summaryrefslogtreecommitdiff
path: root/src
diff options
context:
space:
mode:
Diffstat (limited to 'src')
-rw-r--r--src/gallium/auxiliary/gallivm/gallivm_cpu.cpp5
-rw-r--r--src/gallium/auxiliary/gallivm/instructionssoa.cpp5
-rw-r--r--src/gallium/auxiliary/gallivm/instructionssoa.h1
-rw-r--r--src/gallium/auxiliary/gallivm/storagesoa.cpp71
-rw-r--r--src/gallium/auxiliary/gallivm/storagesoa.h7
-rw-r--r--src/gallium/auxiliary/gallivm/tgsitollvm.cpp10
6 files changed, 78 insertions, 21 deletions
diff --git a/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
index 3a4a41e544..3a2f2878a3 100644
--- a/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
+++ b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
@@ -158,8 +158,8 @@ void gallivm_cpu_jit_compile(struct gallivm_cpu_engine *cpu, struct gallivm_prog
llvm::ExistingModuleProvider *mp = new llvm::ExistingModuleProvider(mod);
llvm::ExecutionEngine *ee = cpu->engine;
assert(ee);
- /*FIXME : remove */
- ee->DisableLazyCompilation();
+ /*FIXME : why was this disabled ? we need it for pow/sqrt/... */
+ ee->DisableLazyCompilation(false);
ee->addModuleProvider(mp);
llvm::Function *func = func_for_shader(prog);
@@ -202,7 +202,6 @@ int gallivm_cpu_vs_exec(struct gallivm_prog *prog,
unsigned int i, j;
unsigned slot;
vertex_shader_runner runner = reinterpret_cast<vertex_shader_runner>(prog->function);
-
assert(runner);
for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.cpp b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
index a658072551..1143ee0b0b 100644
--- a/src/gallium/auxiliary/gallivm/instructionssoa.cpp
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
@@ -171,6 +171,11 @@ std::vector<llvm::Value*> InstructionsSoa::extractVector(llvm::Value *vector)
return res;
}
+llvm::IRBuilder<>* InstructionsSoa::getIRBuilder()
+{
+ return &m_builder;
+}
+
void InstructionsSoa::createFunctionMap()
{
m_functionsMap[TGSI_OPCODE_ABS] = "abs";
diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.h b/src/gallium/auxiliary/gallivm/instructionssoa.h
index 3817fdc904..d6831e0a6b 100644
--- a/src/gallium/auxiliary/gallivm/instructionssoa.h
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.h
@@ -76,6 +76,7 @@ public:
void end();
std::vector<llvm::Value*> extractVector(llvm::Value *vector);
+ llvm::IRBuilder<>* getIRBuilder();
private:
const char * name(const char *prefix) const;
llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y,
diff --git a/src/gallium/auxiliary/gallivm/storagesoa.cpp b/src/gallium/auxiliary/gallivm/storagesoa.cpp
index 78d754371f..646b9d7ca0 100644
--- a/src/gallium/auxiliary/gallivm/storagesoa.cpp
+++ b/src/gallium/auxiliary/gallivm/storagesoa.cpp
@@ -91,19 +91,29 @@ void StorageSoa::declareImmediates()
for (unsigned int i = 0; i < m_immediatesToFlush.size(); ++i) {
std::vector<float> vec = m_immediatesToFlush[i];
std::vector<float> vals(4);
+ float val;
std::vector<Constant*> channelArray;
- vals[0] = vec[0]; vals[1] = vec[0]; vals[2] = vec[0]; vals[3] = vec[0];
- llvm::Constant *xChannel = createConstGlobalVector(vals);
+ val = vec[0];
+ llvm::Constant *xChannel = createConstGlobalFloat(val);
+ val = vec[1];
+ llvm::Constant *yChannel = createConstGlobalFloat(val);
+ val = vec[2];
+ llvm::Constant *zChannel = createConstGlobalFloat(val);
+ val = vec[3];
+ llvm::Constant *wChannel = createConstGlobalFloat(val);
- vals[0] = vec[1]; vals[1] = vec[1]; vals[2] = vec[1]; vals[3] = vec[1];
+// vals[0] = vec[0]; vals[1] = vec[1]; vals[2] = vec[2]; vals[3] = vec[3];
+// llvm::Constant *xChannel = createConstGlobalVector(vec[0]);
+
+/* vals[0] = vec[1]; vals[1] = vec[1]; vals[2] = vec[1]; vals[3] = vec[1];
llvm::Constant *yChannel = createConstGlobalVector(vals);
