GBE: Handle bti allocation for internal buffer used by printf.

1. Move the bti/Register map from gbe::Context to ir::Function.
2. use GlobalVariable instead of 'call' to get internal buffer (used for printf) base address.

Signed-off-by: Ruiling Song <ruiling.song@intel.com>
Reviewed-by: Zhigang Gong <zhigang.gong@linux.intel.com>

16 files changed, 96 insertions, 34 deletions

diff --git a/backend/src/backend/context.cpp b/backend/src/backend/context.cpp
index dcf3f65..e09a309 100644
--- a/backend/src/backend/context.cpp
+++ b/backend/src/backend/context.cpp
@@ -437,10 +437,7 @@ namespace gbe
     curbeRegs.insert(std::make_pair(reg, offset));
   }
   ir::Register Context::getSurfaceBaseReg(unsigned char bti) {
-    map<unsigned char, ir::Register>::iterator iter;
-    iter = btiRegMap.find(bti);
-    GBE_ASSERT(iter != btiRegMap.end());
-    return iter->second;
+    return fn.getSurfaceBaseReg(bti);
   }
 
   void Context::buildArgList(void) {
@@ -449,8 +446,6 @@ namespace gbe
       kernel->args = GBE_NEW_ARRAY_NO_ARG(KernelArgument, kernel->argNum);
     else
       kernel->args = NULL;
-    btiRegMap.clear();
-    btiRegMap.insert(std::make_pair(1, ir::ocl::stackbuffer));
     for (uint32_t argID = 0; argID < kernel->argNum; ++argID) {
       const auto &arg = fn.getArg(argID);
 
@@ -466,7 +461,6 @@ namespace gbe
           kernel->args[argID].type = GBE_ARG_GLOBAL_PTR;
           kernel->args[argID].size = sizeof(void*);
           kernel->args[argID].bti = arg.bti;
-          btiRegMap.insert(std::make_pair(arg.bti, arg.reg));
           break;
         case ir::FunctionArgument::CONSTANT_POINTER:
           kernel->args[argID].type = GBE_ARG_CONSTANT_PTR;
diff --git a/backend/src/backend/program.cpp b/backend/src/backend/program.cpp
index 5edb544..553c907 100644
--- a/backend/src/backend/program.cpp
+++ b/backend/src/backend/program.cpp
@@ -1105,6 +1105,18 @@ namespace gbe {
     return kernel->dupPrintfSet();
   }
 
+  static uint8_t kernelGetPrintfBufBTI(void * printf_info) {
+    if (printf_info == NULL) return 0;
+    const ir::PrintfSet *ps = (ir::PrintfSet *)printf_info;
+    return ps->getBufBTI();
+  }
+
+  static uint8_t kernelGetPrintfIndexBufBTI(void * printf_info) {
+    if (printf_info == NULL) return 0;
+    const ir::PrintfSet *ps = (ir::PrintfSet *)printf_info;
+    return ps->getIndexBufBTI();
+  }
+
   static void kernelReleasePrintfSet(void * printf_info) {
     if (printf_info == NULL) return;
     ir::PrintfSet *ps = (ir::PrintfSet *)printf_info;
@@ -1213,6 +1225,8 @@ GBE_EXPORT_SYMBOL gbe_set_image_base_index_cb *gbe_set_image_base_index = NULL;
 GBE_EXPORT_SYMBOL gbe_get_image_base_index_cb *gbe_get_image_base_index = NULL;
 GBE_EXPORT_SYMBOL gbe_get_printf_num_cb *gbe_get_printf_num = NULL;
 GBE_EXPORT_SYMBOL gbe_dup_printfset_cb *gbe_dup_printfset = NULL;
+GBE_EXPORT_SYMBOL gbe_get_printf_buf_bti_cb *gbe_get_printf_buf_bti = NULL;
+GBE_EXPORT_SYMBOL gbe_get_printf_indexbuf_bti_cb *gbe_get_printf_indexbuf_bti = NULL;
 GBE_EXPORT_SYMBOL gbe_release_printf_info_cb *gbe_release_printf_info = NULL;
 GBE_EXPORT_SYMBOL gbe_get_printf_sizeof_size_cb *gbe_get_printf_sizeof_size = NULL;
 GBE_EXPORT_SYMBOL gbe_output_printf_cb *gbe_output_printf = NULL;
@@ -1259,6 +1273,8 @@ namespace gbe
       gbe_get_image_base_index = gbe::getImageBaseIndex;
       gbe_set_image_base_index = gbe::setImageBaseIndex;
       gbe_get_printf_num = gbe::kernelGetPrintfNum;
+      gbe_get_printf_buf_bti = gbe::kernelGetPrintfBufBTI;
+      gbe_get_printf_indexbuf_bti = gbe::kernelGetPrintfIndexBufBTI;
       gbe_dup_printfset = gbe::kernelDupPrintfSet;
       gbe_get_printf_sizeof_size = gbe::kernelGetPrintfSizeOfSize;
       gbe_release_printf_info = gbe::kernelReleasePrintfSet;
diff --git a/backend/src/backend/program.h b/backend/src/backend/program.h
index 330a3de..39ff402 100644
--- a/backend/src/backend/program.h
+++ b/backend/src/backend/program.h
@@ -133,6 +133,13 @@ extern gbe_kernel_get_image_data_cb *gbe_kernel_get_image_data;
 typedef uint32_t (gbe_get_printf_num_cb)(void* printf_info);
 extern gbe_get_printf_num_cb *gbe_get_printf_num;
 
