#include "program/prog_optimize.h"
#include "program/prog_print.h"
#include "program/program.h"
-#include "program/prog_uniform.h"
#include "program/prog_parameter.h"
#include "program/sampler.h"
(1 << PROGRAM_CONSTANT) | \
(1 << PROGRAM_UNIFORM))
+/**
+ * Maximum number of temporary registers.
+ *
+ * It is too big for stack allocated arrays -- it will cause stack overflow on
+ * Windows and likely Mac OS X.
+ */
#define MAX_TEMPS 4096
/* will be 4 for GLSL 4.00 */
int samplers_used;
bool indirect_addr_temps;
bool indirect_addr_consts;
+ int num_clip_distances;
int glsl_version;
bool native_integers;
fp->OriginUpperLeft = ir->origin_upper_left;
fp->PixelCenterInteger = ir->pixel_center_integer;
-
- } else if (strcmp(ir->name, "gl_FragDepth") == 0) {
- struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog;
- switch (ir->depth_layout) {
- case ir_depth_layout_none:
- fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_NONE;
- break;
- case ir_depth_layout_any:
- fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_ANY;
- break;
- case ir_depth_layout_greater:
- fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_GREATER;
- break;
- case ir_depth_layout_less:
- fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_LESS;
- break;
- case ir_depth_layout_unchanged:
- fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_UNCHANGED;
- break;
- default:
- assert(0);
- break;
- }
}
if (ir->mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) {
case ir_unop_floor:
emit(ir, TGSI_OPCODE_FLR, result_dst, op[0]);
break;
+ case ir_unop_round_even:
+ emit(ir, TGSI_OPCODE_ROUND, result_dst, op[0]);
+ break;
case ir_unop_fract:
emit(ir, TGSI_OPCODE_FRC, result_dst, op[0]);
break;
}
case ir_binop_lshift:
if (native_integers) {
- emit(ir, TGSI_OPCODE_SHL, result_dst, op[0]);
+ emit(ir, TGSI_OPCODE_SHL, result_dst, op[0], op[1]);
break;
}
case ir_binop_rshift:
if (native_integers) {
- emit(ir, TGSI_OPCODE_ISHR, result_dst, op[0]);
+ emit(ir, TGSI_OPCODE_ISHR, result_dst, op[0], op[1]);
break;
}
case ir_binop_bit_and:
if (native_integers) {
- emit(ir, TGSI_OPCODE_AND, result_dst, op[0]);
+ emit(ir, TGSI_OPCODE_AND, result_dst, op[0], op[1]);
break;
}
case ir_binop_bit_xor:
if (native_integers) {
- emit(ir, TGSI_OPCODE_XOR, result_dst, op[0]);
+ emit(ir, TGSI_OPCODE_XOR, result_dst, op[0], op[1]);
break;
}
case ir_binop_bit_or:
if (native_integers) {
- emit(ir, TGSI_OPCODE_OR, result_dst, op[0]);
+ emit(ir, TGSI_OPCODE_OR, result_dst, op[0], op[1]);
break;
}
- case ir_unop_round_even:
+
assert(!"GLSL 1.30 features unsupported");
break;
case GLSL_SAMPLER_DIM_BUF:
assert(!"FINISHME: Implement ARB_texture_buffer_object");
break;
+ case GLSL_SAMPLER_DIM_EXTERNAL:
+ inst->tex_target = TEXTURE_EXTERNAL_INDEX;
+ break;
default:
assert(!"Should not get here.");
}
_mesa_update_shader_textures_used(prog);
}
-
-/**
- * Check if the given vertex/fragment/shader program is within the
- * resource limits of the context (number of texture units, etc).
- * If any of those checks fail, record a linker error.
- *
- * XXX more checks are needed...
