Implement arl, lerp opcodes and do a first stab at if/endif

handling and branching support.

5 files changed, 388 insertions, 135 deletions

diff --git a/src/mesa/pipe/llvm/instructions.cpp b/src/mesa/pipe/llvm/instructions.cpp
index 78d2658724..66c34f8101 100644
--- a/src/mesa/pipe/llvm/instructions.cpp
+++ b/src/mesa/pipe/llvm/instructions.cpp
@@ -369,6 +369,11 @@ llvm::Value * Instructions::floor(llvm::Value *in)
                          callFloor(z), callFloor(w));
 }
 
+llvm::Value * Instructions::arl(llvm::Value *in)
+{
+   return floor(in);
+}
+
 llvm::Value * Instructions::frc(llvm::Value *in)
 {
    llvm::Value *flr = floor(in);
@@ -600,6 +605,34 @@ llvm::Function * Instructions::declarePrintf()
   return func_printf;
 }
 
+
+llvm::Value * Instructions::sgt(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(Type::FloatTy, APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   ExtractElementInst *x1 = new ExtractElementInst(in1, unsigned(0), name("x1"), m_block);
+   ExtractElementInst *x2 = new ExtractElementInst(in2, unsigned(0), name("x2"), m_block);
+   FCmpInst *xcmp = new FCmpInst(FCmpInst::FCMP_OGT, x1, x2, name("xcmp"), m_block);
+   SelectInst *x = new SelectInst(xcmp, const1f, const0f, name("xsel"), m_block);
+
+   ExtractElementInst *y1 = new ExtractElementInst(in1, unsigned(1), name("y1"), m_block);
+   ExtractElementInst *y2 = new ExtractElementInst(in2, unsigned(1), name("y2"), m_block);
+   FCmpInst *ycmp = new FCmpInst(FCmpInst::FCMP_OGT, y1, y2, name("ycmp"), m_block);
+   SelectInst *y = new SelectInst(ycmp, const1f, const0f, name("ysel"), m_block);
+
+   ExtractElementInst *z1 = new ExtractElementInst(in1, unsigned(2), name("z1"), m_block);
+   ExtractElementInst *z2 = new ExtractElementInst(in2, unsigned(2), name("z2"), m_block);
+   FCmpInst *zcmp = new FCmpInst(FCmpInst::FCMP_OGT, z1, z2, name("zcmp"), m_block);
+   SelectInst *z = new SelectInst(zcmp, const1f, const0f, name("zsel"), m_block);
+
+   ExtractElementInst *w1 = new ExtractElementInst(in1, unsigned(3), name("w1"), m_block);
+   ExtractElementInst *w2 = new ExtractElementInst(in2, unsigned(3), name("w2"), m_block);
+   FCmpInst *wcmp = new FCmpInst(FCmpInst::FCMP_OGT, w1, w2, name("wcmp"), m_block);
+   SelectInst *w = new SelectInst(wcmp, const1f, const0f, name("wsel"), m_block);
+
+   return vectorFromVals(x, y, z, w);
+}
 llvm::Value * Instructions::sge(llvm::Value *in1, llvm::Value *in2)
 {
    Constant *const1f = ConstantFP::get(Type::FloatTy, APFloat(1.000000e+00f));
@@ -713,6 +746,8 @@ llvm::Value * Instructions::abs(llvm::Value *in)
 }
 
 /*
+FIXME: hand write the lit function. Currently it's
+generated from:
 typedef __attribute__(( ocu_vector_type(4) )) float float4;
 
 extern float powf(float a, float b);
@@ -744,7 +779,7 @@ clang --emit-llvm lit.c |llvm-as|opt -std-compile-opts|llvm2cpp -gen-contents -f
 */
 Function* makeLitFunction(Module *mod) {
 
-// Type Definitions
+   // Type Definitions
    std::vector<const Type*>FuncTy_0_args;
    FuncTy_0_args.push_back(Type::FloatTy);
    FuncTy_0_args.push_back(Type::FloatTy);
@@ -769,123 +804,172 @@ Function* makeLitFunction(Module *mod) {
       /*ParamAttrs=*/FuncTy_3_PAL);
 
 
-// Function Declarations
+   // Function Declarations
 
    Function* func_approx = new Function(
       /*Type=*/FuncTy_0,
       /*Linkage=*/GlobalValue::ExternalLinkage,
-      /*Name=*/"approx", mod); 
+      /*Name=*/"approx", mod);
    func_approx->setCallingConv(CallingConv::C);
 
