refactor: remove not needed code

Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2024-09-16 11:52:32 +00:00
committed by Compute-Runtime-Automation
parent cd55e41ce3
commit ab090b515c
6 changed files with 0 additions and 150 deletions

View File

@@ -28,11 +28,6 @@ class GpgpuWalkerHelper {
using DefaultWalkerType = typename GfxFamily::DefaultWalkerType;
public:
static void applyWADisableLSQCROPERFforOCL(LinearStream *pCommandStream,
const Kernel &kernel,
bool disablePerfMode);
static size_t getSizeForWADisableLSQCROPERFforOCL(const Kernel *pKernel);
static size_t getSizeForWaDisableRccRhwoOptimization(const Kernel *pKernel);
template <typename WalkerType>
@@ -78,12 +73,6 @@ class GpgpuWalkerHelper {
private:
using PIPE_CONTROL = typename GfxFamily::PIPE_CONTROL;
static void addAluReadModifyWriteRegister(
LinearStream *pCommandStream,
uint32_t aluRegister,
AluRegisters operation,
uint32_t mask);
};
template <typename GfxFamily>

View File

@@ -23,103 +23,6 @@
namespace NEO {
// Performs ReadModifyWrite operation on value of a register: Register = Register Operation Mask
template <typename GfxFamily>
void GpgpuWalkerHelper<GfxFamily>::addAluReadModifyWriteRegister(
LinearStream *pCommandStream,
uint32_t aluRegister,
AluRegisters operation,
uint32_t mask) {
// Load "Register" value into CS_GPR_R0
using MI_LOAD_REGISTER_REG = typename GfxFamily::MI_LOAD_REGISTER_REG;
using MI_MATH = typename GfxFamily::MI_MATH;
using MI_MATH_ALU_INST_INLINE = typename GfxFamily::MI_MATH_ALU_INST_INLINE;
auto pCmd = pCommandStream->getSpaceForCmd<MI_LOAD_REGISTER_REG>();
MI_LOAD_REGISTER_REG cmdReg = GfxFamily::cmdInitLoadRegisterReg;
cmdReg.setSourceRegisterAddress(aluRegister);
cmdReg.setDestinationRegisterAddress(RegisterOffsets::csGprR0);
*pCmd = cmdReg;
// Load "Mask" into CS_GPR_R1
LriHelper<GfxFamily>::program(pCommandStream,
RegisterOffsets::csGprR1,
mask,
false,
false);
// Add instruction MI_MATH with 4 MI_MATH_ALU_INST_INLINE operands
auto pCmd3 = reinterpret_cast<uint32_t *>(pCommandStream->getSpace(sizeof(MI_MATH) + RegisterConstants::numAluInstForReadModifyWrite * sizeof(MI_MATH_ALU_INST_INLINE)));
MI_MATH mathCmd;
mathCmd.DW0.Value = 0x0;
mathCmd.DW0.BitField.InstructionType = MI_MATH::COMMAND_TYPE_MI_COMMAND;
mathCmd.DW0.BitField.InstructionOpcode = MI_MATH::MI_COMMAND_OPCODE_MI_MATH;
// 0x3 - 5 Dwords length cmd (-2): 1 for MI_MATH, 4 for MI_MATH_ALU_INST_INLINE
mathCmd.DW0.BitField.DwordLength = RegisterConstants::numAluInstForReadModifyWrite - 1;
*reinterpret_cast<MI_MATH *>(pCmd3) = mathCmd;
pCmd3++;
MI_MATH_ALU_INST_INLINE *pAluParam = reinterpret_cast<MI_MATH_ALU_INST_INLINE *>(pCmd3);
MI_MATH_ALU_INST_INLINE cmdAluParam;
cmdAluParam.DW0.Value = 0x0;
// Setup first operand of MI_MATH - load CS_GPR_R0 into register A
cmdAluParam.DW0.BitField.ALUOpcode =
