AI/rocm-systems
2
0
Forka 0
Codice Problemi Pull Request Actions Pacchetti Progetti Rilasci Wiki Attività

SWDEV-546328 - API definitions for hipMemsetD2D8/D16/D32 (#789)

Sfoglia il codice sorgente 
Co-authored-by: Rahul Manocha <rmanocha@amd.com>

[ROCm/clr commit: 5f86622adc]
This commit is contained in:
Manocha, Rahul
2025-08-14 14:47:28 -07:00
committato da GitHub
parent 208d124f54
commit 2b79ca9435
8 ha cambiato i file con 339 aggiunte e 19 eliminazioni
Scarica il file Patch Scarica il file Diff Expand all files Collapse all files
projects/clr/hipamd/src/amdhip.def
+6
Vedi File
@@ -493,3 +493,9 @@ hipLinkDestroy
hipLaunchKernelExC
hipDrvLaunchKernelEx
hipModuleGetFunctionCount
hipMemsetD2D8
hipMemsetD2D8Async
hipMemsetD2D16
hipMemsetD2D16Async
hipMemsetD2D32
hipMemsetD2D32Async
projects/clr/hipamd/src/hip_api_trace.cpp
+26 -1
Vedi File
@@ -831,6 +831,18 @@ hipError_t hipDrvLaunchKernelEx(const HIP_LAUNCH_CONFIG* config, hipFunction_t f
hipError_t hipMemGetHandleForAddressRange(void* handle, hipDeviceptr_t dptr, size_t size,
hipMemRangeHandleType handleType,
unsigned long long flags);
hipError_t hipMemsetD2D8(hipDeviceptr_t dst, size_t dstPitch, unsigned char value, size_t width,
size_t height);
hipError_t hipMemsetD2D8Async(hipDeviceptr_t dst, size_t dstPitch, unsigned char value, size_t width,
size_t height, hipStream_t stream);
hipError_t hipMemsetD2D16(hipDeviceptr_t dst, size_t dstPitch, unsigned short value, size_t width,
size_t height);
hipError_t hipMemsetD2D16Async(hipDeviceptr_t dst, size_t dstPitch, unsigned short value,
size_t width, size_t height, hipStream_t stream);
hipError_t hipMemsetD2D32(hipDeviceptr_t dst, size_t dstPitch, unsigned int value, size_t width,
size_t height);
hipError_t hipMemsetD2D32Async(hipDeviceptr_t dst, size_t dstPitch, unsigned int value,
size_t width, size_t height, hipStream_t stream);
} // namespace hip
@@ -1348,6 +1360,12 @@ void UpdateDispatchTable(HipDispatchTable* ptrDispatchTable) {
ptrDispatchTable->hipLaunchKernelExC_fn = hip::hipLaunchKernelExC;
ptrDispatchTable->hipDrvLaunchKernelEx_fn = hip::hipDrvLaunchKernelEx;
ptrDispatchTable->hipMemGetHandleForAddressRange_fn = hip::hipMemGetHandleForAddressRange;
ptrDispatchTable->hipMemsetD2D8_fn = hip::hipMemsetD2D8;
ptrDispatchTable->hipMemsetD2D8Async_fn = hip::hipMemsetD2D8Async;
ptrDispatchTable->hipMemsetD2D16_fn = hip::hipMemsetD2D16;
ptrDispatchTable->hipMemsetD2D16Async_fn = hip::hipMemsetD2D16Async;
ptrDispatchTable->hipMemsetD2D32_fn = hip::hipMemsetD2D32;
ptrDispatchTable->hipMemsetD2D32Async_fn = hip::hipMemsetD2D32Async;
}
#if HIP_ROCPROFILER_REGISTER > 0
@@ -1993,13 +2011,20 @@ HIP_ENFORCE_ABI(HipDispatchTable, hipDrvLaunchKernelEx_fn, 475);
HIP_ENFORCE_ABI(HipDispatchTable, hipMemGetHandleForAddressRange_fn, 476);
// HIP_RUNTIME_API_TABLE_STEP_VERSION == 13
HIP_ENFORCE_ABI(HipDispatchTable, hipModuleGetFunctionCount_fn, 477);
HIP_ENFORCE_ABI(HipDispatchTable, hipMemsetD2D8_fn, 478);
HIP_ENFORCE_ABI(HipDispatchTable, hipMemsetD2D8Async_fn, 479);
HIP_ENFORCE_ABI(HipDispatchTable, hipMemsetD2D16_fn, 480);
HIP_ENFORCE_ABI(HipDispatchTable, hipMemsetD2D16Async_fn, 481);
HIP_ENFORCE_ABI(HipDispatchTable, hipMemsetD2D32_fn, 482);
HIP_ENFORCE_ABI(HipDispatchTable, hipMemsetD2D32Async_fn, 483);
// if HIP_ENFORCE_ABI entries are added for each new function pointer in the table, the number below
// will be +1 of the number in the last HIP_ENFORCE_ABI line. E.g.:
