const struct tgsi_token *
tgsi_emulate(const struct tgsi_token *tokens, unsigned flags)
{
struct tgsi_emulation_context ctx;
struct tgsi_token *newtoks;
int newlen;
if (!(flags & (TGSI_EMU_CLAMP_COLOR_OUTPUTS |
TGSI_EMU_PASSTHROUGH_EDGEFLAG |
TGSI_EMU_FORCE_PERSAMPLE_INTERP)))
return NULL;
memset(&ctx, 0, sizeof(ctx));
ctx.flags = flags;
tgsi_scan_shader(tokens, &ctx.info);
if (flags & TGSI_EMU_FORCE_PERSAMPLE_INTERP)
ctx.base.transform_declaration = transform_decl;
if (flags & (TGSI_EMU_CLAMP_COLOR_OUTPUTS |
TGSI_EMU_PASSTHROUGH_EDGEFLAG))
ctx.base.transform_instruction = transform_instr;
newlen = tgsi_num_tokens(tokens) + 20;
newtoks = tgsi_alloc_tokens(newlen);
if (!newtoks)
return NULL;
tgsi_transform_shader(tokens, newtoks, newlen, &ctx.base);
return newtoks;
}
/**
* Create vertex shader state.
* Called via pipe->create_vs_state()
*/
static void *
cell_create_vs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct cell_context *cell = cell_context(pipe);
struct cell_vertex_shader_state *cvs;
cvs = CALLOC_STRUCT(cell_vertex_shader_state);
if (!cvs)
return NULL;
cvs->shader.tokens = tgsi_dup_tokens(templ->tokens);
if (!cvs->shader.tokens) {
FREE(cvs);
return NULL;
}
tgsi_scan_shader(templ->tokens, &cvs->info);
cvs->draw_data = draw_create_vertex_shader(cell->draw, &cvs->shader);
if (cvs->draw_data == NULL) {
FREE( (void *) cvs->shader.tokens );
FREE( cvs );
return NULL;
}
return cvs;
}
static void *
svga_create_gs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct svga_context *svga = svga_context(pipe);
struct svga_geometry_shader *gs = CALLOC_STRUCT(svga_geometry_shader);
if (!gs)
return NULL;
SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_CREATEGS);
gs->base.tokens = tgsi_dup_tokens(templ->tokens);
/* Collect basic info that we'll need later:
*/
tgsi_scan_shader(gs->base.tokens, &gs->base.info);
gs->draw_shader = draw_create_geometry_shader(svga->swtnl.draw, templ);
gs->base.id = svga->debug.shader_id++;
gs->generic_outputs = svga_get_generic_outputs_mask(&gs->base.info);
/* check for any stream output declarations */
if (templ->stream_output.num_outputs) {
gs->base.stream_output = svga_create_stream_output(svga, &gs->base,
&templ->stream_output);
}
SVGA_STATS_TIME_POP(svga_sws(svga));
return gs;
}
static void *
svga_create_fs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct svga_context *svga = svga_context(pipe);
struct svga_fragment_shader *fs;
fs = CALLOC_STRUCT(svga_fragment_shader);
if (!fs)
return NULL;
fs->base.tokens = tgsi_dup_tokens(templ->tokens);
/* Collect basic info that we'll need later:
*/
tgsi_scan_shader(fs->base.tokens, &fs->base.info);
fs->base.id = svga->debug.shader_id++;
fs->generic_inputs = svga_get_generic_inputs_mask(&fs->base.info);
svga_remap_generics(fs->generic_inputs, fs->generic_remap_table);
fs->draw_shader = draw_create_fragment_shader(svga->swtnl.draw, templ);
if (SVGA_DEBUG & DEBUG_TGSI || 0) {
debug_printf("%s id: %u, inputs: %u, outputs: %u\n",
__FUNCTION__, fs->base.id,
fs->base.info.num_inputs, fs->base.info.num_outputs);
}
return fs;
}
static void *
svga_create_fs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct svga_context *svga = svga_context(pipe);
struct svga_screen *svgascreen = svga_screen(pipe->screen);
struct svga_fragment_shader *fs;
fs = CALLOC_STRUCT(svga_fragment_shader);
if (!fs)
return NULL;
fs->base.tokens = tgsi_dup_tokens(templ->tokens);
/* Collect basic info that we'll need later:
*/
tgsi_scan_shader(fs->base.tokens, &fs->base.info);
fs->base.id = svga->debug.shader_id++;
fs->base.use_sm30 = svgascreen->use_ps30;
if (SVGA_DEBUG & DEBUG_TGSI || 0) {
debug_printf("%s id: %u, inputs: %u, outputs: %u\n",
