static void createPushHeapFunction() {
// Saving last BasicBlock;
LLVMBasicBlockRef OldBB = LLVMGetInsertBlock(Builder);
LLVMTypeRef ParamType = LLVMPointerType(RAType, 0);
LLVMTypeRef FunctionType = LLVMFunctionType(LLVMVoidType(), &ParamType, 1, 0);
LLVMValueRef Function = LLVMAddFunction(Module, "push.heap", FunctionType);
LLVMBasicBlockRef Entry = LLVMAppendBasicBlock(Function, "entry");
LLVMPositionBuilderAtEnd(Builder, Entry);
// Function Body
LLVMValueRef HeapMalloc = LLVMBuildMalloc(Builder, HeapType, "heap.malloc");
LLVMValueRef ExPtrIdx[] = { getSConstInt(0), getSConstInt(0) };
LLVMValueRef LastPtrIdx[] = { getSConstInt(0), getSConstInt(1) };
LLVMValueRef ExPtr = LLVMBuildInBoundsGEP(Builder, HeapMalloc, ExPtrIdx, 2, "heap.exec");
LLVMValueRef LastPtr = LLVMBuildInBoundsGEP(Builder, HeapMalloc, LastPtrIdx, 2, "heap.last");
LLVMBuildStore(Builder, LLVMGetParam(Function, 0), ExPtr);
LLVMBuildStore(Builder, LLVMBuildLoad(Builder, HeapHead, "ld.heap.head"), LastPtr);
LLVMBuildStore(Builder, HeapMalloc, HeapHead);
LLVMBuildRetVoid(Builder);
// Restoring last BasicBlock
LLVMPositionBuilderAtEnd(Builder, OldBB);
}
static LLVMTypeRef llvm_type(int type)
{
switch (type) {
case SCM_FOREIGN_TYPE_FLOAT:
return LLVMFloatType();
case SCM_FOREIGN_TYPE_DOUBLE:
return LLVMDoubleType();
case SCM_FOREIGN_TYPE_BOOL:
return LLVMInt1Type();
case SCM_FOREIGN_TYPE_UINT8:
case SCM_FOREIGN_TYPE_INT8:
return LLVMInt8Type();
case SCM_FOREIGN_TYPE_UINT16:
case SCM_FOREIGN_TYPE_INT16:
return LLVMInt16Type();
case SCM_FOREIGN_TYPE_UINT32:
case SCM_FOREIGN_TYPE_INT32:
return LLVMInt32Type();
case SCM_FOREIGN_TYPE_UINT64:
case SCM_FOREIGN_TYPE_INT64:
return LLVMInt64Type();
default:
return LLVMVoidType();
};
}
struct vm_state *
vm_state_create(const char *module_name)
{
struct vm_state *vm;
LLVMTypeRef function_type;
LLVMValueRef function_value;
LLVMBasicBlockRef entry_block;
vm = calloc(1, sizeof(struct vm_state));
if (vm == NULL) {
fprintf(stderr, "Memory allocation request failed.\n");
exit(EXIT_FAILURE);
}
vm->module = LLVMModuleCreateWithName(module_name);
vm->builder = LLVMCreateBuilder();
function_type = LLVMFunctionType(LLVMVoidType(), NULL, 0, 0);
function_value = LLVMAddFunction(vm->module, "main", function_type);
entry_block = LLVMAppendBasicBlock(function_value, "entry");
LLVMPositionBuilderAtEnd(vm->builder, entry_block);
vm->symtab = symbol_table_create();
return vm;
}
int main (void) {
LLVMModuleRef module = LLVMModuleCreateWithName("kal");
LLVMBuilderRef builder = LLVMCreateBuilder();
// LLVMInitializeNativeTarget();
LLVMTypeRef funcType = LLVMFunctionType(LLVMVoidType(), NULL, 0, 0);
LLVMValueRef func = LLVMAddFunction(module, "main", funcType);
LLVMSetLinkage(func, LLVMExternalLinkage);
LLVMBasicBlockRef block = LLVMAppendBasicBlock(func, "entry");
LLVMPositionBuilderAtEnd(builder, block);
LLVMValueRef cond = LLVMBuildICmp(builder, LLVMIntNE, LLVMConstInt(LLVMInt32Type(), 2, 0), LLVMConstInt(LLVMInt32Type(), 1, 0), "ifcond");
LLVMValueRef owning_block = LLVMGetBasicBlockParent(LLVMGetInsertBlock(builder)); //TODO: WRONG??
//LLVMValueRef owning_block = LLVMBasicBlockAsValue(LLVMGetPreviousBasicBlock(LLVMGetInsertBlock(builder)));
// 2. Generate new blocks for cases.
LLVMBasicBlockRef then_ref = LLVMAppendBasicBlock(owning_block, "then");
LLVMBasicBlockRef else_ref = LLVMAppendBasicBlock(owning_block, "else");
LLVMBasicBlockRef merge_ref = LLVMAppendBasicBlock(owning_block, "ifmerge");
// 3. Branch conditionally on then or else.
LLVMBuildCondBr(builder, cond, then_ref, else_ref);
// 4. Build then branch prologue.
LLVMPositionBuilderAtEnd(builder, then_ref);
LLVMValueRef hi1 = LLVMBuildXor(builder, LLVMGetUndef(LLVMInt32Type()), LLVMGetUndef(LLVMInt32Type()), "subtmp");
// 5. Connect then branch to merge block.
LLVMBuildBr(builder, merge_ref);
then_ref = LLVMGetInsertBlock(builder);
// 6. Build else branch prologue.
LLVMPositionBuilderAtEnd(builder, else_ref);
LLVMValueRef hi2 = LLVMBuildXor(builder, LLVMGetUndef(LLVMInt32Type()), LLVMGetUndef(LLVMInt32Type()), "subtmp2");
// 7. Connect else branch to merge block.
LLVMBuildBr(builder, merge_ref);
else_ref = LLVMGetInsertBlock(builder);
// 8. Position ourselves after the merge block.
LLVMPositionBuilderAtEnd(builder, merge_ref);
// 9. Build the phi node.
// LLVMValueRef phi = LLVMBuildPhi(builder, LLVMDoubleType(), "phi");
// 10. Add incoming edges.
// LLVMAddIncoming(phi, &hi1, &then_ref, 1);
// LLVMAddIncoming(phi, &hi2, &else_ref, 1);
LLVMDumpModule(module);
LLVMDisposeBuilder(builder);
LLVMDisposeModule(module);
return 0;
}
static LLVMValueRef
add_conv_test(LLVMModuleRef module,
struct lp_type src_type, unsigned num_srcs,
struct lp_type dst_type, unsigned num_dsts)
{
LLVMTypeRef args[2];
LLVMValueRef func;
LLVMValueRef src_ptr;
LLVMValueRef dst_ptr;
LLVMBasicBlockRef block;
LLVMBuilderRef builder;
LLVMValueRef src[LP_MAX_VECTOR_LENGTH];
LLVMValueRef dst[LP_MAX_VECTOR_LENGTH];
unsigned i;
args[0] = LLVMPointerType(lp_build_vec_type(src_type), 0);
args[1] = LLVMPointerType(lp_build_vec_type(dst_type), 0);
func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidType(), args, 2, 0));
LLVMSetFunctionCallConv(func, LLVMCCallConv);
src_ptr = LLVMGetParam(func, 0);
dst_ptr = LLVMGetParam(func, 1);
block = LLVMAppendBasicBlock(func, "entry");
builder = LLVMCreateBuilder();
LLVMPositionBuilderAtEnd(builder, block);
for(i = 0; i < num_srcs; ++i) {
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
LLVMValueRef ptr = LLVMBuildGEP(builder, src_ptr, &index, 1, "");
src[i] = LLVMBuildLoad(builder, ptr, "");
}
lp_build_conv(builder, src_type, dst_type, src, num_srcs, dst, num_dsts);
for(i = 0; i < num_dsts; ++i) {
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
LLVMValueRef ptr = LLVMBuildGEP(builder, dst_ptr, &index, 1, "");
LLVMBuildStore(builder, dst[i], ptr);
}
LLVMBuildRetVoid(builder);;
LLVMDisposeBuilder(builder);
return func;
}
/**
* lp_build_assert.