//
// HIP_ENFORCE_ABI(<table>, <functor>, 8)
//
// HIP_ENFORCE_ABI_VERSIONING(<table>, 9) <- 8 + 1 = 9
HIP_ENFORCE_ABI_VERSIONING(HipDispatchTable, 478)
HIP_ENFORCE_ABI_VERSIONING(HipDispatchTable, 484)
static_assert(HIP_RUNTIME_API_TABLE_MAJOR_VERSION == 0 && HIP_RUNTIME_API_TABLE_STEP_VERSION == 14,
"If you get this error, add new HIP_ENFORCE_ABI(...) code for the new function "
projects/clr/hipamd/src/hip_hcc.map.in
+7 -1
Vedi File
@@ -609,6 +609,12 @@ local:
hip_7.1 {
global:
hipModuleGetFunctionCount;
hipMemsetD2D8;
hipMemsetD2D8Async;
hipMemsetD2D16;
hipMemsetD2D16Async;
hipMemsetD2D32;
hipMemsetD2D32Async;
local:
*;
} hip_6.5;
} hip_6.5;
projects/clr/hipamd/src/hip_memory.cpp
+51 -8
Vedi File
@@ -3117,7 +3117,7 @@ hipError_t ihipMemset3DCommand(std::vector<amd::Command*> &commands, hipPitchedP
hipError_t ihipMemset3D(hipPitchedPtr pitchedDevPtr, int value, hipExtent extent,
hipStream_t stream, bool isAsync = false) {
hipStream_t stream, bool isAsync = false, size_t elementSize = 1) {
auto sizeBytes = extent.width * extent.height * extent.depth;
if (sizeBytes == 0) {
@@ -3142,7 +3142,7 @@ hipError_t ihipMemset3D(hipPitchedPtr pitchedDevPtr, int value, hipExtent extent
}
hip::Stream* hip_stream = hip::getStream(stream);
std::vector<amd::Command*> commands;
status = ihipMemset3DCommand(commands, pitchedDevPtr, value, extent, hip_stream);
status = ihipMemset3DCommand(commands, pitchedDevPtr, value, extent, hip_stream, elementSize);
if (status != hipSuccess) {
return status;
}
@@ -3157,9 +3157,10 @@ hipError_t ihipMemset3D(hipPitchedPtr pitchedDevPtr, int value, hipExtent extent
}
hipError_t hipMemset2D_common(void* dst, size_t pitch, int value, size_t width,
size_t height, hipStream_t stream=nullptr) {
size_t height, hipStream_t stream=nullptr, size_t elementSize = 1) {
CHECK_STREAM_CAPTURING();
return ihipMemset3D({dst, pitch, width, height}, value, {width, height, 1}, stream);
return ihipMemset3D({dst, pitch, width, height}, value, {width, height, 1}, stream, false,
elementSize);
}
hipError_t hipMemset2D_spt(void* dst, size_t pitch, int value, size_t width, size_t height) {
@@ -3173,11 +3174,11 @@ hipError_t hipMemset2D(void* dst, size_t pitch, int value, size_t width, size_t
HIP_RETURN(hipMemset2D_common(dst, pitch, value, width, height));
}
hipError_t hipMemset2DAsync_common(void* dst, size_t pitch, int value,
size_t width, size_t height, hipStream_t stream) {
hipError_t hipMemset2DAsync_common(void* dst, size_t pitch, int value, size_t width, size_t height,
hipStream_t stream, size_t elementSize = 1) {
STREAM_CAPTURE(hipMemset2DAsync, stream, dst, pitch, value, width, height);
return ihipMemset3D({dst, pitch, width, height}, value, {width, height, 1}, stream, true);
return ihipMemset3D({dst, pitch, width, height}, value, {width, height, 1}, stream, true,
elementSize);
}
hipError_t hipMemset2DAsync(void* dst, size_t pitch, int value,
@@ -3193,6 +3194,48 @@ hipError_t hipMemset2DAsync_spt(void* dst, size_t pitch, int value,
HIP_RETURN(hipMemset2DAsync_common(dst, pitch, value, width, height, stream));
}
hipError_t hipMemsetD2D8(hipDeviceptr_t dst, size_t dstPitch, unsigned char value, size_t width,
size_t height) {
HIP_INIT_API(hipMemsetD2D8, dst, dstPitch, value, width, height);
HIP_RETURN(hipMemset2D_common(dst, dstPitch, value, width, height, nullptr,
sizeof(unsigned char)));
}
