[Offload][AMDGPU] accept generic target (llvm#118919)

Enables generic ISA, e.g., "--offload-arch=gfx11-generic" device code to
run on gfx11-generic ISA capable device.

Executable may contain one ELF that has specific target ISA and another
ELF that has compatible generic ISA.
Under that circumstance, this code should say both ELFs are compatible,
leaving the rest to PluginManager to handle.
Suggestions on how best to address that is welcome.

Author: hidekisaito

offload/DeviceRTL/CMakeLists.txt

@@ -42,12 +42,16 @@ set(devicertl_base_directory ${CMAKE_CURRENT_SOURCE_DIR})
 set(include_directory ${devicertl_base_directory}/include)
 set(source_directory ${devicertl_base_directory}/src)
 
-set(all_capabilities 35 37 50 52 53 60 61 62 70 72 75 80 86 89 87 90)
-set(all_amdgpu_architectures "gfx700;gfx701;gfx801;gfx803;gfx900;gfx902;gfx906"
-                             "gfx908;gfx90a;gfx90c;gfx940;gfx941;gfx942;gfx950;gfx1010"
-                             "gfx1030;gfx1031;gfx1032;gfx1033;gfx1034;gfx1035"
-                             "gfx1036;gfx1100;gfx1101;gfx1102;gfx1103;gfx1150"
-                             "gfx1151;gfx1152;gfx1153;gfx1200;gfx1201")
+set(all_amdgpu_architectures "gfx700;gfx701;gfx801;gfx803"
+                             "gfx9-generic;gfx900;gfx902;gfx906;gfx908"
+                             "gfx90a;gfx90c"
+                             "gfx9-4-generic;gfx940;gfx941;gfx942;gfx950"
+                             "gfx10-1-generic;gfx1010;gfx1012"
+                             "gfx10-3-generic;gfx1030;gfx1031;gfx1032;gfx1033"
+                             "gfx1034;gfx1035;gfx1036"
+                             "gfx11-generic;gfx1100;gfx1101;gfx1102;gfx1103"
+                             "gfx1150;gfx1151;gfx1152;gfx1153"
+                             "gfx12-generic;gfx1200;gfx1201")
 set(all_nvptx_architectures "sm_35;sm_37;sm_50;sm_52;sm_53;sm_60;sm_61;sm_62"
                             "sm_70;sm_72;sm_75;sm_80;sm_86;sm_87;sm_89;sm_90")
 set(all_gpu_architectures

offload/DeviceRTL/include/Platform.h

@@ -123,6 +123,18 @@ extern const inline unsigned __oclc_ISA_version = 11503;
 extern const inline unsigned __oclc_ISA_version = 12000;
 #elif defined(__gfx1201__)
 extern const inline unsigned __oclc_ISA_version = 12001;
+#elif defined(__gfx9_generic__)
+extern const inline unsigned __oclc_ISA_version = 9000;
+#elif defined(__gfx9_4_generic__)
+extern const inline unsigned __oclc_ISA_version = 9402;
+#elif defined(__gfx10_1_generic__)
+extern const inline unsigned __oclc_ISA_version = 10100;
+#elif defined(__gfx10_3_generic__)
+extern const inline unsigned __oclc_ISA_version = 10300;
+#elif defined(__gfx11_generic__)
+extern const inline unsigned __oclc_ISA_version = 11003;
+#elif defined(__gfx12_generic__)
+extern const inline unsigned __oclc_ISA_version = 12000;
 #else
 #error "Unknown AMDGPU architecture"
 #endif

offload/hostexec/CMakeLists.txt

@@ -97,7 +97,7 @@ else()
   return()
 endif()
 
-set(amdgpu_mcpus gfx700 gfx701 gfx801 gfx803 gfx900 gfx902 gfx906 gfx908 gfx90a gfx90c gfx940 gfx941 gfx942 gfx1010 gfx1030 gfx1031 gfx1032 gfx1033 gfx1034 gfx1035 gfx1036 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx1200 gfx1201)
+set(amdgpu_mcpus gfx700 gfx701 gfx801 gfx803 gfx900 gfx902 gfx906 gfx908 gfx90a gfx90c gfx940 gfx941 gfx942 gfx1010 gfx1030 gfx1031 gfx1032 gfx1033 gfx1034 gfx1035 gfx1036 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1153 gfx1200 gfx1201 gfx9-generic gfx9-4-generic gfx10-1-generic gfx10-3-generic gfx11-generic gfx12-generic)
 if (DEFINED LIBOMPTARGET_AMDGCN_GFXLIST)
   set(amdgpu_mcpus ${LIBOMPTARGET_AMDGCN_GFXLIST})
 endif()

offload/plugins-nextgen/amdgpu/src/rtl.cpp

@@ -406,8 +406,8 @@ Error asyncMemCopy(bool UseMultipleSdmaEngines, void *Dst, hsa_agent_t DstAgent,
 #endif
 }
 
