyBuffer(srcMemory, dstMemory,
|
||||
vcmd.srcOrigin(), vcmd.dstOrigin(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().copyBuffer(srcMemory, dstMemory, vcmd.srcOrigin(), vcmd.dstOrigin(),
|
||||
vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
case CL_COMMAND_COPY_BUFFER_RECT:
|
||||
result = blitMgr().copyBufferRect(srcMemory, dstMemory,
|
||||
vcmd.srcRect(), vcmd.dstRect(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().copyBufferRect(srcMemory, dstMemory, vcmd.srcRect(), vcmd.dstRect(),
|
||||
vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
case CL_COMMAND_COPY_IMAGE_TO_BUFFER:
|
||||
result = blitMgr().copyImageToBuffer(srcMemory, dstMemory,
|
||||
vcmd.srcOrigin(), vcmd.dstOrigin(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().copyImageToBuffer(srcMemory, dstMemory, vcmd.srcOrigin(), vcmd.dstOrigin(),
|
||||
vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
case CL_COMMAND_COPY_BUFFER_TO_IMAGE:
|
||||
result = blitMgr().copyBufferToImage(srcMemory, dstMemory,
|
||||
vcmd.srcOrigin(), vcmd.dstOrigin(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().copyBufferToImage(srcMemory, dstMemory, vcmd.srcOrigin(), vcmd.dstOrigin(),
|
||||
vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
case CL_COMMAND_COPY_IMAGE:
|
||||
result = blitMgr().copyImage(srcMemory, dstMemory,
|
||||
vcmd.srcOrigin(), vcmd.dstOrigin(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().copyImage(srcMemory, dstMemory, vcmd.srcOrigin(), vcmd.dstOrigin(),
|
||||
vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
default:
|
||||
LogError("Unsupported command type for memory copy!");
|
||||
break;
|
||||
}
|
||||
LogError("Unsupported command type for memory copy!");
|
||||
break;
|
||||
}
|
||||
|
||||
// Mark cache as clean (CPU works directly on backing store)
|
||||
vcmd.destination().signalWrite(NULL);
|
||||
// Mark cache as clean (CPU works directly on backing store)
|
||||
vcmd.destination().signalWrite(NULL);
|
||||
|
||||
if (!result) {
|
||||
LogError("submitCopyMemory failed!");
|
||||
vcmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
else {
|
||||
vcmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
if (!result) {
|
||||
LogError("submitCopyMemory failed!");
|
||||
vcmd.setStatus(CL_INVALID_OPERATION);
|
||||
} else {
|
||||
vcmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitMapMemory(amd::MapMemoryCommand& cmd)
|
||||
{
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
void VirtualCPU::submitMapMemory(amd::MapMemoryCommand& cmd) {
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
|
||||
if (cmd.mapFlags() & CL_MAP_READ
|
||||
|| cmd.mapFlags() & CL_MAP_WRITE) {
|
||||
LogInfo("cpu::VirtualCPU::submitMapMemory() CL_MAP_READ and CL_MAP_WRITE ignored");
|
||||
}
|
||||
if (cmd.mapFlags() & CL_MAP_READ || cmd.mapFlags() & CL_MAP_WRITE) {
|
||||
LogInfo("cpu::VirtualCPU::submitMapMemory() CL_MAP_READ and CL_MAP_WRITE ignored");
|
||||
}
|
||||
|
||||
// Ensure memory up-to-date
|
||||
cmd.memory().cacheWriteBack();
|
||||
// Ensure memory up-to-date
|
||||
cmd.memory().cacheWriteBack();
|
||||
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitUnmapMemory(amd::UnmapMemoryCommand& cmd)
|
||||
{