hipError_t hipMemsetD2D8Async(hipDeviceptr_t dst, size_t dstPitch, unsigned char value, size_t width,
size_t height, hipStream_t stream) {
HIP_INIT_API(hipMemsetD2D8Async, dst, dstPitch, value, width, height, stream);
HIP_RETURN(hipMemset2DAsync_common(dst, dstPitch, value, width, height, stream,
sizeof(unsigned char)));
}
hipError_t hipMemsetD2D16(hipDeviceptr_t dst, size_t dstPitch, unsigned short value, size_t width,
size_t height) {
HIP_INIT_API(hipMemsetD2D16, dst, dstPitch, value, width, height);
HIP_RETURN(hipMemset2D_common(dst, dstPitch, value, width, height, nullptr,
sizeof(unsigned short)));
}
hipError_t hipMemsetD2D16Async(hipDeviceptr_t dst, size_t dstPitch, unsigned short value,
size_t width, size_t height, hipStream_t stream) {
HIP_INIT_API(hipMemsetD2D16Async, dst, dstPitch, value, width, height, stream);
HIP_RETURN(hipMemset2DAsync_common(dst, dstPitch, value, width, height, stream,
sizeof(unsigned short)));
}
hipError_t hipMemsetD2D32(hipDeviceptr_t dst, size_t dstPitch, unsigned int value, size_t width,
size_t height) {
HIP_INIT_API(hipMemsetD2D32, dst, dstPitch, value, width, height);
HIP_RETURN(hipMemset2D_common(dst, dstPitch, value, width, height, nullptr,
sizeof(unsigned int)));
}
hipError_t hipMemsetD2D32Async(hipDeviceptr_t dst, size_t dstPitch, unsigned int value,
size_t width, size_t height, hipStream_t stream) {
HIP_INIT_API(hipMemsetD2D32Async, dst, dstPitch, value, width, height, stream);
HIP_RETURN(hipMemset2DAsync_common(dst, dstPitch, value, width, height, stream,
sizeof(unsigned int)));
}
// ================================================================================================
hipError_t hipMemset3D_common(hipPitchedPtr pitchedDevPtr, int value, hipExtent extent,
hipStream_t stream=nullptr) {
projects/clr/hipamd/src/hip_table_interface.cpp
+27
Vedi File
@@ -1888,4 +1888,31 @@ hipError_t hipMemGetHandleForAddressRange(void* handle, hipDeviceptr_t dptr, siz
unsigned long long flags) {
return hip::GetHipDispatchTable()->hipMemGetHandleForAddressRange_fn(handle, dptr, size,
handleType, flags);
}
hipError_t hipMemsetD2D8(hipDeviceptr_t dst, size_t dstPitch, unsigned char value, size_t width,
size_t height) {
return hip::GetHipDispatchTable()->hipMemsetD2D8_fn(dst, dstPitch, value, width, height);
}
hipError_t hipMemsetD2D8Async(hipDeviceptr_t dst, size_t dstPitch, unsigned char value, size_t width,
size_t height, hipStream_t stream) {
return hip::GetHipDispatchTable()->hipMemsetD2D8Async_fn(dst, dstPitch, value, width, height,
stream);
}
hipError_t hipMemsetD2D16(hipDeviceptr_t dst, size_t dstPitch, unsigned short value, size_t width,
size_t height) {
return hip::GetHipDispatchTable()->hipMemsetD2D16_fn(dst, dstPitch, value, width, height);
}
hipError_t hipMemsetD2D16Async(hipDeviceptr_t dst, size_t dstPitch, unsigned short value, size_t width,
size_t height, hipStream_t stream) {
return hip::GetHipDispatchTable()->hipMemsetD2D16Async_fn(dst, dstPitch, value, width, height,
stream);
}
hipError_t hipMemsetD2D32(hipDeviceptr_t dst, size_t dstPitch, unsigned int value, size_t width,
size_t height) {
return hip::GetHipDispatchTable()->hipMemsetD2D32_fn(dst, dstPitch, value, width, height);
}
hipError_t hipMemsetD2D32Async(hipDeviceptr_t dst, size_t dstPitch, unsigned int value, size_t width,
size_t height, hipStream_t stream) {
return hip::GetHipDispatchTable()->hipMemsetD2D32Async_fn(dst, dstPitch, value, width, height,
stream);
}