__FUNCTION__, fs->base.id,
fs->base.info.num_inputs, fs->base.info.num_outputs);
}
return fs;
}
LLVMModuleRef r600_tgsi_llvm(
struct radeon_llvm_context * ctx,
const struct tgsi_token * tokens)
{
struct tgsi_shader_info shader_info;
struct lp_build_tgsi_context * bld_base = &ctx->soa.bld_base;
radeon_llvm_context_init(ctx);
tgsi_scan_shader(tokens, &shader_info);
bld_base->info = &shader_info;
bld_base->userdata = ctx;
bld_base->emit_fetch_funcs[TGSI_FILE_CONSTANT] = llvm_fetch_const;
bld_base->emit_prologue = llvm_emit_prologue;
bld_base->emit_epilogue = llvm_emit_epilogue;
ctx->userdata = ctx;
ctx->load_input = llvm_load_input;
bld_base->op_actions[TGSI_OPCODE_DP2] = dot_action;
bld_base->op_actions[TGSI_OPCODE_DP3] = dot_action;
bld_base->op_actions[TGSI_OPCODE_DP4] = dot_action;
bld_base->op_actions[TGSI_OPCODE_DPH] = dot_action;
bld_base->op_actions[TGSI_OPCODE_TEX].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXB].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXD].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXL].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXP].fetch_args = txp_fetch_args;
bld_base->op_actions[TGSI_OPCODE_TXP].emit = llvm_emit_tex;
lp_build_tgsi_llvm(bld_base, tokens);
radeon_llvm_finalize_module(ctx);
return ctx->gallivm.module;
}
LLVMModuleRef r600_tgsi_llvm(
struct radeon_llvm_context * ctx,
const struct tgsi_token * tokens)
{
struct tgsi_shader_info shader_info;
struct lp_build_tgsi_context * bld_base = &ctx->soa.bld_base;
radeon_llvm_context_init(ctx);
#if HAVE_LLVM >= 0x0304
LLVMTypeRef Arguments[32];
unsigned ArgumentsCount = 0;
for (unsigned i = 0; i < ctx->inputs_count; i++)
Arguments[ArgumentsCount++] = LLVMVectorType(bld_base->base.elem_type, 4);
radeon_llvm_create_func(ctx, Arguments, ArgumentsCount);
for (unsigned i = 0; i < ctx->inputs_count; i++) {
LLVMValueRef P = LLVMGetParam(ctx->main_fn, i);
LLVMAddAttribute(P, LLVMInRegAttribute);
}
#else
radeon_llvm_create_func(ctx, NULL, 0);
#endif
tgsi_scan_shader(tokens, &shader_info);
bld_base->info = &shader_info;
bld_base->userdata = ctx;
bld_base->emit_fetch_funcs[TGSI_FILE_CONSTANT] = llvm_fetch_const;
bld_base->emit_prologue = llvm_emit_prologue;
bld_base->emit_epilogue = llvm_emit_epilogue;
ctx->userdata = ctx;
ctx->load_input = llvm_load_input;
ctx->load_system_value = llvm_load_system_value;
bld_base->op_actions[TGSI_OPCODE_DP2] = dot_action;
bld_base->op_actions[TGSI_OPCODE_DP3] = dot_action;
bld_base->op_actions[TGSI_OPCODE_DP4] = dot_action;
bld_base->op_actions[TGSI_OPCODE_DPH] = dot_action;
bld_base->op_actions[TGSI_OPCODE_DDX].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_DDY].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TEX].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TEX2].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXB].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXB2].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXD].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXL].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXL2].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXF].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXQ].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_TXP].emit = llvm_emit_tex;
bld_base->op_actions[TGSI_OPCODE_CMP].emit = emit_cndlt;
lp_build_tgsi_llvm(bld_base, tokens);
radeon_llvm_finalize_module(ctx);
return ctx->gallivm.module;
}
static void *
nv50_gp_state_create(struct pipe_context *pipe,
const struct pipe_shader_state *cso)
{
struct nv50_program *p = CALLOC_STRUCT(nv50_program);
p->pipe.tokens = tgsi_dup_tokens(cso->tokens);
p->type = PIPE_SHADER_GEOMETRY;
