refactor: update Implicit Scaling interface

BartoszDunajski · Compute-Runtime-Automation · commit 7fbf724c727f · 2024-01-18T16:52:48.000+01:00
Signed-off-by: Dunajski, Bartosz &lt;bartosz.dunajski@intel.com&gt;
diff --git a/shared/source/command_container/implicit_scaling.h b/shared/source/command_container/implicit_scaling.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2021-2023 Intel Corporation
+ * Copyright (C) 2021-2024 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -90,6 +90,9 @@ struct ImplicitScalingDispatch {
 
     static bool platformSupportsImplicitScaling(const RootDeviceEnvironment &rootDeviceEnvironment);
 
+    template <typename WalkerType>
+    static void appendWalkerFields(WalkerType &walkerCmd, uint32_t tileCount);
+
   private:
     static bool pipeControlStallRequired;
 };
diff --git a/shared/source/command_container/implicit_scaling_before_xe_hp.inl b/shared/source/command_container/implicit_scaling_before_xe_hp.inl
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2023 Intel Corporation
+ * Copyright (C) 2023-2024 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -71,11 +71,17 @@ inline bool ImplicitScalingDispatch<GfxFamily>::platformSupportsImplicitScaling(
     return false;
 }
 
+template <typename GfxFamily>
+template <typename WalkerType>
+void ImplicitScalingDispatch<GfxFamily>::appendWalkerFields(WalkerType &walkerCmd, uint32_t tileCount) {
+}
+
 template <>
 bool ImplicitScalingDispatch<Family>::pipeControlStallRequired = true;
 
 template struct ImplicitScalingDispatch<Family>;
 template void ImplicitScalingDispatch<Family>::dispatchCommands<Family::DefaultWalkerType>(LinearStream &commandStream, Family::DefaultWalkerType &walkerCmd, void **outWalkerPtr, const DeviceBitfield &devices, NEO::RequiredPartitionDim requiredPartitionDim, uint32_t &partitionCount, bool useSecondaryBatchBuffer, bool apiSelfCleanup, bool dcFlush, bool forceExecutionOnSingleTile, uint64_t workPartitionAllocationGpuVa, const HardwareInfo &hwInfo);
 template size_t ImplicitScalingDispatch<Family>::getSize<Family::DefaultWalkerType>(bool apiSelfCleanup, bool preferStaticPartitioning, const DeviceBitfield &devices, const Vec3<size_t> &groupStart, const Vec3<size_t> &groupCount);
+template void ImplicitScalingDispatch<Family>::appendWalkerFields<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd, uint32_t tileCount);
 
 } // namespace NEO
diff --git a/shared/source/command_container/implicit_scaling_xehp_and_later.inl b/shared/source/command_container/implicit_scaling_xehp_and_later.inl
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2021-2023 Intel Corporation
+ * Copyright (C) 2021-2024 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -254,4 +254,9 @@ inline bool ImplicitScalingDispatch<GfxFamily>::platformSupportsImplicitScaling(
     return false;
 }
 
+template <typename GfxFamily>
+template <typename WalkerType>
+void ImplicitScalingDispatch<GfxFamily>::appendWalkerFields(WalkerType &walkerCmd, uint32_t tileCount) {
+}
+
 } // namespace NEO
diff --git a/shared/source/command_container/walker_partition_xehp_and_later.h b/shared/source/command_container/walker_partition_xehp_and_later.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2021-2023 Intel Corporation
+ * Copyright (C) 2021-2024 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -8,6 +8,7 @@
 #pragma once
 
 #include "shared/source/command_container/command_encoder.h"
+#include "shared/source/command_container/implicit_scaling.h"
 #include "shared/source/command_container/walker_partition_interface.h"
 #include "shared/source/debug_settings/debug_settings_manager.h"
 #include "shared/source/helpers/aligned_memory.h"
@@ -492,6 +493,7 @@ template <typename GfxFamily, typename WalkerType>
 void *programPartitionedWalker(void *&inputAddress, uint32_t &totalBytesProgrammed,
                                WalkerType *inputWalker,
                                uint32_t partitionCount,
+                               uint32_t tileCount,
                                bool forceExecutionOnSingleTile) {
     auto computeWalker = putCommand<WalkerType>(inputAddress, totalBytesProgrammed);
     WalkerType cmd = *inputWalker;
@@ -521,6 +523,9 @@ void *programPartitionedWalker(void *&inputAddress, uint32_t &totalBytesProgramm
             cmd.setPartitionSize(Math::divideAndRoundUp(workgroupCount, partitionCount));
         }
     }
+
+    NEO::ImplicitScalingDispatch<GfxFamily>::appendWalkerFields(cmd, tileCount);
+
     *computeWalker = cmd;
 
     return computeWalker;
@@ -633,7 +638,7 @@ void constructDynamicallyPartitionedCommandBuffer(void *cpuPointer,
         args.secondaryBatchBuffer);
 