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
void VirtualCPU::submitUnmapMemory(amd::UnmapMemoryCommand& cmd) {
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
|
||||
// Mark cache as clean (CPU works directly on backing store)
|
||||
cmd.memory().signalWrite(NULL);
|
||||
// Mark cache as clean (CPU works directly on backing store)
|
||||
cmd.memory().signalWrite(NULL);
|
||||
|
||||
//! @todo:dgladdin: strictly speaking we should check that the mem object was mapped
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
//! @todo:dgladdin: strictly speaking we should check that the mem object was mapped
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitFillMemory(amd::FillMemoryCommand& vcmd)
|
||||
{
|
||||
vcmd.setStatus(CL_RUNNING);
|
||||
void VirtualCPU::submitFillMemory(amd::FillMemoryCommand& vcmd) {
|
||||
vcmd.setStatus(CL_RUNNING);
|
||||
|
||||
device::Memory memory(vcmd.memory());
|
||||
device::Memory memory(vcmd.memory());
|
||||
|
||||
vcmd.memory().cacheWriteBack();
|
||||
vcmd.memory().cacheWriteBack();
|
||||
|
||||
bool result = false;
|
||||
bool result = false;
|
||||
|
||||
// Find the the right fill operation
|
||||
switch (vcmd.type()) {
|
||||
// Find the the right fill operation
|
||||
switch (vcmd.type()) {
|
||||
case CL_COMMAND_FILL_BUFFER:
|
||||
result = blitMgr().fillBuffer(memory, vcmd.pattern(),
|
||||
vcmd.patternSize(), vcmd.origin(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().fillBuffer(memory, vcmd.pattern(), vcmd.patternSize(), vcmd.origin(),
|
||||
vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
case CL_COMMAND_FILL_IMAGE:
|
||||
result = blitMgr().fillImage(memory, vcmd.pattern(),
|
||||
vcmd.origin(), vcmd.size(), vcmd.isEntireMemory());
|
||||
break;
|
||||
result = blitMgr().fillImage(memory, vcmd.pattern(), vcmd.origin(), vcmd.size(),
|
||||
vcmd.isEntireMemory());
|
||||
break;
|
||||
default:
|
||||
LogError("Unsupported command type for FillMemory!");
|
||||
break;
|
||||
}
|
||||
LogError("Unsupported command type for FillMemory!");
|
||||
break;
|
||||
}
|
||||
|
||||
vcmd.memory().signalWrite(NULL);
|
||||
vcmd.memory().signalWrite(NULL);
|
||||
|
||||
if (!result) {
|
||||
LogError("submitFillMemory failed!");
|
||||
vcmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
else {
|
||||
vcmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
if (!result) {
|
||||
LogError("submitFillMemory failed!");
|
||||
vcmd.setStatus(CL_INVALID_OPERATION);
|
||||
} else {
|
||||
vcmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
}
|
||||
|
||||
//! Helper function for forcing a cache sync for all kernel parameters
|
||||
static void syncAllParams(amd::NDRangeKernelCommand& cmd)
|
||||
{
|
||||
const amd::Kernel& kernel = cmd.kernel();
|
||||
const amd::KernelParameters& kernelParam = kernel.parameters();
|
||||
const amd::KernelSignature& signature = kernel.signature();
|
||||
const amd::Device& device = cmd.queue()->device();
|
||||
static void syncAllParams(amd::NDRangeKernelCommand& cmd) {
|
||||
const amd::Kernel& kernel = cmd.kernel();
|
||||
const amd::KernelParameters& kernelParam = kernel.parameters();
|
||||
const amd::KernelSignature& signature = kernel.signature();
|
||||
const amd::Device& device = cmd.queue()->device();
