/**
* If we fail to compile a fragment shader (because it uses too many
* registers, for example) we'll use a dummy/fallback shader that
* simply emits a constant color (red for debug, black for release).
* We hit this with the Unigine/Heaven demo when Shaders = High.
* With black, the demo still looks good.
*/
static const struct tgsi_token *
get_dummy_fragment_shader(void)
{
#ifdef DEBUG
static const float color[4] = { 1.0, 0.0, 0.0, 0.0 }; /* red */
#else
static const float color[4] = { 0.0, 0.0, 0.0, 0.0 }; /* black */
#endif
struct ureg_program *ureg;
const struct tgsi_token *tokens;
struct ureg_src src;
struct ureg_dst dst;
unsigned num_tokens;
ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
if (!ureg)
return NULL;
dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
src = ureg_DECL_immediate(ureg, color, 4);
ureg_MOV(ureg, dst, src);
ureg_END(ureg);
tokens = ureg_get_tokens(ureg, &num_tokens);
ureg_destroy(ureg);
return tokens;
}
/**
* If we fail to compile a vertex shader we'll use a dummy/fallback shader
* that simply emits a (0,0,0,1) vertex position.
*/
static const struct tgsi_token *
get_dummy_vertex_shader(void)
{
static const float zero[4] = { 0.0, 0.0, 0.0, 1.0 };
struct ureg_program *ureg;
const struct tgsi_token *tokens;
struct ureg_src src;
struct ureg_dst dst;
unsigned num_tokens;
ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
if (!ureg)
return NULL;
dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
src = ureg_DECL_immediate(ureg, zero, 4);
ureg_MOV(ureg, dst, src);
ureg_END(ureg);
tokens = ureg_get_tokens(ureg, &num_tokens);
ureg_destroy(ureg);
return tokens;
}
static void r300_dummy_vertex_shader(
struct r300_context* r300,
struct r300_vertex_shader* shader)
{
struct ureg_program *ureg;
struct ureg_dst dst;
struct ureg_src imm;
/* Make a simple vertex shader which outputs (0, 0, 0, 1),
* effectively rendering nothing. */
ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
imm = ureg_imm4f(ureg, 0, 0, 0, 1);
ureg_MOV(ureg, dst, imm);
ureg_END(ureg);
shader->state.tokens = tgsi_dup_tokens(ureg_finalize(ureg));
ureg_destroy(ureg);
shader->dummy = TRUE;
r300_init_vs_outputs(r300, shader);
r300_translate_vertex_shader(r300, shader);
}
static void r300_dummy_fragment_shader(
struct r300_context* r300,
struct r300_fragment_shader_code* shader)
{
struct pipe_shader_state state;
struct ureg_program *ureg;
struct ureg_dst out;
struct ureg_src imm;
/* Make a simple fragment shader which outputs (0, 0, 0, 1) */
ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
imm = ureg_imm4f(ureg, 0, 0, 0, 1);
ureg_MOV(ureg, out, imm);
ureg_END(ureg);
state.tokens = ureg_finalize(ureg);
shader->dummy = TRUE;
r300_translate_fragment_shader(r300, shader, state.tokens);
ureg_destroy(ureg);
}
/**
* Translate a geometry program to create a new variant.
*/
static struct st_gp_variant *
st_translate_geometry_program(struct st_context *st,
struct st_geometry_program *stgp,
const struct st_gp_variant_key *key)
{
GLuint inputMapping[VARYING_SLOT_MAX];
GLuint outputMapping[VARYING_SLOT_MAX];
struct pipe_context *pipe = st->pipe;
GLuint attr;
uint gs_num_inputs = 0;
ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
ubyte gs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
ubyte gs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
uint gs_num_outputs = 0;
GLint i;
struct ureg_program *ureg;
struct pipe_shader_state state = {0};
struct st_gp_variant *gpv;
gpv = CALLOC_STRUCT(st_gp_variant);
if (!gpv)
return NULL;
ureg = ureg_create(TGSI_PROCESSOR_GEOMETRY);
if (ureg == NULL) {
free(gpv);
return NULL;
}
memset(inputMapping, 0, sizeof(inputMapping));
memset(outputMapping, 0, sizeof(outputMapping));
/*
* Convert Mesa program inputs to TGSI input register semantics.
