P4 to Git Change 1108714 by rayxiao@alit_opencl_rayxiao on 2015/01/05 09:58:36

EPR #399415 - Move aclInsertKernelStatistics to utility library. aclutInsertKernelStatistics.

Affected files ...

... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/amdocl.def.in#11 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/amdocl.map.in#12 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/amdoclcl.def.in#9 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/amdoclcl.map.in#8 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/api/v0_8/acl.cpp#23 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/api/v0_8/aclKstatsSI.cpp#1 add
... //depot/stg/opencl/drivers/opencl/compiler/lib/api/v0_8/aclKstatsr800.cpp#1 add
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/scwrapper/devState.cpp#23 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/scwrapper/devState.h#7 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/acl.h#7 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclEnums.h#15 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclStructs.h#14 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/loaders/bif/bifinternal.hpp#8 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/bif_section_labels.hpp#18 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/libUtils.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/libUtils.h#14 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuscsi.cpp#27 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/CLEnumCheck.cpp#39 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/aclAPI.cpp#18 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/aclAPI.hpp#3 edit

rocclr/compiler/lib/amdoclcl.def.in

@@ -37,6 +37,7 @@ aclSetType
 aclConvertType
 aclDisassemble
 aclGetDeviceBinary
+aclInsertKernelStatistics
 oclCompilerInit
 oclCompilerFini
 oclCompilerVersion

rocclr/compiler/lib/amdoclcl.map.in

@@ -82,6 +82,7 @@ global:
   aclSetType;
   aclConvertType;
   aclDisassemble;
+  aclInsertKernelStatistics;
   aclGetDeviceBinary;
   aclLoader_OnLoad;
   aclLoader_OnUnload;

rocclr/compiler/lib/include/v0_8/acl.h

@@ -7,7 +7,6 @@
 extern "C" {
 #endif
 #include "aclTypes.h"
-
 //!--------------------------------------------------------------------------!//
 // Functions that deal with aclCompiler objects.
 //!--------------------------------------------------------------------------!//
@@ -235,7 +234,18 @@ const void* ACL_API_ENTRY
 // Debug functionality
 //!--------------------------------------------------------------------------!//
 void aclDumpBinary(const aclBinary *bin);
-
+// Functions to retrieve kernel information
+void aclGetKstatsR800(const void* shader,
+       aclKernelStats &kstats, const char* chip_id);
+void aclGetKstatsSI(const void* shader,
+       aclKernelStats &kstats);
+acl_error
+  aclInsertKernelStatistics(aclCompiler *cl,
+    aclBinary *bin);
+//! Define hardware info constants for SI and above devices
+const static unsigned SI_sgprs_avail = 102;
+const static unsigned SI_vgprs_avail = 256;
+const static unsigned SI_ldssize_avail = 32*1024;
 
 //!--------------------------------------------------------------------------!//
 // Functions that deal with memory.

rocclr/compiler/lib/include/v0_8/aclEnums.h

@@ -182,7 +182,8 @@ typedef enum _bif_sections_enum_0_8 {
   aclBRIGoprs       = 32,
   aclBRIGstrs       = 33,
   aclHSADEBUG       = 34,
-  aclLAST           = 35
+  aclKSTATS         = 35, // For storing kernel statistics
+  aclLAST           = 36
 } aclSections_0_8;
 
 //! An enumeration that defines what are valid queries for aclQueryInfo.

rocclr/compiler/lib/include/v0_8/aclStructs.h

@@ -311,4 +311,24 @@ typedef struct _acl_compiler_rec_0_8_1 {
   aclLoaderData *apiData; // pointer to data store for the compiler API loader.
 } aclCompilerHandle_0_8_1;
 
+//! Structure to hold kernel statistics obtained from kernel
+typedef struct _acl_kernel_stats_0_8_1 {
+  unsigned int scratchRegs;
+  unsigned int scratchSize;
+  unsigned int availablevgprs;
+  unsigned int availablesgprs;
+  unsigned int usedvgprs;
+  unsigned int usedsgprs;
+  unsigned int availableldssize;
+  unsigned int usedldssize;
+  unsigned int availablestacksize;
+  unsigned int usedstacksize;
+  unsigned int wavefrontsize;
+  unsigned int wavefrontpersimd;
+  unsigned int threadsperworkgroup;
+  unsigned int reqdworkgroup_x;
+  unsigned int reqdworkgroup_y;
+  unsigned int reqdworkgroup_z;
+} aclKernelStats;
+
 #endif // _ACL_STRUCTS_0_8_H_

rocclr/compiler/lib/utils/bif_section_labels.hpp

@@ -36,7 +36,8 @@ typedef enum {
   symDebugilBinary,
   symAsmText,
   symDLL,
-  symLast
+  symLast,
+  symKernelStats
 } oclBIFSymbolID;
 
 struct oclBIFSymbolStruct {
@@ -48,7 +49,7 @@ struct oclBIFSymbolStruct {
 };
 