static_cast<uint32_t>(AluRegisters::opcodeLoad);
cmdAluParam.DW0.BitField.Operand1 =
static_cast<uint32_t>(AluRegisters::srca);
cmdAluParam.DW0.BitField.Operand2 =
static_cast<uint32_t>(AluRegisters::gpr0);
*pAluParam = cmdAluParam;
pAluParam++;
cmdAluParam.DW0.Value = 0x0;
// Setup second operand of MI_MATH - load CS_GPR_R1 into register B
cmdAluParam.DW0.BitField.ALUOpcode =
static_cast<uint32_t>(AluRegisters::opcodeLoad);
cmdAluParam.DW0.BitField.Operand1 =
static_cast<uint32_t>(AluRegisters::srcb);
cmdAluParam.DW0.BitField.Operand2 =
static_cast<uint32_t>(AluRegisters::gpr1);
*pAluParam = cmdAluParam;
pAluParam++;
cmdAluParam.DW0.Value = 0x0;
// Setup third operand of MI_MATH - "Operation" on registers A and B
cmdAluParam.DW0.BitField.ALUOpcode = static_cast<uint32_t>(operation);
cmdAluParam.DW0.BitField.Operand1 = 0;
cmdAluParam.DW0.BitField.Operand2 = 0;
*pAluParam = cmdAluParam;
pAluParam++;
cmdAluParam.DW0.Value = 0x0;
// Setup fourth operand of MI_MATH - store result into CS_GPR_R0
cmdAluParam.DW0.BitField.ALUOpcode =
static_cast<uint32_t>(AluRegisters::opcodeStore);
cmdAluParam.DW0.BitField.Operand1 =
static_cast<uint32_t>(AluRegisters::gpr0);
cmdAluParam.DW0.BitField.Operand2 =
static_cast<uint32_t>(AluRegisters::accu);
*pAluParam = cmdAluParam;
// LOAD value of CS_GPR_R0 into "Register"
auto pCmd4 = pCommandStream->getSpaceForCmd<MI_LOAD_REGISTER_REG>();
cmdReg = GfxFamily::cmdInitLoadRegisterReg;
cmdReg.setSourceRegisterAddress(RegisterOffsets::csGprR0);
cmdReg.setDestinationRegisterAddress(aluRegister);
*pCmd4 = cmdReg;
// Add PIPE_CONTROL to flush caches
auto pCmd5 = pCommandStream->getSpaceForCmd<PIPE_CONTROL>();
PIPE_CONTROL cmdPipeControl = GfxFamily::cmdInitPipeControl;
cmdPipeControl.setCommandStreamerStallEnable(true);
cmdPipeControl.setDcFlushEnable(true);
cmdPipeControl.setTextureCacheInvalidationEnable(true);
cmdPipeControl.setPipeControlFlushEnable(true);
cmdPipeControl.setStateCacheInvalidationEnable(true);
*pCmd5 = cmdPipeControl;
}
template <typename GfxFamily>
void GpgpuWalkerHelper<GfxFamily>::dispatchPerfCountersCommandsStart(
CommandQueue &commandQueue,
@@ -152,15 +55,6 @@ void GpgpuWalkerHelper<GfxFamily>::dispatchPerfCountersCommandsEnd(
pPerformanceCounters->getGpuCommands(commandBufferType, hwPerfCounter, false, size, pBuffer);
}
template <typename GfxFamily>
void GpgpuWalkerHelper<GfxFamily>::applyWADisableLSQCROPERFforOCL(NEO::LinearStream *pCommandStream, const Kernel &kernel, bool disablePerfMode) {
}
template <typename GfxFamily>
size_t GpgpuWalkerHelper<GfxFamily>::getSizeForWADisableLSQCROPERFforOCL(const Kernel *pKernel) {
return (size_t)0;
}
template <typename GfxFamily>
size_t GpgpuWalkerHelper<GfxFamily>::getSizeForWaDisableRccRhwoOptimization(const Kernel *pKernel) {
return 0u;