*
* Build an assertion in LLVM IR by building a function call to the
* lp_assert() function above.
*
* \param condition should be an 'i1' or 'i32' value
* \param msg a string to print if the assertion fails.
*/
LLVMValueRef
lp_build_assert(LLVMBuilderRef builder, LLVMValueRef condition,
const char *msg)
{
LLVMModuleRef module;
LLVMTypeRef arg_types[2];
LLVMValueRef msg_string, assert_func, params[2], r;
module = LLVMGetGlobalParent(LLVMGetBasicBlockParent(
LLVMGetInsertBlock(builder)));
msg_string = lp_build_const_string_variable(module, msg, strlen(msg) + 1);
arg_types[0] = LLVMInt32Type();
arg_types[1] = LLVMPointerType(LLVMInt8Type(), 0);
/* lookup the lp_assert function */
assert_func = LLVMGetNamedFunction(module, "lp_assert");
/* Create the assertion function if not found */
if (!assert_func) {
LLVMTypeRef func_type =
LLVMFunctionType(LLVMVoidType(), arg_types, 2, 0);
assert_func = LLVMAddFunction(module, "lp_assert", func_type);
LLVMSetFunctionCallConv(assert_func, LLVMCCallConv);
LLVMSetLinkage(assert_func, LLVMExternalLinkage);
LLVMAddGlobalMapping(lp_build_engine, assert_func,
func_to_pointer((func_pointer)lp_assert));
}
assert(assert_func);
/* build function call param list */
params[0] = LLVMBuildZExt(builder, condition, arg_types[0], "");
params[1] = LLVMBuildBitCast(builder, msg_string, arg_types[1], "");
/* check arg types */
assert(LLVMTypeOf(params[0]) == arg_types[0]);
assert(LLVMTypeOf(params[1]) == arg_types[1]);
r = LLVMBuildCall(builder, assert_func, params, 2, "");
return r;
}
static LLVMValueRef zephir_get_add_function(zephir_context *context)
{
LLVMValueRef function;
LLVMTypeRef arg_tys[3];
function = LLVMGetNamedFunction(context->module, "add_function");
if (!function) {
arg_tys[0] = context->types.zval_pointer_type;
arg_tys[1] = context->types.zval_pointer_type;
arg_tys[2] = context->types.zval_pointer_type;
function = LLVMAddFunction(context->module, "add_function", LLVMFunctionType(LLVMVoidType(), arg_tys, 3, 0));
if (!function) {
zend_error(E_ERROR, "Cannot register add_function");
}
LLVMAddGlobalMapping(context->engine, function, add_function);
LLVMSetFunctionCallConv(function, LLVMCCallConv);
LLVMAddFunctionAttr(function, LLVMNoUnwindAttribute);
}
return function;
}
static void createPopHeapFunction() {
// Saving last BasicBlock;
LLVMBasicBlockRef OldBB = LLVMGetInsertBlock(Builder);
LLVMTypeRef FunctionType = LLVMFunctionType(LLVMVoidType(), NULL, 0, 0);
LLVMValueRef Function = LLVMAddFunction(Module, "pop.heap", FunctionType);
LLVMBasicBlockRef Entry = LLVMAppendBasicBlock(Function, "entry");
LLVMPositionBuilderAtEnd(Builder, Entry);
// Function Body
LLVMValueRef HeapHdPtr = LLVMBuildLoad(Builder, HeapHead, "");
LLVMValueRef LastPtrIdx[] = { getSConstInt(0), getSConstInt(1) };
LLVMValueRef LastPtr = LLVMBuildInBoundsGEP(Builder, HeapHdPtr, LastPtrIdx, 2, "heap.last");
LLVMValueRef LastPtrLd = LLVMBuildLoad(Builder, LastPtr, "ld.heap.last");
LLVMBuildStore(Builder, LastPtrLd, HeapHead);
LLVMBuildRetVoid(Builder);
// Restoring last BasicBlock
LLVMPositionBuilderAtEnd(Builder, OldBB);
}
static LLVMValueRef
add_blend_test(LLVMModuleRef module,
const struct pipe_blend_state *blend,
enum vector_mode mode,
struct lp_type type)
{
LLVMTypeRef ret_type;
LLVMTypeRef vec_type;
LLVMTypeRef args[4];
LLVMValueRef func;
LLVMValueRef src_ptr;
LLVMValueRef dst_ptr;
LLVMValueRef const_ptr;
LLVMValueRef res_ptr;
LLVMBasicBlockRef block;
LLVMBuilderRef builder;
ret_type = LLVMInt64Type();
vec_type = lp_build_vec_type(type);
args[3] = args[2] = args[1] = args[0] = LLVMPointerType(vec_type, 0);
func = LLVMAddFunction(module, "test", LLVMFunctionType(LLVMVoidType(), args, 4, 0));
LLVMSetFunctionCallConv(func, LLVMCCallConv);
src_ptr = LLVMGetParam(func, 0);
dst_ptr = LLVMGetParam(func, 1);
const_ptr = LLVMGetParam(func, 2);
res_ptr = LLVMGetParam(func, 3);
block = LLVMAppendBasicBlock(func, "entry");
builder = LLVMCreateBuilder();
LLVMPositionBuilderAtEnd(builder, block);
if (mode == AoS) {
LLVMValueRef src;
LLVMValueRef dst;
LLVMValueRef con;
LLVMValueRef res;
src = LLVMBuildLoad(builder, src_ptr, "src");
dst = LLVMBuildLoad(builder, dst_ptr, "dst");
con = LLVMBuildLoad(builder, const_ptr, "const");
res = lp_build_blend_aos(builder, blend, type, src, dst, con, 3);
lp_build_name(res, "res");
LLVMBuildStore(builder, res, res_ptr);
}
if (mode == SoA) {
LLVMValueRef src[4];
LLVMValueRef dst[4];
LLVMValueRef con[4];
LLVMValueRef res[4];
unsigned i;
for(i = 0; i < 4; ++i) {
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
src[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, src_ptr, &index, 1, ""), "");
dst[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dst_ptr, &index, 1, ""), "");
con[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, const_ptr, &index, 1, ""), "");
lp_build_name(src[i], "src.%c", "rgba"[i]);
lp_build_name(con[i], "con.%c", "rgba"[i]);
lp_build_name(dst[i], "dst.%c", "rgba"[i]);
}
lp_build_blend_soa(builder, blend, type, src, dst, con, res);
for(i = 0; i < 4; ++i) {
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
lp_build_name(res[i], "res.%c", "rgba"[i]);
LLVMBuildStore(builder, res[i], LLVMBuildGEP(builder, res_ptr, &index, 1, ""));
}
}
LLVMBuildRetVoid(builder);;
LLVMDisposeBuilder(builder);
return func;
}
/**
* Generate the runtime callable function for the whole fragment pipeline.
