#include "main/imports.h"
#include "main/mtypes.h"
+#include "shader/prog_print.h"
+#include "shader/programopt.h"
#include "pipe/p_context.h"
#include "pipe/p_defines.h"
-#include "pipe/tgsi/mesa/mesa_to_tgsi.h"
-#include "pipe/tgsi/exec/tgsi_core.h"
+#include "pipe/p_shader_tokens.h"
+#include "draw/draw_context.h"
+#include "tgsi/util/tgsi_dump.h"
#include "st_context.h"
-#include "st_cache.h"
#include "st_atom.h"
#include "st_program.h"
+#include "st_mesa_to_tgsi.h"
+#include "cso_cache/cso_context.h"
#define TGSI_DEBUG 0
+/** XXX we should use the version of this from p_util.h but including
+ * that header causes symbol collisions.
+ */
+static INLINE void *
+mem_dup(const void *src, uint size)
+{
+ void *dup = MALLOC(size);
+ if (dup)
+ memcpy(dup, src, size);
+ return dup;
+}
+
+
+
/**
* Translate a Mesa vertex shader into a TGSI shader.
* \param outputMapping to map vertex program output registers to TGSI
* \param tokensOut destination for TGSI tokens
* \return pointer to cached pipe_shader object.
*/
-const struct cso_vertex_shader *
+void
st_translate_vertex_program(struct st_context *st,
struct st_vertex_program *stvp,
- const GLuint outputMapping[],
- struct tgsi_token *tokensOut,
- GLuint maxTokens)
+ const GLuint outputMapping[])
{
+ struct pipe_context *pipe = st->pipe;
+ struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
GLuint defaultOutputMapping[VERT_RESULT_MAX];
struct pipe_shader_state vs;
- const struct cso_vertex_shader *cso;
GLuint attr, i;
+ GLuint num_generic = 0;
+ GLuint num_tokens;
+
+ ubyte vs_input_semantic_name[PIPE_MAX_SHADER_INPUTS];
+ ubyte vs_input_semantic_index[PIPE_MAX_SHADER_INPUTS];
+ uint vs_num_inputs = 0;
+
+ ubyte vs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
+ ubyte vs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
+ uint vs_num_outputs = 0;
memset(&vs, 0, sizeof(vs));
+ if (stvp->Base.IsPositionInvariant)
+ _mesa_insert_mvp_code(st->ctx, &stvp->Base);
+
/*
* Determine number of inputs, the mappings between VERT_ATTRIB_x
* and TGSI generic input indexes, plus input attrib semantic info.
*/
for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
if (stvp->Base.Base.InputsRead & (1 << attr)) {
- const GLuint slot = vs.num_inputs;
+ const GLuint slot = vs_num_inputs;
- vs.num_inputs++;
+ vs_num_inputs++;
stvp->input_to_index[attr] = slot;
stvp->index_to_input[slot] = attr;
switch (attr) {
case VERT_ATTRIB_POS:
- vs.input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
- vs.input_semantic_index[slot] = 0;
+ vs_input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
+ vs_input_semantic_index[slot] = 0;
break;
case VERT_ATTRIB_WEIGHT:
/* fall-through */
case VERT_ATTRIB_NORMAL:
/* just label as a generic */
- vs.input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
- vs.input_semantic_index[slot] = 0;
+ vs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
+ vs_input_semantic_index[slot] = 0;
break;
case VERT_ATTRIB_COLOR0:
- vs.input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- vs.input_semantic_index[slot] = 0;
+ vs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
+ vs_input_semantic_index[slot] = 0;
break;
case VERT_ATTRIB_COLOR1:
- vs.input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- vs.input_semantic_index[slot] = 1;
+ vs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
+ vs_input_semantic_index[slot] = 1;
break;
case VERT_ATTRIB_FOG:
- vs.input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
- vs.input_semantic_index[slot] = 0;
+ vs_input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
+ vs_input_semantic_index[slot] = 0;
break;
case VERT_ATTRIB_TEX0:
case VERT_ATTRIB_TEX1:
case VERT_ATTRIB_TEX5:
case VERT_ATTRIB_TEX6:
case VERT_ATTRIB_TEX7:
- vs.input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
- vs.input_semantic_index[slot] = attr - VERT_ATTRIB_TEX0;
+ vs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
+ vs_input_semantic_index[slot] = num_generic++;
break;
case VERT_ATTRIB_GENERIC0:
case VERT_ATTRIB_GENERIC1:
case VERT_ATTRIB_GENERIC6:
case VERT_ATTRIB_GENERIC7:
assert(attr < VERT_ATTRIB_MAX);
- vs.input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
- vs.input_semantic_index[slot] = attr - VERT_ATTRIB_GENERIC0;
+ vs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
+ vs_input_semantic_index[slot] = num_generic++;
break;
default:
assert(0);
/* initialize output semantics to defaults */
for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
- vs.output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
- vs.output_semantic_index[i] = 0;
+ vs_output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
+ vs_output_semantic_index[i] = 0;
}
+ num_generic = 0;
/*
* Determine number of outputs, the (default) output register
* mapping and the semantic information for each output.