vals[0] = vec[2]; vals[1] = vec[2]; vals[2] = vec[2]; vals[3] = vec[2];
llvm::Constant *zChannel = createConstGlobalVector(vals);
vals[0] = vec[3]; vals[1] = vec[3]; vals[2] = vec[3]; vals[3] = vec[3];
- llvm::Constant *wChannel = createConstGlobalVector(vals);
+ llvm::Constant *wChannel = createConstGlobalVector(vals);*/
channelArray.push_back(xChannel);
channelArray.push_back(yChannel);
channelArray.push_back(zChannel);
@@ -144,22 +154,54 @@ std::vector<llvm::Value*> StorageSoa::inputElement(llvm::Value *idx)
return res;
}
-std::vector<llvm::Value*> StorageSoa::constElement(llvm::Value *idx)
+llvm::Value* StorageSoa::unpackConstElement(llvm::IRBuilder<>* m_builder, llvm::Value* vector, int cc)
+{
+ std::vector<llvm::Value*> x(4);
+ x[0] = m_builder->CreateExtractElement(vector,
+ constantInt(cc),
+ name("x"));
+
+ VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+ Constant *constVector = Constant::getNullValue(vectorType);
+ Value *res = m_builder->CreateInsertElement(constVector, x[0],
+ constantInt(0),
+ name("vecx"));
+ res = m_builder->CreateInsertElement(res, x[0], constantInt(1),
+ name("vecxx"));
+ res = m_builder->CreateInsertElement(res, x[0], constantInt(2),
+ name("vecxxx"));
+ res = m_builder->CreateInsertElement(res, x[0], constantInt(3),
+ name("vecxxxx"));
+ return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::constElement(llvm::IRBuilder<>* m_builder, llvm::Value *idx)
{
std::vector<llvm::Value*> res(4);
+ std::vector<llvm::Value*> res2(4);
llvm::Value *xChannel, *yChannel, *zChannel, *wChannel;
xChannel = elementPointer(m_consts, idx, 0);
- yChannel = elementPointer(m_consts, idx, 1);
+/* yChannel = elementPointer(m_consts, idx, 1);
zChannel = elementPointer(m_consts, idx, 2);
- wChannel = elementPointer(m_consts, idx, 3);
+ wChannel = elementPointer(m_consts, idx, 3);*/
res[0] = alignedArrayLoad(xChannel);
+/* res[1] = alignedArrayLoad(xChannel);
+ res[2] = alignedArrayLoad(xChannel);
+ res[3] = alignedArrayLoad(xChannel);*/
+
+
+ res2[0]=unpackConstElement(m_builder, res[0],0);
+ res2[1]=unpackConstElement(m_builder, res[0],1);
+ res2[2]=unpackConstElement(m_builder, res[0],2);
+ res2[3]=unpackConstElement(m_builder, res[0],3);
+/*res[0] = alignedArrayLoad(xChannel);
res[1] = alignedArrayLoad(yChannel);
res[2] = alignedArrayLoad(zChannel);
- res[3] = alignedArrayLoad(wChannel);
+ res[3] = alignedArrayLoad(wChannel);*/
- return res;
+ return res2;
}
std::vector<llvm::Value*> StorageSoa::outputElement(llvm::Value *idx)
@@ -260,6 +302,12 @@ llvm::Module * StorageSoa::currentModule() const
return m_block->getParent()->getParent();
}
+llvm::Constant * StorageSoa::createConstGlobalFloat(const float val)
+{
+ Constant*c = ConstantFP::get(APFloat(val));
+ return c;
+}
+
llvm::Constant * StorageSoa::createConstGlobalVector(const std::vector<float> &vec)
{
VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
@@ -278,7 +326,7 @@ llvm::Constant * StorageSoa::createConstGlobalVector(const std::vector<float> &v
}
std::vector<llvm::Value*> StorageSoa::load(enum tgsi_file_type type, int idx, int swizzle,
- llvm::Value *indIdx)
+ llvm::IRBuilder<>* m_builder,llvm::Value *indIdx)
{
std::vector<llvm::Value*> val(4);
@@ -302,7 +350,8 @@ std::vector<llvm::Value*> StorageSoa::load(enum tgsi_file_type type, int idx, in
val = tempElement(realIndex);
break;
case TGSI_FILE_CONSTANT:
- val = constElement(realIndex);
+ val = constElement(m_builder, realIndex);
+ printf("constant COUCOU index %d\n",realIndex);
break;
case TGSI_FILE_IMMEDIATE:
val = immediateElement(realIndex);
diff --git a/src/gallium/auxiliary/gallivm/storagesoa.h b/src/gallium/auxiliary/gallivm/storagesoa.h
index ae2fc7c6ae..f21ca6ec43 100644
--- a/src/gallium/auxiliary/gallivm/storagesoa.h
