Draw first triangle. Start on the llvm builder.

3 files changed, 237 insertions, 24 deletions

diff --git a/src/mesa/pipe/draw/draw_vertex_shader_llvm.c b/src/mesa/pipe/draw/draw_vertex_shader_llvm.c
index 97a0480b5c..2ff35ace24 100644
--- a/src/mesa/pipe/draw/draw_vertex_shader_llvm.c
+++ b/src/mesa/pipe/draw/draw_vertex_shader_llvm.c
@@ -98,6 +98,21 @@ void vertex_fetch(struct draw_context *draw,
    }
 }
 
+static INLINE unsigned
+compute_clipmask(const float *clip, const float (*plane)[4], unsigned nr)
+{
+   unsigned mask = 0;
+   unsigned i;
+
+   for (i = 0; i < nr; i++) {
+      if (dot4(clip, plane[i]) < 0) 
+         mask |= (1<<i);
+   }
+
+   return mask;
+}
+
+
 /**
  * Called by the draw module when the vertx cache needs to be flushed.
  * This involves running the vertex shader.
@@ -109,8 +124,10 @@ void draw_vertex_shader_queue_flush_llvm(struct draw_context *draw)
    struct vertex_header *dests[VS_QUEUE_LENGTH];
    float                 inputs[VS_QUEUE_LENGTH][PIPE_MAX_SHADER_INPUTS][4];
    float                 outputs[VS_QUEUE_LENGTH][PIPE_MAX_SHADER_INPUTS][4];
-   float (*consts)[4] = (float (*)[4]) draw->mapped_constants;
-   struct ga_llvm_prog *prog = draw->vertex_shader->state->llvm_prog;
+   float (*consts)[4]          = (float (*)[4]) draw->mapped_constants;
+   struct ga_llvm_prog  *prog  = (struct ga_llvm_prog *)draw->vertex_shader->state->llvm_prog;
+   const float          *scale = draw->viewport.scale;
+   const float          *trans = draw->viewport.translate;
 
    fprintf(stderr, "--- XX q(%d) \n", draw->vs.queue_nr);
 
@@ -129,5 +146,58 @@ void draw_vertex_shader_queue_flush_llvm(struct draw_context *draw)
    /* FIXME: finish conversion */
    /* dests = outputs */
 
+   /* store machine results */
+   for (int i = 0; i < draw->vs.queue_nr; ++i) {
+      unsigned slot;
+      float x, y, z, w;
+      struct vertex_header *vOut = draw->vs.queue[i].dest;
+      float (*dests)[4] = outputs[i];
+
+      /* Handle attr[0] (position) specially:
+       *
+       * XXX: Computing the clipmask should be done in the vertex
+       * program as a set of DP4 instructions appended to the
+       * user-provided code.
+       */
+      x = vOut->clip[0] = dests[0][0];
+      y = vOut->clip[1] = dests[0][1];
+      z = vOut->clip[2] = dests[0][2];
+      w = vOut->clip[3] = dests[0][3];
+      printf("output %d: %f %f %f %f\n", 0, x, y, z, w);
+
+      vOut->clipmask = compute_clipmask(vOut->clip, draw->plane, draw->nr_planes);
+      vOut->edgeflag = 1;
+
+      /* divide by w */
+      w = 1.0f / w;
+      x *= w;
+      y *= w;
+      z *= w;
+
+      /* Viewport mapping */
+      vOut->data[0][0] = x * scale[0] + trans[0];
+      vOut->data[0][1] = y * scale[1] + trans[1];
+      vOut->data[0][2] = z * scale[2] + trans[2];
+      vOut->data[0][3] = w;
+
+      /* Remaining attributes are packed into sequential post-transform
+       * vertex attrib slots.
+       * Skip 0 since we just did it above.
+       * Subtract two because of the VERTEX_HEADER, CLIP_POS attribs.
+       */
+      for (slot = 1; slot < draw->vertex_info.num_attribs - 2; slot++) {
+         vOut->data[slot][0] = dests[slot][0];
+         vOut->data[slot][1] = dests[slot][1];
+         vOut->data[slot][2] = dests[slot][2];
+         vOut->data[slot][3] = dests[slot][3];
+
+         printf("output %d: %f %f %f %f\n", slot,
+                vOut->data[slot][0],
+                vOut->data[slot][1],
+                vOut->data[slot][2],
+                vOut->data[slot][3]);
+      }
+   } /* loop over vertices */
+
    draw->vs.queue_nr = 0;
 }
diff --git a/src/mesa/pipe/llvm/llvmtgsi.cpp b/src/mesa/pipe/llvm/llvmtgsi.cpp
index 4ff6281c74..375e61c0db 100644
--- a/src/mesa/pipe/llvm/llvmtgsi.cpp
+++ b/src/mesa/pipe/llvm/llvmtgsi.cpp
@@ -32,6 +32,7 @@
 