-Expected<std::string> getTargetTripleAndFeatures(hsa_agent_t Agent) {
-  std::string Target;
+Error getTargetTripleAndFeatures(hsa_agent_t Agent,
+                                 SmallVector<SmallString<32>> &Targets) {
   auto Err = utils::iterateAgentISAs(Agent, [&](hsa_isa_t ISA) {
     uint32_t Length;
     hsa_status_t Status;
@@ -421,13 +421,13 @@ Expected<std::string> getTargetTripleAndFeatures(hsa_agent_t Agent) {
       return Status;
 
     llvm::StringRef TripleTarget(ISAName.begin(), Length);
-    if (TripleTarget.consume_front("amdgcn-amd-amdhsa"))
-      Target = TripleTarget.ltrim('-').rtrim('\0').str();
-    return HSA_STATUS_INFO_BREAK;
+    if (TripleTarget.consume_front("amdgcn-amd-amdhsa")) {
+      auto Target = TripleTarget.ltrim('-').rtrim('\0');
+      Targets.push_back(Target);
+    }
+    return HSA_STATUS_SUCCESS;
   });
-  if (Err)
-    return Err;
-  return Target;
+  return Err;
 }
 
 } // namespace utils
@@ -2928,12 +2928,10 @@ struct AMDGPUDeviceTy : public GenericDeviceTy, AMDGenericDeviceTy {
     DP("The number of XGMI Engines: %i\n", NumXGmiEngines);
 
     // Detect if XNACK is enabled
-    auto TargeTripleAndFeaturesOrError =
-        utils::getTargetTripleAndFeatures(Agent);
-    if (!TargeTripleAndFeaturesOrError)
-      return TargeTripleAndFeaturesOrError.takeError();
-    if (static_cast<StringRef>(*TargeTripleAndFeaturesOrError)
-            .contains("xnack+"))
+    SmallVector<SmallString<32>> Targets;
+    if (auto Err = utils::getTargetTripleAndFeatures(Agent, Targets))
+      return Err;
+    if (!Targets.empty() && Targets[0].str().contains("xnack+"))
       IsXnackEnabled = true;
 
     // detect if device is an APU.
@@ -4666,13 +4664,16 @@ struct AMDGPUPluginTy final : public GenericPluginTy {
     if (!Processor)
       return false;
 
-    auto TargeTripleAndFeaturesOrError =
-        utils::getTargetTripleAndFeatures(getKernelAgent(DeviceId));
-    if (!TargeTripleAndFeaturesOrError)
-      return TargeTripleAndFeaturesOrError.takeError();
-    return utils::isImageCompatibleWithEnv(Processor ? *Processor : "",
-                                           ElfOrErr->getPlatformFlags(),
-                                           *TargeTripleAndFeaturesOrError);
+    SmallVector<SmallString<32>> Targets;
+    if (auto Err = utils::getTargetTripleAndFeatures(
+            getKernelAgent(DeviceId), Targets))
+      return Err;
+    for (auto &Target : Targets)
+      if (utils::isImageCompatibleWithEnv(
+              Processor ? *Processor : "", ElfOrErr->getPlatformFlags(),
+              Target.str()))
+        return true;
+    return false;
   }
 
   bool isDataExchangable(int32_t SrcDeviceId, int32_t DstDeviceId) override {

offload/plugins-nextgen/common/src/Utils/ELF.cpp

@@ -70,7 +70,8 @@ checkMachineImpl(const object::ELFObjectFile<ELFT> &ELFObj, uint16_t EMachine) {
         Header.e_ident[EI_ABIVERSION] != ELFABIVERSION_AMDGPU_HSA_V6)
       return createError("Invalid AMD ABI version, must be version above 4");
     if ((Header.e_flags & EF_AMDGPU_MACH) < EF_AMDGPU_MACH_AMDGCN_GFX700 ||
-        (Header.e_flags & EF_AMDGPU_MACH) > EF_AMDGPU_MACH_AMDGCN_GFX1201)
+        (Header.e_flags & EF_AMDGPU_MACH) >
+            EF_AMDGPU_MACH_AMDGCN_GFX9_4_GENERIC)
       return createError("Unsupported AMDGPU architecture");
   } else if (Header.e_machine == EM_CUDA) {
     if (~Header.e_flags & EF_CUDA_64BIT_ADDRESS)