+/*! Get the printf buffer bti */
+typedef uint8_t (gbe_get_printf_buf_bti_cb)(void* printf_info);
+extern gbe_get_printf_buf_bti_cb *gbe_get_printf_buf_bti;
+
+typedef uint8_t (gbe_get_printf_indexbuf_bti_cb)(void* printf_info);
+extern gbe_get_printf_indexbuf_bti_cb *gbe_get_printf_indexbuf_bti;
+
 /*! Release the printfset */
 typedef void (gbe_release_printf_info_cb)(void* printf_info);
 extern gbe_release_printf_info_cb *gbe_release_printf_info;
diff --git a/backend/src/backend/program.hpp b/backend/src/backend/program.hpp
index a6303b9..8f5f125 100644
--- a/backend/src/backend/program.hpp
+++ b/backend/src/backend/program.hpp
@@ -145,6 +145,15 @@ namespace gbe {
       void* ptr = printfSet ? (void *)(new ir::PrintfSet(*printfSet)) : NULL;
       return ptr;
     }
+    uint8_t getPrintfBufBTI() const {
+      GBE_ASSERT(printfSet);
+      return printfSet->getBufBTI();
+    }
+
+    uint8_t getPrintfIndexBufBTI() const {
+      GBE_ASSERT(printfSet);
+      return printfSet->getIndexBufBTI();
+    }
 
     void outputPrintf(void* index_addr, void* buf_addr, size_t global_wk_sz0,
                       size_t global_wk_sz1, size_t global_wk_sz2) {
diff --git a/backend/src/gbe_bin_interpreter.cpp b/backend/src/gbe_bin_interpreter.cpp
index 2f02b34..ffca1f5 100644
--- a/backend/src/gbe_bin_interpreter.cpp
+++ b/backend/src/gbe_bin_interpreter.cpp
@@ -66,6 +66,8 @@ struct BinInterpCallBackInitializer
     gbe_get_image_base_index = gbe::getImageBaseIndex;
     gbe_set_image_base_index = gbe::setImageBaseIndex;
     gbe_get_printf_num = gbe::kernelGetPrintfNum;
+    gbe_get_printf_buf_bti = gbe::kernelGetPrintfBufBTI;
+    gbe_get_printf_indexbuf_bti = gbe::kernelGetPrintfIndexBufBTI;
     gbe_dup_printfset = gbe::kernelDupPrintfSet;
     gbe_get_printf_sizeof_size = gbe::kernelGetPrintfSizeOfSize;
     gbe_release_printf_info = gbe::kernelReleasePrintfSet;
diff --git a/backend/src/ir/context.hpp b/backend/src/ir/context.hpp
index d426aee..cd09413 100644
--- a/backend/src/ir/context.hpp
+++ b/backend/src/ir/context.hpp
@@ -206,6 +206,7 @@ namespace ir {
       GBE_ASSERT(valueNum > 0);
       this->STORE(type, index, offset, space, valueNum, dwAligned, bti);
     }
+    void appendSurface(uint8_t bti, Register reg) { fn->appendSurface(bti, reg); }
 