|
||||
|
||||
for (size_t i = 0; i < signature.numParameters(); ++i) {
|
||||
const amd::KernelParameterDescriptor& desc = signature.at(i);
|
||||
if (desc.type_ == T_POINTER && desc.size_ > 0 &&
|
||||
!kernelParam.boundToSvmPointer(device, cmd.parameters(), i)) {
|
||||
address ptr = (address) (cmd.parameters() + desc.offset_);
|
||||
amd::Memory* memArg = *(amd::Memory**)ptr;
|
||||
for (size_t i = 0; i < signature.numParameters(); ++i) {
|
||||
const amd::KernelParameterDescriptor& desc = signature.at(i);
|
||||
if (desc.type_ == T_POINTER && desc.size_ > 0 &&
|
||||
!kernelParam.boundToSvmPointer(device, cmd.parameters(), i)) {
|
||||
address ptr = (address)(cmd.parameters() + desc.offset_);
|
||||
amd::Memory* memArg = *(amd::Memory**)ptr;
|
||||
|
||||
if (memArg != NULL) {
|
||||
memArg->cacheWriteBack();
|
||||
memArg->signalWrite(NULL);
|
||||
}
|
||||
}
|
||||
if (memArg != NULL) {
|
||||
memArg->cacheWriteBack();
|
||||
memArg->signalWrite(NULL);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::computeLocalSizes(amd::NDRangeKernelCommand& command,
|
||||
amd::NDRange& local) {
|
||||
bool uniformSize = (OPENCL_MAJOR < 2) ||
|
||||
command.kernel().getDeviceKernel(device())->getUniformWorkGroupSize();
|
||||
void VirtualCPU::computeLocalSizes(amd::NDRangeKernelCommand& command, amd::NDRange& local) {
|
||||
bool uniformSize =
|
||||
(OPENCL_MAJOR < 2) || command.kernel().getDeviceKernel(device())->getUniformWorkGroupSize();
|
||||
|
||||
const amd::NDRangeContainer& sizes = command.sizes();
|
||||
const size_t numCores = device().info().maxComputeUnits_;
|
||||
|
||||
const size_t globalSize1D = sizes.global().product();
|
||||
const size_t targetNumOperations =
|
||||
std::min(globalSize1D, numCores * 4);
|
||||
size_t localSize1D =
|
||||
std::min(globalSize1D / targetNumOperations,
|
||||
device().info().maxWorkGroupSize_);
|
||||
|
||||
const size_t targetNumOperations = std::min(globalSize1D, numCores * 4);
|
||||
size_t localSize1D =
|
||||
std::min(globalSize1D / targetNumOperations, device().info().maxWorkGroupSize_);
|
||||
|
||||
for (size_t i = 0; i < local.dimensions(); ++i) {
|
||||
const size_t globalSize = sizes.global()[i];
|
||||
size_t localSize =
|
||||
std::min(std::min(localSize1D, globalSize),
|
||||
device().info().maxWorkItemSizes_[i]);
|
||||
|
||||
std::min(std::min(localSize1D, globalSize), device().info().maxWorkItemSizes_[i]);
|
||||
|
||||
// local must exactly divide global if uniform size is required
|
||||
// For non uniform size, we could use the work group size hint
|
||||
if (uniformSize && globalSize % localSize != 0) {
|
||||
@@ -395,222 +356,181 @@ VirtualCPU::computeLocalSizes(amd::NDRangeKernelCommand& command,
|
||||
}
|
||||
|
||||
|
||||
static
|
||||
amd::NDRange computeRemainders(const amd::NDRange& global,
|
||||
const amd::NDRange& local)
|
||||
{
|
||||
static amd::NDRange computeRemainders(const amd::NDRange& global, const amd::NDRange& local) {
|
||||
amd::NDRange remainders(local.dimensions());
|
||||
|
||||
for (size_t i = 0; i < local.dimensions(); ++i) {
|
||||
remainders[i] = (global[i] % local[i] != 0) ? 1 : 0;
|
||||
remainders[i] = (global[i] % local[i] != 0) ? 1 : 0;
|
||||
}
|
||||
|
||||