assert(slot != ~0);
}
else {
- slot = vs.num_outputs;
- vs.num_outputs++;
+ slot = vs_num_outputs;
+ vs_num_outputs++;
defaultOutputMapping[attr] = slot;
}
switch (attr) {
case VERT_RESULT_HPOS:
- vs.output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
- vs.output_semantic_index[slot] = 0;
+ assert(slot == 0);
+ vs_output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
+ vs_output_semantic_index[slot] = 0;
break;
case VERT_RESULT_COL0:
- vs.output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- vs.output_semantic_index[slot] = 0;
+ vs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
+ vs_output_semantic_index[slot] = 0;
break;
case VERT_RESULT_COL1:
- vs.output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- vs.output_semantic_index[slot] = 1;
+ vs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
+ vs_output_semantic_index[slot] = 1;
break;
case VERT_RESULT_BFC0:
- vs.output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
- vs.output_semantic_index[slot] = 0;
+ vs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
+ vs_output_semantic_index[slot] = 0;
break;
case VERT_RESULT_BFC1:
- vs.output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
- vs.output_semantic_index[slot] = 1;
+ vs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
+ vs_output_semantic_index[slot] = 1;
break;
case VERT_RESULT_FOGC:
- vs.output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
- vs.output_semantic_index[slot] = 0;
+ vs_output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
+ vs_output_semantic_index[slot] = 0;
break;
case VERT_RESULT_PSIZ:
- vs.output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
- vs.output_semantic_index[slot] = 0;
+ vs_output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
+ vs_output_semantic_index[slot] = 0;
break;
case VERT_RESULT_EDGE:
assert(0);
case VERT_RESULT_TEX5:
case VERT_RESULT_TEX6:
case VERT_RESULT_TEX7:
- vs.output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
- vs.output_semantic_index[slot] = attr - VERT_RESULT_TEX0;
+ vs_output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
+ vs_output_semantic_index[slot] = num_generic++;
break;
case VERT_RESULT_VAR0:
/* fall-through */
default:
assert(attr - VERT_RESULT_VAR0 < MAX_VARYING);
- vs.output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
- vs.output_semantic_index[slot] = attr - VERT_RESULT_VAR0;
+ vs_output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
+ vs_output_semantic_index[slot] = num_generic++;
}
}
}
+ assert(vs_output_semantic_name[0] == TGSI_SEMANTIC_POSITION);
+
if (outputMapping) {
- /* find max output slot referenced to compute vs.num_outputs */
+ /* find max output slot referenced to compute vs_num_outputs */
GLuint maxSlot = 0;
for (attr = 0; attr < VERT_RESULT_MAX; attr++) {
if (outputMapping[attr] != ~0 && outputMapping[attr] > maxSlot)
maxSlot = outputMapping[attr];
}
- vs.num_outputs = maxSlot + 1;
+ vs_num_outputs = maxSlot + 1;
}
else {
outputMapping = defaultOutputMapping;
}
