diff options
author | Zack Rusin <[email protected]> | 2007-10-16 05:41:06 -0400 |
---|---|---|
committer | Zack Rusin <[email protected]> | 2007-10-24 11:21:03 -0400 |
commit | 5e0205023e8e6a08b0eb61286e15eb095f32ab3d (patch) | |
tree | 6261e444496113f5558a8bb041c57f10c8d0a2dd | |
parent | 5ffdada717466a78b5b148764ba23c7a95098887 (diff) |
Cleanup some of the testing code. Implement first pass at actually
running shaders in llvm.
-rw-r--r-- | src/mesa/pipe/draw/draw_vertex_shader.c | 1 | ||||
-rw-r--r-- | src/mesa/pipe/draw/draw_vertex_shader_llvm.c | 29 | ||||
-rw-r--r-- | src/mesa/pipe/llvm/llvm_builtins.c | 19 | ||||
-rw-r--r-- | src/mesa/pipe/llvm/llvmtgsi.cpp | 195 | ||||
-rw-r--r-- | src/mesa/pipe/llvm/llvmtgsi.h | 6 |
5 files changed, 208 insertions, 42 deletions
diff --git a/src/mesa/pipe/draw/draw_vertex_shader.c b/src/mesa/pipe/draw/draw_vertex_shader.c index 874259e727f..e36ecdc849e 100644 --- a/src/mesa/pipe/draw/draw_vertex_shader.c +++ b/src/mesa/pipe/draw/draw_vertex_shader.c @@ -132,6 +132,7 @@ run_vertex_program(struct draw_context *draw, y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j]; z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j]; w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j]; + printf("output %d: %f %f %f %f\n", 0, x, y, z, w); vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes); vOut[j]->edgeflag = 1; diff --git a/src/mesa/pipe/draw/draw_vertex_shader_llvm.c b/src/mesa/pipe/draw/draw_vertex_shader_llvm.c index 37315f75662..97a0480b5c2 100644 --- a/src/mesa/pipe/draw/draw_vertex_shader_llvm.c +++ b/src/mesa/pipe/draw/draw_vertex_shader_llvm.c @@ -98,11 +98,6 @@ void vertex_fetch(struct draw_context *draw, } } -void execute_shader(void *dests, void *inputs, void *consts) -{ - fprintf(stderr, "EXECUTING--\n"); -} - /** * Called by the draw module when the vertx cache needs to be flushed. * This involves running the vertex shader. @@ -118,28 +113,6 @@ void draw_vertex_shader_queue_flush_llvm(struct draw_context *draw) struct ga_llvm_prog *prog = draw->vertex_shader->state->llvm_prog; fprintf(stderr, "--- XX q(%d) \n", draw->vs.queue_nr); - /* Consts does not require 16 byte alignment. */ - fprintf(stderr, "0AAAAA = %f %f %f %f\n", - consts[0][0], - consts[0][1], - consts[0][2], - consts[0][3]); - - fprintf(stderr, "1AAAAA = %f %f %f %f\n", - consts[1][0], - consts[1][1], - consts[1][2], - consts[1][3]); - fprintf(stderr, "2AAAAA = %f %f %f %f\n", - consts[2][0], - consts[2][1], - consts[2][2], - consts[2][3]); - fprintf(stderr, "3AAAAA = %f %f %f %f\n", - consts[3][0], - consts[3][1], - consts[3][2], - consts[3][3]); /* fetch the inputs */ for (i = 0; i < draw->vs.queue_nr; ++i) { @@ -151,7 +124,7 @@ void draw_vertex_shader_queue_flush_llvm(struct draw_context *draw) /* batch execute the shaders on all the vertices */ ga_llvm_prog_exec(prog, inputs, outputs, consts, draw->vs.queue_nr, - draw->vertex_info.num_attribs); + draw->vertex_shader->state->num_inputs); /* FIXME: finish conversion */ /* dests = outputs */ diff --git a/src/mesa/pipe/llvm/llvm_builtins.c b/src/mesa/pipe/llvm/llvm_builtins.c index c7a9ea0d5a1..e135b4781d6 100644 --- a/src/mesa/pipe/llvm/llvm_builtins.c +++ b/src/mesa/pipe/llvm/llvm_builtins.c @@ -92,7 +92,7 @@ inline void