 using namespace llvm;
 #include "llvm_base_shader.cpp"
+#include "tgsillvmbuilder.cpp"
 
 
 static inline void addPass(PassManager &PM, Pass *P) {
@@ -116,6 +117,7 @@ translate_immediate(llvm::Module *module,
 
 static void
 translate_instruction(llvm::Module *module,
+                      VertexShaderBuilder *builder,
                       struct tgsi_full_instruction *inst,
                       struct tgsi_full_instruction *fi)
 {
@@ -135,6 +137,7 @@ translate_instruction(llvm::Module *module,
    case TGSI_OPCODE_LOG:
       break;
    case TGSI_OPCODE_MUL:
+      
       break;
    case TGSI_OPCODE_ADD:
       break;
@@ -435,6 +438,7 @@ tgsi_to_llvm(const struct tgsi_token *tokens)
 
    tgsi_parse_init(&parse, tokens);
 
+   Function* func_printf = mod->getFunction("printf");
    //parse.FullHeader.Processor.Processor
 
    //parse.FullVersion.Version.MajorVersion
@@ -446,7 +450,7 @@ tgsi_to_llvm(const struct tgsi_token *tokens)
 
    fi = tgsi_default_full_instruction();
    fd = tgsi_default_full_declaration();
-
+   VertexShaderBuilder builder(label_entry, ptr_IN, ptr_CONST);
    while(!tgsi_parse_end_of_tokens(&parse)) {
       tgsi_parse_token(&parse);
 
@@ -465,7 +469,7 @@ tgsi_to_llvm(const struct tgsi_token *tokens)
          break;
 
       case TGSI_TOKEN_TYPE_INSTRUCTION:
-         translate_instruction(mod,
+         translate_instruction(mod, &builder,
                                &parse.FullToken.FullInstruction,
                                &fi);
          break;
@@ -475,29 +479,26 @@ tgsi_to_llvm(const struct tgsi_token *tokens)
       }
    }
 
-
-   Function* func_printf = mod->getFunction("printf");
-#if 1
-     
+#if 0
    // Type Definitions
    ArrayType* ArrayTy_0 = ArrayType::get(IntegerType::get(8), 19);
-  
+
    PointerType* PointerTy_1 = PointerType::get(ArrayTy_0);
-  
+
    VectorType* VectorTy_4 = VectorType::get(Type::FloatTy, 4);
-  
+
    PointerType* PointerTy_3 = PointerType::get(VectorTy_4);
-  
-  
+
+
    VectorType* VectorTy_5 = VectorType::get(IntegerType::get(32), 4);
-  
-  
+
+
    PointerType* PointerTy_8 = PointerType::get(IntegerType::get(8));
-  
-  
+
+
    // Global Variable Declarations
 