-Function* func_powf = new Function(
-  /*Type=*/FuncTy_0,
-  /*Linkage=*/GlobalValue::ExternalLinkage,
-  /*Name=*/"powf", mod); // (external, no body)
-func_powf->setCallingConv(CallingConv::C);
-
-Function* func_lit = new Function(
-  /*Type=*/FuncTy_3,
-  /*Linkage=*/GlobalValue::ExternalLinkage,
-  /*Name=*/"lit", mod); 
-func_lit->setCallingConv(CallingConv::C);
-
-// Global Variable Declarations
-
-
-// Constant Definitions
-ConstantFP* const_float_4 = ConstantFP::get(Type::FloatTy, APFloat(-1.280000e+02f));
-ConstantFP* const_float_5 = ConstantFP::get(Type::FloatTy, APFloat(1.280000e+02f));
-Constant* const_float_6 = Constant::getNullValue(Type::FloatTy);
-Constant* const_int32_7 = Constant::getNullValue(IntegerType::get(32));
-std::vector<Constant*> const_packed_8_elems;
-ConstantFP* const_float_9 = ConstantFP::get(Type::FloatTy, APFloat(1.000000e+00f));
-const_packed_8_elems.push_back(const_float_9);
-UndefValue* const_float_10 = UndefValue::get(Type::FloatTy);
-const_packed_8_elems.push_back(const_float_10);
-const_packed_8_elems.push_back(const_float_10);
-const_packed_8_elems.push_back(const_float_9);
-Constant* const_packed_8 = ConstantVector::get(VectorTy_2, const_packed_8_elems);
-ConstantInt* const_int32_11 = ConstantInt::get(APInt(32,  "1", 10));
-ConstantInt* const_int32_12 = ConstantInt::get(APInt(32,  "3", 10));
-ConstantInt* const_int32_13 = ConstantInt::get(APInt(32,  "2", 10));
-std::vector<Constant*> const_packed_14_elems;
-const_packed_14_elems.push_back(const_float_9);
-const_packed_14_elems.push_back(const_float_6);
-const_packed_14_elems.push_back(const_float_6);
-const_packed_14_elems.push_back(const_float_9);
-Constant* const_packed_14 = ConstantVector::get(VectorTy_2, const_packed_14_elems);
-
-// Global Variable Definitions
-
-// Function Definitions
-
-// Function: approx (func_approx)
-{
-  Function::arg_iterator args = func_approx->arg_begin();
-  Value* float_a = args++;
-  float_a->setName("a");
-  Value* float_b = args++;
-  float_b->setName("b");
-  
-  BasicBlock* label_entry = new BasicBlock("entry",func_approx,0);
-  
-  // Block entry (label_entry)
-  FCmpInst* int1_cmp = new FCmpInst(FCmpInst::FCMP_OLT, float_b, const_float_4, "cmp", label_entry);
-  SelectInst* float_b_addr_0 = new SelectInst(int1_cmp, const_float_4, float_b, "b.addr.0", label_entry);
-  FCmpInst* int1_cmp3 = new FCmpInst(FCmpInst::FCMP_OGT, float_b_addr_0, const_float_5, "cmp3", label_entry);
-  SelectInst* float_b_addr_1 = new SelectInst(int1_cmp3, const_float_5, float_b_addr_0, "b.addr.1", label_entry);
-  FCmpInst* int1_cmp7 = new FCmpInst(FCmpInst::FCMP_OLT, float_a, const_float_6, "cmp7", label_entry);
-  SelectInst* float_a_addr_0 = new SelectInst(int1_cmp7, const_float_6, float_a, "a.addr.0", label_entry);
-  std::vector<Value*> float_call_params;
-  float_call_params.push_back(float_a_addr_0);
-  float_call_params.push_back(float_b_addr_1);
-  CallInst* float_call = new CallInst(func_powf, float_call_params.begin(), float_call_params.end(), "call", label_entry);
-  float_call->setCallingConv(CallingConv::C);
-  float_call->setTailCall(true);
-  new ReturnInst(float_call, label_entry);
-  
-}
-
-// Function: lit (func_lit)
-{
-  Function::arg_iterator args = func_lit->arg_begin();
-  Value* packed_tmp = args++;
-  packed_tmp->setName("tmp");
-  
-  BasicBlock* label_entry_16 = new BasicBlock("entry",func_lit,0);
-  BasicBlock* label_ifthen = new BasicBlock("ifthen",func_lit,0);
-  BasicBlock* label_UnifiedReturnBlock = new BasicBlock("UnifiedReturnBlock",func_lit,0);
-  
-  // Block entry (label_entry_16)
-  ExtractElementInst* float_tmp7 = new ExtractElementInst(packed_tmp, const_int32_7, "tmp7", label_entry_16);
-  FCmpInst* int1_cmp_17 = new FCmpInst(FCmpInst::FCMP_OGT, float_tmp7, const_float_6, "cmp", label_entry_16);
-  new BranchInst(label_ifthen, label_UnifiedReturnBlock, int1_cmp_17, label_entry_16);
-  
-  // Block ifthen (label_ifthen)
-  InsertElementInst* packed_tmp12 = new InsertElementInst(const_packed_8, float_tmp7, const_int32_11, "tmp12", label_ifthen);
-  ExtractElementInst* float_tmp14 = new ExtractElementInst(packed_tmp, const_int32_11, "tmp14", label_ifthen);
-  ExtractElementInst* float_tmp16 = new ExtractElementInst(packed_tmp, const_int32_12, "tmp16", label_ifthen);
-  FCmpInst* int1_cmp_i = new FCmpInst(FCmpInst::FCMP_OLT, float_tmp16, const_float_4, "cmp.i", label_ifthen);
-  SelectInst* float_b_addr_0_i = new SelectInst(int1_cmp_i, const_float_4, float_tmp16, "b.addr.0.i", label_ifthen);
-  FCmpInst* int1_cmp3_i = new FCmpInst(FCmpInst::FCMP_OGT, float_b_addr_0_i, const_float_5, "cmp3.i", label_ifthen);
-  SelectInst* float_b_addr_1_i = new SelectInst(int1_cmp3_i, const_float_5, float_b_addr_0_i, "b.addr.1.i", label_ifthen);
-  FCmpInst* int1_cmp7_i = new FCmpInst(FCmpInst::FCMP_OLT, float_tmp14, const_float_6, "cmp7.i", label_ifthen);
-  SelectInst* float_a_addr_0_i = new SelectInst(int1_cmp7_i, const_float_6, float_tmp14, "a.addr.0.i", label_ifthen);
-  std::vector<Value*> float_call_i_params;
-  float_call_i_params.push_back(float_a_addr_0_i);
-  float_call_i_params.push_back(float_b_addr_1_i);
-  CallInst* float_call_i = new CallInst(func_powf, float_call_i_params.begin(), float_call_i_params.end(), "call.i", label_ifthen);
-  float_call_i->setCallingConv(CallingConv::C);
-  float_call_i->setTailCall(true);
-  InsertElementInst* packed_tmp18 = new InsertElementInst(packed_tmp12, float_call_i, const_int32_13, "tmp18", label_ifthen);
-  new ReturnInst(packed_tmp18, label_ifthen);
-  
-  // Block UnifiedReturnBlock (label_UnifiedReturnBlock)
-  new ReturnInst(const_packed_14, label_UnifiedReturnBlock);
-  
-}
+   Function* func_powf = new Function(
+      /*Type=*/FuncTy_0,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Name=*/"powf", mod); // (external, no body)