* Note that the function which we generate operates on a block of 16
* pixels at at time. The block contains 2x2 quads. Each quad contains
* 2x2 pixels.
*/
static void
generate_fragment(struct llvmpipe_context *lp,
struct lp_fragment_shader *shader,
struct lp_fragment_shader_variant *variant,
unsigned do_tri_test)
{
struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
const struct lp_fragment_shader_variant_key *key = &variant->key;
struct lp_type fs_type;
struct lp_type blend_type;
LLVMTypeRef fs_elem_type;
LLVMTypeRef fs_vec_type;
LLVMTypeRef fs_int_vec_type;
LLVMTypeRef blend_vec_type;
LLVMTypeRef blend_int_vec_type;
LLVMTypeRef arg_types[14];
LLVMTypeRef func_type;
LLVMTypeRef int32_vec4_type = lp_build_int32_vec4_type();
LLVMValueRef context_ptr;
LLVMValueRef x;
LLVMValueRef y;
LLVMValueRef a0_ptr;
LLVMValueRef dadx_ptr;
LLVMValueRef dady_ptr;
LLVMValueRef color_ptr_ptr;
LLVMValueRef depth_ptr;
LLVMValueRef c0, c1, c2, step0_ptr, step1_ptr, step2_ptr;
LLVMBasicBlockRef block;
LLVMBuilderRef builder;
LLVMValueRef x0;
LLVMValueRef y0;
struct lp_build_sampler_soa *sampler;
struct lp_build_interp_soa_context interp;
LLVMValueRef fs_mask[LP_MAX_VECTOR_LENGTH];
LLVMValueRef fs_out_color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][LP_MAX_VECTOR_LENGTH];
LLVMValueRef blend_mask;
LLVMValueRef blend_in_color[NUM_CHANNELS];
LLVMValueRef function;
unsigned num_fs;
unsigned i;
unsigned chan;
unsigned cbuf;
/* TODO: actually pick these based on the fs and color buffer
* characteristics. */
memset(&fs_type, 0, sizeof fs_type);
fs_type.floating = TRUE; /* floating point values */
fs_type.sign = TRUE; /* values are signed */
fs_type.norm = FALSE; /* values are not limited to [0,1] or [-1,1] */
fs_type.width = 32; /* 32-bit float */
fs_type.length = 4; /* 4 elements per vector */
num_fs = 4; /* number of quads per block */
memset(&blend_type, 0, sizeof blend_type);
blend_type.floating = FALSE; /* values are integers */
blend_type.sign = FALSE; /* values are unsigned */
blend_type.norm = TRUE; /* values are in [0,1] or [-1,1] */
blend_type.width = 8; /* 8-bit ubyte values */
blend_type.length = 16; /* 16 elements per vector */
/*
* Generate the function prototype. Any change here must be reflected in
* lp_jit.h's lp_jit_frag_func function pointer type, and vice-versa.
*/
fs_elem_type = lp_build_elem_type(fs_type);
fs_vec_type = lp_build_vec_type(fs_type);
fs_int_vec_type = lp_build_int_vec_type(fs_type);
blend_vec_type = lp_build_vec_type(blend_type);
blend_int_vec_type = lp_build_int_vec_type(blend_type);
arg_types[0] = screen->context_ptr_type; /* context */
arg_types[1] = LLVMInt32Type(); /* x */
arg_types[2] = LLVMInt32Type(); /* y */
arg_types[3] = LLVMPointerType(fs_elem_type, 0); /* a0 */
arg_types[4] = LLVMPointerType(fs_elem_type, 0); /* dadx */
arg_types[5] = LLVMPointerType(fs_elem_type, 0); /* dady */
arg_types[6] = LLVMPointerType(LLVMPointerType(blend_vec_type, 0), 0); /* color */
arg_types[7] = LLVMPointerType(fs_int_vec_type, 0); /* depth */
arg_types[8] = LLVMInt32Type(); /* c0 */
arg_types[9] = LLVMInt32Type(); /* c1 */
arg_types[10] = LLVMInt32Type(); /* c2 */
/* Note: the step arrays are built as int32[16] but we interpret
* them here as int32_vec4[4].
*/
arg_types[11] = LLVMPointerType(int32_vec4_type, 0);/* step0 */
arg_types[12] = LLVMPointerType(int32_vec4_type, 0);/* step1 */
arg_types[13] = LLVMPointerType(int32_vec4_type, 0);/* step2 */
func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
function = LLVMAddFunction(screen->module, "shader", func_type);
LLVMSetFunctionCallConv(function, LLVMCCallConv);
variant->function[do_tri_test] = function;
/* XXX: need to propagate noalias down into color param now we are
* passing a pointer-to-pointer?
*/
for(i = 0; i < Elements(arg_types); ++i)
if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
LLVMAddAttribute(LLVMGetParam(function, i), LLVMNoAliasAttribute);
context_ptr = LLVMGetParam(function, 0);
x = LLVMGetParam(function, 1);
y = LLVMGetParam(function, 2);
a0_ptr = LLVMGetParam(function, 3);
dadx_ptr = LLVMGetParam(function, 4);
dady_ptr = LLVMGetParam(function, 5);
color_ptr_ptr = LLVMGetParam(function, 6);
depth_ptr = LLVMGetParam(function, 7);
c0 = LLVMGetParam(function, 8);
c1 = LLVMGetParam(function, 9);
c2 = LLVMGetParam(function, 10);
step0_ptr = LLVMGetParam(function, 11);
step1_ptr = LLVMGetParam(function, 12);
step2_ptr = LLVMGetParam(function, 13);
lp_build_name(context_ptr, "context");
lp_build_name(x, "x");
lp_build_name(y, "y");
lp_build_name(a0_ptr, "a0");
lp_build_name(dadx_ptr, "dadx");
lp_build_name(dady_ptr, "dady");
lp_build_name(color_ptr_ptr, "color_ptr");
lp_build_name(depth_ptr, "depth");
lp_build_name(c0, "c0");
lp_build_name(c1, "c1");
lp_build_name(c2, "c2");
lp_build_name(step0_ptr, "step0");
lp_build_name(step1_ptr, "step1");
lp_build_name(step2_ptr, "step2");
/*
* Function body
*/
block = LLVMAppendBasicBlock(function, "entry");
builder = LLVMCreateBuilder();
LLVMPositionBuilderAtEnd(builder, block);
generate_pos0(builder, x, y, &x0, &y0);
lp_build_interp_soa_init(&interp,
shader->base.tokens,
key->flatshade,
builder, fs_type,
a0_ptr, dadx_ptr, dady_ptr,
x0, y0);
/* code generated texture sampling */
sampler = lp_llvm_sampler_soa_create(key->sampler, context_ptr);
/* loop over quads in the block */
for(i = 0; i < num_fs; ++i) {
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
LLVMValueRef out_color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS];
LLVMValueRef depth_ptr_i;
int cbuf;
if(i != 0)
lp_build_interp_soa_update(&interp, i);
depth_ptr_i = LLVMBuildGEP(builder, depth_ptr, &index, 1, "");
generate_fs(lp, shader, key,
builder,
fs_type,
context_ptr,
i,
&interp,
sampler,
&fs_mask[i], /* output */
out_color,
depth_ptr_i,
do_tri_test,
c0, c1, c2,
step0_ptr, step1_ptr, step2_ptr);
for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++)
for(chan = 0; chan < NUM_CHANNELS; ++chan)
fs_out_color[cbuf][chan][i] = out_color[cbuf][chan];
}
sampler->destroy(sampler);
/* Loop over color outputs / color buffers to do blending.