return remainders;
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitKernel(amd::NDRangeKernelCommand& command)
|
||||
{
|
||||
const amd::NDRangeContainer& sizes = command.sizes();
|
||||
const size_t numCores = device().info().maxComputeUnits_;
|
||||
void VirtualCPU::submitKernel(amd::NDRangeKernelCommand& command) {
|
||||
const amd::NDRangeContainer& sizes = command.sizes();
|
||||
const size_t numCores = device().info().maxComputeUnits_;
|
||||
|
||||
amd::NDRange local = sizes.local();
|
||||
amd::NDRange local = sizes.local();
|
||||
|
||||
if (local == 0) {
|
||||
computeLocalSizes(command, local);
|
||||
}
|
||||
amd::NDRange remainders = computeRemainders(sizes.global(), local);
|
||||
if (local == 0) {
|
||||
computeLocalSizes(command, local);
|
||||
}
|
||||
amd::NDRange remainders = computeRemainders(sizes.global(), local);
|
||||
|
||||
// number of groups in each dimensions
|
||||
const amd::NDRange numGroups = (sizes.global() / local) + remainders;
|
||||
// number of groups in each dimensions
|
||||
const amd::NDRange numGroups = (sizes.global() / local) + remainders;
|
||||
|
||||
size_t numOperations = numGroups.product();
|
||||
if (numOperations == 0) {
|
||||
command.setStatus(CL_COMPLETE);
|
||||
return;
|
||||
}
|
||||
|
||||
syncAllParams(command);
|
||||
// retain the command here instead of retaining in NDRangeKernelBatch' ctor
|
||||
command.retain();
|
||||
|
||||
size_t batchCount = std::min(numOperations, numCores);
|
||||
NDRangeKernelBatch batch(command, *this, numGroups, batchCount);
|
||||
|
||||
Operation::Counter counter(command, batchCount);
|
||||
command.setData(&counter);
|
||||
|
||||
for (size_t coreId = 0; coreId < batchCount; ++coreId) {
|
||||
batch.setCoreId(coreId);
|
||||
cores_[coreId]->enqueue(batch);
|
||||
cores_[coreId]->flush();
|
||||
}
|
||||
|
||||
command.awaitCompletion();
|
||||
command.release();
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitNativeFn(amd::NativeFnCommand& command)
|
||||
{
|
||||
NativeFn fn(command);
|
||||
cores_[0]->enqueue(fn);
|
||||
cores_[0]->flush();
|
||||
command.awaitCompletion();
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitMarker(amd::Marker& command)
|
||||
{
|
||||
size_t numOperations = numGroups.product();
|
||||
if (numOperations == 0) {
|
||||
command.setStatus(CL_COMPLETE);
|
||||
return;
|
||||
}
|
||||
|
||||
syncAllParams(command);
|
||||
// retain the command here instead of retaining in NDRangeKernelBatch' ctor
|
||||
command.retain();
|
||||
|
||||
size_t batchCount = std::min(numOperations, numCores);
|
||||
NDRangeKernelBatch batch(command, *this, numGroups, batchCount);
|
||||
|
||||
Operation::Counter counter(command, batchCount);
|
||||
command.setData(&counter);
|
||||
|
||||
for (size_t coreId = 0; coreId < batchCount; ++coreId) {
|
||||
batch.setCoreId(coreId);
|
||||
cores_[coreId]->enqueue(batch);
|
||||
cores_[coreId]->flush();
|
||||
}
|
||||
|
||||
command.awaitCompletion();
|
||||
command.release();
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitAcquireExtObjects(amd::AcquireExtObjectsCommand& cmd)
|
||||
{
|
||||
//! @todo [odintsov]: create an AcquireExtObjectsOperation and enqueue it
|
||||
//! to a core when a core scheduler is around.