tgsi_scan_shader(p->pipe.tokens, &p->info);
return (void *)p;
}
static int
nv30_fp(int chipset, struct tgsi_token tokens[],
unsigned *size, unsigned **code) {
struct nv30_fragprog fp;
memset(&fp, 0, sizeof(fp));
fp.pipe.tokens = tokens;
tgsi_scan_shader(fp.pipe.tokens, &fp.info);
_nvfx_fragprog_translate(chipset >= 0x40 ? 0x4097 : 0x3097, &fp);
*size = fp.insn_len * 4;
*code = fp.insn;
return !fp.translated;
}
struct draw_geometry_shader *
draw_create_geometry_shader(struct draw_context *draw,
const struct pipe_shader_state *state)
{
struct draw_geometry_shader *gs;
int i;
gs = CALLOC_STRUCT(draw_geometry_shader);
if (!gs)
return NULL;
gs->draw = draw;
gs->state = *state;
gs->state.tokens = tgsi_dup_tokens(state->tokens);
if (!gs->state.tokens) {
FREE(gs);
return NULL;
}
tgsi_scan_shader(state->tokens, &gs->info);
/* setup the defaults */
gs->input_primitive = PIPE_PRIM_TRIANGLES;
gs->output_primitive = PIPE_PRIM_TRIANGLE_STRIP;
gs->max_output_vertices = 32;
for (i = 0; i < gs->info.num_properties; ++i) {
if (gs->info.properties[i].name ==
TGSI_PROPERTY_GS_INPUT_PRIM)
gs->input_primitive = gs->info.properties[i].data[0];
else if (gs->info.properties[i].name ==
TGSI_PROPERTY_GS_OUTPUT_PRIM)
gs->output_primitive = gs->info.properties[i].data[0];
else if (gs->info.properties[i].name ==
TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES)
gs->max_output_vertices = gs->info.properties[i].data[0];
}
gs->machine = draw->gs.tgsi.machine;
if (gs)
{
uint i;
for (i = 0; i < gs->info.num_outputs; i++) {
if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_POSITION &&
gs->info.output_semantic_index[i] == 0)
gs->position_output = i;
}
}
return gs;
}
static void *
nv30_vp_state_create(struct pipe_context *pipe,
const struct pipe_shader_state *cso)
{
struct nv30_vertprog *vp = CALLOC_STRUCT(nv30_vertprog);
if (!vp)
return NULL;
vp->pipe.tokens = tgsi_dup_tokens(cso->tokens);
tgsi_scan_shader(vp->pipe.tokens, &vp->info);
return vp;
}
static void *
nv30_fp_state_create(struct pipe_context *pipe,
const struct pipe_shader_state *cso)
{
struct nv30_fragprog *fp = CALLOC_STRUCT(nv30_fragprog);
if (!fp)
return NULL;
fp->pipe.tokens = tgsi_dup_tokens(cso->tokens);
tgsi_scan_shader(fp->pipe.tokens, &fp->info);
return fp;
}
static void *
nv40_fp_state_create(struct pipe_context *pipe,
const struct pipe_shader_state *cso)
{
struct nv40_fragment_program *fp;
fp = CALLOC(1, sizeof(struct nv40_fragment_program));
fp->pipe.tokens = tgsi_dup_tokens(cso->tokens);
tgsi_scan_shader(fp->pipe.tokens, &fp->info);
return (void *)fp;
}
static int
nv30_vp(int chipset, struct tgsi_token tokens[],
unsigned *size, unsigned **code) {
struct nv30_vertprog vp;
memset(&vp, 0, sizeof(vp));
vp.pipe.tokens = tokens;
tgsi_scan_shader(vp.pipe.tokens, &vp.info);
_nvfx_vertprog_translate(chipset >= 0x40 ? 0x4097 : 0x3097, &vp);
*size = vp.nr_insns * 16;
*code = (unsigned *)vp.insns;
return !vp.translated;
}
struct draw_fragment_shader *
draw_create_fragment_shader(struct draw_context *draw,
const struct pipe_shader_state *shader)
{
struct draw_fragment_shader *dfs;
dfs = CALLOC_STRUCT(draw_fragment_shader);
if (dfs) {
dfs->base = *shader;
tgsi_scan_shader(shader->tokens, &dfs->info);
}
return dfs;
}
/* Create fragment shader state. */
static void* r300_create_fs_state(struct pipe_context* pipe,
const struct pipe_shader_state* shader)
{
struct r300_fragment_shader* fs = NULL;
fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader);
/* Copy state directly into shader. */
fs->state = *shader;
fs->state.tokens = tgsi_dup_tokens(shader->tokens);
tgsi_scan_shader(shader->tokens, &fs->info);
return (void*)fs;
}
/**
* Create a new fragment shader variant.