*/
for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
LLVMValueRef color_ptr;
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), cbuf, 0);
/*
* Convert the fs's output color and mask to fit to the blending type.
*/
for(chan = 0; chan < NUM_CHANNELS; ++chan) {
lp_build_conv(builder, fs_type, blend_type,
fs_out_color[cbuf][chan], num_fs,
&blend_in_color[chan], 1);
lp_build_name(blend_in_color[chan], "color%d.%c", cbuf, "rgba"[chan]);
}
lp_build_conv_mask(builder, fs_type, blend_type,
fs_mask, num_fs,
&blend_mask, 1);
color_ptr = LLVMBuildLoad(builder,
LLVMBuildGEP(builder, color_ptr_ptr, &index, 1, ""),
"");
lp_build_name(color_ptr, "color_ptr%d", cbuf);
/*
* Blending.
*/
generate_blend(&key->blend,
builder,
blend_type,
context_ptr,
blend_mask,
blend_in_color,
color_ptr);
}
LLVMBuildRetVoid(builder);
LLVMDisposeBuilder(builder);
/* Verify the LLVM IR. If invalid, dump and abort */
#ifdef DEBUG
if(LLVMVerifyFunction(function, LLVMPrintMessageAction)) {
if (1)
LLVMDumpValue(function);
abort();
}
#endif
/* Apply optimizations to LLVM IR */
if (1)
LLVMRunFunctionPassManager(screen->pass, function);
if (LP_DEBUG & DEBUG_JIT) {
/* Print the LLVM IR to stderr */
LLVMDumpValue(function);
debug_printf("\n");
}
/*
* Translate the LLVM IR into machine code.
*/
variant->jit_function[do_tri_test] = (lp_jit_frag_func)LLVMGetPointerToGlobal(screen->engine, function);
if (LP_DEBUG & DEBUG_ASM)
lp_disassemble(variant->jit_function[do_tri_test]);
}
/*
* Create a function that deforms a tuple of type desc up to natts columns.
*/
LLVMValueRef
slot_compile_deform(LLVMJitContext *context, TupleDesc desc, int natts)
{
char *funcname;
LLVMModuleRef mod;
LLVMBuilderRef b;
LLVMTypeRef deform_sig;
LLVMValueRef v_deform_fn;
LLVMBasicBlockRef b_entry;
LLVMBasicBlockRef b_adjust_unavail_cols;
LLVMBasicBlockRef b_find_start;
LLVMBasicBlockRef b_out;
LLVMBasicBlockRef b_dead;
LLVMBasicBlockRef *attcheckattnoblocks;
LLVMBasicBlockRef *attstartblocks;
LLVMBasicBlockRef *attisnullblocks;
LLVMBasicBlockRef *attcheckalignblocks;
LLVMBasicBlockRef *attalignblocks;
LLVMBasicBlockRef *attstoreblocks;
LLVMValueRef v_offp;
LLVMValueRef v_tupdata_base;
LLVMValueRef v_tts_values;
LLVMValueRef v_tts_nulls;
LLVMValueRef v_slotoffp;
LLVMValueRef v_slowp;
LLVMValueRef v_nvalidp;
LLVMValueRef v_nvalid;
LLVMValueRef v_maxatt;
LLVMValueRef v_slot;
LLVMValueRef v_tupleheaderp;
LLVMValueRef v_tuplep;
LLVMValueRef v_infomask1;
LLVMValueRef v_infomask2;
LLVMValueRef v_bits;
LLVMValueRef v_hoff;
LLVMValueRef v_hasnulls;
/* last column (0 indexed) guaranteed to exist */
int guaranteed_column_number = -1;
/* current known alignment */
int known_alignment = 0;
/* if true, known_alignment describes definite offset of column */
bool attguaranteedalign = true;
int attnum;
mod = llvm_mutable_module(context);
funcname = llvm_expand_funcname(context, "deform");
/*
* Check which columns do have to exist, so we don't have to check the
* rows natts unnecessarily.
*/
for (attnum = 0; attnum < desc->natts; attnum++)
{
Form_pg_attribute att = TupleDescAttr(desc, attnum);
/*
* If the column is possibly missing, we can't rely on its (or
* subsequent) NOT NULL constraints to indicate minimum attributes in
* the tuple, so stop here.
*/
if (att->atthasmissing)
break;
/*
* Column is NOT NULL and there've been no preceding missing columns,
* it's guaranteed that all columns up to here exist at least in the
* NULL bitmap.
*/
if (att->attnotnull)
guaranteed_column_number = attnum;
}
/* Create the signature and function */
{
LLVMTypeRef param_types[1];
param_types[0] = l_ptr(StructTupleTableSlot);
deform_sig = LLVMFunctionType(LLVMVoidType(), param_types,
lengthof(param_types), 0);
}
v_deform_fn = LLVMAddFunction(mod, funcname, deform_sig);
LLVMSetLinkage(v_deform_fn, LLVMInternalLinkage);
LLVMSetParamAlignment(LLVMGetParam(v_deform_fn, 0), MAXIMUM_ALIGNOF);
llvm_copy_attributes(AttributeTemplate, v_deform_fn);
b_entry =
LLVMAppendBasicBlock(v_deform_fn, "entry");
b_adjust_unavail_cols =
LLVMAppendBasicBlock(v_deform_fn, "adjust_unavail_cols");
b_find_start =
LLVMAppendBasicBlock(v_deform_fn, "find_startblock");
b_out =
LLVMAppendBasicBlock(v_deform_fn, "outblock");
b_dead =
LLVMAppendBasicBlock(v_deform_fn, "deadblock");
b = LLVMCreateBuilder();
attcheckattnoblocks = palloc(sizeof(LLVMBasicBlockRef) * natts);
attstartblocks = palloc(sizeof(LLVMBasicBlockRef) * natts);
attisnullblocks = palloc(sizeof(LLVMBasicBlockRef) * natts);
attcheckalignblocks = palloc(sizeof(LLVMBasicBlockRef) * natts);
attalignblocks = palloc(sizeof(LLVMBasicBlockRef) * natts);
attstoreblocks = palloc(sizeof(LLVMBasicBlockRef) * natts);
known_alignment = 0;
LLVMPositionBuilderAtEnd(b, b_entry);
/* perform allocas first, llvm only converts those to registers */
v_offp = LLVMBuildAlloca(b, TypeSizeT, "v_offp");
v_slot = LLVMGetParam(v_deform_fn, 0);
v_tts_values =
l_load_struct_gep(b, v_slot, FIELDNO_TUPLETABLESLOT_VALUES,
"tts_values");
v_tts_nulls =
l_load_struct_gep(b, v_slot, FIELDNO_TUPLETABLESLOT_ISNULL,
"tts_ISNULL");
v_slotoffp = LLVMBuildStructGEP(b, v_slot, FIELDNO_TUPLETABLESLOT_OFF, "");