+   func_powf->setCallingConv(CallingConv::C);
+
+   Function* func_lit = new Function(
+      /*Type=*/FuncTy_3,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Name=*/"lit", mod);
+   func_lit->setCallingConv(CallingConv::C);
+
+   // Global Variable Declarations
+
+
+   // Constant Definitions
+   ConstantFP* const_float_4 = ConstantFP::get(Type::FloatTy, APFloat(-1.280000e+02f));
+   ConstantFP* const_float_5 = ConstantFP::get(Type::FloatTy, APFloat(1.280000e+02f));
+   Constant* const_float_6 = Constant::getNullValue(Type::FloatTy);
+   Constant* const_int32_7 = Constant::getNullValue(IntegerType::get(32));
+   std::vector<Constant*> const_packed_8_elems;
+   ConstantFP* const_float_9 = ConstantFP::get(Type::FloatTy, APFloat(1.000000e+00f));
+   const_packed_8_elems.push_back(const_float_9);
+   UndefValue* const_float_10 = UndefValue::get(Type::FloatTy);
+   const_packed_8_elems.push_back(const_float_10);
+   const_packed_8_elems.push_back(const_float_10);
+   const_packed_8_elems.push_back(const_float_9);
+   Constant* const_packed_8 = ConstantVector::get(VectorTy_2, const_packed_8_elems);
+   ConstantInt* const_int32_11 = ConstantInt::get(APInt(32,  "1", 10));
+   ConstantInt* const_int32_12 = ConstantInt::get(APInt(32,  "3", 10));
+   ConstantInt* const_int32_13 = ConstantInt::get(APInt(32,  "2", 10));
+   std::vector<Constant*> const_packed_14_elems;
+   const_packed_14_elems.push_back(const_float_9);
+   const_packed_14_elems.push_back(const_float_6);
+   const_packed_14_elems.push_back(const_float_6);
+   const_packed_14_elems.push_back(const_float_9);
+   Constant* const_packed_14 = ConstantVector::get(VectorTy_2, const_packed_14_elems);
+
+   // Global Variable Definitions
+
+   // Function Definitions
+
+   // Function: approx (func_approx)
+   {
+      Function::arg_iterator args = func_approx->arg_begin();
+      Value* float_a = args++;
+      float_a->setName("a");
+      Value* float_b = args++;
+      float_b->setName("b");
+
+      BasicBlock* label_entry = new BasicBlock("entry",func_approx,0);
+
+      // Block entry (label_entry)
+      FCmpInst* int1_cmp = new FCmpInst(FCmpInst::FCMP_OLT, float_b, const_float_4, "cmp", label_entry);
+      SelectInst* float_b_addr_0 = new SelectInst(int1_cmp, const_float_4, float_b, "b.addr.0", label_entry);
+      FCmpInst* int1_cmp3 = new FCmpInst(FCmpInst::FCMP_OGT, float_b_addr_0, const_float_5, "cmp3", label_entry);
+      SelectInst* float_b_addr_1 = new SelectInst(int1_cmp3, const_float_5, float_b_addr_0, "b.addr.1", label_entry);
+      FCmpInst* int1_cmp7 = new FCmpInst(FCmpInst::FCMP_OLT, float_a, const_float_6, "cmp7", label_entry);
+      SelectInst* float_a_addr_0 = new SelectInst(int1_cmp7, const_float_6, float_a, "a.addr.0", label_entry);
+      std::vector<Value*> float_call_params;
+      float_call_params.push_back(float_a_addr_0);
+      float_call_params.push_back(float_b_addr_1);
+      CallInst* float_call = new CallInst(func_powf, float_call_params.begin(), float_call_params.end(), "call", label_entry);
+      float_call->setCallingConv(CallingConv::C);
+      float_call->setTailCall(true);
+      new ReturnInst(float_call, label_entry);
+   }
+
+   // Function: lit (func_lit)
+   {
+      Function::arg_iterator args = func_lit->arg_begin();
+      Value* packed_tmp = args++;
+      packed_tmp->setName("tmp");
+
+      BasicBlock* label_entry_16 = new BasicBlock("entry",func_lit,0);
+      BasicBlock* label_ifthen = new BasicBlock("ifthen",func_lit,0);
+      BasicBlock* label_UnifiedReturnBlock = new BasicBlock("UnifiedReturnBlock",func_lit,0);
+
+      // Block entry (label_entry_16)
+      ExtractElementInst* float_tmp7 = new ExtractElementInst(packed_tmp, const_int32_7, "tmp7", label_entry_16);
+      FCmpInst* int1_cmp_17 = new FCmpInst(FCmpInst::FCMP_OGT, float_tmp7, const_float_6, "cmp", label_entry_16);
+      new BranchInst(label_ifthen, label_UnifiedReturnBlock, int1_cmp_17, label_entry_16);
+
+      // Block ifthen (label_ifthen)
+      InsertElementInst* packed_tmp12 = new InsertElementInst(const_packed_8, float_tmp7, const_int32_11, "tmp12", label_ifthen);
+      ExtractElementInst* float_tmp14 = new ExtractElementInst(packed_tmp, const_int32_11, "tmp14", label_ifthen);
+      ExtractElementInst* float_tmp16 = new ExtractElementInst(packed_tmp, const_int32_12, "tmp16", label_ifthen);
+      FCmpInst* int1_cmp_i = new FCmpInst(FCmpInst::FCMP_OLT, float_tmp16, const_float_4, "cmp.i", label_ifthen);
+      SelectInst* float_b_addr_0_i = new SelectInst(int1_cmp_i, const_float_4, float_tmp16, "b.addr.0.i", label_ifthen);
+      FCmpInst* int1_cmp3_i = new FCmpInst(FCmpInst::FCMP_OGT, float_b_addr_0_i, const_float_5, "cmp3.i", label_ifthen);
+      SelectInst* float_b_addr_1_i = new SelectInst(int1_cmp3_i, const_float_5, float_b_addr_0_i, "b.addr.1.i", label_ifthen);
+      FCmpInst* int1_cmp7_i = new FCmpInst(FCmpInst::FCMP_OLT, float_tmp14, const_float_6, "cmp7.i", label_ifthen);
+      SelectInst* float_a_addr_0_i = new SelectInst(int1_cmp7_i, const_float_6, float_tmp14, "a.addr.0.i", label_ifthen);
+      std::vector<Value*> float_call_i_params;
+      float_call_i_params.push_back(float_a_addr_0_i);
+      float_call_i_params.push_back(float_b_addr_1_i);
+      CallInst* float_call_i = new CallInst(func_powf, float_call_i_params.begin(), float_call_i_params.end(), "call.i", label_ifthen);
+      float_call_i->setCallingConv(CallingConv::C);
+      float_call_i->setTailCall(true);
+      InsertElementInst* packed_tmp18 = new InsertElementInst(packed_tmp12, float_call_i, const_int32_13, "tmp18", label_ifthen);
+      new ReturnInst(packed_tmp18, label_ifthen);
+
+      // Block UnifiedReturnBlock (label_UnifiedReturnBlock)
+      new ReturnInst(const_packed_14, label_UnifiedReturnBlock);
+
+   }
    return func_lit;
 }
 