*/
static struct sp_fragment_shader_variant *
create_fs_variant(struct softpipe_context *softpipe,
struct sp_fragment_shader *fs,
const struct sp_fragment_shader_variant_key *key)
{
struct sp_fragment_shader_variant *var;
struct pipe_shader_state *curfs = &fs->shader;
/* codegen, create variant object */
var = softpipe_create_fs_variant_exec(softpipe);
if (var) {
var->key = *key;
#if DO_PSTIPPLE_IN_HELPER_MODULE
if (key->polygon_stipple) {
/* get new shader that implements polygon stippling */
var->tokens =
util_pstipple_create_fragment_shader(curfs->tokens,
&var->stipple_sampler_unit, 0,
TGSI_FILE_INPUT);
}
else
#endif
{
var->tokens = tgsi_dup_tokens(curfs->tokens);
var->stipple_sampler_unit = 0;
}
tgsi_scan_shader(var->tokens, &var->info);
/* See comments elsewhere about draw fragment shaders */
#if 0
/* draw's fs state */
var->draw_shader = draw_create_fragment_shader(softpipe->draw,
&fs->shader);
if (!var->draw_shader) {
var->delete(var);
FREE((void *) var->tokens);
return NULL;
}
#endif
/* insert variant into linked list */
var->next = fs->variants;
fs->variants = var;
}
static void *brw_create_fs_state( struct pipe_context *pipe,
const struct pipe_shader_state *shader )
{
struct brw_context *brw = brw_context(pipe);
struct brw_fragment_shader *fs;
int i;
fs = CALLOC_STRUCT(brw_fragment_shader);
if (fs == NULL)
return NULL;
/* Duplicate tokens, scan shader
*/
fs->id = brw->program_id++;
fs->has_flow_control = has_flow_control(&fs->info);
fs->tokens = tgsi_dup_tokens(shader->tokens);
if (fs->tokens == NULL)
goto fail;
tgsi_scan_shader(fs->tokens, &fs->info);
scan_immediates(fs->tokens, &fs->info, &fs->immediates);
fs->signature.nr_inputs = fs->info.num_inputs;
for (i = 0; i < fs->info.num_inputs; i++) {
fs->signature.input[i].interp = fs->info.input_interpolate[i];
fs->signature.input[i].semantic = fs->info.input_semantic_name[i];
fs->signature.input[i].semantic_index = fs->info.input_semantic_index[i];
}
for (i = 0; i < fs->info.num_inputs; i++)
if (fs->info.input_semantic_name[i] == TGSI_SEMANTIC_POSITION)
fs->uses_depth = 1;
if (fs->info.uses_kill)
fs->iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
if (fs->info.writes_z)
fs->iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
return (void *)fs;
fail:
FREE(fs);
return NULL;
}
static void *
i915_create_fs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct i915_context *i915 = i915_context(pipe);
struct i915_fragment_shader *ifs = CALLOC_STRUCT(i915_fragment_shader);
if (!ifs)
return NULL;
ifs->state.tokens = tgsi_dup_tokens(templ->tokens);
tgsi_scan_shader(templ->tokens, &ifs->info);
/* The shader's compiled to i915 instructions here */
i915_translate_fragment_program(i915, ifs);
return ifs;
}
static unsigned
dd_num_active_viewports(struct dd_context *dctx)
{
struct tgsi_shader_info info;
const struct tgsi_token *tokens;
if (dctx->shaders[PIPE_SHADER_GEOMETRY])
tokens = dctx->shaders[PIPE_SHADER_GEOMETRY]->state.shader.tokens;
else if (dctx->shaders[PIPE_SHADER_TESS_EVAL])
tokens = dctx->shaders[PIPE_SHADER_TESS_EVAL]->state.shader.tokens;
else if (dctx->shaders[PIPE_SHADER_VERTEX])
tokens = dctx->shaders[PIPE_SHADER_VERTEX]->state.shader.tokens;
else
return 1;
tgsi_scan_shader(tokens, &info);
return info.writes_viewport_index ? PIPE_MAX_VIEWPORTS : 1;
}
void *
llvmpipe_create_fs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct lp_fragment_shader *shader;