View File

@@ -86,7 +86,6 @@ size_t EnqueueOperation<GfxFamily>::getSizeRequiredCSKernel(bool reserveProfilin
size += 2 * sizeof(PIPE_CONTROL) + 2 * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
}
size += PerformanceCounters::getGpuCommandsSize(commandQueue.getPerfCounters(), commandQueue.getGpgpuEngine().osContext->getEngineType(), reservePerfCounters);
size += GpgpuWalkerHelper<GfxFamily>::getSizeForWADisableLSQCROPERFforOCL(pKernel);
size += GpgpuWalkerHelper<GfxFamily>::getSizeForWaDisableRccRhwoOptimization(pKernel);
return size;
@@ -97,10 +96,6 @@ size_t EnqueueOperation<GfxFamily>::getSizeRequiredForTimestampPacketWrite() {
return sizeof(PIPE_CONTROL);
}
template <typename GfxFamily>
void GpgpuWalkerHelper<GfxFamily>::adjustMiStoreRegMemMode(MI_STORE_REG_MEM<GfxFamily> *storeCmd) {
}
template <typename GfxFamily>
void GpgpuWalkerHelper<GfxFamily>::dispatchProfilingCommandsStart(
TagNodeBase &hwTimeStamps,

View File

@@ -28,24 +28,6 @@ inline void HardwareInterface<GfxFamily>::getDefaultDshSpace(
dsh->getSpace(totalInterfaceDescriptorTableSize);
}
template <typename GfxFamily>
inline void HardwareInterface<GfxFamily>::dispatchWorkarounds(
LinearStream *commandStream,
CommandQueue &commandQueue,
Kernel &kernel,
const bool &enable) {
if (enable) {
PreemptionHelper::applyPreemptionWaCmdsBegin<GfxFamily>(commandStream, commandQueue.getDevice());
// Implement enabling special WA DisableLSQCROPERFforOCL if needed
GpgpuWalkerHelper<GfxFamily>::applyWADisableLSQCROPERFforOCL(commandStream, kernel, enable);
} else {
// Implement disabling special WA DisableLSQCROPERFforOCL if needed
GpgpuWalkerHelper<GfxFamily>::applyWADisableLSQCROPERFforOCL(commandStream, kernel, enable);
PreemptionHelper::applyPreemptionWaCmdsEnd<GfxFamily>(commandStream, commandQueue.getDevice());
}
}
template <typename GfxFamily>
template <typename WalkerType>
inline void HardwareInterface<GfxFamily>::programWalker(

View File

@@ -18,11 +18,6 @@ inline cl_command_queue_capabilities_intel ClGfxCoreHelperHw<GfxFamily>::getAddi
return 0;
}
template <typename GfxFamily>
cl_device_feature_capabilities_intel ClGfxCoreHelperHw<GfxFamily>::getSupportedDeviceFeatureCapabilities(const RootDeviceEnvironment &rootDeviceEnvironment) const {
return 0;
}
static const std::vector<cl_image_format> redescribeFormats = {
{CL_R, CL_UNSIGNED_INT8},
{CL_R, CL_UNSIGNED_INT16},

View File

@@ -25,11 +25,6 @@ void ProductHelperHw<gfxProduct>::enableCompression(HardwareInfo *hwInfo) const
hwInfo->capabilityTable.ftrRenderCompressedBuffers = hwInfo->featureTable.flags.ftrE2ECompression;
}
template <PRODUCT_FAMILY gfxProduct>
uint32_t ProductHelperHw<gfxProduct>::getMaxThreadsForWorkgroupInDSSOrSS(const HardwareInfo &hwInfo, uint32_t maxNumEUsPerSubSlice, uint32_t maxNumEUsPerDualSubSlice) const {
return getMaxThreadsForWorkgroup(hwInfo, maxNumEUsPerSubSlice);
}
template <PRODUCT_FAMILY gfxProduct>
bool ProductHelperHw<gfxProduct>::obtainBlitterPreference(const HardwareInfo &hwInfo) const {
return false;