     // Walker section
-    auto walkerPtr = programPartitionedWalker<GfxFamily, WalkerType>(currentBatchBufferPointer, totalBytesProgrammed, inputWalker, args.partitionCount, args.forceExecutionOnSingleTile);
+    auto walkerPtr = programPartitionedWalker<GfxFamily, WalkerType>(currentBatchBufferPointer, totalBytesProgrammed, inputWalker, args.partitionCount, args.tileCount, args.forceExecutionOnSingleTile);
     if (outWalkerPtr) {
         *outWalkerPtr = walkerPtr;
     }
@@ -727,7 +732,7 @@ void constructStaticallyPartitionedCommandBuffer(void *cpuPointer,
     if (args.initializeWparidRegister) {
         programMiLoadRegisterMem<GfxFamily>(currentBatchBufferPointer, totalBytesProgrammed, args.workPartitionAllocationGpuVa, wparidCCSOffset);
     }
-    auto walkerPtr = programPartitionedWalker<GfxFamily>(currentBatchBufferPointer, totalBytesProgrammed, inputWalker, args.partitionCount, args.forceExecutionOnSingleTile);
+    auto walkerPtr = programPartitionedWalker<GfxFamily>(currentBatchBufferPointer, totalBytesProgrammed, inputWalker, args.partitionCount, args.tileCount, args.forceExecutionOnSingleTile);
     if (outWalkerPtr) {
         *outWalkerPtr = walkerPtr;
     }
diff --git a/shared/source/xe_hpc_core/implicit_scaling_xe_hpc_core.cpp b/shared/source/xe_hpc_core/implicit_scaling_xe_hpc_core.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2021-2023 Intel Corporation
+ * Copyright (C) 2021-2024 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -32,5 +32,6 @@ bool ImplicitScalingDispatch<Family>::platformSupportsImplicitScaling(const Root
 template struct ImplicitScalingDispatch<Family>;
 template void ImplicitScalingDispatch<Family>::dispatchCommands<Family::DefaultWalkerType>(LinearStream &commandStream, Family::DefaultWalkerType &walkerCmd, void **outWalkerPtr, const DeviceBitfield &devices, NEO::RequiredPartitionDim requiredPartitionDim, uint32_t &partitionCount, bool useSecondaryBatchBuffer, bool apiSelfCleanup, bool dcFlush, bool forceExecutionOnSingleTile, uint64_t workPartitionAllocationGpuVa, const HardwareInfo &hwInfo);
 template size_t ImplicitScalingDispatch<Family>::getSize<Family::DefaultWalkerType>(bool apiSelfCleanup, bool preferStaticPartitioning, const DeviceBitfield &devices, const Vec3<size_t> &groupStart, const Vec3<size_t> &groupCount);
+template void ImplicitScalingDispatch<Family>::appendWalkerFields<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd, uint32_t tileCount);
 
 } // namespace NEO
diff --git a/shared/source/xe_hpg_core/implicit_scaling_xe_hpg_core.cpp b/shared/source/xe_hpg_core/implicit_scaling_xe_hpg_core.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2021-2023 Intel Corporation
+ * Copyright (C) 2021-2024 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -19,5 +19,6 @@ bool ImplicitScalingDispatch<Family>::pipeControlStallRequired = true;
 template struct ImplicitScalingDispatch<Family>;
 template void ImplicitScalingDispatch<Family>::dispatchCommands<Family::DefaultWalkerType>(LinearStream &commandStream, Family::DefaultWalkerType &walkerCmd, void **outWalkerPtr, const DeviceBitfield &devices, NEO::RequiredPartitionDim requiredPartitionDim, uint32_t &partitionCount, bool useSecondaryBatchBuffer, bool apiSelfCleanup, bool dcFlush, bool forceExecutionOnSingleTile, uint64_t workPartitionAllocationGpuVa, const HardwareInfo &hwInfo);
 template size_t ImplicitScalingDispatch<Family>::getSize<Family::DefaultWalkerType>(bool apiSelfCleanup, bool preferStaticPartitioning, const DeviceBitfield &devices, const Vec3<size_t> &groupStart, const Vec3<size_t> &groupCount);
+template void ImplicitScalingDispatch<Family>::appendWalkerFields<Family::DefaultWalkerType>(Family::DefaultWalkerType &walkerCmd, uint32_t tileCount);
 
