Implement dot4 opcode

diff --git a/src/mesa/pipe/llvm/instructions.cpp b/src/mesa/pipe/llvm/instructions.cpp
index 31729d0f5861102a1d59c40d1ccab31519bc7525..9e3b989600b685a62e07cfb17d25ef2f711523bf 100644 (file)
--- a/src/mesa/pipe/llvm/instructions.cpp
+++ b/src/mesa/pipe/llvm/instructions.cpp
@@ -226,3 +226,30 @@ llvm::Value * Instructions::rcp(llvm::Value *in1)
    return vectorFromVals(res, res, res, res);
 }
 
+llvm::Value * Instructions::dp4(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *mulRes = mul(in1, in2);
+   ExtractElementInst *x = new ExtractElementInst(mulRes, unsigned(0),
+                                                  name("extractx"),
+                                                  m_block);
+   ExtractElementInst *y = new ExtractElementInst(mulRes, unsigned(1),
+                                                  name("extracty"),
+                                                  m_block);
+   ExtractElementInst *z = new ExtractElementInst(mulRes, unsigned(2),
+                                                  name("extractz"),
+                                                  m_block);
+   ExtractElementInst *w = new ExtractElementInst(mulRes, unsigned(3),
+                                                  name("extractw"),
+                                                  m_block);
+   BinaryOperator *xy = BinaryOperator::create(Instruction::Add, x, y,
+                                                name("xy"),
+                                                m_block);
+   BinaryOperator *xyz = BinaryOperator::create(Instruction::Add, xy, z,
+                                                name("xyz"),
+                                                m_block);
+   BinaryOperator *dot4 = BinaryOperator::create(Instruction::Add, xyz, w,
+                                                 name("dot4"),
+                                                 m_block);
+   return vectorFromVals(dot4, dot4, dot4, dot4);
+}
+

diff --git a/src/mesa/pipe/llvm/instructions.h b/src/mesa/pipe/llvm/instructions.h
index 18b5f91131ec0a11332dfa07cb55fd37e4ad3034..c35cdb75f01f8cceb1bce335f103081d5283c96e 100644 (file)
--- a/src/mesa/pipe/llvm/instructions.h
+++ b/src/mesa/pipe/llvm/instructions.h
@@ -16,6 +16,7 @@ public:
 
    llvm::Value *add(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 *lit(llvm::Value *in1);
    llvm::Value *madd(llvm::Value *in1, llvm::Value *in2,
                      llvm::Value *in2);

diff --git a/src/mesa/pipe/llvm/llvmtgsi.cpp b/src/mesa/pipe/llvm/llvmtgsi.cpp
index 45114abe4e32209ed088221c7ce3d6b370a391b9..45bc96cb1afdae2eb1df3b24c234511a430ed56d 100644 (file)
--- a/src/mesa/pipe/llvm/llvmtgsi.cpp
+++ b/src/mesa/pipe/llvm/llvmtgsi.cpp
@@ -205,7 +205,9 @@ translate_instruction(llvm::Module *module,
       out = instr->dp3(inputs[0], inputs[1]);
    }
       break;
-   case TGSI_OPCODE_DP4:
+   case TGSI_OPCODE_DP4: {
+      out = instr->dp4(inputs[0], inputs[1]);
+   }
       break;
    case TGSI_OPCODE_DST:
       break;
@@ -767,6 +769,10 @@ ga_llvm_from_tgsi(struct pipe_context *pipe, const struct tgsi_token *tokens)
    std::cout << "Creating llvm " <<std::endl;
    struct ga_llvm_prog *ga_llvm =
       (struct ga_llvm_prog *)malloc(sizeof(struct ga_llvm_prog));
+   fprintf(stderr, "DUMPX \n");
+   //tgsi_dump(tokens, TGSI_DUMP_VERBOSE);
+   tgsi_dump(tokens, 0);
+   fprintf(stderr, "DUMPEND \n");
    llvm::Module *mod = tgsi_to_llvm(tokens);
 
    /* Run optimization passes over it */
@@ -789,10 +795,6 @@ ga_llvm_from_tgsi(struct pipe_context *pipe, const struct tgsi_token *tokens)
       ee->addModuleProvider(mp);
    }
    ga_llvm->module = mod;
-   fprintf(stderr, "DUMPX \n");
-   //tgsi_dump(tokens, TGSI_DUMP_VERBOSE);
-   tgsi_dump(tokens, 0);
-   fprintf(stderr, "DUMPEND \n");
 
    Function *func = mod->getFunction("run_vertex_shader");
    std::cout << "run_vertex_shader  = "<<func<<std::endl;