+void Instructions::ifop(llvm::Value *in)
+{
+   BasicBlock *ifthen = new BasicBlock(name("ifthen"), m_func,0);
+   BasicBlock *ifend = new BasicBlock(name("ifend"), m_func,0);
+
+   //BasicBlock *yblock = new BasicBlock(name("yblock"), m_func,0);
+   //BasicBlock *zblock = new BasicBlock(name("zblock"), m_func,0);
+   //BasicBlock *wblock = new BasicBlock(name("wblock"), m_func,0);
+
+   Constant *float0 = Constant::getNullValue(Type::FloatTy);
+
+   ExtractElementInst *x = new ExtractElementInst(in, unsigned(0), name("extractx"),
+                                                  m_block);
+   FCmpInst *xcmp = new FCmpInst(FCmpInst::FCMP_UNE, x, float0,
+                                 name("xcmp"), m_block);
+   new BranchInst(ifend, ifthen, xcmp, m_block);
+   //m_block = yblock;
+
+
+   m_block = ifthen;
+   m_ifStack.push(ifend);
+}
+
+llvm::BasicBlock * Instructions::currentBlock() const
+{
+   return m_block;
+}
 
+void Instructions::endif()
+{
+   assert(!m_ifStack.empty());
+   new BranchInst(m_ifStack.top(), m_block);
+   m_block = m_ifStack.top();
+   m_ifStack.pop();
+}
+
+llvm::Value * Instructions::lerp(llvm::Value *in1, llvm::Value *in2,
+                                 llvm::Value *in3)
+{
+   llvm::Value *m = mul(in1, in2);
+   llvm::Value *vec1 = vectorFromVals(ConstantFP::get(Type::FloatTy,
+                                                      APFloat(1.f)),
+                                      ConstantFP::get(Type::FloatTy,
+                                                      APFloat(1.f)),
+                                      ConstantFP::get(Type::FloatTy,
+                                                      APFloat(1.f)),
+                                      ConstantFP::get(Type::FloatTy,
+                                                      APFloat(1.f)));
+   llvm::Value *s = sub(vec1, in1);
+   return add(m, mul(s, in3));
+}
 