collect_results(float4 *results, struct vertex_header *vOut, } #endif -void from_array(float4 (*res)[32], float (*ainputs)[32][4], +void from_array(float4 (*res)[16], float (*ainputs)[16][4], int count, int num_attribs) { for (int i = 0; i < count; ++i) { @@ -102,6 +102,7 @@ void from_array(float4 (*res)[32], float (*ainputs)[32][4], vec.y = ainputs[i][j][1]; vec.z = ainputs[i][j][2]; vec.w = ainputs[i][j][3]; + //printf("FAR(%d %d) %f %f %f %f\n", i, j, vec.x, vec.y, vec.z, vec.w); res[i][j] = vec; } } @@ -116,6 +117,7 @@ void from_consts(float4 *res, float (*ainputs)[4], vec.y = ainputs[i][1]; vec.z = ainputs[i][2]; vec.w = ainputs[i][3]; + //printf("VCONST %f %f %f %f\n", vec.x, vec.y, vec.z, vec.w); res[i] = vec; } } @@ -125,6 +127,7 @@ void to_array(float (*dests)[4], float4 *in, int num_attribs) for (int i = 0; i < num_attribs; ++i) { float *rd = dests[i]; float4 ri = in[i]; + printf("DEST = %f %f %f %f\n", ri.x, ri.y, ri.z, ri.w); rd[0] = ri.x; rd[1] = ri.y; rd[2] = ri.z; @@ -135,23 +138,23 @@ void to_array(float (*dests)[4], float4 *in, int num_attribs) extern void execute_shader(float4 *dests, float4 *inputs, float4 *consts); -void run_vertex_shader(float (*ainputs)[32][4], - float (*dests)[32][4], +void run_vertex_shader(float (*ainputs)[16][4], + float (*dests)[16][4], float (*aconsts)[4], int count, int num_attribs) { - float4 inputs[16*32*4][32]; - float4 consts[32]; - float4 results[16*32*4][32]; + float4 inputs[16*32*4][16]; + float4 consts[16]; + float4 results[16*32*4][16]; printf("XXXXXXXXXXX run_vertex_shader\n"); from_array(inputs, ainputs, count, num_attribs); - from_consts(consts, aconsts, 32); + from_consts(consts, aconsts, 5); for (int i = 0; i < count; ++i) { float4 *in = inputs[i]; float4 *res = results[i]; - to_array(dests[i], results[i], num_attribs); execute_shader(res, in, consts); + to_array(dests[i], res, num_attribs); } } diff --git a/src/mesa/pipe/llvm/llvmtgsi.cpp b/src/mesa/pipe/llvm/llvmtgsi.cpp index a616355f8fd..4ff6281c744 100644 --- a/src/mesa/pipe/llvm/llvmtgsi.cpp +++ b/src/mesa/pipe/llvm/llvmtgsi.cpp @@ -420,6 +420,19 @@ tgsi_to_llvm(const struct tgsi_token *tokens) struct tgsi_full_instruction fi; struct tgsi_full_declaration fd; + Function* shader = mod->getFunction("execute_shader"); + shader->setName("execute_shader_2"); + + Function::arg_iterator args = shader->arg_begin(); + Value *ptr_OUT = args++; + ptr_OUT->setName("OUT"); + Value *ptr_IN = args++; + ptr_IN->setName("IN"); + Value *ptr_CONST = args++; + ptr_CONST->setName("CONST"); + + BasicBlock *label_entry = new BasicBlock("entry", shader, 0); + tgsi_parse_init(&parse, tokens); //parse.FullHeader.Processor.Processor @@ -462,10 +475,184 @@ tgsi_to_llvm(const struct tgsi_token *tokens) } } + + Function* func_printf = mod->getFunction("printf"); +#if 1 + + // 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, + /*Linkage=*/GlobalValue::InternalLinkage, + /*Initializer=*/0, // has initializer, specified below + /*Name=*/".str", + mod); + + GlobalVariable* gvar_array__str1 = new GlobalVariable( + /*Type=*/ArrayTy_0, + /*isConstant=*/true, + /*Linkage=*/GlobalValue::InternalLinkage, + /*Initializer=*/0, // has initializer, specified below + /*Name=*/".str1", + mod); + + GlobalVariable* gvar_array__str2 = new