   protected:
     /*! A block must be started with a label */
diff --git a/backend/src/ir/function.cpp b/backend/src/ir/function.cpp
index 988ea2b..85e7934 100644
--- a/backend/src/ir/function.cpp
+++ b/backend/src/ir/function.cpp
@@ -222,6 +222,15 @@ namespace ir {
     const uint32_t specialNum = this->getSpecialRegNum();
     return ID >= firstID && ID < firstID + specialNum;
   }
+  Register Function::getSurfaceBaseReg(uint8_t bti) const {
+    map<uint8_t, Register>::const_iterator iter = btiRegMap.find(bti);
+    GBE_ASSERT(iter != btiRegMap.end());
+    return iter->second;
+  }
+
+  void Function::appendSurface(uint8_t bti, Register reg) {
+    btiRegMap.insert(std::make_pair(bti, reg));
+  }
 
   void Function::computeCFG(void) {
     // Clear possible previously computed CFG and compute the direct
diff --git a/backend/src/ir/function.hpp b/backend/src/ir/function.hpp
index 0f84aa0..2f90c2f 100644
--- a/backend/src/ir/function.hpp
+++ b/backend/src/ir/function.hpp
@@ -355,6 +355,9 @@ namespace ir {
     /*! add the loop info for later liveness analysis */
     void addLoop(const vector<LabelIndex> &bbs, const vector<std::pair<LabelIndex, LabelIndex>> &exits);
     INLINE const vector<Loop * > &getLoops() { return loops; }
+    /*! Get surface starting address register from bti */
+    Register getSurfaceBaseReg(uint8_t bti) const;
+    void appendSurface(uint8_t bti, Register reg);
   private:
     friend class Context;           //!< Can freely modify a function
     std::string name;               //!< Function name
@@ -365,6 +368,7 @@ namespace ir {
     vector<Immediate> immediates;   //!< All immediate values in the function
     vector<BasicBlock*> blocks;     //!< All chained basic blocks
     vector<Loop *> loops;           //!< Loops info of the function
+    map<uint8_t, Register> btiRegMap;//!< map bti to surface base address
     RegisterFile file;              //!< RegisterDatas used by the instructions
     Profile profile;                //!< Current function profile
     PushMap pushMap;                //!< Pushed function arguments (reg->loc)
diff --git a/backend/src/ir/printf.hpp b/backend/src/ir/printf.hpp
index 1aef767..4db7245 100644
--- a/backend/src/ir/printf.hpp
+++ b/backend/src/ir/printf.hpp
@@ -155,6 +155,8 @@ namespace gbe
         }
 
         sizeOfSize = other.sizeOfSize;
+        btiBuf = other.btiBuf;
+        btiIndexBuf = other.btiIndexBuf;
       }
 
       PrintfSet(void) = default;
@@ -180,6 +182,11 @@ namespace gbe
         return sizeOfSize;
       }
 
+      void setBufBTI(uint8_t b)      { btiBuf = b; }
+      void setIndexBufBTI(uint8_t b) { btiIndexBuf = b; }
+      uint8_t getBufBTI() const      { return btiBuf; }
+      uint8_t getIndexBufBTI() const { return btiIndexBuf; }
+
       uint32_t getPrintfBufferElementSize(uint32_t i) {
         PrintfSlot* slot = slots[i];
         int vec_num = 1;
@@ -226,6 +233,8 @@ namespace gbe
       vector<PrintfSlot*> slots;
       uint32_t sizeOfSize; // Total sizeof size.
       friend struct LockOutput;
+      uint8_t btiBuf;
+      uint8_t btiIndexBuf;
       static pthread_mutex_t lock;
       GBE_CLASS(PrintfSet);
     };
diff --git a/backend/src/llvm/llvm_gen_backend.cpp b/backend/src/llvm/llvm_gen_backend.cpp
index ed19054..39ae384 100644
--- a/backend/src/llvm/llvm_gen_backend.cpp
+++ b/backend/src/llvm/llvm_gen_backend.cpp
@@ -1287,6 +1287,7 @@ namespace gbe
         argNameNode = attrNode;
       }
     }
+    ctx.appendSurface(1, ir::ocl::stackbuffer);
 