diff --git a/src/mesa/pipe/llvm/instructions.h b/src/mesa/pipe/llvm/instructions.h
index f7bc4a5839..ce2ce796d9 100644
--- a/src/mesa/pipe/llvm/instructions.h
+++ b/src/mesa/pipe/llvm/instructions.h
@@ -5,6 +5,8 @@
 #include <llvm/Module.h>
 #include <llvm/Value.h>
 
+#include <stack>
+
 namespace llvm {
    class VectorType;
    class Function;
@@ -15,16 +17,23 @@ class Instructions
 public:
    Instructions(llvm::Module *mod, llvm::Function *func, llvm::BasicBlock *block);
 
+   llvm::BasicBlock *currentBlock() const;
+
    llvm::Value *abs(llvm::Value *in1);
+   llvm::Value *arl(llvm::Value *in1);
    llvm::Value *add(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *cross(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *dp3(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *dp4(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *dph(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *dst(llvm::Value *in1, llvm::Value *in2);
+   void         endif();
    llvm::Value *ex2(llvm::Value *in);
    llvm::Value *floor(llvm::Value *in);
    llvm::Value *frc(llvm::Value *in);
+   void         ifop(llvm::Value *in);
+   llvm::Value *lerp(llvm::Value *in1, llvm::Value *in2,
+                     llvm::Value *in3);
    llvm::Value *lit(llvm::Value *in);
    llvm::Value *lg2(llvm::Value *in);
    llvm::Value *madd(llvm::Value *in1, llvm::Value *in2,
@@ -36,6 +45,7 @@ public:
    llvm::Value *rcp(llvm::Value *in);
    llvm::Value *rsq(llvm::Value *in);
    llvm::Value *sge(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sgt(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *slt(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *sub(llvm::Value *in1, llvm::Value *in2);
 
@@ -70,6 +80,8 @@ private:
    llvm::Function   *m_llvmLit;
 
    llvm::Constant   *m_fmtPtr;
+
+   std::stack<llvm::BasicBlock*> m_ifStack;
 };
 