shader = CALLOC_STRUCT(lp_fragment_shader);
if (!shader)
return NULL;
/* get/save the summary info for this shader */
tgsi_scan_shader(templ->tokens, &shader->info);
/* we need to keep a local copy of the tokens */
shader->base.tokens = tgsi_dup_tokens(templ->tokens);
return shader;
}
static void *
llvmpipe_create_fs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct lp_fragment_shader *shader;
int nr_samplers;
shader = CALLOC_STRUCT(lp_fragment_shader);
if (!shader)
return NULL;
shader->no = fs_no++;
make_empty_list(&shader->variants);
/* get/save the summary info for this shader */
tgsi_scan_shader(templ->tokens, &shader->info);
/* we need to keep a local copy of the tokens */
shader->base.tokens = tgsi_dup_tokens(templ->tokens);
nr_samplers = shader->info.file_max[TGSI_FILE_SAMPLER] + 1;
shader->variant_key_size = Offset(struct lp_fragment_shader_variant_key,
sampler[nr_samplers]);
if (LP_DEBUG & DEBUG_TGSI) {
unsigned attrib;
debug_printf("llvmpipe: Create fragment shader #%u %p:\n", shader->no, (void *) shader);
tgsi_dump(templ->tokens, 0);
debug_printf("usage masks:\n");
for (attrib = 0; attrib < shader->info.num_inputs; ++attrib) {
unsigned usage_mask = shader->info.input_usage_mask[attrib];
debug_printf(" IN[%u].%s%s%s%s\n",
attrib,
usage_mask & TGSI_WRITEMASK_X ? "x" : "",
usage_mask & TGSI_WRITEMASK_Y ? "y" : "",
usage_mask & TGSI_WRITEMASK_Z ? "z" : "",
usage_mask & TGSI_WRITEMASK_W ? "w" : "");
}
debug_printf("\n");
}
return shader;
}
static void *
svga_create_vs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct svga_context *svga = svga_context(pipe);
struct svga_vertex_shader *vs = CALLOC_STRUCT(svga_vertex_shader);
if (!vs)
return NULL;
SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_CREATEVS);
/* substitute a debug shader?
*/
vs->base.tokens = tgsi_dup_tokens(substitute_vs(svga->debug.shader_id,
templ->tokens));
/* Collect basic info that we'll need later:
*/
tgsi_scan_shader(vs->base.tokens, &vs->base.info);
{
/* Need to do construct a new template in case we substitued a
* debug shader.
*/
struct pipe_shader_state tmp2 = *templ;
tmp2.tokens = vs->base.tokens;
vs->draw_shader = draw_create_vertex_shader(svga->swtnl.draw, &tmp2);
}
vs->base.id = svga->debug.shader_id++;
vs->generic_outputs = svga_get_generic_outputs_mask(&vs->base.info);
/* check for any stream output declarations */
if (templ->stream_output.num_outputs) {
vs->base.stream_output = svga_create_stream_output(svga, &vs->base,
&templ->stream_output);
}
SVGA_STATS_TIME_POP(svga_sws(svga));
return vs;
}
static void* r300_create_vs_state(struct pipe_context* pipe,
const struct pipe_shader_state* shader)
{
struct r300_context* r300 = r300_context(pipe);
if (r300_screen(pipe->screen)->caps->has_tcl) {
struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
/* Copy state directly into shader. */
vs->state = *shader;
vs->state.tokens = tgsi_dup_tokens(shader->tokens);
tgsi_scan_shader(shader->tokens, &vs->info);
/* Appease Draw. */
vs->draw = draw_create_vertex_shader(r300->draw, shader);
return (void*)vs;
} else {
return draw_create_vertex_shader(r300->draw, shader);
}
}
static void *
svga_create_vs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct svga_context *svga = svga_context(pipe);
struct svga_screen *svgascreen = svga_screen(pipe->screen);
struct svga_vertex_shader *vs = CALLOC_STRUCT(svga_vertex_shader);
if (!vs)
return NULL;
/* substitute a debug shader?
