Add gfx1102 support

Change-Id: I39cbda81a7a999aa2ecfad7a3e720000f7ca3408
Signed-off-by: David Yat Sin <David.YatSin@amd.com>
This commit is contained in:
David Yat Sin
2022-01-21 07:51:52 -05:00
committed by David Yat Sin
parent 446c5e9672
commit cc3bd31591
6 changed files with 10 additions and 2 deletions
+1 -1
View File
@@ -49,7 +49,7 @@ set ( CPACK_DEBIAN_PACKAGE_HOMEPAGE "https://github.com/RadeonOpenCompute/ROCR-R
set ( CPACK_PACKAGE_HOMEPAGE_URL "https://github.com/RadeonOpenCompute/ROCR-Runtime" )
set(DEFAULT_TARGETS "gfx700;gfx701;gfx702;gfx801;gfx802;gfx803;gfx805;gfx810;gfx900;gfx902;gfx904;gfx906;gfx908;gfx909;gfx90a;gfx90c;gfx1010;gfx1011;gfx1012;gfx1013;gfx1030;gfx1031;gfx1032;gfx1033;gfx1034;gfx1035;gfx1036;gfx1100")
set(DEFAULT_TARGETS "gfx700;gfx701;gfx702;gfx801;gfx802;gfx803;gfx805;gfx810;gfx900;gfx902;gfx904;gfx906;gfx908;gfx909;gfx90a;gfx90c;gfx1010;gfx1011;gfx1012;gfx1013;gfx1030;gfx1031;gfx1032;gfx1033;gfx1034;gfx1035;gfx1036;gfx1100;gfx1102")
#
# Currently support for Windows platform is not present
+1
View File
@@ -316,6 +316,7 @@ constexpr size_t hsa_name_size = 63;
ISAREG_ENTRY_GEN("gfx1035", 10, 3, 5, unsupported, unsupported)
ISAREG_ENTRY_GEN("gfx1036", 10, 3, 6, unsupported, unsupported)
ISAREG_ENTRY_GEN("gfx1100", 11, 0, 0, unsupported, unsupported)
ISAREG_ENTRY_GEN("gfx1102", 11, 0, 2, unsupported, unsupported)
#undef ISAREG_ENTRY_GEN
return supported_isas;
}
@@ -97,6 +97,7 @@ extern uint8_t ocl_blit_object_gfx1034[];
extern uint8_t ocl_blit_object_gfx1035[];
extern uint8_t ocl_blit_object_gfx1036[];
extern uint8_t ocl_blit_object_gfx1100[];
extern uint8_t ocl_blit_object_gfx1102[];
// Arguments inserted by OCL compiler, all zero here.
struct OCLHiddenArgs {
@@ -1028,6 +1029,8 @@ hsa_status_t BlitKernel::GetPatchedBlitObject(const char* agent_name,
*blit_code_object = ocl_blit_object_gfx1036;
} else if (sname == "gfx1100") {
*blit_code_object = ocl_blit_object_gfx1100;
} else if (sname == "gfx1102") {
*blit_code_object = ocl_blit_object_gfx1102;
} else {
return HSA_STATUS_ERROR_INVALID_ISA_NAME;
}
@@ -47,7 +47,7 @@ find_package(Clang REQUIRED HINTS ${CMAKE_PREFIX_PATH}/llvm PATHS /opt/rocm/llvm
# Determine the target devices if not specified
if (NOT DEFINED TARGET_DEVICES)
set (TARGET_DEVICES "gfx700;gfx701;gfx702;gfx801;gfx802;gfx803;gfx805;gfx810;gfx900;gfx902;gfx904;gfx906;gfx908;gfx909;gfx90a;gfx90c;gfx1010;gfx1011;gfx1012;gfx1013;gfx1030;gfx1031;gfx1032;gfx1033;gfx1034;gfx1035;gfx1036;gfx1100")
set (TARGET_DEVICES "gfx700;gfx701;gfx702;gfx801;gfx802;gfx803;gfx805;gfx810;gfx900;gfx902;gfx904;gfx906;gfx908;gfx909;gfx90a;gfx90c;gfx1010;gfx1011;gfx1012;gfx1013;gfx1030;gfx1031;gfx1032;gfx1033;gfx1034;gfx1035;gfx1036;gfx1100;gfx1102")
endif()
set( TARGET_DEVICES ${TARGET_DEVICES} CACHE STRING "Build targets" FORCE )
+1
View File
@@ -123,6 +123,7 @@ enum : unsigned {
EF_AMDGPU_MACH_AMDGCN_GFX1100 = 0x041,
EF_AMDGPU_MACH_AMDGCN_GFX1013 = 0x042,
EF_AMDGPU_MACH_AMDGCN_GFX1036 = 0x045,
EF_AMDGPU_MACH_AMDGCN_GFX1102 = 0x047,
// First/last AMDGCN-based processors.
EF_AMDGPU_MACH_AMDGCN_FIRST = EF_AMDGPU_MACH_AMDGCN_GFX600,
@@ -586,6 +586,7 @@ namespace code {
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1035: name = "gfx1035"; xnack_supported = false; sramecc_supported = false; break;
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1036: name = "gfx1036"; xnack_supported = false; sramecc_supported = false; break;
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1100: name = "gfx1100"; xnack_supported = false; sramecc_supported = false; break;
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1102: name = "gfx1102"; xnack_supported = false; sramecc_supported = false; break;
default: return false;
}
return true;
@@ -636,6 +637,8 @@ namespace code {
mach = ELF::EF_AMDGPU_MACH_AMDGCN_GFX90C;
else if (old_name == "AMD:AMDGPU:11:0:0")
mach = ELF::EF_AMDGPU_MACH_AMDGCN_GFX1100;
else if (old_name == "AMD:AMDGPU:11:0:2")
mach = ELF::EF_AMDGPU_MACH_AMDGCN_GFX1102;
else {
// Code object v2 only supports asics up to gfx906 plus gfx90c. Do NOT
// add handling of new asics into this if-else-if* block.