|
||||
//
|
||||
// cores_[0]->enqueue(new AcquireExtObjectsOperation(cmd));
|
||||
// the code below will be moved to AcquireExtObjectsOperation::execute()
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
|
||||
//
|
||||
// AcquireExtObjects execution starts here
|
||||
//
|
||||
bool bError = false;
|
||||
|
||||
//! Go through ext objects by one and call member function to execute
|
||||
//! a sequence of external graphics API commands for each external object
|
||||
for(std::vector<amd::Memory*>::const_iterator itr = cmd.getMemList().begin();
|
||||
itr != cmd.getMemList().end(); itr++) {
|
||||
if(*itr) {
|
||||
bError |= !((*itr)->mapExtObjectInCQThread());
|
||||
}
|
||||
}
|
||||
if(bError) {
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
else {
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
void VirtualCPU::submitNativeFn(amd::NativeFnCommand& command) {
|
||||
NativeFn fn(command);
|
||||
cores_[0]->enqueue(fn);
|
||||
cores_[0]->flush();
|
||||
command.awaitCompletion();
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitReleaseExtObjects(amd::ReleaseExtObjectsCommand& cmd)
|
||||
{
|
||||
//! @todo [odintsov]: create a ReleaseExtObjectsOperation and enqueue it
|
||||
//! to a core when a core scheduler is around.
|
||||
//
|
||||
// cores_[i]->enqueue(new ReleaseExtObjectsOperation(cmd));
|
||||
// the code below will be moved to ReleaseExtObjectsOperation::execute()
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
void VirtualCPU::submitMarker(amd::Marker& command) { command.setStatus(CL_COMPLETE); }
|
||||
|
||||
bool bError = false;
|
||||
void VirtualCPU::submitAcquireExtObjects(amd::AcquireExtObjectsCommand& cmd) {
|
||||
//! @todo [odintsov]: create an AcquireExtObjectsOperation and enqueue it
|
||||
//! to a core when a core scheduler is around.
|
||||
//
|
||||
// cores_[0]->enqueue(new AcquireExtObjectsOperation(cmd));
|
||||
// the code below will be moved to AcquireExtObjectsOperation::execute()
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
|
||||
for(std::vector<amd::Memory*>::const_iterator itr = cmd.getMemList().begin();
|
||||
itr != cmd.getMemList().end(); itr++) {
|
||||
if(*itr) {
|
||||
bError |= !((*itr)->unmapExtObjectInCQThread());
|
||||
}
|
||||
}
|
||||
if(bError) {
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
else {
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
}
|
||||
//
|
||||
// AcquireExtObjects execution starts here
|
||||
//
|
||||
bool bError = false;
|
||||
|
||||
void VirtualCPU::submitPerfCounter(amd::PerfCounterCommand& cmd)
|
||||
{
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
LogError("We don't support HW perf counters on CPU");
|
||||
//! Go through ext objects by one and call member function to execute
|
||||
//! a sequence of external graphics API commands for each external object
|
||||
for (std::vector<amd::Memory*>::const_iterator itr = cmd.getMemList().begin();
|
||||
itr != cmd.getMemList().end(); itr++) {
|
||||
if (*itr) {
|
||||
bError |= !((*itr)->mapExtObjectInCQThread());
|
||||
}
|
||||
}
|
||||
if (bError) {
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
} else {
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
}
|
||||
|
||||
void VirtualCPU::submitThreadTraceMemObjects(amd::ThreadTraceMemObjectsCommand& cmd)
|
||||
{
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
LogError("We don't support thread trace on CPU");
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
void VirtualCPU::submitReleaseExtObjects(amd::ReleaseExtObjectsCommand& cmd) {
|
||||
//! @todo [odintsov]: create a ReleaseExtObjectsOperation and enqueue it
|
||||
//! to a core when a core scheduler is around.