+++ b/src/gallium/auxiliary/gallivm/storagesoa.h
@@ -29,6 +29,7 @@
#define STORAGESOA_H
#include <pipe/p_shader_tokens.h>
+#include <llvm/Support/IRBuilder.h>
#include <vector>
#include <list>
@@ -56,7 +57,7 @@ public:
std::vector<llvm::Value*> load(enum tgsi_file_type type, int idx, int swizzle,
- llvm::Value *indIdx =0);
+ llvm::IRBuilder<>* m_builder, llvm::Value *indIdx =0);
void store(enum tgsi_file_type type, int idx, const std::vector<llvm::Value*> &val,
int mask);
@@ -76,10 +77,12 @@ private:
const char *name(const char *prefix) const;
llvm::Value *alignedArrayLoad(llvm::Value *val);
llvm::Module *currentModule() const;
+ llvm::Constant *createConstGlobalFloat(const float val);
llvm::Constant *createConstGlobalVector(const std::vector<float> &vec);
std::vector<llvm::Value*> inputElement(llvm::Value *indIdx);
- std::vector<llvm::Value*> constElement(llvm::Value *indIdx);
+ llvm::Value* unpackConstElement(llvm::IRBuilder<>* m_builder, llvm::Value *indIdx, int cc);
+ std::vector<llvm::Value*> constElement(llvm::IRBuilder<>* m_builder, llvm::Value *indIdx);
std::vector<llvm::Value*> outputElement(llvm::Value *indIdx);
std::vector<llvm::Value*> tempElement(llvm::Value *indIdx);
std::vector<llvm::Value*> immediateElement(llvm::Value *indIdx);
diff --git a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
index 7292c0e366..1191a6cae9 100644
--- a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
+++ b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
@@ -52,7 +52,7 @@ static inline FunctionType *vertexShaderFunctionType()
// pass are castable to the following:
// [4 x <4 x float>] inputs,
// [4 x <4 x float>] output,
- // [4 x [4 x float]] consts,
+ // [4 x [1 x float]] consts,
// [4 x <4 x float>] temps
std::vector<const Type*> funcArgs;
@@ -61,7 +61,7 @@ static inline FunctionType *vertexShaderFunctionType()
PointerType *vectorArrayPtr = PointerType::get(vectorArray, 0);
ArrayType *floatArray = ArrayType::get(Type::FloatTy, 4);
- ArrayType *constsArray = ArrayType::get(floatArray, 4);
+ ArrayType *constsArray = ArrayType::get(floatArray, 1);
PointerType *constsArrayPtr = PointerType::get(constsArray, 0);
funcArgs.push_back(vectorArrayPtr);//inputs
@@ -246,6 +246,7 @@ translate_instruction(llvm::Module *module,
val = storage->constElement(src->SrcRegister.Index, indIdx);
} else if (src->SrcRegister.File == TGSI_FILE_INPUT) {
val = storage->inputElement(src->SrcRegister.Index, indIdx);
+ // FIXME we should not be generating elements for temporaries, this creates useless memory writes
} else if (src->SrcRegister.File == TGSI_FILE_TEMPORARY) {
val = storage->tempElement(src->SrcRegister.Index);
} else if (src->SrcRegister.File == TGSI_FILE_OUTPUT) {
@@ -676,6 +677,7 @@ translate_instruction(llvm::Module *module,
if (dst->DstRegister.File == TGSI_FILE_OUTPUT) {
storage->setOutputElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+ // FIXME we should not be generating elements for temporaries, this creates useless memory writes
} 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) {
@@ -707,9 +709,8 @@ translate_instructionir(llvm::Module *module,
if (src->SrcRegister.Indirect) {
indIdx = storage->addrElement(src->SrcRegisterInd.Index);
}
-
val = storage->load((enum tgsi_file_type)src->SrcRegister.File,
- src->SrcRegister.Index, swizzle, indIdx);
+ src->SrcRegister.Index, swizzle, instr->getIRBuilder(), indIdx);
inputs[i] = val;
}
@@ -1025,7 +1026,6 @@ translate_instructionir(llvm::Module *module,
/* store results */
for (int i = 0; i < inst->Instruction.NumDstRegs; ++i) {
struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
-
storage->store((enum tgsi_file_type)dst->DstRegister.File,
dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-03 05:07:32 +0000