 // These are the symbols that are defined by the BIF 3.0 spec
-static const oclBIFSymbolStruct BIF30[26] =
+static const oclBIFSymbolStruct BIF30[27] =
 {
   // 0: BIF 3.0 compiler options, .comment section via library support.
   {symOpenclCompilerOptions,
@@ -106,6 +107,8 @@ static const oclBIFSymbolStruct BIF30[26] =
   {symDebugilBinary, { "__debugil_binary", "" }, {aclINTERNAL, aclLAST}},
   {symAsmText, {"", ""}, {aclLAST, aclCODEGEN}},
   {symDLL, {"", ""}, {aclLAST, aclTEXT}},
+  // 26: BIF 3.0 HSAIL kernel statistics
+  {symKernelStats, { "__HSAIL_", "_kernel_statistics" }, {aclKSTATS, aclLAST}},
 }; // BIF30
 
 

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

@@ -3,6 +3,7 @@
 //
 #include "acl.h"
 #include "aclTypes.h"
+#include "api/v0_8/aclValidation.h"
 #include "v0_7/clTypes.h"
 #include "v0_7/clCompiler.h"
 #include "libUtils.h"
@@ -10,6 +11,7 @@
 #include "utils/target_mappings.h"
 #include "utils/versions.hpp"
 #include "utils/options.hpp"
+#include "backends/gpu/scwrapper/devState.h"
 #include <cassert>
 #include <cstring>
 #include "bif/bif.hpp"
@@ -373,6 +375,43 @@ aclutCopyBinOpts(aclBinaryOptions *dst, const aclBinaryOptions *src, bool is64)
   }
 }
 
+acl_error
+aclutInsertKernelStatistics(aclCompiler *cl, aclBinary *bin)
+{
+    if (!aclValidateCompiler(cl, true)) {
+        return ACL_INVALID_COMPILER;
+    }
+    if (!aclValidateBinary(bin)) {
+        return ACL_INVALID_BINARY;
+    }
+    size_t len = 0;
+    acl_error err = ACL_SUCCESS;
+    const void *isa = aclExtractSection(cl, bin, &len, aclTEXT, &err);
+    if (err != ACL_SUCCESS)
+        return err;
+    aclTargetInfo *tgtInfo = aclutGetTargetInfo(bin);
+    const char* chipName = aclGetChip(*tgtInfo);
+    unsigned family = getFamilyEnum(tgtInfo);
+    unsigned chip = getChipEnum(tgtInfo);
+    // Non-GPU devices have family_enum set to 1 and do not qualify. Need to update.
+    if (family >= FAMILY_R600 &&
+	    family <= FAMILY_CZ) {
+	  aclKernelStats kstats = {0};
+      if (family < FAMILY_SI) {
+        aclGetKstatsR800(isa, kstats, chipName);
+      }
+      else {
+        aclGetKstatsSI(isa, kstats);
+      }
+	  kstats.wavefrontsize = amdcl::GetWavefrontSize(family, chip);
+	  const oclBIFSymbolStruct* symbol = findBIF30SymStruct(symKernelStats);
+	  assert(symbol && "symbol not found");
+	  std::string symName = std::string(symbol->str[PRE]) + std::string(symbol->str[POST]);
+	  err = aclInsertSymbol(cl, bin, reinterpret_cast<void*>(&kstats), sizeof(kstats), aclKSTATS, symName.c_str());
+    }
+    return err;
+}
+
 void initElfDeviceCaps(aclBinary *elf)
 {
   if (aclutGetCaps(elf)->encryptCode) {

rocclr/compiler/lib/utils/v0_8/libUtils.h

@@ -108,6 +108,10 @@ aclutCopyBinOpts(aclBinaryOptions *dst,
     const aclBinaryOptions *src,
     bool is64bit);
 
+// Retrieve kernel statistics from binary
+// and insert to elf as symbol
+acl_error aclutInsertKernelStatistics(aclCompiler*, aclBinary*);
+
 // Helper function that returns the
 // allocation function from the binary.
 AllocFunc

rocclr/runtime/device/gpu/gpuscsi.cpp

@@ -56,10 +56,10 @@ NullKernel::siCreateHwInfo(const void* shader, AMUabiAddEncoding& encoding)
     newInfos[i].value = cShader->common.numSgprs;
     i++;
     newInfos[i].address = AMU_ABI_SI_NUM_SGPRS_AVAIL;
-    newInfos[i].value = 104-2; //512;//options.NumSGPRsAvailable;
+    newInfos[i].value = SI_sgprs_avail; //512;//options.NumSGPRsAvailable;
     i++;
     newInfos[i].address = AMU_ABI_SI_NUM_VGPRS_AVAIL;
-    newInfos[i].value = 256;//options.NumVGPRsAvailable;
+    newInfos[i].value = SI_vgprs_avail;//options.NumVGPRsAvailable;
     i++;
 
     newInfos[i].address = AMU_ABI_SI_FLOAT_MODE;
@@ -110,7 +110,7 @@ NullKernel::siCreateHwInfo(const void* shader, AMUabiAddEncoding& encoding)
     i++;
 
     newInfos[i].address = AMU_ABI_LDS_SIZE_AVAIL;
-    newInfos[i].value = 32*1024;    //options.LDSSize;
+    newInfos[i].value = SI_ldssize_avail;    //options.LDSSize;
     i++;
 
     COMPUTE_PGM_RSRC2 computePgmRsrc2;