*/
vs->base.tokens = tgsi_dup_tokens(substitute_vs(svga->debug.shader_id,
templ->tokens));
/* Collect basic info that we'll need later:
*/
tgsi_scan_shader(vs->base.tokens, &vs->base.info);
{
/* Need to do construct a new template in case we substitued a
* debug shader.
*/
struct pipe_shader_state tmp2 = *templ;
tmp2.tokens = vs->base.tokens;
vs->draw_shader = draw_create_vertex_shader(svga->swtnl.draw, &tmp2);
}
vs->base.id = svga->debug.shader_id++;
vs->base.use_sm30 = svgascreen->use_vs30;
if (SVGA_DEBUG & DEBUG_TGSI || 0) {
debug_printf("%s id: %u, inputs: %u, outputs: %u\n",
__FUNCTION__, vs->base.id,
vs->base.info.num_inputs, vs->base.info.num_outputs);
}
return vs;
}
/*
* A post-process step in the draw call to fix texture targets and
* insert code for fog.
*/
const struct tgsi_token *
st_fixup_atifs(const struct tgsi_token *tokens,
const struct st_fp_variant_key *key)
{
struct tgsi_atifs_transform ctx;
struct tgsi_token *newtoks;
int newlen;
memset(&ctx, 0, sizeof(ctx));
ctx.base.transform_declaration = transform_decl;
ctx.base.transform_instruction = transform_instr;
ctx.key = key;
tgsi_scan_shader(tokens, &ctx.info);
newlen = tgsi_num_tokens(tokens) + 30;
newtoks = tgsi_alloc_tokens(newlen);
if (!newtoks)
return NULL;
tgsi_transform_shader(tokens, newtoks, newlen, &ctx.base);
return newtoks;
}
/**
* Create fragment shader state.
* Called via pipe->create_fs_state()
*/
static void *
cell_create_fs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct cell_context *cell = cell_context(pipe);
struct cell_fragment_shader_state *cfs;
cfs = CALLOC_STRUCT(cell_fragment_shader_state);
if (!cfs)
return NULL;
cfs->shader.tokens = tgsi_dup_tokens(templ->tokens);
if (!cfs->shader.tokens) {
FREE(cfs);
return NULL;
}
tgsi_scan_shader(templ->tokens, &cfs->info);
cell_gen_fragment_program(cell, cfs->shader.tokens, &cfs->code);
return cfs;
}
struct draw_geometry_shader *
draw_create_geometry_shader(struct draw_context *draw,
const struct pipe_shader_state *state)
{
#ifdef HAVE_LLVM
boolean use_llvm = draw_get_option_use_llvm();
struct llvm_geometry_shader *llvm_gs;
#endif
struct draw_geometry_shader *gs;
unsigned i;
#ifdef HAVE_LLVM
if (use_llvm) {
llvm_gs = CALLOC_STRUCT(llvm_geometry_shader);
if (llvm_gs == NULL)
return NULL;
gs = &llvm_gs->base;
make_empty_list(&llvm_gs->variants);
} else
#endif
{
gs = CALLOC_STRUCT(draw_geometry_shader);
}
if (!gs)
return NULL;
gs->draw = draw;
gs->state = *state;
gs->state.tokens = tgsi_dup_tokens(state->tokens);
if (!gs->state.tokens) {
FREE(gs);
return NULL;
}
tgsi_scan_shader(state->tokens, &gs->info);
/* setup the defaults */
gs->input_primitive = PIPE_PRIM_TRIANGLES;
gs->output_primitive = PIPE_PRIM_TRIANGLE_STRIP;
gs->max_output_vertices = 32;
gs->max_out_prims = 0;
#ifdef HAVE_LLVM
if (use_llvm) {
/* TODO: change the input array to handle the following
vector length, instead of the currently hardcoded
TGSI_NUM_CHANNELS
gs->vector_length = lp_native_vector_width / 32;*/
gs->vector_length = TGSI_NUM_CHANNELS;
} else
#endif
{
gs->vector_length = 1;
}
for (i = 0; i < gs->info.num_properties; ++i) {
if (gs->info.properties[i].name ==
TGSI_PROPERTY_GS_INPUT_PRIM)
gs->input_primitive = gs->info.properties[i].data[0];
else if (gs->info.properties[i].name ==
TGSI_PROPERTY_GS_OUTPUT_PRIM)
gs->output_primitive = gs->info.properties[i].data[0];
else if (gs->info.properties[i].name ==
TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES)
gs->max_output_vertices = gs->info.properties[i].data[0];
}
/* Primitive boundary is bigger than max_output_vertices by one, because
* the specification says that the geometry shader should exit if the
* number of emitted vertices is bigger or equal to max_output_vertices and
* we can't do that because we're running in the SoA mode, which means that
* our storing routines will keep getting called on channels that have
* overflown.