- */
-static void
-check_resources(const struct gl_context *ctx,
- struct gl_shader_program *shader_program,
- glsl_to_tgsi_visitor *prog,
- struct gl_program *proginfo)
-{
- switch (proginfo->Target) {
- case GL_VERTEX_PROGRAM_ARB:
- if (_mesa_bitcount(prog->samplers_used) >
- ctx->Const.MaxVertexTextureImageUnits) {
- fail_link(shader_program, "Too many vertex shader texture samplers");
- }
- if (proginfo->Parameters->NumParameters > MAX_UNIFORMS) {
- fail_link(shader_program, "Too many vertex shader constants");
- }
- break;
- case MESA_GEOMETRY_PROGRAM:
- if (_mesa_bitcount(prog->samplers_used) >
- ctx->Const.MaxGeometryTextureImageUnits) {
- fail_link(shader_program, "Too many geometry shader texture samplers");
- }
- if (proginfo->Parameters->NumParameters >
- MAX_GEOMETRY_UNIFORM_COMPONENTS / 4) {
- fail_link(shader_program, "Too many geometry shader constants");
- }
- break;
- case GL_FRAGMENT_PROGRAM_ARB:
- if (_mesa_bitcount(prog->samplers_used) >
- ctx->Const.MaxTextureImageUnits) {
- fail_link(shader_program, "Too many fragment shader texture samplers");
- }
- if (proginfo->Parameters->NumParameters > MAX_UNIFORMS) {
- fail_link(shader_program, "Too many fragment shader constants");
- }
- break;
- default:
- _mesa_problem(ctx, "unexpected program type in check_resources()");
- }
-}
-
-
static void
set_uniform_initializer(struct gl_context *ctx, void *mem_ctx,
struct gl_shader_program *shader_program,
element_type->matrix_columns,
element_type->vector_elements,
loc, 1, GL_FALSE, (GLfloat *)values);
- loc += element_type->matrix_columns;
} else {
_mesa_uniform(ctx, shader_program, loc, element_type->matrix_columns,
values, element_type->gl_type);
- loc += type_size(element_type);
}
+
+ loc++;
}
}
GLint outputMap[VERT_RESULT_MAX];
GLint outputTypes[VERT_RESULT_MAX];
GLuint numVaryingReads = 0;
- GLboolean usedTemps[MAX_TEMPS];
+ GLboolean *usedTemps;
GLuint firstTemp = 0;
+ usedTemps = new GLboolean[MAX_TEMPS];
+ if (!usedTemps) {
+ return;
+ }
_mesa_find_used_registers(prog, PROGRAM_TEMPORARY,
usedTemps, MAX_TEMPS);
}
}
+ delete [] usedTemps;
+
if (numVaryingReads == 0)
return; /* nothing to be done */
void
glsl_to_tgsi_visitor::simplify_cmp(void)
{
- unsigned tempWrites[MAX_TEMPS];
+ unsigned *tempWrites;
unsigned outputWrites[MAX_PROGRAM_OUTPUTS];
+ tempWrites = new unsigned[MAX_TEMPS];
+ if (!tempWrites) {
+ return;
+ }
memset(tempWrites, 0, sizeof(tempWrites));
memset(outputWrites, 0, sizeof(outputWrites));
inst->op == TGSI_OPCODE_END ||
inst->op == TGSI_OPCODE_ENDSUB ||
inst->op == TGSI_OPCODE_RET) {
- return;
+ break;
}
if (inst->dst.file == PROGRAM_OUTPUT) {
inst->src[0] = inst->src[1];
}
}
+
+ delete [] tempWrites;
}
/* Replaces all references to a temporary register index with another index. */
switch (inst->op) {
case TGSI_OPCODE_BGNLOOP:
case TGSI_OPCODE_ENDLOOP:
+ case TGSI_OPCODE_CONT:
+ case TGSI_OPCODE_BRK:
/* End of a basic block, clear the write array entirely.
- * FIXME: This keeps us from killing dead code when the writes are
+ *
+ * This keeps us from killing dead code when the writes are
* on either side of a loop, even when the register isn't touched
- * inside the loop.
+ * inside the loop. However, glsl_to_tgsi_visitor doesn't seem to emit
+ * dead code of this type, so it shouldn't make a difference as long as
+ * the dead code elimination pass in the GLSL compiler does its job.
*/
memset(writes, 0, sizeof(*writes) * this->next_temp * 4);
break;
case TGSI_OPCODE_ENDIF:
- --level;
- break;
-
case TGSI_OPCODE_ELSE:
- /* Clear all channels written inside the preceding if block from the
- * write array, but leave those that were not touched.
- *
- * FIXME: This destroys opportunities to remove dead code inside of
- * IF blocks that are followed by an ELSE block.
+ /* Promote the recorded level of all channels written inside the
+ * preceding if or else block to the level above the if/else block.