-  
+
    GlobalVariable* gvar_array__str = new GlobalVariable(
       /*Type=*/ArrayTy_0,
       /*isConstant=*/true,
@@ -505,7 +506,7 @@ tgsi_to_llvm(const struct tgsi_token *tokens)
       /*Initializer=*/0, // has initializer, specified below
       /*Name=*/".str",
       mod);
-  
+
    GlobalVariable* gvar_array__str1 = new GlobalVariable(
       /*Type=*/ArrayTy_0,
       /*isConstant=*/true,
@@ -513,7 +514,7 @@ tgsi_to_llvm(const struct tgsi_token *tokens)
       /*Initializer=*/0, // has initializer, specified below
       /*Name=*/".str1",
       mod);
-  
+
    GlobalVariable* gvar_array__str2 = new GlobalVariable(
       /*Type=*/ArrayTy_0,
       /*isConstant=*/true,
@@ -521,7 +522,7 @@ tgsi_to_llvm(const struct tgsi_token *tokens)
       /*Initializer=*/0, // has initializer, specified below
       /*Name=*/".str2",
       mod);
-  
+
    // Constant Definitions
    Constant* const_array_9 = ConstantArray::get("const %f %f %f %f\x0A", true);
    Constant* const_array_10 = ConstantArray::get("resul %f %f %f %f\x0A", true);
@@ -563,14 +564,14 @@ tgsi_to_llvm(const struct tgsi_token *tokens)
    const_ptr_24_indices.push_back(const_int32_21);
    const_ptr_24_indices.push_back(const_int32_21);
    Constant* const_ptr_24 = ConstantExpr::getGetElementPtr(gvar_array__str2, &const_ptr_24_indices[0], const_ptr_24_indices.size() );
-  
+
    // Global Variable Definitions
    gvar_array__str->setInitializer(const_array_9);
    gvar_array__str1->setInitializer(const_array_10);
    gvar_array__str2->setInitializer(const_array_11);
-  
+
    // Function Definitions
-  
+
    // Function: execute_shader (func_execute_shader)
    {
       // Block entry (label_entry)
@@ -725,5 +726,13 @@ int ga_llvm_prog_exec(struct ga_llvm_prog *prog,
    runner(inputs, dests, consts, count, num_attribs);
 
