Support for bindless mode in L0 - improvements

Related-To: NEO-6448

- add new IGC compilation flag when bindless mode used
- fix SBA programming of BindlessSurfaceStateSize -
always set maximum surface state count
- fix residency of global DSH heap on gen9 - gen12lp
in bindless mode
- add L0 aub test with bindless kernel - disabled
- partial fixes in OCL aub tests


Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>

shared/source/built_ins/kernels/CMakeLists.txt

@@ -26,7 +26,7 @@ set(BUILTIN_OPTIONS_STATELESS
 )
 
 set(bindless_OPTIONS
-    -internal_options "-cl-intel-use-bindless-mode"
+    -internal_options "-cl-intel-use-bindless-mode -cl-intel-use-bindless-advanced-mode"
 )
 
 set(bindful_OPTIONS

shared/source/command_container/command_encoder_bdw_and_later.inl

@@ -113,9 +113,6 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container,
                                                                     kernelDescriptor.payloadMappings.samplerTable.borderColor,
                                                                     dispatchInterface->getDynamicStateHeapData(),
                                                                     device->getBindlessHeapsHelper(), device->getHardwareInfo());
-        if (ApiSpecificConfig::getBindlessConfiguration()) {
-            container.getResidencyContainer().push_back(device->getBindlessHeapsHelper()->getHeap(NEO::BindlessHeapsHelper::BindlesHeapType::GLOBAL_DSH)->getGraphicsAllocation());
-        }
     }
 
     idd.setSamplerStatePointer(samplerStateOffset);
@@ -220,6 +217,10 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container,
 
     cmd.setPredicateEnable(isPredicate);
 
+    if (ApiSpecificConfig::getBindlessConfiguration()) {
+        container.getResidencyContainer().push_back(device->getBindlessHeapsHelper()->getHeap(NEO::BindlessHeapsHelper::BindlesHeapType::GLOBAL_DSH)->getGraphicsAllocation());
+    }
+
     EncodeDispatchKernel<Family>::adjustInterfaceDescriptorData(idd, hwInfo);
 
     PreemptionHelper::applyPreemptionWaCmdsBegin<Family>(listCmdBufferStream, *device);

shared/source/compiler_interface/compiler_options/compiler_options_base.h

@@ -26,7 +26,7 @@ static constexpr ConstStringRef fastRelaxedMath = "-cl-fast-relaxed-math";
 static constexpr ConstStringRef preserveVec3Type = "-fpreserve-vec3-type";
 static constexpr ConstStringRef createLibrary = "-create-library";
 static constexpr ConstStringRef generateDebugInfo = "-g";
-static constexpr ConstStringRef bindlessMode = "-cl-intel-use-bindless-mode";
+static constexpr ConstStringRef bindlessMode = "-cl-intel-use-bindless-mode -cl-intel-use-bindless-advanced-mode";
 static constexpr ConstStringRef uniformWorkgroupSize = "-cl-uniform-work-group-size";
 static constexpr ConstStringRef forceEmuInt32DivRem = "-cl-intel-force-emu-int32divrem";
 static constexpr ConstStringRef forceEmuInt32DivRemSP = "-cl-intel-force-emu-sp-int32divrem";

shared/source/helpers/state_base_address.h

@@ -57,5 +57,7 @@ struct StateBaseAddressHelper {
     static void appendExtraCacheSettings(STATE_BASE_ADDRESS *stateBaseAddress, GmmHelper *gmmHelper);
 
     static void programBindingTableBaseAddress(LinearStream &commandStream, const IndirectHeap &ssh, GmmHelper *gmmHelper);
+
+    static uint32_t getMaxBindlessSurfaceStates();
 };
 } // namespace NEO

shared/source/helpers/state_base_address_base.inl

@@ -37,6 +37,10 @@ void StateBaseAddressHelper<GfxFamily>::programStateBaseAddress(
 
     *stateBaseAddress = GfxFamily::cmdInitStateBaseAddress;
     bool overrideBindlessSurfaceStateBase = true;
+
+    const auto surfaceStateCount = getMaxBindlessSurfaceStates();
+    stateBaseAddress->setBindlessSurfaceStateSize(surfaceStateCount);
+
     if (useGlobalHeapsBaseAddress) {
         stateBaseAddress->setDynamicStateBaseAddressModifyEnable(true);
         stateBaseAddress->setDynamicStateBufferSizeModifyEnable(true);
@@ -48,7 +52,6 @@ void StateBaseAddressHelper<GfxFamily>::programStateBaseAddress(
 
         stateBaseAddress->setBindlessSurfaceStateBaseAddressModifyEnable(true);
         stateBaseAddress->setBindlessSurfaceStateBaseAddress(globalHeapsBaseAddress);
-        stateBaseAddress->setBindlessSurfaceStateSize(MemoryConstants::sizeOf4GBinPageEntities);
 
         overrideBindlessSurfaceStateBase = false;
     } else {

shared/source/helpers/state_base_address_bdw.inl

@@ -24,4 +24,9 @@ void StateBaseAddressHelper<GfxFamily>::appendStateBaseAddressParameters(
     appendExtraCacheSettings(stateBaseAddress, gmmHelper);
 }
 
+template <typename GfxFamily>
+uint32_t StateBaseAddressHelper<GfxFamily>::getMaxBindlessSurfaceStates() {
+    return 0;
+}
+
 } // namespace NEO

shared/source/helpers/state_base_address_icllp_and_later.inl

@@ -40,4 +40,10 @@ void StateBaseAddressHelper<GfxFamily>::appendStateBaseAddressParameters(
 
     appendExtraCacheSettings(stateBaseAddress, gmmHelper);
 }
+
+template <typename GfxFamily>
+uint32_t StateBaseAddressHelper<GfxFamily>::getMaxBindlessSurfaceStates() {
+    return (1 << 20) - 1;
+}
+
 } // namespace NEO

shared/source/helpers/state_base_address_skl.inl

@@ -32,4 +32,8 @@ void StateBaseAddressHelper<GfxFamily>::appendStateBaseAddressParameters(
     appendExtraCacheSettings(stateBaseAddress, gmmHelper);
 }
 
+template <typename GfxFamily>
+uint32_t StateBaseAddressHelper<GfxFamily>::getMaxBindlessSurfaceStates() {
+    return (1 << 20) - 1;
+}
 } // namespace NEO

shared/source/helpers/state_base_address_xehp_and_later.inl

@@ -114,4 +114,8 @@ template <typename GfxFamily>
 void StateBaseAddressHelper<GfxFamily>::appendIohParameters(STATE_BASE_ADDRESS *stateBaseAddress, const IndirectHeap *ioh, bool useGlobalHeapsBaseAddress, uint64_t indirectObjectHeapBaseAddress) {
 }
 
+template <typename GfxFamily>
+uint32_t StateBaseAddressHelper<GfxFamily>::getMaxBindlessSurfaceStates() {
+    return std::numeric_limits<uint32_t>::max();
+}
 } // namespace NEO