P4 to Git Change 1325492 by smekhano@stas-mwamba-rocm on 2016/10/12 12:36:24

SWDEV-104875 - ROCm/HSA: use the finalizer from source tree
	Right now ROCm/HSA uses finalizer from the HSA RT installed.
	This finalizer version has outdated stale SC sources.
	At the same time source tree has fresh finalizer sources matching ORCA.
	The offline tool amdhsafin is built from that sources.
	This change switches from HSA RT finalizer to the in tree finalizer.

	Testing: precheckin
	Reviewed by Laurent Morichetti and Evgeny Mankov

Affected files ...

... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/libUtils.cpp#20 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#39 edit


[ROCm/clr commit: 9385a3778f]

projects/clr/rocclr/compiler/lib/utils/v0_8/libUtils.cpp

@@ -18,6 +18,8 @@ static const std::string sgfx700 = "AMD:AMDGPU:7:0:0";
 static const std::string sgfx701 = "AMD:AMDGPU:7:0:1";
 static const std::string sgfx800 = "AMD:AMDGPU:8:0:0";
 static const std::string sgfx801 = "AMD:AMDGPU:8:0:1";
+static const std::string sgfx802 = "AMD:AMDGPU:8:0:2";
+static const std::string sgfx803 = "AMD:AMDGPU:8:0:3";
 static const std::string sgfx804 = "AMD:AMDGPU:8:0:4";
 static const std::string sgfx810 = "AMD:AMDGPU:8:1:0";
 static const std::string sgfx900 = "AMD:AMDGPU:9:0:0";
@@ -518,6 +520,8 @@ const std::string &getIsaTypeName(const aclTargetInfo *target)
   case 701: return sgfx701;
   case 800: return sgfx800;
   case 801: return sgfx801;
+  case 802: return sgfx802;
+  case 803: return sgfx803;
   case 804: return sgfx804;
   case 810: return sgfx810;
   case 900: return sgfx900;
@@ -553,11 +557,19 @@ int getIsaType(const aclTargetInfo *target)
     case FAMILY_VI:
       switch (Mapping.chip_enum) {
         default: return 800;
-        case VI_ICELAND_M_A0:
-        case VI_TONGA_P_A0: return 800;
+        case VI_TONGA_P_A0:
+#if defined(BUILD_HSA_TARGET)
+          return 802; // Special case - Boltzmann Tonga
+          // Otherwise fall through
+#endif
+        case VI_ICELAND_M_A0: return 800;
+        case VI_FIJI_P_A0:
+#if defined(BUILD_HSA_TARGET)
+          return 803; // Special case - Boltzmann Fiji
+          // Otherwise fall through
+#endif
         case VI_ELLESMERE_P_A0:
         case VI_BAFFIN_M_A0:
-        case VI_FIJI_P_A0:
         case VI_LEXA_V_A0: return 804;
       }
     case FAMILY_CZ:

projects/clr/rocclr/runtime/device/rocm/rocprogram.cpp

@@ -1216,6 +1216,7 @@ HSAILProgram::linkImpl(amd::option::Options *options)
         buildLog_ += "Failed to create Brig Container";
         return false;
     }
+#if defined(USE_HSART_FINALIZER)
     // Create a program.
     hsa_status_t status = hsa_ext_program_create(
         HSA_MACHINE_MODEL_LARGE,
@@ -1274,6 +1275,36 @@ HSAILProgram::linkImpl(amd::option::Options *options)
         return false;
     }
 
+#else // ! USE_HSART_FINALIZER
+    std::string fin_options(options->origOptionStr);
+    // Append an option so that we can selectively enable a SCOption on CZ
+    // whenever IOMMUv2 is enabled.
+    if (dev().isFineGrainedSystem(true)) {
+        fin_options.append(" -sc-xnack-iommu");
+    }
+    errorCode = aclCompile(dev().compiler(), binaryElf_,
+        fin_options.c_str(), ACL_TYPE_CG, ACL_TYPE_ISA, logFunction);
+    buildLog_ += aclGetCompilerLog(dev().compiler());
+    if (errorCode != ACL_SUCCESS) {
+        buildLog_ += "Error: BRIG finalization to ISA failed.\n";
+        return false;
+    }
+    size_t secSize;
+    void *data = (void*)aclExtractSection(device().compiler(),
+        binaryElf_, &secSize, aclTEXT, &errorCode);
+    if (errorCode != ACL_SUCCESS) {
+        buildLog_ += "Error: cannot extract ISA from compiled binary.\n";
+        return false;
+    }
+    hsa_status_t status = hsa_code_object_deserialize(data, secSize,
+        NULL, &hsaProgramCodeObject_ );
+    if (status != HSA_STATUS_SUCCESS) {
+      buildLog_ += "Error: failed to load finalized code object.\n";
+      return false;
+    }
+
+#endif // USE_HSART_FINALIZER
+
     // HLC always generates full profile
     hsa_profile_t profile = HSA_PROFILE_FULL;
 
@@ -1431,6 +1462,7 @@ HSAILProgram::linkImpl(amd::option::Options *options)
     saveBinaryAndSetType(TYPE_EXECUTABLE);
     buildLog_ += g_complibApi._aclGetCompilerLog(device().compiler());
 
+#if defined(USE_HSART_FINALIZER)
     if (options->isDumpFlagSet(amd::option::DUMP_O) || options->isDumpFlagSet(amd::option::DUMP_ISA)) {
         amd::hsa::code::AmdHsaCode code;
         if (!code.InitAsHandle(hsaProgramCodeObject_)) {
@@ -1446,7 +1478,7 @@ HSAILProgram::linkImpl(amd::option::Options *options)
             }
         }
     }
-
+#endif // defined(USE_HSART_FINALIZER)
 #endif // !defined(WITH_LIGHTNING_COMPILER)
     return true;
 }