    std::cout << "---- END LLVM Execution "<<std::endl;
+
+   for (int i = 0; i < count; ++i) {
+      for (int j = 0; j < num_attribs; ++j) {
+         printf("OUT(%d, %d) [%f, %f, %f, %f]\n", i, j,
+                dests[i][j][0], dests[i][j][1],
+                dests[i][j][2], dests[i][j][3]);
+      }
+   }
    return 0;
 }
diff --git a/src/mesa/pipe/llvm/tgsillvmbuilder.cpp b/src/mesa/pipe/llvm/tgsillvmbuilder.cpp
new file mode 100644
index 0000000000..4088da11f8
--- /dev/null
+++ b/src/mesa/pipe/llvm/tgsillvmbuilder.cpp
@@ -0,0 +1,134 @@
+
+#include <map>
+
+class VertexShaderBuilder
+{
+   typedef std::map<int, llvm::LoadInst*> LoadMap;
+public:
+   VertexShaderBuilder(llvm::BasicBlock *block, llvm::Value *in, llvm::Value *consts);
+
+   llvm::ConstantInt *constantInt(int);
+   llvm::Constant *shuffleMask(int vec);
+   llvm::Value *inputElement(int idx);
+   llvm::Value *constElement(int idx);
+
+   llvm::Value *shuffleVector(llvm::Value *vec, int shuffle);
+
+
+private:
+   llvm::BasicBlock *m_block;
+   llvm::Value *m_IN;
+   llvm::Value *m_CONST;
+
+   std::map<int, llvm::ConstantInt*> m_constInts;
+   std::map<int, llvm::Constant*> m_intVecs;
+   LoadMap    m_inputs;
+   LoadMap    m_consts;
+
+   VectorType *m_floatVecType;
+   VectorType *m_intVecType;
+
+   Value      *m_undefFloatVec;
+   Value      *m_undefIntVec;
+
+   int         m_shuffleId;
+};
+
+VertexShaderBuilder::VertexShaderBuilder(llvm::BasicBlock *block, llvm::Value *in, llvm::Value *consts)
+   : m_block(block), m_IN(in), m_CONST(consts)
+{
+   m_floatVecType = VectorType::get(Type::FloatTy, 4);
+   m_intVecType   = VectorType::get(IntegerType::get(32), 4);
+
+   m_undefFloatVec = UndefValue::get(m_floatVecType);
+   m_undefIntVec   = UndefValue::get(m_intVecType);
+
+   m_shuffleId = 0;
+}
+
+//can only build vectors with all members in the [0, 9] range
+llvm::Constant *VertexShaderBuilder::shuffleMask(int vec)
+{
+   if (m_intVecs.find(vec) != m_intVecs.end()) {
+      return m_intVecs[vec];
+   }
+   int origVec = vec;
+   Constant* const_vec = 0;
+   if (origVec == 0) {
+      const_vec = Constant::getNullValue(m_intVecType);
+   } else {
+      int x = vec / 1000; vec -= x * 1000;
+      int y = vec / 100;  vec -= y * 100;
+      int z = vec / 10;   vec -= z * 10;
+      int w = vec;
+      std::vector<Constant*> elems;
+      elems.push_back(constantInt(x));
+      elems.push_back(constantInt(y));
+      elems.push_back(constantInt(z));
+      elems.push_back(constantInt(w));
+      const_vec = ConstantVector::get(m_intVecType, elems);
+   }
+
+   m_intVecs[origVec] = const_vec;
+   return const_vec;
+}
+
+llvm::ConstantInt *VertexShaderBuilder::constantInt(int idx)
+{
+   if (m_constInts.find(idx) != m_constInts.end()) {
+      return m_constInts[idx];
+   }
+   ConstantInt *const_int = ConstantInt::get(APInt(32,  idx));
+   m_constInts[idx] = const_int;
+   return const_int;
+}
+
+llvm::Value *VertexShaderBuilder::inputElement(int idx)
+{
+   if (m_inputs.find(idx) != m_inputs.end()) {
+      return m_inputs[idx];
+   }
+   char ptrName[13];
+   char name[9];
+   snprintf(ptrName, 13, "input_ptr%d", idx);
+   snprintf(name, 9, "input%d", idx);
+   GetElementPtrInst *getElem = new GetElementPtrInst(m_IN,
+                                                      constantInt(idx),
+                                                      ptrName,
+                                                      m_block);
+   LoadInst *load = new LoadInst(getElem, name,
+                                 false, m_block);
+   m_inputs[idx] = load;
+   return load;
+}
+
+llvm::Value *VertexShaderBuilder::constElement(int idx)
+{
+   if (m_consts.find(idx) != m_consts.end()) {
+      return m_consts[idx];
+   }
+   char ptrName[13];
+   char name[9];
+   snprintf(ptrName, 13, "const_ptr%d", idx);
+   snprintf(name, 9, "const%d", idx);
+   GetElementPtrInst *getElem = new GetElementPtrInst(m_CONST,
+                                                      constantInt(idx),
+                                                      ptrName,
+                                                      m_block);
+   LoadInst *load = new LoadInst(getElem, name,
+                                 false, m_block);
+   m_consts[idx] = load;
+   return load;
+}
+
+llvm::Value *VertexShaderBuilder::shuffleVector(llvm::Value *vec, int shuffle)
+{
+   Constant *mask = shuffleMask(shuffle);
+   ++m_shuffleId;
+   char name[11];
+   snprintf(name, 11, "shuffle%d", m_shuffleId);
+   ShuffleVectorInst *res =
+      new ShuffleVectorInst(vec, m_undefFloatVec, mask,
+                            name, m_block);
+   return res;
+}