GlobalVariable( + /*Type=*/ArrayTy_0, + /*isConstant=*/true, + /*Linkage=*/GlobalValue::InternalLinkage, + /*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); + Constant* const_array_11 = ConstantArray::get("outpu %f %f %f %f\x0A", true); + UndefValue* const_packed_12 = UndefValue::get(VectorTy_4); + Constant* const_packed_13 = Constant::getNullValue(VectorTy_5); + std::vector<Constant*> const_packed_14_elems; + ConstantInt* const_int32_15 = ConstantInt::get(APInt(32, "1", 10)); + const_packed_14_elems.push_back(const_int32_15); + const_packed_14_elems.push_back(const_int32_15); + const_packed_14_elems.push_back(const_int32_15); + const_packed_14_elems.push_back(const_int32_15); + Constant* const_packed_14 = ConstantVector::get(VectorTy_5, const_packed_14_elems); + std::vector<Constant*> const_packed_16_elems; + ConstantInt* const_int32_17 = ConstantInt::get(APInt(32, "2", 10)); + const_packed_16_elems.push_back(const_int32_17); + const_packed_16_elems.push_back(const_int32_17); + const_packed_16_elems.push_back(const_int32_17); + const_packed_16_elems.push_back(const_int32_17); + Constant* const_packed_16 = ConstantVector::get(VectorTy_5, const_packed_16_elems); + std::vector<Constant*> const_packed_18_elems; + ConstantInt* const_int32_19 = ConstantInt::get(APInt(32, "3", 10)); + const_packed_18_elems.push_back(const_int32_19); + const_packed_18_elems.push_back(const_int32_19); + const_packed_18_elems.push_back(const_int32_19); + const_packed_18_elems.push_back(const_int32_19); + Constant* const_packed_18 = ConstantVector::get(VectorTy_5, const_packed_18_elems); + std::vector<Constant*> const_ptr_20_indices; + Constant* const_int32_21 = Constant::getNullValue(IntegerType::get(32)); + const_ptr_20_indices.push_back(const_int32_21); + const_ptr_20_indices.push_back(const_int32_21); + Constant* const_ptr_20 = ConstantExpr::getGetElementPtr(gvar_array__str, &const_ptr_20_indices[0], const_ptr_20_indices.size() ); + UndefValue* const_double_22 = UndefValue::get(Type::DoubleTy); + std::vector<Constant*> const_ptr_23_indices; + const_ptr_23_indices.push_back(const_int32_21); + const_ptr_23_indices.push_back(const_int32_21); + Constant* const_ptr_23 = ConstantExpr::getGetElementPtr(gvar_array__str1, &const_ptr_23_indices[0], const_ptr_23_indices.size() ); + std::vector<Constant*> const_ptr_24_indices; + 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) + LoadInst* packed_tmp1 = new LoadInst(ptr_IN, "tmp1", false, label_entry); + ShuffleVectorInst* packed_tmp3 = new ShuffleVectorInst(packed_tmp1, const_packed_12, const_packed_13, "tmp3", label_entry); + LoadInst* packed_tmp6 = new LoadInst(ptr_CONST, "tmp6", false, label_entry); + BinaryOperator* packed_mul = BinaryOperator::create(Instruction::Mul, packed_tmp3, packed_tmp6, "mul", label_entry); + ShuffleVectorInst* packed_tmp8 = new ShuffleVectorInst(packed_tmp1, const_packed_12, const_packed_14, "tmp8", label_entry); + GetElementPtrInst* ptr_arrayidx10 = new GetElementPtrInst(ptr_CONST, const_int32_15, "arrayidx10", label_entry); + LoadInst* packed_tmp11 = new LoadInst(ptr_arrayidx10, "tmp11", false, label_entry); + BinaryOperator* packed_mul12 = BinaryOperator::create(Instruction::Mul, packed_tmp8, packed_tmp11, "mul12", label_entry); + BinaryOperator* packed_add = BinaryOperator::create(Instruction::Add, packed_mul12, packed_mul, "add", label_entry); + ShuffleVectorInst* packed_tmp15 = new ShuffleVectorInst(packed_tmp1, const_packed_12, const_packed_16, "tmp15", label_entry); + GetElementPtrInst* ptr_arrayidx17 = new GetElementPtrInst(ptr_CONST, const_int32_17, "arrayidx17", label_entry); + LoadInst* packed_tmp18 = new LoadInst(ptr_arrayidx17, "tmp18", false, label_entry); + BinaryOperator* packed_mul19 = BinaryOperator::create(Instruction::Mul, packed_tmp15, packed_tmp18, "mul19", label_entry); + BinaryOperator* packed_add21 = BinaryOperator::create(Instruction::Add, packed_mul19, packed_add, "add21", label_entry); + ShuffleVectorInst* packed_tmp25 = new ShuffleVectorInst(packed_tmp1, const_packed_12, const_packed_18, "tmp25", label_entry); + GetElementPtrInst* ptr_arrayidx27 = new GetElementPtrInst(ptr_CONST, const_int32_19, "arrayidx27", label_entry); + LoadInst* packed_tmp28 = new LoadInst(ptr_arrayidx27, "tmp28", false, label_entry); + BinaryOperator* packed_mul29 = BinaryOperator::create(Instruction::Mul, packed_tmp25, packed_tmp28, "mul29", label_entry); + BinaryOperator* packed_add31 = BinaryOperator::create(Instruction::Add, packed_mul29, packed_add21, "add31", label_entry); + StoreInst* void_25 = new StoreInst(packed_add31, ptr_OUT, false, label_entry); + GetElementPtrInst* ptr_arrayidx33 = new GetElementPtrInst(ptr_OUT, const_int32_15, "arrayidx33", label_entry); + GetElementPtrInst* ptr_arrayidx35 = new GetElementPtrInst(ptr_IN, const_int32_15, "arrayidx35", label_entry); + LoadInst* packed_tmp36 = new LoadInst(ptr_arrayidx35, "tmp36", false, label_entry); + StoreInst* void_26 = new StoreInst(packed_tmp36, ptr_arrayidx33, false, label_entry); + std::vector<Value*> int32_call_params; + int32_call_params.push_back(const_ptr_20); + int32_call_params.push_back(const_double_22); + int32_call_params.push_back(const_double_22); + int32_call_params.push_back(const_double_22); + int32_call_params.push_back(const_double_22); + //CallInst* int32_call = new CallInst(func_printf, int32_call_params.begin(), int32_call_params.end(), "call", label_entry); + //int32_call->setCallingConv(CallingConv::C); + //int32_call->setTailCall(true); + ExtractElementInst* float_tmp52 = new ExtractElementInst(packed_tmp1, const_int32_21, "tmp52", label_entry); + CastInst* double_conv53 = new FPExtInst(float_tmp52, Type::DoubleTy, "conv53", label_entry); + ExtractElementInst* float_tmp55 = new ExtractElementInst(packed_tmp1, const_int32_15, "tmp55", label_entry); + CastInst* double_conv56 = new FPExtInst(float_tmp55, Type::DoubleTy, "conv56", label_entry); + ExtractElementInst* float_tmp58 = new ExtractElementInst(packed_tmp1, const_int32_17, "tmp58", label_entry); + CastInst* double_conv59 = new FPExtInst(float_tmp58, Type::DoubleTy, "conv59", label_entry); + ExtractElementInst* float_tmp61 = new ExtractElementInst(packed_tmp1, const_int32_19, "tmp61", label_entry); + CastInst* double_conv62 = new FPExtInst(float_tmp61, Type::DoubleTy, "conv62", label_entry); + std::vector<Value*> int32_call63_params; + int32_call63_params.push_back(const_ptr_23); + int32_call63_params.push_back(double_conv53); + int32_call63_params.push_back(double_conv56); + int32_call63_params.push_back(double_conv59); + int32_call63_params.push_back(double_conv62); + //CallInst* int32_call63 = new CallInst(func_printf, int32_call63_params.begin(), int32_call63_params.end(), "call63", label_entry); + //int32_call63->setCallingConv(CallingConv::C); + //int32_call63->setTailCall(true); + ExtractElementInst* float_tmp65 = new ExtractElementInst(packed_add31, const_int32_21, "tmp65", label_entry); + CastInst* double_conv66 = new FPExtInst(float_tmp65, Type::DoubleTy, "conv66", label_entry); + ExtractElementInst* float_tmp68 = new ExtractElementInst(packed_add31, const_int32_15, "tmp68", label_entry); + CastInst* double_conv69 = new FPExtInst(float_tmp68, Type::DoubleTy, "conv69", label_entry); + ExtractElementInst* float_tmp71 = new ExtractElementInst(packed_add31, const_int32_17, "tmp71", label_entry); + CastInst* double_conv72 = new FPExtInst(float_tmp71, Type::DoubleTy, "conv72", label_entry); + ExtractElementInst* float_tmp74 = new ExtractElementInst(packed_add31, const_int32_19, "tmp74", label_entry); + CastInst* double_conv75 = new FPExtInst(float_tmp74, Type::DoubleTy, "conv75", label_entry); + std::vector<Value*> int32_call76_params; + int32_call76_params.push_back(const_ptr_24); + int32_call76_params.push_back(double_conv66); + int32_call76_params.push_back(double_conv69); + int32_call76_params.push_back(double_conv72); + int32_call76_params.push_back(double_conv75); + //CallInst* int32_call76 = new CallInst(func_printf, int32_call76_params.begin(), int32_call76_params.end(), "call76", label_entry); + //int32_call76->setCallingConv(CallingConv::C); + //int32_call76->setTailCall(true); + } +#endif + + new ReturnInst(label_entry); + //TXT("\ntgsi-dump end -------------------\n"); tgsi_parse_free(&parse); + std::cout<<"XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX"<<std::endl; + std::cout<<*mod<<std::endl; + std::cout<<"YYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYY"<<std::endl; return mod; } @@ -520,15 +707,15 @@ void ga_llvm_prog_delete(struct ga_llvm_prog *prog) free(prog); } -typedef void (*vertex_shader_runner)(float (*ainputs)[32][4], - float (*dests)[32][4], +typedef void (*vertex_shader_runner)(float (*ainputs)[PIPE_MAX_SHADER_INPUTS][4], + float (*dests)[PIPE_MAX_SHADER_INPUTS][4], float (*aconsts)[4], int count, int num_attribs); int ga_llvm_prog_exec(struct ga_llvm_prog *prog, - float (*inputs)[32][4], - float (*dests)[32][4], + float (*inputs)[PIPE_MAX_SHADER_INPUTS][4], + float (*dests)[PIPE_MAX_SHADER_INPUTS][4], float (*consts)[4], int count, int num_attribs) diff --git a/src/mesa/pipe/llvm/llvmtgsi.h b/src/mesa/pipe/llvm/llvmtgsi.h index f78fbce4c61..f957a69a72a 100644 --- a/src/mesa/pipe/llvm/llvmtgsi.h +++ b/src/mesa/pipe/llvm/llvmtgsi.h @@ -5,6 +5,8 @@ extern "C" { #endif +#include "pipe/p_state.h" + struct tgsi_exec_machine; struct tgsi_token; struct tgsi_sampler; @@ -21,8 +23,8 @@ ga_llvm_from_tgsi(struct pipe_context *pipe, const struct tgsi_token *tokens); void ga_llvm_prog_delete(struct ga_llvm_prog *prog); int ga_llvm_prog_exec(struct ga_llvm_prog *prog, - float (*inputs)[32][4], - float (*dests)[32][4], + float (*inputs)[PIPE_MAX_SHADER_INPUTS][4], + float (*dests)[PIPE_MAX_SHADER_INPUTS][4], float (*consts)[4], int count, int num_attribs); |