* So we need some scratch area where we can keep writing the overflown
* vertices without overwriting anything important or crashing.
*/
gs->primitive_boundary = gs->max_output_vertices + 1;
for (i = 0; i < gs->info.num_outputs; i++) {
if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_POSITION &&
gs->info.output_semantic_index[i] == 0)
gs->position_output = i;
if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_VIEWPORT_INDEX)
gs->viewport_index_output = i;
if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_CLIPDIST) {
debug_assert(gs->info.output_semantic_index[i] <
PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT);
gs->clipdistance_output[gs->info.output_semantic_index[i]] = i;
}
if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_CULLDIST) {
debug_assert(gs->info.output_semantic_index[i] <
PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT);
gs->culldistance_output[gs->info.output_semantic_index[i]] = i;
}
}
gs->machine = draw->gs.tgsi.machine;
#ifdef HAVE_LLVM
if (use_llvm) {
int vector_size = gs->vector_length * sizeof(float);
gs->gs_input = align_malloc(sizeof(struct draw_gs_inputs), 16);
memset(gs->gs_input, 0, sizeof(struct draw_gs_inputs));
gs->llvm_prim_lengths = 0;
gs->llvm_emitted_primitives = align_malloc(vector_size, vector_size);
gs->llvm_emitted_vertices = align_malloc(vector_size, vector_size);
gs->llvm_prim_ids = align_malloc(vector_size, vector_size);
gs->fetch_outputs = llvm_fetch_gs_outputs;
gs->fetch_inputs = llvm_fetch_gs_input;
gs->prepare = llvm_gs_prepare;
gs->run = llvm_gs_run;
gs->jit_context = &draw->llvm->gs_jit_context;
llvm_gs->variant_key_size =
draw_gs_llvm_variant_key_size(
MAX2(gs->info.file_max[TGSI_FILE_SAMPLER]+1,
gs->info.file_max[TGSI_FILE_SAMPLER_VIEW]+1));
} else
#endif
{
gs->fetch_outputs = tgsi_fetch_gs_outputs;
gs->fetch_inputs = tgsi_fetch_gs_input;
gs->prepare = tgsi_gs_prepare;
gs->run = tgsi_gs_run;
}
return gs;
}
static void *si_create_compute_state(
struct pipe_context *ctx,
const struct pipe_compute_state *cso)
{
struct si_context *sctx = (struct si_context *)ctx;
struct si_screen *sscreen = (struct si_screen *)ctx->screen;
struct si_compute *program = CALLOC_STRUCT(si_compute);
struct si_shader *shader = &program->shader;
program->ir_type = cso->ir_type;
program->local_size = cso->req_local_mem;
program->private_size = cso->req_private_mem;
program->input_size = cso->req_input_mem;
program->use_code_object_v2 = HAVE_LLVM >= 0x0400 &&
cso->ir_type == PIPE_SHADER_IR_NATIVE;
if (cso->ir_type == PIPE_SHADER_IR_TGSI) {
struct si_shader_selector sel;
bool scratch_enabled;
memset(&sel, 0, sizeof(sel));
sel.tokens = tgsi_dup_tokens(cso->prog);
if (!sel.tokens) {
FREE(program);
return NULL;
}
tgsi_scan_shader(cso->prog, &sel.info);
sel.type = PIPE_SHADER_COMPUTE;
sel.local_size = cso->req_local_mem;
p_atomic_inc(&sscreen->b.num_shaders_created);
program->shader.selector = &sel;
if (si_shader_create(sscreen, sctx->tm, &program->shader,
&sctx->b.debug)) {
FREE(sel.tokens);
FREE(program);
return NULL;
}
scratch_enabled = shader->config.scratch_bytes_per_wave > 0;