+ /* free old shader state, if any */
+ if (stvp->state.tokens) {
+ FREE((void *) stvp->state.tokens);
+ stvp->state.tokens = NULL;
+ }
+ if (stvp->driver_shader) {
+ cso_delete_vertex_shader(st->cso_context, stvp->driver_shader);
+ stvp->driver_shader = NULL;
+ }
+
/* XXX: fix static allocation of tokens:
*/
- tgsi_mesa_compile_vp_program( &stvp->Base,
- /* inputs */
- vs.num_inputs,
- stvp->input_to_index,
- vs.input_semantic_name,
- vs.input_semantic_index,
- /* outputs */
- vs.num_outputs,
- outputMapping,
- vs.output_semantic_name,
- vs.output_semantic_index,
- /* tokenized result */
- tokensOut, maxTokens);
-
- vs.tokens = tokensOut;
- cso = st_cached_vs_state(st, &vs);
- stvp->vs = cso;
+ num_tokens = tgsi_translate_mesa_program( TGSI_PROCESSOR_VERTEX,
+ &stvp->Base.Base,
+ /* inputs */
+ vs_num_inputs,
+ stvp->input_to_index,
+ vs_input_semantic_name,
+ vs_input_semantic_index,
+ NULL,
+ /* outputs */
+ vs_num_outputs,
+ outputMapping,
+ vs_output_semantic_name,
+ vs_output_semantic_index,
+ /* tokenized result */
+ tokens, ST_MAX_SHADER_TOKENS);
+
+ vs.tokens = (struct tgsi_token *)
+ mem_dup(tokens, num_tokens * sizeof(tokens[0]));
+
+ stvp->num_inputs = vs_num_inputs;
+ stvp->state = vs; /* struct copy */
+ stvp->driver_shader = pipe->create_vs_state(pipe, &vs);
+
+ if (0)
+ _mesa_print_program(&stvp->Base.Base);
if (TGSI_DEBUG)
- tgsi_dump( tokensOut, 0 );
-
- return cso;
+ tgsi_dump( vs.tokens, 0 );
}
* \param tokensOut destination for TGSI tokens
* \return pointer to cached pipe_shader object.
*/
-const struct cso_fragment_shader *
+void
st_translate_fragment_program(struct st_context *st,
struct st_fragment_program *stfp,
- const GLuint inputMapping[],
- struct tgsi_token *tokensOut,
- GLuint maxTokens)
+ const GLuint inputMapping[])
{
+ struct pipe_context *pipe = st->pipe;
+ struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
GLuint outputMapping[FRAG_RESULT_MAX];
GLuint defaultInputMapping[FRAG_ATTRIB_MAX];
struct pipe_shader_state fs;
- const struct cso_fragment_shader *cso;
GLuint interpMode[16]; /* XXX size? */
GLuint attr;
- GLbitfield inputsRead = stfp->Base.Base.InputsRead;
+ const GLbitfield inputsRead = stfp->Base.Base.InputsRead;
+ GLuint vslot = 0;
+ GLuint num_generic = 0;
+ GLuint num_tokens;
- /* Check if all fragment programs need the fragment position (in order
- * to do perspective-corrected interpolation).
- */
- /* XXX temporary! */
- if (st->pipe->get_param(st->pipe, PIPE_PARAM_FS_NEEDS_POS))
- inputsRead |= FRAG_BIT_WPOS;
+ ubyte fs_input_semantic_name[PIPE_MAX_SHADER_INPUTS];
+ ubyte fs_input_semantic_index[PIPE_MAX_SHADER_INPUTS];
+ uint fs_num_inputs = 0;
+
+ ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
+ ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
+ uint fs_num_outputs = 0;
memset(&fs, 0, sizeof(fs));
+ /* which vertex output goes to the first fragment input: */
+ if (inputsRead & FRAG_BIT_WPOS)
+ vslot = 0;
+ else
+ vslot = 1;
+
/*
* Convert Mesa program inputs to TGSI input register semantics.