 #endif
diff --git a/src/mesa/pipe/llvm/llvmtgsi.cpp b/src/mesa/pipe/llvm/llvmtgsi.cpp
index 8983f16297..ade4573fb8 100644
--- a/src/mesa/pipe/llvm/llvmtgsi.cpp
+++ b/src/mesa/pipe/llvm/llvmtgsi.cpp
@@ -146,10 +146,16 @@ translate_instruction(llvm::Module *module,
    for (int i = 0; i < inst->Instruction.NumSrcRegs; ++i) {
       struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
       llvm::Value *val = 0;
+      llvm::Value *indIdx = 0;
+
+      if (src->SrcRegister.Indirect) {
+         indIdx = storage->addrElement(src->SrcRegisterInd.Index);
+         indIdx = storage->extractIndex(indIdx);
+      }
       if (src->SrcRegister.File == TGSI_FILE_CONSTANT) {
-         val = storage->constElement(src->SrcRegister.Index);
+         val = storage->constElement(src->SrcRegister.Index, indIdx);
       } else if (src->SrcRegister.File == TGSI_FILE_INPUT) {
-         val = storage->inputElement(src->SrcRegister.Index);
+         val = storage->inputElement(src->SrcRegister.Index, indIdx);
       } else if (src->SrcRegister.File == TGSI_FILE_TEMPORARY) {
          val = storage->tempElement(src->SrcRegister.Index);
       } else {
@@ -204,7 +210,9 @@ translate_instruction(llvm::Module *module,
    instr->printVector(inputs[1]);*/
    llvm::Value *out = 0;
    switch (inst->Instruction.Opcode) {
-   case TGSI_OPCODE_ARL:
+   case TGSI_OPCODE_ARL: {
+      out = instr->arl(inputs[0]);
+   }
       break;
    case TGSI_OPCODE_MOV: {
       out = inputs[0];
@@ -270,7 +278,9 @@ translate_instruction(llvm::Module *module,
       out = instr->sub(inputs[0], inputs[1]);
    }
       break;
-   case TGSI_OPCODE_LERP:
+   case TGSI_OPCODE_LERP: {
+      out = instr->lerp(inputs[0], inputs[1], inputs[2]);
+   }
       break;
    case TGSI_OPCODE_CND:
       break;
@@ -344,7 +354,9 @@ translate_instruction(llvm::Module *module,
       break;
    case TGSI_OPCODE_SFL:
       break;
-   case TGSI_OPCODE_SGT:
+   case TGSI_OPCODE_SGT: {
+      out = instr->sgt(inputs[0], inputs[1]);
+   }
       break;
    case TGSI_OPCODE_SIN:
       break;
@@ -398,7 +410,11 @@ translate_instruction(llvm::Module *module,
       break;
    case TGSI_OPCODE_BRK:
       break;
-   case TGSI_OPCODE_IF:
+   case TGSI_OPCODE_IF: {
+      instr->ifop(inputs[0]);
+      storage->setCurrentBlock(instr->currentBlock());
+      return;  //just update the state
+   }
       break;
    case TGSI_OPCODE_LOOP:
       break;
@@ -406,7 +422,12 @@ translate_instruction(llvm::Module *module,
       break;
    case TGSI_OPCODE_ELSE:
       break;
-   case TGSI_OPCODE_ENDIF:
+   case TGSI_OPCODE_ENDIF: {
+      instr->endif();
+      storage->setCurrentBlock(instr->currentBlock());
+      storage->popPhiNode();
+      return; //just update the state
+   }
       break;
    case TGSI_OPCODE_ENDLOOP:
       break;
@@ -554,8 +575,11 @@ translate_instruction(llvm::Module *module,
          storage->store(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
       } else if (dst->DstRegister.File == TGSI_FILE_TEMPORARY) {
          storage->setTempElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      } else if (dst->DstRegister.File == TGSI_FILE_ADDRESS) {
+         storage->setAddrElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
       } else {
          fprintf(stderr, "ERROR: unsupported LLVM destination!");
+         assert(!"wrong destination");
       }
    }
 }
@@ -617,7 +641,7 @@ tgsi_to_llvm(struct ga_llvm_prog *prog, const struct tgsi_token *tokens)
       }
    }
 
-   new ReturnInst(label_entry);
+   new ReturnInst(instr.currentBlock());
 
    tgsi_parse_free(&parse);
 
