From 2c6bcfbf8d2747c9c18a12e1ee0e21628610a94f Mon Sep 17 00:00:00 2001 From: foreman Date: Thu, 17 Nov 2016 17:25:47 -0500 Subject: [PATCH] P4 to Git Change 1342976 by todli@todli-win-opencl-kv1 on 2016/11/17 17:13:47 SWDEV-107271 - [OpenCL][GFXIP9 Bring up] add support for Raven(gfx901) Review: http://ocltc.amd.com/reviews/r/11809/ Affected files ... ... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/linker.cpp#154 edit ... //depot/stg/opencl/drivers/opencl/compiler/lib/loaders/bif/amuabi/amuABI.h#19 edit ... //depot/stg/opencl/drivers/opencl/compiler/lib/loaders/bif/amuabi/amuABIMultiBinary.cpp#19 edit ... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/libUtils.cpp#22 edit ... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/target_mappings.h#49 edit ... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/target_mappings_hsail.h#40 edit ... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/target_mappings_hsail64.h#35 edit ... //depot/stg/opencl/drivers/opencl/compiler/tools/driver/driver.cpp#65 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudefs.hpp#144 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#559 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.cpp#348 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp#176 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/caltarget.h#4 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldefs.hpp#14 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#35 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.cpp#11 edit ... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/binary/BIFEnumCheck.cpp#26 edit ... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/math/OCLMathFunc.cpp#28 edit ... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/runtime/OCLDeviceQueries.cpp#42 edit ... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/runtime/OCLOfflineCompilation.cpp#18 edit ... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/runtime/OCLPerfCounters.cpp#35 edit --- rocclr/compiler/lib/backends/common/linker.cpp | 3 ++- rocclr/compiler/lib/utils/v0_8/libUtils.cpp | 5 +++++ rocclr/compiler/lib/utils/v0_8/target_mappings.h | 2 +- rocclr/compiler/lib/utils/v0_8/target_mappings_hsail.h | 3 +++ rocclr/compiler/lib/utils/v0_8/target_mappings_hsail64.h | 3 +++ rocclr/runtime/device/gpu/gpudefs.hpp | 1 + rocclr/runtime/device/gpu/gpudevice.cpp | 1 + rocclr/runtime/device/gpu/gpusettings.cpp | 3 +++ rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp | 4 ++++ rocclr/runtime/device/gpu/gslbe/src/rt/caltarget.h | 3 ++- rocclr/runtime/device/pal/paldefs.hpp | 3 +++ rocclr/runtime/device/pal/paldevice.cpp | 7 ++++++- rocclr/runtime/device/pal/palsettings.cpp | 9 +++++---- 13 files changed, 39 insertions(+), 8 deletions(-) diff --git a/rocclr/compiler/lib/backends/common/linker.cpp b/rocclr/compiler/lib/backends/common/linker.cpp index e2b33830b3..d1d1604882 100644 --- a/rocclr/compiler/lib/backends/common/linker.cpp +++ b/rocclr/compiler/lib/backends/common/linker.cpp @@ -582,7 +582,8 @@ amdcl::OCLLinker::link(llvm::Module* input, std::vectorgetContext().setAMDLLVMContextHook(&hookup_); diff --git a/rocclr/compiler/lib/utils/v0_8/libUtils.cpp b/rocclr/compiler/lib/utils/v0_8/libUtils.cpp index 42a45ff36a..3ae71322a0 100644 --- a/rocclr/compiler/lib/utils/v0_8/libUtils.cpp +++ b/rocclr/compiler/lib/utils/v0_8/libUtils.cpp @@ -585,6 +585,11 @@ int getIsaType(const aclTargetInfo *target) default: return 900; case AI_GREENLAND_P_A0: return 900; } + case FAMILY_RV: + switch (Mapping.chip_enum) { + default: return 901; + case RAVEN_A0: return 901; + } } } diff --git a/rocclr/compiler/lib/utils/v0_8/target_mappings.h b/rocclr/compiler/lib/utils/v0_8/target_mappings.h index a7be39f755..5b86ccd4de 