v_slowp = LLVMBuildStructGEP(b, v_slot, FIELDNO_TUPLETABLESLOT_SLOW, "");
v_nvalidp = LLVMBuildStructGEP(b, v_slot, FIELDNO_TUPLETABLESLOT_NVALID, "");
v_tupleheaderp =
l_load_struct_gep(b, v_slot, FIELDNO_TUPLETABLESLOT_TUPLE,
"tupleheader");
v_tuplep =
l_load_struct_gep(b, v_tupleheaderp, FIELDNO_HEAPTUPLEDATA_DATA,
"tuple");
v_bits =
LLVMBuildBitCast(b,
LLVMBuildStructGEP(b, v_tuplep,
FIELDNO_HEAPTUPLEHEADERDATA_BITS,
""),
l_ptr(LLVMInt8Type()),
"t_bits");
v_infomask1 =
l_load_struct_gep(b, v_tuplep,
FIELDNO_HEAPTUPLEHEADERDATA_INFOMASK,
"infomask1");
v_infomask2 =
l_load_struct_gep(b,
v_tuplep, FIELDNO_HEAPTUPLEHEADERDATA_INFOMASK2,
"infomask2");
/* t_infomask & HEAP_HASNULL */
v_hasnulls =
LLVMBuildICmp(b, LLVMIntNE,
LLVMBuildAnd(b,
l_int16_const(HEAP_HASNULL),
v_infomask1, ""),
l_int16_const(0),
"hasnulls");
/* t_infomask2 & HEAP_NATTS_MASK */
v_maxatt = LLVMBuildAnd(b,
l_int16_const(HEAP_NATTS_MASK),
v_infomask2,
"maxatt");
v_hoff =
l_load_struct_gep(b, v_tuplep,
FIELDNO_HEAPTUPLEHEADERDATA_HOFF,
"t_hoff");
v_tupdata_base =
LLVMBuildGEP(b,
LLVMBuildBitCast(b,
v_tuplep,
l_ptr(LLVMInt8Type()),
""),
&v_hoff, 1,
"v_tupdata_base");
/*
* Load tuple start offset from slot. Will be reset below in case there's
* no existing deformed columns in slot.
*/
{
LLVMValueRef v_off_start;
v_off_start = LLVMBuildLoad(b, v_slotoffp, "v_slot_off");
v_off_start = LLVMBuildZExt(b, v_off_start, TypeSizeT, "");
LLVMBuildStore(b, v_off_start, v_offp);
}
/* build the basic block for each attribute, need them as jump target */
for (attnum = 0; attnum < natts; attnum++)
{
attcheckattnoblocks[attnum] =
l_bb_append_v(v_deform_fn, "block.attr.%d.attcheckattno", attnum);
attstartblocks[attnum] =
l_bb_append_v(v_deform_fn, "block.attr.%d.start", attnum);
attisnullblocks[attnum] =
l_bb_append_v(v_deform_fn, "block.attr.%d.attisnull", attnum);
attcheckalignblocks[attnum] =
l_bb_append_v(v_deform_fn, "block.attr.%d.attcheckalign", attnum);
attalignblocks[attnum] =
l_bb_append_v(v_deform_fn, "block.attr.%d.align", attnum);
attstoreblocks[attnum] =
l_bb_append_v(v_deform_fn, "block.attr.%d.store", attnum);
}
/*
* Check if's guaranteed the all the desired attributes are available in
* tuple. If so, we can start deforming. If not, need to make sure to
* fetch the missing columns.
*/
if ((natts - 1) <= guaranteed_column_number)
{
/* just skip through unnecessary blocks */
LLVMBuildBr(b, b_adjust_unavail_cols);
LLVMPositionBuilderAtEnd(b, b_adjust_unavail_cols);
LLVMBuildBr(b, b_find_start);
}
else
{
LLVMValueRef v_params[3];
/* branch if not all columns available */
LLVMBuildCondBr(b,
LLVMBuildICmp(b, LLVMIntULT,
v_maxatt,
l_int16_const(natts),
""),
b_adjust_unavail_cols,
b_find_start);
/* if not, memset tts_isnull of relevant cols to true */
LLVMPositionBuilderAtEnd(b, b_adjust_unavail_cols);
v_params[0] = v_slot;
v_params[1] = LLVMBuildZExt(b, v_maxatt, LLVMInt32Type(), "");
v_params[2] = l_int32_const(natts);
LLVMBuildCall(b, llvm_get_decl(mod, FuncSlotGetmissingattrs),
v_params, lengthof(v_params), "");
LLVMBuildBr(b, b_find_start);
}
LLVMPositionBuilderAtEnd(b, b_find_start);
v_nvalid = LLVMBuildLoad(b, v_nvalidp, "");
/*
* Build switch to go from nvalid to the right startblock. Callers
* currently don't have the knowledge, but it'd be good for performance to
* avoid this check when it's known that the slot is empty (e.g. in scan
* nodes).
*/
if (true)
{
LLVMValueRef v_switch = LLVMBuildSwitch(b, v_nvalid,
b_dead, natts);
for (attnum = 0; attnum < natts; attnum++)
{
LLVMValueRef v_attno = l_int32_const(attnum);
LLVMAddCase(v_switch, v_attno, attcheckattnoblocks[attnum]);
}
}
else
{
/* jump from entry block to first block */
LLVMBuildBr(b, attcheckattnoblocks[0]);
}
LLVMPositionBuilderAtEnd(b, b_dead);
LLVMBuildUnreachable(b);
/*
* Iterate over each attribute that needs to be deformed, build code to
* deform it.
*/
for (attnum = 0; attnum < natts; attnum++)
{
Form_pg_attribute att = TupleDescAttr(desc, attnum);
LLVMValueRef v_incby;
int alignto;
LLVMValueRef l_attno = l_int16_const(attnum);
LLVMValueRef v_attdatap;
LLVMValueRef v_resultp;
/* build block checking whether we did all the necessary attributes */
LLVMPositionBuilderAtEnd(b, attcheckattnoblocks[attnum]);
/*
* If this is the first attribute, slot->tts_nvalid was 0. Therefore
* reset offset to 0 to, it be from a previous execution.
*/
if (attnum == 0)
{
LLVMBuildStore(b, l_sizet_const(0), v_offp);
}
/*
* Build check whether column is available (i.e. whether the tuple has
* that many columns stored). We can avoid the branch if we know
* there's a subsequent NOT NULL column.
*/
if (attnum <= guaranteed_column_number)
{
LLVMBuildBr(b, attstartblocks[attnum]);
}
else
{
LLVMValueRef v_islast;
v_islast = LLVMBuildICmp(b, LLVMIntUGE,
l_attno,
v_maxatt,
"heap_natts");
LLVMBuildCondBr(b, v_islast, b_out, attstartblocks[attnum]);
}
LLVMPositionBuilderAtEnd(b, attstartblocks[attnum]);
/*
* Check for nulls if necessary. No need to take missing attributes
* into account, because in case they're present the heaptuple's natts
* would have indicated that a slot_getmissingattrs() is needed.