*/
for (int r = 0; r < this->next_temp; r++) {
for (int c = 0; c < 4; c++) {
if (!writes[4 * r + c])
continue;
- if (write_level[4 * r + c] >= level)
- writes[4 * r + c] = NULL;
+ if (write_level[4 * r + c] == level)
+ write_level[4 * r + c] = level-1;
}
}
+
+ if(inst->op == TGSI_OPCODE_ENDIF)
+ --level;
+
break;
case TGSI_OPCODE_IF:
/* Copy attributes of the glsl_to_tgsi_visitor in the original shader. */
v->ctx = original->ctx;
v->prog = prog;
+ v->shader_program = NULL;
v->glsl_version = original->glsl_version;
v->native_integers = original->native_integers;
v->options = original->options;
inst->sampler = 0;
inst->tex_target = TEXTURE_2D_INDEX;
- prog->InputsRead |= (1 << FRAG_ATTRIB_TEX0);
+ prog->InputsRead |= FRAG_BIT_TEX0;
prog->SamplersUsed |= (1 << 0); /* mark sampler 0 as used */
v->samplers_used |= (1 << 0);
* new visitor. */
foreach_iter(exec_list_iterator, iter, original->instructions) {
glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *)iter.get();
+ glsl_to_tgsi_instruction *newinst;
st_src_reg src_regs[3];
if (inst->dst.file == PROGRAM_OUTPUT)
src_regs[i].index = src0.index;
}
else if (src_regs[i].file == PROGRAM_INPUT)
- prog->InputsRead |= (1 << src_regs[i].index);
+ prog->InputsRead |= BITFIELD64_BIT(src_regs[i].index);
}
- v->emit(NULL, inst->op, inst->dst, src_regs[0], src_regs[1], src_regs[2]);
+ newinst = v->emit(NULL, inst->op, inst->dst, src_regs[0], src_regs[1], src_regs[2]);
+ newinst->tex_target = inst->tex_target;
}
/* Make modifications to fragment program info. */
/* Copy attributes of the glsl_to_tgsi_visitor in the original shader. */
v->ctx = original->ctx;
v->prog = prog;
+ v->shader_program = NULL;
v->glsl_version = original->glsl_version;
v->native_integers = original->native_integers;
v->options = original->options;
inst->sampler = samplerIndex;
inst->tex_target = TEXTURE_2D_INDEX;
- prog->InputsRead |= (1 << FRAG_ATTRIB_TEX0);
+ prog->InputsRead |= FRAG_BIT_TEX0;
prog->SamplersUsed |= (1 << samplerIndex); /* mark sampler as used */
v->samplers_used |= (1 << samplerIndex);
* new visitor. */
foreach_iter(exec_list_iterator, iter, original->instructions) {
glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *)iter.get();
+ glsl_to_tgsi_instruction *newinst;
st_src_reg src_regs[3];
if (inst->dst.file == PROGRAM_OUTPUT)
for (int i=0; i<3; i++) {
src_regs[i] = inst->src[i];
if (src_regs[i].file == PROGRAM_INPUT)
- prog->InputsRead |= (1 << src_regs[i].index);
+ prog->InputsRead |= BITFIELD64_BIT(src_regs[i].index);
}
- v->emit(NULL, inst->op, inst->dst, src_regs[0], src_regs[1], src_regs[2]);
+ newinst = v->emit(NULL, inst->op, inst->dst, src_regs[0], src_regs[1], src_regs[2]);
+ newinst->tex_target = inst->tex_target;
}
/* Make modifications to fragment program info. */
/** Map Mesa's SYSTEM_VALUE_x to TGSI_SEMANTIC_x */
static unsigned mesa_sysval_to_semantic[SYSTEM_VALUE_MAX] = {
TGSI_SEMANTIC_FACE,
+ TGSI_SEMANTIC_VERTEXID,
TGSI_SEMANTIC_INSTANCEID
};
const ubyte outputSemanticIndex[],
boolean passthrough_edgeflags)
{
- struct st_translate translate, *t;
+ struct st_translate *t;
unsigned i;
enum pipe_error ret = PIPE_OK;
assert(numInputs <= Elements(t->inputs));
assert(numOutputs <= Elements(t->outputs));
- t = &translate;
+ t = CALLOC_STRUCT(st_translate);
+ if (!t) {
+ ret = PIPE_ERROR_OUT_OF_MEMORY;
+ goto out;
+ }
+
memset(t, 0, sizeof *t);
t->procType = procType;
t->pointSizeOutIndex = -1;
t->prevInstWrotePointSize = GL_FALSE;
+ if (program->shader_program) {
+ for (i = 0; i < program->shader_program->NumUserUniformStorage; i++) {
+ struct gl_uniform_storage *const storage =
+ &program->shader_program->UniformStorage[i];
+
+ _mesa_uniform_detach_all_driver_storage(storage);
+ }
+ }
+
/*
* Declare input attributes.