*/
for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) {
if (inputsRead & (1 << attr)) {
- const GLuint slot = fs.num_inputs;
-
- fs.num_inputs++;
+ const GLuint slot = fs_num_inputs;
defaultInputMapping[attr] = slot;
+ stfp->input_map[slot] = vslot++;
+
+ fs_num_inputs++;
+
switch (attr) {
case FRAG_ATTRIB_WPOS:
- fs.input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
- fs.input_semantic_index[slot] = 0;
- interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
+ fs_input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
+ fs_input_semantic_index[slot] = 0;
+ interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
break;
case FRAG_ATTRIB_COL0:
- fs.input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- fs.input_semantic_index[slot] = 0;
+ fs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
+ fs_input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
break;
case FRAG_ATTRIB_COL1:
- fs.input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
- fs.input_semantic_index[slot] = 1;
+ fs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
+ fs_input_semantic_index[slot] = 1;
interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
break;
case FRAG_ATTRIB_FOGC:
- fs.input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
- fs.input_semantic_index[slot] = 0;
+ fs_input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
+ fs_input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
break;
case FRAG_ATTRIB_TEX0:
case FRAG_ATTRIB_TEX5:
case FRAG_ATTRIB_TEX6:
case FRAG_ATTRIB_TEX7:
- fs.input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
- fs.input_semantic_index[slot] = attr - FRAG_ATTRIB_TEX0;
+ fs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
+ fs_input_semantic_index[slot] = num_generic++;
interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
break;
case FRAG_ATTRIB_VAR0:
/* fall-through */
default:
- fs.input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
- fs.input_semantic_index[slot] = attr - FRAG_ATTRIB_VAR0;
+ fs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
+ fs_input_semantic_index[slot] = num_generic++;
interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
}
}
}
/*
- * Semantics for outputs
+ * Semantics and mapping for outputs
*/
- for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
- if (stfp->Base.Base.OutputsWritten & (1 << attr)) {
- switch (attr) {
- case FRAG_RESULT_DEPR:
- fs.output_semantic_name[fs.num_outputs] = TGSI_SEMANTIC_POSITION;
- outputMapping[attr] = fs.num_outputs;
- break;
- case FRAG_RESULT_COLR:
- fs.output_semantic_name[fs.num_outputs] = TGSI_SEMANTIC_COLOR;
- outputMapping[attr] = fs.num_outputs;
- break;
- default:
- assert(0);
+ {
+ uint numColors = 0;
+ GLbitfield outputsWritten = stfp->Base.Base.OutputsWritten;
+
+ /* if z is written, emit that first */
+ if (outputsWritten & (1 << FRAG_RESULT_DEPR)) {
+ fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
+ fs_output_semantic_index[fs_num_outputs] = 0;
+ outputMapping[FRAG_RESULT_DEPR] = fs_num_outputs;
+ fs_num_outputs++;
+ outputsWritten &= ~(1 << FRAG_RESULT_DEPR);
+ }
+
+ /* handle remaning outputs (color) */
+ for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
+ if (outputsWritten & (1 << attr)) {
+ switch (attr) {
+ case FRAG_RESULT_DEPR:
+ /* handled above */
+ assert(0);
+ break;
+ case FRAG_RESULT_COLR:
+ fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
+ fs_output_semantic_index[fs_num_outputs] = numColors;
+ outputMapping[attr] = fs_num_outputs;
+ numColors++;
+ break;
+ default:
+ assert(0);
+ }
+ fs_num_outputs++;
}
- fs.num_outputs++;
}
}
/* XXX: fix static allocation of tokens:
*/
- tgsi_mesa_compile_fp_program( &stfp->Base,
- /* inputs */
- fs.num_inputs,
- inputMapping,
- fs.input_semantic_name,
- fs.input_semantic_index,
- interpMode,
- /* outputs */
- outputMapping,
- /* tokenized result */
- tokensOut, maxTokens);
-
-
- fs.tokens = tokensOut;
-
- cso = st_cached_fs_state(st, &fs);
- stfp->fs = cso;
+ num_tokens = tgsi_translate_mesa_program( TGSI_PROCESSOR_FRAGMENT,
+ &stfp->Base.Base,
+ /* inputs */
+ fs_num_inputs,
+ inputMapping,
+ fs_input_semantic_name,
+ fs_input_semantic_index,
+ interpMode,
+ /* outputs */
+ fs_num_outputs,
+ outputMapping,
+ fs_output_semantic_name,
+ fs_output_semantic_index,
+ /* tokenized result */
+ tokens, ST_MAX_SHADER_TOKENS);
+
+ fs.tokens = (struct tgsi_token *)
+ mem_dup(tokens, num_tokens * sizeof(tokens[0]));
+
+ stfp->state = fs; /* struct copy */
+ stfp->driver_shader = pipe->create_fs_state(pipe, &fs);
+
+ if (0)
+ _mesa_print_program(&stfp->Base.Base);
if (TGSI_DEBUG)
- tgsi_dump( tokensOut, 0/*TGSI_DUMP_VERBOSE*/ );
-
-#if defined(__i386__) || defined(__386__)
- if (stfp->sse2_program.csr == stfp->sse2_program.store)
- tgsi_emit_sse2_fs( tokensOut, &stfp->sse2_program );
-
- if (!cso->state.executable)
- ((struct cso_fragment_shader*)cso)->state.executable = (void *) x86_get_func( &stfp->sse2_program );
-#endif
-
- return cso;
+ tgsi_dump( fs.tokens, 0/*TGSI_DUMP_VERBOSE*/ );
}
projects / mesa.git / blobdiff