*/
if (!att->attnotnull)
{
LLVMBasicBlockRef b_ifnotnull;
LLVMBasicBlockRef b_ifnull;
LLVMBasicBlockRef b_next;
LLVMValueRef v_attisnull;
LLVMValueRef v_nullbyteno;
LLVMValueRef v_nullbytemask;
LLVMValueRef v_nullbyte;
LLVMValueRef v_nullbit;
b_ifnotnull = attcheckalignblocks[attnum];
b_ifnull = attisnullblocks[attnum];
if (attnum + 1 == natts)
b_next = b_out;
else
b_next = attcheckattnoblocks[attnum + 1];
v_nullbyteno = l_int32_const(attnum >> 3);
v_nullbytemask = l_int8_const(1 << ((attnum) & 0x07));
v_nullbyte = l_load_gep1(b, v_bits, v_nullbyteno, "attnullbyte");
v_nullbit = LLVMBuildICmp(b,
LLVMIntEQ,
LLVMBuildAnd(b, v_nullbyte, v_nullbytemask, ""),
l_int8_const(0),
"attisnull");
v_attisnull = LLVMBuildAnd(b, v_hasnulls, v_nullbit, "");
LLVMBuildCondBr(b, v_attisnull, b_ifnull, b_ifnotnull);
LLVMPositionBuilderAtEnd(b, b_ifnull);
/* store null-byte */
LLVMBuildStore(b,
l_int8_const(1),
LLVMBuildGEP(b, v_tts_nulls, &l_attno, 1, ""));
/* store zero datum */
LLVMBuildStore(b,
l_sizet_const(0),
LLVMBuildGEP(b, v_tts_values, &l_attno, 1, ""));
LLVMBuildBr(b, b_next);
attguaranteedalign = false;
}
else
{
/**
* Generate the runtime callable function for the whole fragment pipeline.
*/
static struct lp_fragment_shader_variant *
generate_fragment(struct llvmpipe_context *lp,
struct lp_fragment_shader *shader,
const struct lp_fragment_shader_variant_key *key)
{
struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
struct lp_fragment_shader_variant *variant;
struct lp_type fs_type;
struct lp_type blend_type;
LLVMTypeRef fs_elem_type;
LLVMTypeRef fs_vec_type;
LLVMTypeRef fs_int_vec_type;
LLVMTypeRef blend_vec_type;
LLVMTypeRef blend_int_vec_type;
LLVMTypeRef arg_types[9];
LLVMTypeRef func_type;
LLVMValueRef context_ptr;
LLVMValueRef x;
LLVMValueRef y;
LLVMValueRef a0_ptr;
LLVMValueRef dadx_ptr;
LLVMValueRef dady_ptr;
LLVMValueRef mask_ptr;
LLVMValueRef color_ptr;
LLVMValueRef depth_ptr;
LLVMBasicBlockRef block;
LLVMBuilderRef builder;
LLVMValueRef x0;
LLVMValueRef y0;
struct lp_build_sampler_soa *sampler;
struct lp_build_interp_soa_context interp;
LLVMValueRef fs_mask[LP_MAX_VECTOR_LENGTH];
LLVMValueRef fs_out_color[NUM_CHANNELS][LP_MAX_VECTOR_LENGTH];
LLVMValueRef blend_mask;
LLVMValueRef blend_in_color[NUM_CHANNELS];
unsigned num_fs;
unsigned i;
unsigned chan;
#ifdef DEBUG
tgsi_dump(shader->base.tokens, 0);
if(key->depth.enabled) {
debug_printf("depth.format = %s\n", pf_name(key->zsbuf_format));
debug_printf("depth.func = %s\n", debug_dump_func(key->depth.func, TRUE));
debug_printf("depth.writemask = %u\n", key->depth.writemask);
}
if(key->alpha.enabled) {
debug_printf("alpha.func = %s\n", debug_dump_func(key->alpha.func, TRUE));
debug_printf("alpha.ref_value = %f\n", key->alpha.ref_value);
}
if(key->blend.logicop_enable) {
debug_printf("blend.logicop_func = %u\n", key->blend.logicop_func);
}
else if(key->blend.blend_enable) {
debug_printf("blend.rgb_func = %s\n", debug_dump_blend_func (key->blend.rgb_func, TRUE));
debug_printf("rgb_src_factor = %s\n", debug_dump_blend_factor(key->blend.rgb_src_factor, TRUE));
debug_printf("rgb_dst_factor = %s\n", debug_dump_blend_factor(key->blend.rgb_dst_factor, TRUE));
debug_printf("alpha_func = %s\n", debug_dump_blend_func (key->blend.alpha_func, TRUE));
debug_printf("alpha_src_factor = %s\n", debug_dump_blend_factor(key->blend.alpha_src_factor, TRUE));
debug_printf("alpha_dst_factor = %s\n", debug_dump_blend_factor(key->blend.alpha_dst_factor, TRUE));
}
debug_printf("blend.colormask = 0x%x\n", key->blend.colormask);
for(i = 0; i < PIPE_MAX_SAMPLERS; ++i) {
if(key->sampler[i].format) {
debug_printf("sampler[%u] = \n", i);
debug_printf(" .format = %s\n",
pf_name(key->sampler[i].format));
debug_printf(" .target = %s\n",
debug_dump_tex_target(key->sampler[i].target, TRUE));
debug_printf(" .pot = %u %u %u\n",
key->sampler[i].pot_width,
key->sampler[i].pot_height,
key->sampler[i].pot_depth);
debug_printf(" .wrap = %s %s %s\n",
debug_dump_tex_wrap(key->sampler[i].wrap_s, TRUE),
debug_dump_tex_wrap(key->sampler[i].wrap_t, TRUE),
debug_dump_tex_wrap(key->sampler[i].wrap_r, TRUE));
debug_printf(" .min_img_filter = %s\n",
debug_dump_tex_filter(key->sampler[i].min_img_filter, TRUE));
debug_printf(" .min_mip_filter = %s\n",
debug_dump_tex_mipfilter(key->sampler[i].min_mip_filter, TRUE));
debug_printf(" .mag_img_filter = %s\n",
debug_dump_tex_filter(key->sampler[i].mag_img_filter, TRUE));
if(key->sampler[i].compare_mode)
debug_printf(" .compare_mode = %s\n", debug_dump_func(key->sampler[i].compare_func, TRUE));
debug_printf(" .normalized_coords = %u\n", key->sampler[i].normalized_coords);
debug_printf(" .prefilter = %u\n", key->sampler[i].prefilter);
}
}
#endif
variant = CALLOC_STRUCT(lp_fragment_shader_variant);
if(!variant)
return NULL;
variant->shader = shader;
memcpy(&variant->key, key, sizeof *key);
/* TODO: actually pick these based on the fs and color buffer
* characteristics. */
memset(&fs_type, 0, sizeof fs_type);
fs_type.floating = TRUE; /* floating point values */
fs_type.sign = TRUE; /* values are signed */
fs_type.norm = FALSE; /* values are not limited to [0,1] or [-1,1] */
fs_type.width = 32; /* 32-bit float */
fs_type.length = 4; /* 4 element per vector */
num_fs = 4;
memset(&blend_type, 0, sizeof blend_type);
blend_type.floating = FALSE; /* values are integers */
blend_type.sign = FALSE; /* values are unsigned */
blend_type.norm = TRUE; /* values are in [0,1] or [-1,1] */
blend_type.width = 8; /* 8-bit ubyte values */
blend_type.length = 16; /* 16 elements per vector */
/*
* Generate the function prototype. Any change here must be reflected in
* lp_jit.h's lp_jit_frag_func function pointer type, and vice-versa.