*/
break;
default:
assert(!"fragment shader outputs must be POSITION/STENCIL/COLOR");
- return PIPE_ERROR_BAD_INPUT;
+ ret = PIPE_ERROR_BAD_INPUT;
+ goto out;
}
}
}
}
for (i = 0; i < numOutputs; i++) {
- t->outputs[i] = ureg_DECL_output(ureg,
- outputSemanticName[i],
- outputSemanticIndex[i]);
+ if (outputSemanticName[i] == TGSI_SEMANTIC_CLIPDIST) {
+ int mask = ((1 << (program->num_clip_distances - 4*outputSemanticIndex[i])) - 1) & TGSI_WRITEMASK_XYZW;
+ t->outputs[i] = ureg_DECL_output_masked(ureg,
+ outputSemanticName[i],
+ outputSemanticIndex[i],
+ mask);
+ } else {
+ t->outputs[i] = ureg_DECL_output(ureg,
+ outputSemanticName[i],
+ outputSemanticIndex[i]);
+ }
if ((outputSemanticName[i] == TGSI_SEMANTIC_PSIZE) && proginfo->Id) {
/* Writing to the point size result register requires special
* handling to implement clamping.
t->insn[t->labels[i].branch_target]);
}
+ if (program->shader_program) {
+ /* This has to be done last. Any operation the can cause
+ * prog->ParameterValues to get reallocated (e.g., anything that adds a
+ * program constant) has to happen before creating this linkage.
+ */
+ for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
+ if (program->shader_program->_LinkedShaders[i] == NULL)
+ continue;
+
+ _mesa_associate_uniform_storage(ctx, program->shader_program,
+ program->shader_program->_LinkedShaders[i]->Program->Parameters);
+ }
+ }
+
out:
- FREE(t->insn);
- FREE(t->labels);
- FREE(t->constants);
- FREE(t->immediates);
+ if (t) {
+ FREE(t->insn);
+ FREE(t->labels);
+ FREE(t->constants);
+ FREE(t->immediates);
+
+ if (t->error) {
+ debug_printf("%s: translate error flag set\n", __FUNCTION__);
+ }
- if (t->error) {
- debug_printf("%s: translate error flag set\n", __FUNCTION__);
+ FREE(t);
}
return ret;
static struct gl_program *
get_mesa_program(struct gl_context *ctx,
struct gl_shader_program *shader_program,
- struct gl_shader *shader)
+ struct gl_shader *shader,
+ int num_clip_distances)
{
glsl_to_tgsi_visitor* v = new glsl_to_tgsi_visitor();
struct gl_program *prog;
+ struct pipe_screen * screen = st_context(ctx)->pipe->screen;
+ unsigned pipe_shader_type;
GLenum target;
const char *target_string;
bool progress;
case GL_VERTEX_SHADER:
target = GL_VERTEX_PROGRAM_ARB;
target_string = "vertex";
+ pipe_shader_type = PIPE_SHADER_VERTEX;
break;
case GL_FRAGMENT_SHADER:
target = GL_FRAGMENT_PROGRAM_ARB;
target_string = "fragment";
+ pipe_shader_type = PIPE_SHADER_FRAGMENT;
break;
case GL_GEOMETRY_SHADER:
target = GL_GEOMETRY_PROGRAM_NV;
target_string = "geometry";
+ pipe_shader_type = PIPE_SHADER_GEOMETRY;
break;
default:
assert(!"should not be reached");
v->options = options;
v->glsl_version = ctx->Const.GLSLVersion;
v->native_integers = ctx->Const.NativeIntegers;
+ v->num_clip_distances = num_clip_distances;
_mesa_generate_parameters_list_for_uniforms(shader_program, shader,
prog->Parameters);
+ if (!screen->get_shader_param(screen, pipe_shader_type,
+ PIPE_SHADER_CAP_OUTPUT_READ)) {
+ /* Remove reads to output registers, and to varyings in vertex shaders. */
+ lower_output_reads(shader->ir);
+ }
+
+
/* Emit intermediate IR for main(). */
visit_exec_list(shader->ir, v);
}
#endif
- /* Remove reads to output registers, and to varyings in vertex shaders. */
- v->remove_output_reads(PROGRAM_OUTPUT);
- if (target == GL_VERTEX_PROGRAM_ARB)
- v->remove_output_reads(PROGRAM_VARYING);
-
/* Perform optimizations on the instructions in the glsl_to_tgsi_visitor. */
v->simplify_cmp();
v->copy_propagate();
_mesa_print_ir(shader->ir, NULL);
printf("\n");
printf("\n");
+ fflush(stdout);
}
prog->Instructions = NULL;
prog->NumInstructions = 0;
- do_set_program_inouts(shader->ir, prog);
+ do_set_program_inouts(shader->ir, prog, shader->Type == GL_FRAGMENT_SHADER);
count_resources(v, prog);
- check_resources(ctx, shader_program, v, prog);
-
_mesa_reference_program(ctx, &shader->Program, prog);
+ /* This has to be done last. Any operation the can cause
+ * prog->ParameterValues to get reallocated (e.g., anything that adds a
+ * program constant) has to happen before creating this linkage.