100644 --- a/rocclr/compiler/lib/utils/v0_8/target_mappings.h +++ b/rocclr/compiler/lib/utils/v0_8/target_mappings.h @@ -206,7 +206,7 @@ static const char* calTargetMapping[] = { "Hainan", "Hawaii", "Iceland", "Tonga", "Mullins", "Fiji", "Carrizo", "Ellesmere", "Baffin", - "gfx900", "Stoney", "gfx804", + "gfx900", "Stoney", "gfx804", "gfx901", }; #include "utils/v0_8/target_mappings_amdil.h" diff --git a/rocclr/compiler/lib/utils/v0_8/target_mappings_hsail.h b/rocclr/compiler/lib/utils/v0_8/target_mappings_hsail.h index 47fa56fcaf..6d5df7b195 100644 --- a/rocclr/compiler/lib/utils/v0_8/target_mappings_hsail.h +++ b/rocclr/compiler/lib/utils/v0_8/target_mappings_hsail.h @@ -9,6 +9,7 @@ #include "inc/asic_reg/ci_id.h" #include "inc/asic_reg/cz_id.h" #include "inc/asic_reg/ai_id.h" +#include "inc/asic_reg/rv_id.h" #include "inc/asic_reg/atiid.h" static const TargetMapping HSAILTargetMapping_0_8[] = { @@ -32,6 +33,8 @@ static const TargetMapping HSAILTargetMapping_0_8[] = { { "VI", "Ellesmere", "GFX8", amd::GPU_Library_HSAIL, VI_ELLESMERE_P_A0, F_VI_BASE, true, true, FAMILY_VI }, { "AI", "gfx900", "GFX9", amd::GPU_Library_HSAIL, AI_GREENLAND_P_A0, F_AI_BASE, true, true, FAMILY_AI }, { "VI", "gfx804", "GFX8", amd::GPU_Library_HSAIL, VI_LEXA_V_A0, F_VI_BASE, true, true, FAMILY_VI }, + + UnknownTarget, InvalidTarget }; diff --git a/rocclr/compiler/lib/utils/v0_8/target_mappings_hsail64.h b/rocclr/compiler/lib/utils/v0_8/target_mappings_hsail64.h index f2ab80d60b..98b0dab4d4 100644 --- a/rocclr/compiler/lib/utils/v0_8/target_mappings_hsail64.h +++ b/rocclr/compiler/lib/utils/v0_8/target_mappings_hsail64.h @@ -8,6 +8,7 @@ #include "inc/asic_reg/kv_id.h" #include "inc/asic_reg/ci_id.h" #include "inc/asic_reg/ai_id.h" +#include "inc/asic_reg/rv_id.h" #include "inc/asic_reg/atiid.h" static const TargetMapping HSAIL64TargetMapping_0_8[] = { @@ -31,6 +32,8 @@ static const TargetMapping HSAIL64TargetMapping_0_8[] = { { "VI", "Ellesmere", "GFX8", amd::GPU_Library_HSAIL, VI_ELLESMERE_P_A0, F_VI_BASE, true, true, FAMILY_VI }, { "AI", "gfx900", "GFX9", amd::GPU_Library_HSAIL, AI_GREENLAND_P_A0, F_AI_BASE, true, true, FAMILY_AI }, { "VI", "gfx804", "GFX8", amd::GPU_Library_HSAIL, VI_LEXA_V_A0, F_VI_BASE, true, true, FAMILY_VI }, + + UnknownTarget, InvalidTarget }; diff --git a/rocclr/runtime/device/gpu/gpudefs.hpp b/rocclr/runtime/device/gpu/gpudefs.hpp index 1e62dd4386..6bfdc6ef0e 100644 --- a/rocclr/runtime/device/gpu/gpudefs.hpp +++ b/rocclr/runtime/device/gpu/gpudefs.hpp @@ -128,6 +128,7 @@ static const AMDDeviceInfo DeviceInfo[] = { /* CAL_TARGET_GREENLAND */ { ED_ATI_CAL_MACHINE_GREENLAND_ISA, "gfx900", "gfx900", 4, 16, 1, 256, 64 * Ki, 32, 900 }, /* CAL_TARGET_STONEY */ { ED_ATI_CAL_MACHINE_STONEY_ISA, "Stoney", "stoney", 4, 16, 1, 256, 64 * Ki, 32, 800 }, /* CAL_TARGET_LEXA */ { ED_ATI_CAL_MACHINE_LEXA_ISA, "gfx804", "gfx804", 4, 16, 1, 256, 64 * Ki, 32, 800 }, +/* CAL_TARGET_RAVEN */ { ED_ATI_CAL_MACHINE_RAVEN_ISA, "gfx901", "gfx901", 4, 16, 1, 256, 64 * Ki, 32, 900 }, }; enum gfx_handle { diff --git a/rocclr/runtime/device/gpu/gpudevice.cpp b/rocclr/runtime/device/gpu/gpudevice.cpp index 04b2ae9706..6e20ca09c2 100644 --- a/rocclr/runtime/device/gpu/gpudevice.cpp +++ b/rocclr/runtime/device/gpu/gpudevice.cpp @@ -155,6 +155,7 @@ NullDevice::create(CALtarget target) case CAL_TARGET_GREENLAND: case CAL_TARGET_STONEY: case CAL_TARGET_LEXA: + case CAL_TARGET_RAVEN: calAttr.doublePrecision = CAL_TRUE; calAttr.isOpenCL200Device = CAL_TRUE; break; diff --git a/rocclr/runtime/device/gpu/gpusettings.cpp b/rocclr/runtime/device/gpu/gpusettings.cpp index caba9cf5cc..c90e2f5867 100644 --- a/rocclr/runtime/device/gpu/gpusettings.cpp +++ b/rocclr/runtime/device/gpu/gpusettings.cpp @@ -167,6 +167,9 @@ Settings::create( ModifyMaxWorkload modifyMaxWorkload = {0}; switch (target) { + case CAL_TARGET_RAVEN: + // APU systems for AI + apuSystem_ = true; case CAL_TARGET_GREENLAND: //TODO: specific codes for AI aiPlus_ = true; diff --git a/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp b/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp index 2b0f6a2a90..40a5cbd1cf 100644 --- a/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp +++ b/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp @@ -505,6 +505,10 @@ CALGSLDevice::SetupContext(int32 &asic_id) m_target = CAL_TARGET_LEXA; m_elfmachine = ED_ATI_CAL_MACHINE_LEXA_ISA; break; + case GSL_ATIASIC_ID_RAVEN: + m_target = CAL_TARGET_RAVEN; + m_elfmachine = ED_ATI_CAL_MACHINE_RAVEN_ISA; + break; default: // 6XX is not supported m_adp->deleteContext(temp_cs); diff --git a/rocclr/runtime/device/gpu/gslbe/src/rt/caltarget.h b/rocclr/runtime/device/gpu/gslbe/src/rt/caltarget.h index c6053e258a..0ef16010ee 100644 --- a/rocclr/runtime/device/gpu/gslbe/src/rt/caltarget.h +++ b/rocclr/runtime/device/gpu/gslbe/src/rt/caltarget.h @@ -46,7 +46,8 @@ typedef enum CALtargetEnum { CAL_TARGET_GREENLAND, /**< GREENLAND GPU ISA*/ CAL_TARGET_STONEY, /**< STONEY GPU ISA*/ CAL_TARGET_LEXA, /**< LEXA GPU ISA*/ - CAL_TARGET_LAST = CAL_TARGET_LEXA, /**< last */ + CAL_TARGET_RAVEN, /**< RAVEN GPU ISA*/ + CAL_TARGET_LAST = CAL_TARGET_RAVEN, /**< last */ //##END_PRIVATE## } CALtarget; diff --git a/rocclr/runtime/device/pal/paldefs.hpp b/rocclr/runtime/device/pal/paldefs.hpp index 421bf3115a..9e38daada4 100644 --- a/rocclr/runtime/device/pal/paldefs.hpp +++ b/rocclr/runtime/device/pal/paldefs.hpp @@ -156,6 +156,9 @@ static const AMDDeviceInfo GfxIpDeviceInfo[] = { /* GFX900 */ { "gfx900", "gfx900", 4, 16, 1, 256, 64 * Ki, 32, 900 }, }; +static const AMDDeviceInfo Gfx901DeviceInfo = +/* GFX901 */ { "gfx901", "gfx901", 4, 16, 1, 256, 64 * Ki, 32, 901 }; + enum gfx_handle { gfx700 = 700, gfx701 = 701, diff --git a/rocclr/runtime/device/pal/paldevice.cpp b/rocclr/runtime/device/pal/paldevice.cpp index 9b1dd982ae..2d36aa0290 100644 --- a/rocclr/runtime/device/pal/paldevice.cpp +++ b/rocclr/runtime/device/pal/paldevice.cpp @@ -700,7 +700,12 @@ Device::create(Pal::IDevice* device) hwInfo_ = &DeviceInfo[static_cast(properties().revision)]; } else if (ipLevel_ >= Pal::GfxIpLevel::GfxIp9) { - hwInfo_ = &GfxIpDeviceInfo[static_cast(ipLevel_)]; + if (properties().gpuType == Pal::GpuType::Integrated) { + hwInfo_ = &Gfx901DeviceInfo; + } + else { + hwInfo_ = &GfxIpDeviceInfo[static_cast(ipLevel_)]; + } } else { return false; diff --git a/rocclr/runtime/device/pal/palsettings.cpp b/rocclr/runtime/device/pal/palsettings.cpp index 44f47a418c..7df2023cbb 100644 --- a/rocclr/runtime/device/pal/palsettings.cpp +++ b/rocclr/runtime/device/pal/palsettings.cpp @@ -163,6 +163,11 @@ Settings::create( // Update GPU specific settings and info structure if we have any ModifyMaxWorkload modifyMaxWorkload = {0}; + // APU systems + if (palProp.gpuType == Pal::GpuType::Integrated) { + apuSystem_ = true; + } + switch (palProp.revision) { case Pal::AsicRevision::Unknown: switch (palProp.gfxLevel) { @@ -177,8 +182,6 @@ Settings::create( case Pal::AsicRevision::Carrizo: case Pal::AsicRevision::Stoney: if (!aiPlus_) { - // APU systems for VI - apuSystem_ = true; // Fix BSOD/TDR issues observed on Stoney Win7/8.1/10 minWorkloadTime_ = 1000; modifyMaxWorkload.time = 1000; // Decided by experiment @@ -204,8 +207,6 @@ Settings::create( case Pal::AsicRevision::Kalindi: case Pal::AsicRevision::Spectre: if (!viPlus_) { - // APU systems for CI - apuSystem_ = true; // Fix BSOD/TDR issues observed on Kaveri Win7 (EPR#416903) modifyMaxWorkload.time = 250000; // 250ms modifyMaxWorkload.minorVersion = 1; // Win 7