|
||||
//
|
||||
// cores_[i]->enqueue(new ReleaseExtObjectsOperation(cmd));
|
||||
// the code below will be moved to ReleaseExtObjectsOperation::execute()
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
|
||||
void VirtualCPU::submitThreadTrace(amd::ThreadTraceCommand& cmd)
|
||||
{
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
LogError("We don't support thread trace on CPU");
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
bool bError = false;
|
||||
|
||||
void
|
||||
VirtualCPU::flush(amd::Command* list, bool wait)
|
||||
{
|
||||
amd::Command* head = list;
|
||||
|
||||
// Release all commands from the link list
|
||||
while (head != NULL) {
|
||||
amd::Command * it = head->getNext();
|
||||
head->release();
|
||||
head = it;
|
||||
for (std::vector<amd::Memory*>::const_iterator itr = cmd.getMemList().begin();
|
||||
itr != cmd.getMemList().end(); itr++) {
|
||||
if (*itr) {
|
||||
bError |= !((*itr)->unmapExtObjectInCQThread());
|
||||
}
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitSignal(amd::SignalCommand & cmd)
|
||||
{
|
||||
}
|
||||
if (bError) {
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
} else {
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitMakeBuffersResident(amd::MakeBuffersResidentCommand & cmd)
|
||||
{
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
void VirtualCPU::submitPerfCounter(amd::PerfCounterCommand& cmd) {
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
LogError("We don't support HW perf counters on CPU");
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitSvmFreeMemory(amd::SvmFreeMemoryCommand& cmd)
|
||||
{
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
if (cmd.pfnFreeFunc() == NULL) {
|
||||
// pointers allocated using clSVMAlloc
|
||||
for (cl_uint i = 0; i < cmd.svmPointers().size(); i++) {
|
||||
amd::SvmBuffer::free(cmd.context(), cmd.svmPointers()[i]);
|
||||
}
|
||||
void VirtualCPU::submitThreadTraceMemObjects(amd::ThreadTraceMemObjectsCommand& cmd) {
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
LogError("We don't support thread trace on CPU");
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
|
||||
void VirtualCPU::submitThreadTrace(amd::ThreadTraceCommand& cmd) {
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
LogError("We don't support thread trace on CPU");
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
|
||||
void VirtualCPU::flush(amd::Command* list, bool wait) {
|
||||
amd::Command* head = list;
|
||||
|
||||
// Release all commands from the link list
|
||||
while (head != NULL) {
|
||||
amd::Command* it = head->getNext();
|
||||
head->release();
|
||||
head = it;
|
||||
}
|
||||
}
|
||||
|
||||
void VirtualCPU::submitSignal(amd::SignalCommand& cmd) { cmd.setStatus(CL_INVALID_OPERATION); }
|
||||
|
||||
void VirtualCPU::submitMakeBuffersResident(amd::MakeBuffersResidentCommand& cmd) {
|
||||
cmd.setStatus(CL_INVALID_OPERATION);
|
||||
}
|
||||
|
||||
void VirtualCPU::submitSvmFreeMemory(amd::SvmFreeMemoryCommand& cmd) {
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
if (cmd.pfnFreeFunc() == NULL) {
|
||||
// pointers allocated using clSVMAlloc
|
||||
for (cl_uint i = 0; i < cmd.svmPointers().size(); i++) {
|
||||
amd::SvmBuffer::free(cmd.context(), cmd.svmPointers()[i]);
|
||||
}
|
||||
else {
|
||||
cmd.pfnFreeFunc()(as_cl(cmd.queue()->asCommandQueue()), cmd.svmPointers().size(),
|
||||
(void**) (&(cmd.svmPointers()[0])), cmd.userData());
|
||||
}
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
} else {
|
||||