+ */
+ _mesa_associate_uniform_storage(ctx, shader_program, prog->Parameters);
+ if (!shader_program->LinkStatus) {
+ return NULL;
+ }
+
struct st_vertex_program *stvp;
struct st_fragment_program *stfp;
struct st_geometry_program *stgp;
return prog;
}
+/**
+ * Searches through the IR for a declaration of gl_ClipDistance and returns the
+ * declared size of the gl_ClipDistance array. Returns 0 if gl_ClipDistance is
+ * not declared in the IR.
+ */
+int get_clip_distance_size(exec_list *ir)
+{
+ foreach_iter (exec_list_iterator, iter, *ir) {
+ ir_instruction *inst = (ir_instruction *)iter.get();
+ ir_variable *var = inst->as_variable();
+ if (var == NULL) continue;
+ if (!strcmp(var->name, "gl_ClipDistance")) {
+ return var->type->length;
+ }
+ }
+
+ return 0;
+}
+
extern "C" {
struct gl_shader *
GLboolean
st_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
{
+ int num_clip_distances[MESA_SHADER_TYPES];
assert(prog->LinkStatus);
for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
const struct gl_shader_compiler_options *options =
&ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(prog->_LinkedShaders[i]->Type)];
+ /* We have to determine the length of the gl_ClipDistance array before
+ * the array is lowered to two vec4s by lower_clip_distance().
+ */
+ num_clip_distances[i] = get_clip_distance_size(ir);
+
do {
progress = false;
progress = do_lower_jumps(ir, true, true, options->EmitNoMainReturn, options->EmitNoCont, options->EmitNoLoops) || progress;
- progress = do_common_optimization(ir, true, options->MaxUnrollIterations) || progress;
+ progress = do_common_optimization(ir, true, true,
+ options->MaxUnrollIterations)
+ || progress;
progress = lower_quadop_vector(ir, false) || progress;
+ progress = lower_clip_distance(ir) || progress;
if (options->MaxIfDepth == 0)
progress = lower_discard(ir) || progress;
if (prog->_LinkedShaders[i] == NULL)
continue;
- linked_prog = get_mesa_program(ctx, prog, prog->_LinkedShaders[i]);
+ linked_prog = get_mesa_program(ctx, prog, prog->_LinkedShaders[i],
+ num_clip_distances[i]);
if (linked_prog) {
static const GLenum targets[] = {
return GL_TRUE;
}
+void
+st_translate_stream_output_info(struct glsl_to_tgsi_visitor *glsl_to_tgsi,
+ const GLuint outputMapping[],
+ struct pipe_stream_output_info *so)
+{
+ static unsigned comps_to_mask[] = {
+ 0,
+ TGSI_WRITEMASK_X,
+ TGSI_WRITEMASK_XY,
+ TGSI_WRITEMASK_XYZ,
+ TGSI_WRITEMASK_XYZW
+ };
+ unsigned i;
+ struct gl_transform_feedback_info *info =
+ &glsl_to_tgsi->shader_program->LinkedTransformFeedback;
+
+ for (i = 0; i < info->NumOutputs; i++) {
+ assert(info->Outputs[i].NumComponents < Elements(comps_to_mask));
+ so->output[i].register_index =
+ outputMapping[info->Outputs[i].OutputRegister];
+ so->output[i].register_mask =
+ comps_to_mask[info->Outputs[i].NumComponents]
+ << info->Outputs[i].ComponentOffset;
+ so->output[i].output_buffer = info->Outputs[i].OutputBuffer;
+ }
+ so->num_outputs = info->NumOutputs;
+}
+
} /* extern "C" */
projects / mesa.git / blobdiff