     ctx.getFunction().setCompileWorkGroupSize(reqd_wg_sz[0], reqd_wg_sz[1], reqd_wg_sz[2]);
     // Loop over the arguments and output registers for them
@@ -1361,6 +1362,7 @@ namespace gbe
               switch (addrSpace) {
               case ir::MEM_GLOBAL:
                 globalPointer.insert(std::make_pair(I, btiBase));
+                ctx.appendSurface(btiBase, reg);
                 ctx.input(argName, ir::FunctionArgument::GLOBAL_POINTER, reg, llvmInfo, ptrSize, align, btiBase);
                 btiBase++;
               break;
@@ -1716,18 +1718,16 @@ namespace gbe
         GBE_ASSERT(con.getName() == v.getName());
         ctx.LOADI(ir::TYPE_S32, reg, ctx.newIntegerImmediate(con.getOffset(), ir::TYPE_S32));
       } else {
-	if(v.getName().str().substr(0, 4) == ".str") {
-          /* When there are multi printf statements in multi kernel fucntions within the same
-             translate unit, if they have the same sting parameter, such as
-             kernel_func1 () {
-               printf("Line is %d\n", line_num1);
-             }
-             kernel_func2 () {
-               printf("Line is %d\n", line_num2);
-             }
-             The Clang will just generate one global string named .strXXX to represent "Line is %d\n"
-             So when translating the kernel_func1, we can not unref that global var, so we will
-             get here. Just ignore it to avoid assert. */
+        if(v.getName().equals(StringRef("__gen_ocl_printf_buf"))) {
+          ctx.appendSurface(btiBase, ir::ocl::printfbptr);
+          ctx.getFunction().getPrintfSet()->setBufBTI(btiBase);
+          globalPointer.insert(std::make_pair(&v, btiBase++));
+          regTranslator.newScalarProxy(ir::ocl::printfbptr, const_cast<GlobalVariable*>(&v));
+        } else if(v.getName().equals(StringRef("__gen_ocl_printf_index_buf"))) {
+          ctx.appendSurface(btiBase, ir::ocl::printfiptr);
+          ctx.getFunction().getPrintfSet()->setIndexBufBTI(btiBase);
+          globalPointer.insert(std::make_pair(&v, btiBase++));
+          regTranslator.newScalarProxy(ir::ocl::printfiptr, const_cast<GlobalVariable*>(&v));
         } else {
           GBE_ASSERT(0);
         }
@@ -3367,7 +3367,7 @@ handle_write_image:
       bool isPrivate = false;
       p = candidates[idx];
 
-      while (dyn_cast<User>(p)) {
+      while (dyn_cast<User>(p) && !dyn_cast<GlobalVariable>(p)) {
 
         if (processed.find(p) == processed.end()) {
           processed.insert(p);
diff --git a/backend/src/llvm/llvm_printf_parser.cpp b/backend/src/llvm/llvm_printf_parser.cpp
index 00aa4a3..00e1ef8 100644
--- a/backend/src/llvm/llvm_printf_parser.cpp
+++ b/backend/src/llvm/llvm_printf_parser.cpp
@@ -467,7 +467,7 @@ error:
     // (index_offset + offset)* sizeof(int)
     op0 = builder->CreateMul(op0, ConstantInt::get(intTy, sizeof(int)));
     // Final index address = index_buf_ptr + (index_offset + offset)* sizeof(int)
-    op0 = builder->CreateAdd(op0, index_buf_ptr);
+    op0 = builder->CreateAdd(index_buf_ptr, op0);
     Value* index_addr = builder->CreateIntToPtr(op0, Type::getInt32PtrTy(module->getContext(), 1));
     builder->CreateStore(ConstantInt::get(intTy, 1), index_addr);// The flag
 