cmd.pfnFreeFunc()(as_cl(cmd.queue()->asCommandQueue()), cmd.svmPointers().size(),
|
||||
(void**)(&(cmd.svmPointers()[0])), cmd.userData());
|
||||
}
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitSvmCopyMemory(amd::SvmCopyMemoryCommand& cmd)
|
||||
{
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
amd::SvmBuffer::memFill(cmd.dst(), cmd.src(), cmd.srcSize(), 1);
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
void VirtualCPU::submitSvmCopyMemory(amd::SvmCopyMemoryCommand& cmd) {
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
amd::SvmBuffer::memFill(cmd.dst(), cmd.src(), cmd.srcSize(), 1);
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitSvmFillMemory(amd::SvmFillMemoryCommand& cmd)
|
||||
{
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
amd::SvmBuffer::memFill(cmd.dst(), cmd.pattern(), cmd.patternSize(), cmd.times());
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
void VirtualCPU::submitSvmFillMemory(amd::SvmFillMemoryCommand& cmd) {
|
||||
cmd.setStatus(CL_RUNNING);
|
||||
amd::SvmBuffer::memFill(cmd.dst(), cmd.pattern(), cmd.patternSize(), cmd.times());
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitSvmMapMemory(amd::SvmMapMemoryCommand& cmd)
|
||||
{
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
void VirtualCPU::submitSvmMapMemory(amd::SvmMapMemoryCommand& cmd) { cmd.setStatus(CL_COMPLETE); }
|
||||
|
||||
void VirtualCPU::submitSvmUnmapMemory(amd::SvmUnmapMemoryCommand& cmd) {
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
|
||||
void
|
||||
VirtualCPU::submitSvmUnmapMemory(amd::SvmUnmapMemoryCommand& cmd)
|
||||
{
|
||||
cmd.setStatus(CL_COMPLETE);
|
||||
}
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
@@ -17,60 +17,52 @@ namespace cpu {
|
||||
class WorkerThread;
|
||||
class Device;
|
||||
|
||||
class VirtualCPU : public device::VirtualDevice
|
||||
{
|
||||
private:
|
||||
WorkerThread** cores_; //!< Pointer to array of Worker threads
|
||||
static amd::Atomic<size_t> numWorkerThreads_; //!< Current Worker Threads number
|
||||
bool acceptingCommands_;
|
||||
class VirtualCPU : public device::VirtualDevice {
|
||||
private:
|
||||
WorkerThread** cores_; //!< Pointer to array of Worker threads
|
||||
static amd::Atomic<size_t> numWorkerThreads_; //!< Current Worker Threads number
|
||||
bool acceptingCommands_;
|
||||
|
||||
public:
|
||||
VirtualCPU(cpu::Device& device);
|
||||
~VirtualCPU();
|
||||
bool terminate();
|
||||
public:
|
||||
VirtualCPU(cpu::Device& device);
|
||||
~VirtualCPU();
|
||||
bool terminate();
|
||||
|
||||
WorkerThread* getWorkerThread(size_t id) { return cores_[id]; }
|
||||
WorkerThread* getWorkerThread(size_t id) { return cores_[id]; }
|
||||
|
||||
bool acceptingCommands() const { return acceptingCommands_; }
|
||||
bool acceptingCommands() const { return acceptingCommands_; }
|
||||
|
||||
virtual void submitReadMemory(amd::ReadMemoryCommand& command);
|
||||
virtual void submitWriteMemory(amd::WriteMemoryCommand& command);
|
||||
virtual void submitCopyMemory(amd::CopyMemoryCommand& command);
|
||||
virtual void submitMapMemory(amd::MapMemoryCommand& command);
|
||||
virtual void submitUnmapMemory(amd::UnmapMemoryCommand& command);
|
||||
virtual void submitKernel(amd::NDRangeKernelCommand& command);
|
||||
virtual void submitNativeFn(amd::NativeFnCommand& command);
|
||||
virtual void submitMarker(amd::Marker& command);
|
||||
virtual void submitFillMemory(amd::FillMemoryCommand& command);
|
||||
virtual void submitMigrateMemObjects(amd::MigrateMemObjectsCommand& cmd) {}