shader->config.rsrc1 =
S_00B848_VGPRS((shader->config.num_vgprs - 1) / 4) |
S_00B848_SGPRS((shader->config.num_sgprs - 1) / 8) |
S_00B848_DX10_CLAMP(1) |
S_00B848_FLOAT_MODE(shader->config.float_mode);
shader->config.rsrc2 = S_00B84C_USER_SGPR(SI_CS_NUM_USER_SGPR) |
S_00B84C_SCRATCH_EN(scratch_enabled) |
S_00B84C_TGID_X_EN(1) | S_00B84C_TGID_Y_EN(1) |
S_00B84C_TGID_Z_EN(1) | S_00B84C_TIDIG_COMP_CNT(2) |
S_00B84C_LDS_SIZE(shader->config.lds_size);
FREE(sel.tokens);
} else {
const struct pipe_llvm_program_header *header;
const char *code;
header = cso->prog;
code = cso->prog + sizeof(struct pipe_llvm_program_header);
radeon_elf_read(code, header->num_bytes, &program->shader.binary);
if (program->use_code_object_v2) {
const amd_kernel_code_t *code_object =
si_compute_get_code_object(program, 0);
code_object_to_config(code_object, &program->shader.config);
} else {
si_shader_binary_read_config(&program->shader.binary,
&program->shader.config, 0);
}
si_shader_dump(sctx->screen, &program->shader, &sctx->b.debug,
PIPE_SHADER_COMPUTE, stderr);
si_shader_binary_upload(sctx->screen, &program->shader);
}
return program;
}
/**
* Detect any direct relationship between the output color
*/
void
lp_build_tgsi_info(const struct tgsi_token *tokens,
struct lp_tgsi_info *info)
{
struct tgsi_parse_context parse;
struct analysis_context ctx;
unsigned index;
unsigned chan;
memset(info, 0, sizeof *info);
tgsi_scan_shader(tokens, &info->base);
memset(&ctx, 0, sizeof ctx);
ctx.info = info;
tgsi_parse_init(&parse, tokens);
while (!tgsi_parse_end_of_tokens(&parse)) {
tgsi_parse_token(&parse);
switch (parse.FullToken.Token.Type) {
case TGSI_TOKEN_TYPE_DECLARATION:
break;
case TGSI_TOKEN_TYPE_INSTRUCTION:
{
struct tgsi_full_instruction *inst =
&parse.FullToken.FullInstruction;
if (inst->Instruction.Opcode == TGSI_OPCODE_END ||
inst->Instruction.Opcode == TGSI_OPCODE_BGNSUB) {
/* We reached the end of main function body. */
goto finished;
}
analyse_instruction(&ctx, inst);
}
break;
case TGSI_TOKEN_TYPE_IMMEDIATE:
{
const unsigned size =
parse.FullToken.FullImmediate.Immediate.NrTokens - 1;
assert(size <= 4);
if (ctx.num_imms < Elements(ctx.imm)) {
for (chan = 0; chan < size; ++chan) {
float value = parse.FullToken.FullImmediate.u[chan].Float;
ctx.imm[ctx.num_imms][chan] = value;
if (value < 0.0f || value > 1.0f) {
info->unclamped_immediates = TRUE;
}
}
++ctx.num_imms;
}
}
break;
case TGSI_TOKEN_TYPE_PROPERTY:
break;
default:
assert(0);
}
}
finished:
tgsi_parse_free(&parse);
/*
* Link the output color values.
*/
for (index = 0; index < PIPE_MAX_COLOR_BUFS; ++index) {
const struct lp_tgsi_channel_info null_output[4];
info->cbuf[index] = null_output;
}
for (index = 0; index < info->base.num_outputs; ++index) {
unsigned semantic_name = info->base.output_semantic_name[index];
unsigned semantic_index = info->base.output_semantic_index[index];
if (semantic_name == TGSI_SEMANTIC_COLOR &&
semantic_index < PIPE_MAX_COLOR_BUFS) {
info->cbuf[semantic_index] = info->output[index];
}
}
if (gallivm_debug & GALLIVM_DEBUG_TGSI) {
dump_info(tokens, info);
}
}