@@ -635,7 +659,8 @@ ga_llvm_from_tgsi(struct pipe_context *pipe, const struct tgsi_token *tokens)
    ga_llvm->id = GLOBAL_ID;
    tgsi_dump(tokens, 0);
    llvm::Module *mod = tgsi_to_llvm(ga_llvm, tokens);
-
+   ga_llvm->module = mod;
+   ga_llvm_prog_dump(ga_llvm, 0);
    /* Run optimization passes over it */
    PassManager passes;
    passes.add(new TargetData(mod));
@@ -715,8 +740,14 @@ void ga_llvm_prog_dump(struct ga_llvm_prog *prog, const char *file_prefix)
       out << (*mod);
       out.close();
    } else {
+      std::ostringstream stream;
+      stream << "execute_shader";
+      stream << prog->id;
+      std::string func_name = stream.str();
+      llvm::Function *func = mod->getFunction(func_name.c_str());
+      assert(func);
       std::cout<<"; ---------- Start shader "<<prog->id<<std::endl;
-      std::cout<<*mod<<std::endl;
+      std::cout<<*func<<std::endl;
       std::cout<<"; ---------- End shader "<<prog->id<<std::endl;
    }
 }
diff --git a/src/mesa/pipe/llvm/storage.cpp b/src/mesa/pipe/llvm/storage.cpp
index 3bf37c7567..5b1f05190e 100644
--- a/src/mesa/pipe/llvm/storage.cpp
+++ b/src/mesa/pipe/llvm/storage.cpp
@@ -17,7 +17,8 @@ Storage::Storage(llvm::BasicBlock *block, llvm::Value *out,
                                          llvm::Value *in, llvm::Value *consts)
    : m_block(block), m_OUT(out),
      m_IN(in), m_CONST(consts),
-     m_temps(32), m_dstCache(32),
+     m_temps(32), m_addrs(32),
+     m_dstCache(32),
      m_idx(0)
 {
    m_floatVecType = VectorType::get(Type::FloatTy, 4);
@@ -88,15 +89,29 @@ llvm::ConstantInt *Storage::constantInt(int idx)
    return const_int;
 }
 
-llvm::Value *Storage::inputElement(int idx)
+llvm::Value *Storage::inputElement(int idx, llvm::Value *indIdx)
 {
-   if (m_inputs.find(idx) != m_inputs.end()) {
+   if (!indIdx && m_inputs.find(idx) != m_inputs.end()) {
       return m_inputs[idx];
    }
-   GetElementPtrInst *getElem = new GetElementPtrInst(m_IN,
-                                                      constantInt(idx),
-                                                      name("input_ptr"),
-                                                      m_block);
+   GetElementPtrInst *getElem = 0;
+
+   if (indIdx) {
+      getElem = new GetElementPtrInst(m_IN,
+                                      BinaryOperator::create(Instruction::Add,
+                                                             indIdx,
+                                                             constantInt(idx),
+                                                             name("add"),
+                                                             m_block),
+                                      name("input_ptr"),
+                                      m_block);
+   } else {
+      getElem = new GetElementPtrInst(m_IN,
+                                      constantInt(idx),
+                                      name("input_ptr"),
+                                      m_block);
+   }
+
    LoadInst *load = new LoadInst(getElem, name("input"),
                                  false, m_block);
    load->setAlignment(8);
@@ -105,16 +120,29 @@ llvm::Value *Storage::inputElement(int idx)
    return load;
 }
 
-llvm::Value *Storage::constElement(int idx)
+llvm::Value *Storage::constElement(int idx, llvm::Value *indIdx)
 {
    m_numConsts = ((idx + 1) > m_numConsts) ? (idx + 1) : m_numConsts;
-   if (m_consts.find(idx) != m_consts.end()) {
+   if (!indIdx && m_consts.find(idx) != m_consts.end()) {
       return m_consts[idx];
    }
-   GetElementPtrInst *getElem = new GetElementPtrInst(m_CONST,
-                                                      constantInt(idx),
-                                                      name("const_ptr"),
-                                                      m_block);
+
+   GetElementPtrInst *getElem = 0;
+
+   if (indIdx)
+      getElem = new GetElementPtrInst(m_CONST,
+                                      BinaryOperator::create(Instruction::Add,
+                                                             indIdx,
+                                                             constantInt(idx),
+                                                             name("add"),
+                                                             m_block),
+                                      name("const_ptr"),
+                                      m_block);
+   else
+      getElem = new GetElementPtrInst(m_CONST,
+                                      constantInt(idx),
+                                      name("const_ptr"),
+                                      m_block);
    LoadInst *load = new LoadInst(getElem, name("const"),
                                  false, m_block);
    load->setAlignment(8);
@@ -146,7 +174,17 @@ void Storage::setTempElement(int idx, llvm::Value *val, int mask)
       llvm::Value *templ = m_temps[idx];
       val = maskWrite(val, mask, templ);
    }
+   llvm::Value *templ = m_temps[idx];
+   if (templ) {
+      BasicBlock *block = m_varBlocks[templ];
+      if (block != m_block) {
+         addPhiNode(idx, val, m_block, templ, block);
+      } else
+         updatePhiNode(idx, val);
+   }
+
    m_temps[idx] = val;
+   m_varBlocks[val] = m_block;
 }
 