 } // namespace NEO
diff --git a/shared/test/unit_test/encoders/walker_partition_tests_xehp_and_later_2.cpp b/shared/test/unit_test/encoders/walker_partition_tests_xehp_and_later_2.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2021-2023 Intel Corporation
+ * Copyright (C) 2021-2024 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -426,7 +426,7 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, WalkerPartitionTests, givenProgramComputeWalkerWhen
 
     walker.setPartitionType(WalkerType::PARTITION_TYPE::PARTITION_TYPE_X);
     void *walkerCommandAddress = cmdBufferAddress;
-    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 2u, false);
+    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 2u, 2, false);
     auto walkerCommand = genCmdCast<WalkerType *>(walkerCommandAddress);
 
     ASSERT_NE(nullptr, walkerCommand);
@@ -437,7 +437,7 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, WalkerPartitionTests, givenProgramComputeWalkerWhen
 
     walker.setPartitionType(WalkerType::PARTITION_TYPE::PARTITION_TYPE_Y);
     walkerCommandAddress = cmdBufferAddress;
-    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 2u, false);
+    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 2u, 2, false);
     walkerCommand = genCmdCast<WalkerType *>(walkerCommandAddress);
 
     ASSERT_NE(nullptr, walkerCommand);
@@ -446,7 +446,7 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, WalkerPartitionTests, givenProgramComputeWalkerWhen
 
     walker.setPartitionType(WalkerType::PARTITION_TYPE::PARTITION_TYPE_Z);
     walkerCommandAddress = cmdBufferAddress;
-    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 2u, false);
+    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 2u, 2, false);
     walkerCommand = genCmdCast<WalkerType *>(walkerCommandAddress);
 
     ASSERT_NE(nullptr, walkerCommand);
@@ -456,7 +456,7 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, WalkerPartitionTests, givenProgramComputeWalkerWhen
     // if we program with partition Count == 1 then do not trigger partition stuff
     walker.setPartitionType(WalkerType::PARTITION_TYPE::PARTITION_TYPE_DISABLED);
     walkerCommandAddress = cmdBufferAddress;
-    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 1u, false);
+    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 1u, 2, false);
     walkerCommand = genCmdCast<WalkerType *>(walkerCommandAddress);
 
     ASSERT_NE(nullptr, walkerCommand);
@@ -1758,7 +1758,7 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, WalkerPartitionTests, givenForceExecutionOnSingleTi
     bool forceExecutionOnSingleTile = false;
     walker.setPartitionType(WalkerType::PARTITION_TYPE::PARTITION_TYPE_X);
     void *walkerCommandAddress = cmdBufferAddress;
-    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 2u, forceExecutionOnSingleTile);
+    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 2u, 2, forceExecutionOnSingleTile);
     auto walkerCommand = genCmdCast<WalkerType *>(walkerCommandAddress);
 
     ASSERT_NE(nullptr, walkerCommand);
@@ -1768,7 +1768,7 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, WalkerPartitionTests, givenForceExecutionOnSingleTi
 
     forceExecutionOnSingleTile = true;
     walkerCommandAddress = cmdBufferAddress;
-    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 2u, forceExecutionOnSingleTile);
+    programPartitionedWalker<FamilyType>(cmdBufferAddress, totalBytesProgrammed, &walker, 2u, 2, forceExecutionOnSingleTile);
     walkerCommand = genCmdCast<WalkerType *>(walkerCommandAddress);
 
     ASSERT_NE(nullptr, walkerCommand);

Original file line number	Diff line number	Diff line change
`@@ -1,5 +1,5 @@`
`1`	`1`	`/*`
`2`		`- * Copyright (C) 2021-2023 Intel Corporation`
	`2`	`+ * Copyright (C) 2021-2024 Intel Corporation`
`3`	`3`	`*`
`4`	`4`	`* SPDX-License-Identifier: MIT`
`5`	`5`	`*`
`@@ -90,6 +90,9 @@ struct ImplicitScalingDispatch {`
`90`	`90`
`91`	`91`	`static bool platformSupportsImplicitScaling(const RootDeviceEnvironment &rootDeviceEnvironment);`
`92`	`92`
	`93`	`+ template <typename WalkerType>`
	`94`	`+ static void appendWalkerFields(WalkerType &walkerCmd, uint32_t tileCount);`
	`95`	`+`
`93`	`96`	`private:`
`94`	`97`	`static bool pipeControlStallRequired;`
`95`	`98`	`};`