@@ -507,7 +507,7 @@ error:
       //offset * sizeof(specify)
       val = builder->CreateMul(offset, ConstantInt::get(intTy, sizeof_size));
       //data_offset + pbuf_ptr
-      op0 = builder->CreateAdd(op0, pbuf_ptr);
+      op0 = builder->CreateAdd(pbuf_ptr, op0);
       op0 = builder->CreateAdd(op0, val);
       data_addr = builder->CreateIntToPtr(op0, dst_type);
       builder->CreateStore(out_arg, data_addr);
@@ -575,15 +575,14 @@ error:
 
         if (!pbuf_ptr) {
           /* alloc a new buffer ptr to collect the print output. */
-          pbuf_ptr = builder->CreateCall(cast<llvm::Function>(module->getOrInsertFunction(
-                                           "__gen_ocl_printf_get_buf_addr", Type::getInt32Ty(module->getContext()),
-                                           NULL)));
+          Type *ptrTy = Type::getInt32PtrTy(module->getContext());
+          llvm::Constant * pBuf = module->getOrInsertGlobal(StringRef("__gen_ocl_printf_buf"), ptrTy);
+          pbuf_ptr = builder->CreatePtrToInt(pBuf, Type::getInt32Ty(module->getContext()));
         }
         if (!index_buf_ptr) {
-          /* alloc a new buffer ptr to collect the print valid index. */
-          index_buf_ptr = builder->CreateCall(cast<llvm::Function>(module->getOrInsertFunction(
-                                                "__gen_ocl_printf_get_index_buf_addr", Type::getInt32Ty(module->getContext()),
-                                                NULL)));
+          Type *ptrTy = Type::getInt32PtrTy(module->getContext());
+          llvm::Constant * pBuf = module->getOrInsertGlobal(StringRef("__gen_ocl_printf_index_buf"), ptrTy);
+          index_buf_ptr = builder->CreatePtrToInt(pBuf, Type::getInt32Ty(module->getContext()));
         }
 
         deadprintfs.push_back(PrintfInst(cast<Instruction>(call),parseOnePrintfInstruction(call)));
diff --git a/src/cl_command_queue_gen7.c b/src/cl_command_queue_gen7.c
index a4c8af7..330f0f9 100644
--- a/src/cl_command_queue_gen7.c
+++ b/src/cl_command_queue_gen7.c
@@ -264,7 +264,7 @@ cl_bind_printf(cl_gpgpu gpgpu, cl_kernel ker, void* printf_info, int printf_num,
   int32_t offset = interp_kernel_get_curbe_offset(ker->opaque, value, 0);
   size_t buf_size = global_sz * sizeof(int) * printf_num;
   if (offset > 0) {
-    if (cl_gpgpu_set_printf_buffer(gpgpu, 0, buf_size, offset) != 0)
+    if (cl_gpgpu_set_printf_buffer(gpgpu, 0, buf_size, offset, interp_get_printf_indexbuf_bti(printf_info)) != 0)
       return -1;
   }
 
@@ -272,7 +272,7 @@ cl_bind_printf(cl_gpgpu gpgpu, cl_kernel ker, void* printf_info, int printf_num,
   offset = interp_kernel_get_curbe_offset(ker->opaque, value, 0);
   buf_size = interp_get_printf_sizeof_size(printf_info) * global_sz;
   if (offset > 0) {
-    if (cl_gpgpu_set_printf_buffer(gpgpu, 1, buf_size, offset) != 0)
+    if (cl_gpgpu_set_printf_buffer(gpgpu, 1, buf_size, offset, interp_get_printf_buf_bti(printf_info)) != 0)
       return -1;
   }
   return 0;
diff --git a/src/cl_driver.h b/src/cl_driver.h
index 461c11e..9cdba98 100644
--- a/src/cl_driver.h
+++ b/src/cl_driver.h
@@ -226,7 +226,7 @@ typedef void (cl_gpgpu_unref_batch_buf_cb)(void*);
 extern cl_gpgpu_unref_batch_buf_cb *cl_gpgpu_unref_batch_buf;
 
 /* Set the printf buffer */
-typedef int (cl_gpgpu_set_printf_buffer_cb)(cl_gpgpu, uint32_t, uint32_t, uint32_t);
+typedef int (cl_gpgpu_set_printf_buffer_cb)(cl_gpgpu, uint32_t, uint32_t, uint32_t, uint8_t);
 extern cl_gpgpu_set_printf_buffer_cb *cl_gpgpu_set_printf_buffer;
 