|
||||
virtual void submitAcquireExtObjects(amd::AcquireExtObjectsCommand& cmd);
|
||||
virtual void submitReleaseExtObjects(amd::ReleaseExtObjectsCommand& cmd);
|
||||
virtual void submitPerfCounter(amd::PerfCounterCommand& cmd);
|
||||
virtual void submitThreadTraceMemObjects(amd::ThreadTraceMemObjectsCommand& cmd);
|
||||
virtual void submitThreadTrace(amd::ThreadTraceCommand& cmd);
|
||||
virtual void flush(amd::Command* list = NULL, bool wait = false);
|
||||
virtual void submitSignal(amd::SignalCommand & cmd);
|
||||
virtual void submitMakeBuffersResident(amd::MakeBuffersResidentCommand & cmd);
|
||||
virtual void submitSvmFreeMemory(amd::SvmFreeMemoryCommand& cmd);
|
||||
virtual void submitSvmCopyMemory(amd::SvmCopyMemoryCommand& cmd);
|
||||
virtual void submitSvmFillMemory(amd::SvmFillMemoryCommand& cmd);
|
||||
virtual void submitSvmMapMemory(amd::SvmMapMemoryCommand& cmd);
|
||||
virtual void submitSvmUnmapMemory(amd::SvmUnmapMemoryCommand& cmd);
|
||||
virtual void submitReadMemory(amd::ReadMemoryCommand& command);
|
||||
virtual void submitWriteMemory(amd::WriteMemoryCommand& command);
|
||||
virtual void submitCopyMemory(amd::CopyMemoryCommand& command);
|
||||
virtual void submitMapMemory(amd::MapMemoryCommand& command);
|
||||
virtual void submitUnmapMemory(amd::UnmapMemoryCommand& command);
|
||||
virtual void submitKernel(amd::NDRangeKernelCommand& command);
|
||||
virtual void submitNativeFn(amd::NativeFnCommand& command);
|
||||
virtual void submitMarker(amd::Marker& command);
|
||||
virtual void submitFillMemory(amd::FillMemoryCommand& command);
|
||||
virtual void submitMigrateMemObjects(amd::MigrateMemObjectsCommand& cmd) {}
|
||||
virtual void submitAcquireExtObjects(amd::AcquireExtObjectsCommand& cmd);
|
||||
virtual void submitReleaseExtObjects(amd::ReleaseExtObjectsCommand& cmd);
|
||||
virtual void submitPerfCounter(amd::PerfCounterCommand& cmd);
|
||||
virtual void submitThreadTraceMemObjects(amd::ThreadTraceMemObjectsCommand& cmd);
|
||||
virtual void submitThreadTrace(amd::ThreadTraceCommand& cmd);
|
||||
virtual void flush(amd::Command* list = NULL, bool wait = false);
|
||||
virtual void submitSignal(amd::SignalCommand& cmd);
|
||||
virtual void submitMakeBuffersResident(amd::MakeBuffersResidentCommand& cmd);
|
||||
virtual void submitSvmFreeMemory(amd::SvmFreeMemoryCommand& cmd);
|
||||
virtual void submitSvmCopyMemory(amd::SvmCopyMemoryCommand& cmd);
|
||||
virtual void submitSvmFillMemory(amd::SvmFillMemoryCommand& cmd);
|
||||
virtual void submitSvmMapMemory(amd::SvmMapMemoryCommand& cmd);
|
||||
virtual void submitSvmUnmapMemory(amd::SvmUnmapMemoryCommand& cmd);
|
||||
|
||||
virtual void computeLocalSizes(amd::NDRangeKernelCommand& command,
|
||||
amd::NDRange& local);
|
||||
virtual void computeLocalSizes(amd::NDRangeKernelCommand& command, amd::NDRange& local);
|
||||
|
||||
static bool fillImage(
|
||||
amd::Image& image,
|
||||
address fillMem,
|
||||
const void* pattern,
|
||||
const amd::Coord3D& origin,
|
||||
const amd::Coord3D& region,
|
||||
size_t rowPitch,
|
||||
size_t slicePitch,
|
||||
size_t elementSize);
|
||||
static bool fillImage(amd::Image& image, address fillMem, const void* pattern,
|
||||
const amd::Coord3D& origin, const amd::Coord3D& region, size_t rowPitch,
|
||||
size_t slicePitch, size_t elementSize);
|
||||
};
|
||||
|
||||
} // namespace cpu
|
||||
} // namespace cpu
|
||||
|
||||
#endif // CPUVIRTUAL_HPP_
|
||||
#endif // CPUVIRTUAL_HPP_
|
||||
|
||||
Reference in New Issue
Block a user