 void Storage::store(int dstIdx, llvm::Value *val, int mask)
@@ -208,3 +246,69 @@ int Storage::numConsts() const
 {
    return m_numConsts;
 }
+
+llvm::Value * Storage::addrElement(int idx) const
+{
+   Value *ret = m_addrs[idx];
+   if (!ret)
+      return m_undefFloatVec;
+   return ret;
+}
+
+void Storage::setAddrElement(int idx, llvm::Value *val, int mask)
+{
+   if (mask != TGSI_WRITEMASK_XYZW) {
+      llvm::Value *templ = m_addrs[idx];
+      val = maskWrite(val, mask, templ);
+   }
+   m_addrs[idx] = val;
+}
+
+llvm::Value * Storage::extractIndex(llvm::Value *vec)
+{
+   llvm::Value *x = new ExtractElementInst(vec, unsigned(0),
+                                           name("x"), m_block);
+   return new FPToSIInst(x, IntegerType::get(32), name("intidx"), m_block);
+}
+
+void Storage::setCurrentBlock(llvm::BasicBlock *block)
+{
+   m_block = block;
+}
+
+void Storage::addPhiNode(int idx, llvm::Value *val1, llvm::BasicBlock *blk1,
+                         llvm::Value *val2, llvm::BasicBlock *blk2)
+{
+   PhiNode node;
+   node.val1 = val1;
+   node.block1 = blk1;
+   node.val2 = val2;
+   node.block2 = blk2;
+   m_phiNodes[idx] = node;
+}
+
+void Storage::updatePhiNode(int idx, llvm::Value *val1)
+{
+   if (m_phiNodes.find(idx) == m_phiNodes.end())
+      return;
+   PhiNode node = m_phiNodes[idx];
+   node.val1 = val1;
+   m_phiNodes[idx] = node;
+}
+
+void Storage::popPhiNode()
+{
+   if (!m_phiNodes.empty()) {
+      std::map<int, PhiNode>::const_iterator itr;
+      for (itr = m_phiNodes.begin(); itr != m_phiNodes.end(); ++itr) {
+         PhiNode node = (*itr).second;
+         PHINode *dest = new PHINode(m_floatVecType,
+                                     name("phiDest"), m_block);
+         dest->reserveOperandSpace(2);
+         dest->addIncoming(node.val1, node.block1);
+         dest->addIncoming(node.val2, node.block2);
+         m_temps[(*itr).first] = dest;
+      }
+   }
+   m_phiNodes.clear();
+}
diff --git a/src/mesa/pipe/llvm/storage.h b/src/mesa/pipe/llvm/storage.h
index da7b0dbe01..919fbc9ded 100644
--- a/src/mesa/pipe/llvm/storage.h
+++ b/src/mesa/pipe/llvm/storage.h
@@ -21,19 +21,27 @@ public:
            llvm::Value *out,
            llvm::Value *in, llvm::Value *consts);
 
+   void setCurrentBlock(llvm::BasicBlock *block);
+
    llvm::ConstantInt *constantInt(int);
    llvm::Constant *shuffleMask(int vec);
-   llvm::Value *inputElement(int idx);
-   llvm::Value *constElement(int idx);
+   llvm::Value *inputElement(int idx, llvm::Value *indIdx =0);
+   llvm::Value *constElement(int idx, llvm::Value *indIdx =0);
 
    llvm::Value *tempElement(int idx) const;
    void setTempElement(int idx, llvm::Value *val, int mask);
 
+   llvm::Value *addrElement(int idx) const;
+   void setAddrElement(int idx, llvm::Value *val, int mask);
+
    llvm::Value *shuffleVector(llvm::Value *vec, int shuffle);
 
+   llvm::Value *extractIndex(llvm::Value *vec);
 
    void store(int dstIdx, llvm::Value *val, int mask);
 
+   void popPhiNode();
+
    int numConsts() const;
 private:
    llvm::Value *maskWrite(llvm::Value *src, int mask, llvm::Value *templ);
@@ -47,6 +55,7 @@ private:
    std::map<int, llvm::ConstantInt*> m_constInts;
    std::map<int, llvm::Constant*>    m_intVecs;
    std::vector<llvm::Value*>         m_temps;
+   std::vector<llvm::Value*>         m_addrs;
    std::vector<llvm::Value*>         m_dstCache;
    LoadMap                           m_inputs;
    LoadMap                           m_consts;
@@ -63,6 +72,19 @@ private:
    int         m_idx;
 
    int         m_numConsts;
+
+   void addPhiNode(int, llvm::Value*, llvm::BasicBlock*,
+                   llvm::Value*, llvm::BasicBlock*);
+   void updatePhiNode(int, llvm::Value*);
+   struct PhiNode {
+      llvm::Value      *val1;
+      llvm::BasicBlock *block1;
+      llvm::Value      *val2;
+      llvm::BasicBlock *block2;
+   };
+
+   std::map<llvm::Value*, llvm::BasicBlock*> m_varBlocks;
+   std::map<int, PhiNode> m_phiNodes;
 };
 
 #endif