*/
fs_elem_type = lp_build_elem_type(fs_type);
fs_vec_type = lp_build_vec_type(fs_type);
fs_int_vec_type = lp_build_int_vec_type(fs_type);
blend_vec_type = lp_build_vec_type(blend_type);
blend_int_vec_type = lp_build_int_vec_type(blend_type);
arg_types[0] = screen->context_ptr_type; /* context */
arg_types[1] = LLVMInt32Type(); /* x */
arg_types[2] = LLVMInt32Type(); /* y */
arg_types[3] = LLVMPointerType(fs_elem_type, 0); /* a0 */
arg_types[4] = LLVMPointerType(fs_elem_type, 0); /* dadx */
arg_types[5] = LLVMPointerType(fs_elem_type, 0); /* dady */
arg_types[6] = LLVMPointerType(fs_int_vec_type, 0); /* mask */
arg_types[7] = LLVMPointerType(blend_vec_type, 0); /* color */
arg_types[8] = LLVMPointerType(fs_int_vec_type, 0); /* depth */
func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
variant->function = LLVMAddFunction(screen->module, "shader", func_type);
LLVMSetFunctionCallConv(variant->function, LLVMCCallConv);
for(i = 0; i < Elements(arg_types); ++i)
if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
LLVMAddAttribute(LLVMGetParam(variant->function, i), LLVMNoAliasAttribute);
context_ptr = LLVMGetParam(variant->function, 0);
x = LLVMGetParam(variant->function, 1);
y = LLVMGetParam(variant->function, 2);
a0_ptr = LLVMGetParam(variant->function, 3);
dadx_ptr = LLVMGetParam(variant->function, 4);
dady_ptr = LLVMGetParam(variant->function, 5);
mask_ptr = LLVMGetParam(variant->function, 6);
color_ptr = LLVMGetParam(variant->function, 7);
depth_ptr = LLVMGetParam(variant->function, 8);
lp_build_name(context_ptr, "context");
lp_build_name(x, "x");
lp_build_name(y, "y");
lp_build_name(a0_ptr, "a0");
lp_build_name(dadx_ptr, "dadx");
lp_build_name(dady_ptr, "dady");
lp_build_name(mask_ptr, "mask");
lp_build_name(color_ptr, "color");
lp_build_name(depth_ptr, "depth");
/*
* Function body
*/
block = LLVMAppendBasicBlock(variant->function, "entry");
builder = LLVMCreateBuilder();
LLVMPositionBuilderAtEnd(builder, block);
generate_pos0(builder, x, y, &x0, &y0);
lp_build_interp_soa_init(&interp, shader->base.tokens, builder, fs_type,
a0_ptr, dadx_ptr, dady_ptr,
x0, y0, 2, 0);
#if 0
/* C texture sampling */
sampler = lp_c_sampler_soa_create(context_ptr);
#else
/* code generated texture sampling */
sampler = lp_llvm_sampler_soa_create(key->sampler, context_ptr);
#endif
for(i = 0; i < num_fs; ++i) {
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
LLVMValueRef out_color[NUM_CHANNELS];
LLVMValueRef depth_ptr_i;
if(i != 0)
lp_build_interp_soa_update(&interp);
fs_mask[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, mask_ptr, &index, 1, ""), "");
depth_ptr_i = LLVMBuildGEP(builder, depth_ptr, &index, 1, "");
generate_fs(lp, shader, key,
builder,
fs_type,
context_ptr,
i,
&interp,
sampler,
&fs_mask[i],
out_color,
depth_ptr_i);
for(chan = 0; chan < NUM_CHANNELS; ++chan)
fs_out_color[chan][i] = out_color[chan];
}
sampler->destroy(sampler);
/*
* Convert the fs's output color and mask to fit to the blending type.
*/
for(chan = 0; chan < NUM_CHANNELS; ++chan) {
lp_build_conv(builder, fs_type, blend_type,
fs_out_color[chan], num_fs,
&blend_in_color[chan], 1);
lp_build_name(blend_in_color[chan], "color.%c", "rgba"[chan]);
}
lp_build_conv_mask(builder, fs_type, blend_type,
fs_mask, num_fs,
&blend_mask, 1);
/*
* Blending.
*/
generate_blend(&key->blend,
builder,
blend_type,
context_ptr,
blend_mask,
blend_in_color,
color_ptr);
LLVMBuildRetVoid(builder);
LLVMDisposeBuilder(builder);
/*
* Translate the LLVM IR into machine code.
*/
if(LLVMVerifyFunction(variant->function, LLVMPrintMessageAction)) {
LLVMDumpValue(variant->function);
abort();
}
LLVMRunFunctionPassManager(screen->pass, variant->function);
#ifdef DEBUG
LLVMDumpValue(variant->function);
debug_printf("\n");
#endif
variant->jit_function = (lp_jit_frag_func)LLVMGetPointerToGlobal(screen->engine, variant->function);
#ifdef DEBUG
lp_disassemble(variant->jit_function);
#endif
variant->next = shader->variants;
shader->variants = variant;
return variant;
}
/**
* Generate the runtime callable function for the whole fragment pipeline.
* Note that the function which we generate operates on a block of 16
* pixels at at time. The block contains 2x2 quads. Each quad contains
* 2x2 pixels.
*/
static void
generate_fragment(struct llvmpipe_context *lp,
struct lp_fragment_shader *shader,
struct lp_fragment_shader_variant *variant,
unsigned partial_mask)
{
struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
const struct lp_fragment_shader_variant_key *key = &variant->key;
char func_name[256];
struct lp_type fs_type;
struct lp_type blend_type;
LLVMTypeRef fs_elem_type;
LLVMTypeRef fs_int_vec_type;
LLVMTypeRef blend_vec_type;
LLVMTypeRef arg_types[11];
LLVMTypeRef func_type;
LLVMValueRef context_ptr;
LLVMValueRef x;
LLVMValueRef y;
LLVMValueRef a0_ptr;
LLVMValueRef dadx_ptr;
LLVMValueRef dady_ptr;
LLVMValueRef color_ptr_ptr;
LLVMValueRef depth_ptr;
LLVMValueRef mask_input;
LLVMValueRef counter = NULL;
LLVMBasicBlockRef block;
LLVMBuilderRef builder;
struct lp_build_sampler_soa *sampler;
struct lp_build_interp_soa_context interp;
LLVMValueRef fs_mask[LP_MAX_VECTOR_LENGTH];
LLVMValueRef fs_out_color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][LP_MAX_VECTOR_LENGTH];
LLVMValueRef blend_mask;
LLVMValueRef function;
LLVMValueRef facing;
unsigned num_fs;
unsigned i;
unsigned chan;
unsigned cbuf;
/* TODO: actually pick these based on the fs and color buffer
* characteristics. */
memset(&fs_type, 0, sizeof fs_type);
fs_type.floating = TRUE; /* floating point values */
fs_type.sign = TRUE; /* values are signed */
fs_type.norm = FALSE; /* values are not limited to [0,1] or [-1,1] */
fs_type.width = 32; /* 32-bit float */
fs_type.length = 4; /* 4 elements per vector */
num_fs = 4; /* number of quads per block */
memset(&blend_type, 0, sizeof blend_type);
blend_type.floating = FALSE; /* values are integers */
blend_type.sign = FALSE; /* values are unsigned */
blend_type.norm = TRUE; /* values are in [0,1] or [-1,1] */
blend_type.width = 8; /* 8-bit ubyte values */
blend_type.length = 16; /* 16 elements per vector */
/*
* Generate the function prototype. Any change here must be reflected in
* lp_jit.h's lp_jit_frag_func function pointer type, and vice-versa.