 /* get the printf buffer offset in the apeture*/
diff --git a/src/cl_gbe_loader.cpp b/src/cl_gbe_loader.cpp
index 5f2f9ce..9d609c7 100644
--- a/src/cl_gbe_loader.cpp
+++ b/src/cl_gbe_loader.cpp
@@ -66,6 +66,8 @@ gbe_kernel_get_image_data_cb *interp_kernel_get_image_data = NULL;
 gbe_set_image_base_index_cb *interp_set_image_base_index = NULL;
 gbe_get_image_base_index_cb *interp_get_image_base_index = NULL;
 gbe_get_printf_num_cb* interp_get_printf_num = NULL;
+gbe_get_printf_buf_bti_cb* interp_get_printf_buf_bti = NULL;
+gbe_get_printf_indexbuf_bti_cb* interp_get_printf_indexbuf_bti = NULL;
 gbe_dup_printfset_cb* interp_dup_printfset = NULL;
 gbe_get_printf_sizeof_size_cb* interp_get_printf_sizeof_size = NULL;
 gbe_release_printf_info_cb* interp_release_printf_info = NULL;
@@ -220,6 +222,14 @@ struct GbeLoaderInitializer
     if (interp_get_printf_num == NULL)
       return false;
 
+    interp_get_printf_buf_bti = *(gbe_get_printf_buf_bti_cb**)dlsym(dlhInterp, "gbe_get_printf_buf_bti");
+    if (interp_get_printf_buf_bti == NULL)
+      return false;
+
+    interp_get_printf_indexbuf_bti = *(gbe_get_printf_indexbuf_bti_cb**)dlsym(dlhInterp, "gbe_get_printf_indexbuf_bti");
+    if (interp_get_printf_indexbuf_bti == NULL)
+      return false;
+
     interp_dup_printfset = *(gbe_dup_printfset_cb**)dlsym(dlhInterp, "gbe_dup_printfset");
     if (interp_dup_printfset == NULL)
       return false;
diff --git a/src/cl_gbe_loader.h b/src/cl_gbe_loader.h
index d095240..6cbc99e 100644
--- a/src/cl_gbe_loader.h
+++ b/src/cl_gbe_loader.h
@@ -66,6 +66,8 @@ extern gbe_kernel_get_image_data_cb *interp_kernel_get_image_data;
 extern gbe_set_image_base_index_cb *interp_set_image_base_index;
 extern gbe_get_image_base_index_cb *interp_get_image_base_index;
 extern gbe_get_printf_num_cb* interp_get_printf_num;
+extern gbe_get_printf_buf_bti_cb* interp_get_printf_buf_bti;
+extern gbe_get_printf_indexbuf_bti_cb* interp_get_printf_indexbuf_bti;
 extern gbe_dup_printfset_cb* interp_dup_printfset;
 extern gbe_get_printf_sizeof_size_cb* interp_get_printf_sizeof_size;
 extern gbe_release_printf_info_cb* interp_release_printf_info;
diff --git a/src/intel/intel_gpgpu.c b/src/intel/intel_gpgpu.c
index 1df0255..867ab4c 100644
--- a/src/intel/intel_gpgpu.c
+++ b/src/intel/intel_gpgpu.c
@@ -1353,7 +1353,7 @@ intel_gpgpu_event_get_exec_timestamp(intel_gpgpu_t* gpgpu, intel_event_t *event,
 }
 
 static int
-intel_gpgpu_set_printf_buf(intel_gpgpu_t *gpgpu, uint32_t i, uint32_t size, uint32_t offset)
+intel_gpgpu_set_printf_buf(intel_gpgpu_t *gpgpu, uint32_t i, uint32_t size, uint32_t offset, uint8_t bti)
 {
   drm_intel_bo *bo = NULL;
   if (i == 0) { // the index buffer.
@@ -1378,7 +1378,7 @@ intel_gpgpu_set_printf_buf(intel_gpgpu_t *gpgpu, uint32_t i, uint32_t size, uint
   }
   memset(bo->virtual, 0, size);
   drm_intel_bo_unmap(bo);
-  intel_gpgpu_bind_buf(gpgpu, bo, offset, 0, size, 0);
+  intel_gpgpu_bind_buf(gpgpu, bo, offset, 0, size, bti);
   return 0;
 }