*/
fs_elem_type = lp_build_elem_type(fs_type);
fs_int_vec_type = lp_build_int_vec_type(fs_type);
blend_vec_type = lp_build_vec_type(blend_type);
util_snprintf(func_name, sizeof(func_name), "fs%u_variant%u_%s",
shader->no, variant->no, partial_mask ? "partial" : "whole");
arg_types[0] = screen->context_ptr_type; /* context */
arg_types[1] = LLVMInt32Type(); /* x */
arg_types[2] = LLVMInt32Type(); /* y */
arg_types[3] = LLVMFloatType(); /* facing */
arg_types[4] = LLVMPointerType(fs_elem_type, 0); /* a0 */
arg_types[5] = LLVMPointerType(fs_elem_type, 0); /* dadx */
arg_types[6] = LLVMPointerType(fs_elem_type, 0); /* dady */
arg_types[7] = LLVMPointerType(LLVMPointerType(blend_vec_type, 0), 0); /* color */
arg_types[8] = LLVMPointerType(fs_int_vec_type, 0); /* depth */
arg_types[9] = LLVMInt32Type(); /* mask_input */
arg_types[10] = LLVMPointerType(LLVMInt32Type(), 0);/* counter */
func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
function = LLVMAddFunction(screen->module, func_name, func_type);
LLVMSetFunctionCallConv(function, LLVMCCallConv);
variant->function[partial_mask] = function;
/* XXX: need to propagate noalias down into color param now we are
* passing a pointer-to-pointer?
*/
for(i = 0; i < Elements(arg_types); ++i)
if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
LLVMAddAttribute(LLVMGetParam(function, i), LLVMNoAliasAttribute);
context_ptr = LLVMGetParam(function, 0);
x = LLVMGetParam(function, 1);
y = LLVMGetParam(function, 2);
facing = LLVMGetParam(function, 3);
a0_ptr = LLVMGetParam(function, 4);
dadx_ptr = LLVMGetParam(function, 5);
dady_ptr = LLVMGetParam(function, 6);
color_ptr_ptr = LLVMGetParam(function, 7);
depth_ptr = LLVMGetParam(function, 8);
mask_input = LLVMGetParam(function, 9);
lp_build_name(context_ptr, "context");
lp_build_name(x, "x");
lp_build_name(y, "y");
lp_build_name(a0_ptr, "a0");
lp_build_name(dadx_ptr, "dadx");
lp_build_name(dady_ptr, "dady");
lp_build_name(color_ptr_ptr, "color_ptr_ptr");
lp_build_name(depth_ptr, "depth");
lp_build_name(mask_input, "mask_input");
if (key->occlusion_count) {
counter = LLVMGetParam(function, 10);
lp_build_name(counter, "counter");
}
/*
* Function body
*/
block = LLVMAppendBasicBlock(function, "entry");
builder = LLVMCreateBuilder();
LLVMPositionBuilderAtEnd(builder, block);
/*
* The shader input interpolation info is not explicitely baked in the
* shader key, but everything it derives from (TGSI, and flatshade) is
* already included in the shader key.
*/
lp_build_interp_soa_init(&interp,
lp->num_inputs,
lp->inputs,
builder, fs_type,
a0_ptr, dadx_ptr, dady_ptr,
x, y);
/* code generated texture sampling */
sampler = lp_llvm_sampler_soa_create(key->sampler, context_ptr);
/* loop over quads in the block */
for(i = 0; i < num_fs; ++i) {
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
LLVMValueRef out_color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS];
LLVMValueRef depth_ptr_i;
if(i != 0)
lp_build_interp_soa_update(&interp, i);
depth_ptr_i = LLVMBuildGEP(builder, depth_ptr, &index, 1, "");
generate_fs(lp, shader, key,
builder,
fs_type,
context_ptr,
i,
&interp,
sampler,
&fs_mask[i], /* output */
out_color,
depth_ptr_i,
facing,
partial_mask,
mask_input,
counter);
for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++)
for(chan = 0; chan < NUM_CHANNELS; ++chan)
fs_out_color[cbuf][chan][i] = out_color[cbuf][chan];
}
sampler->destroy(sampler);
/* Loop over color outputs / color buffers to do blending.
*/
for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
LLVMValueRef color_ptr;
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), cbuf, 0);
LLVMValueRef blend_in_color[NUM_CHANNELS];
unsigned rt;
/*
* Convert the fs's output color and mask to fit to the blending type.
*/
for(chan = 0; chan < NUM_CHANNELS; ++chan) {
lp_build_conv(builder, fs_type, blend_type,
fs_out_color[cbuf][chan], num_fs,
&blend_in_color[chan], 1);
lp_build_name(blend_in_color[chan], "color%d.%c", cbuf, "rgba"[chan]);
}
if (partial_mask || !variant->opaque) {
lp_build_conv_mask(builder, fs_type, blend_type,
fs_mask, num_fs,
&blend_mask, 1);
} else {
blend_mask = lp_build_const_int_vec(blend_type, ~0);
}
color_ptr = LLVMBuildLoad(builder,
LLVMBuildGEP(builder, color_ptr_ptr, &index, 1, ""),
"");
lp_build_name(color_ptr, "color_ptr%d", cbuf);
/* which blend/colormask state to use */
rt = key->blend.independent_blend_enable ? cbuf : 0;
/*
* Blending.
*/
generate_blend(&key->blend,
rt,
builder,
blend_type,
context_ptr,
blend_mask,
blend_in_color,
color_ptr);
}
#ifdef PIPE_ARCH_X86
/* Avoid corrupting the FPU stack on 32bit OSes. */
lp_build_intrinsic(builder, "llvm.x86.mmx.emms", LLVMVoidType(), NULL, 0);
#endif
LLVMBuildRetVoid(builder);
LLVMDisposeBuilder(builder);
/* Verify the LLVM IR. If invalid, dump and abort */
#ifdef DEBUG
if(LLVMVerifyFunction(function, LLVMPrintMessageAction)) {
if (1)
lp_debug_dump_value(function);
abort();
}
#endif
/* Apply optimizations to LLVM IR */
LLVMRunFunctionPassManager(screen->pass, function);
if (gallivm_debug & GALLIVM_DEBUG_IR) {
/* Print the LLVM IR to stderr */
lp_debug_dump_value(function);
debug_printf("\n");
}
/*
* Translate the LLVM IR into machine code.
*/
{
void *f = LLVMGetPointerToGlobal(screen->engine, function);
variant->jit_function[partial_mask] = (lp_jit_frag_func)pointer_to_func(f);
if (gallivm_debug & GALLIVM_DEBUG_ASM) {
lp_disassemble(f);
}
lp_func_delete_body(function);
}
}
/* create new void type */
struct type *type_new_void(void) {
struct type *r = malloc(sizeof *r);
r->val = TYPE_VOID;
r->llvm_type = LLVMVoidType();
return r;
}