*/
for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
if ((stgp->Base.Base.InputsRead & BITFIELD64_BIT(attr)) != 0) {
const GLuint slot = gs_num_inputs++;
inputMapping[attr] = slot;
switch (attr) {
case VARYING_SLOT_PRIMITIVE_ID:
input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
input_semantic_index[slot] = 0;
break;
case VARYING_SLOT_POS:
input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
input_semantic_index[slot] = 0;
break;
case VARYING_SLOT_COL0:
input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
input_semantic_index[slot] = 0;
break;
case VARYING_SLOT_COL1:
input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
input_semantic_index[slot] = 1;
break;
case VARYING_SLOT_FOGC:
input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
input_semantic_index[slot] = 0;
break;
case VARYING_SLOT_CLIP_VERTEX:
input_semantic_name[slot] = TGSI_SEMANTIC_CLIPVERTEX;
input_semantic_index[slot] = 0;
break;
case VARYING_SLOT_CLIP_DIST0:
input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
input_semantic_index[slot] = 0;
break;
case VARYING_SLOT_CLIP_DIST1:
input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
input_semantic_index[slot] = 1;
break;
case VARYING_SLOT_PSIZ:
input_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
input_semantic_index[slot] = 0;
break;
case VARYING_SLOT_TEX0:
case VARYING_SLOT_TEX1:
case VARYING_SLOT_TEX2:
case VARYING_SLOT_TEX3:
case VARYING_SLOT_TEX4:
case VARYING_SLOT_TEX5:
case VARYING_SLOT_TEX6:
case VARYING_SLOT_TEX7:
if (st->needs_texcoord_semantic) {
input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
input_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
break;
}
/* fall through */
case VARYING_SLOT_VAR0:
default:
assert(attr >= VARYING_SLOT_VAR0 && attr < VARYING_SLOT_MAX);
input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
input_semantic_index[slot] =
st_get_generic_varying_index(st, attr);
break;
}
}
}
/* initialize output semantics to defaults */
for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
gs_output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
gs_output_semantic_index[i] = 0;
}
/*
* Determine number of outputs, the (default) output register
* mapping and the semantic information for each output.
*/
for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
if (stgp->Base.Base.OutputsWritten & BITFIELD64_BIT(attr)) {
GLuint slot = gs_num_outputs++;
outputMapping[attr] = slot;
switch (attr) {
case VARYING_SLOT_POS:
assert(slot == 0);
gs_output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
gs_output_semantic_index[slot] = 0;
break;
case VARYING_SLOT_COL0:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
gs_output_semantic_index[slot] = 0;
break;
case VARYING_SLOT_COL1:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
gs_output_semantic_index[slot] = 1;
break;
case VARYING_SLOT_BFC0:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
gs_output_semantic_index[slot] = 0;
break;
case VARYING_SLOT_BFC1:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
gs_output_semantic_index[slot] = 1;
break;
case VARYING_SLOT_FOGC:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
gs_output_semantic_index[slot] = 0;
break;
case VARYING_SLOT_PSIZ:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
gs_output_semantic_index[slot] = 0;
break;
case VARYING_SLOT_CLIP_VERTEX:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_CLIPVERTEX;
gs_output_semantic_index[slot] = 0;
break;
case VARYING_SLOT_CLIP_DIST0:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
gs_output_semantic_index[slot] = 0;
break;
case VARYING_SLOT_CLIP_DIST1:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
gs_output_semantic_index[slot] = 1;
break;
case VARYING_SLOT_LAYER:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_LAYER;
gs_output_semantic_index[slot] = 0;
break;
case VARYING_SLOT_PRIMITIVE_ID:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
gs_output_semantic_index[slot] = 0;
break;
case VARYING_SLOT_VIEWPORT:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX;
gs_output_semantic_index[slot] = 0;
break;
case VARYING_SLOT_TEX0:
case VARYING_SLOT_TEX1:
case VARYING_SLOT_TEX2:
case VARYING_SLOT_TEX3:
case VARYING_SLOT_TEX4:
case VARYING_SLOT_TEX5:
case VARYING_SLOT_TEX6:
case VARYING_SLOT_TEX7:
if (st->needs_texcoord_semantic) {
gs_output_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
gs_output_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
break;
}
/* fall through */
case VARYING_SLOT_VAR0:
default:
assert(slot < ARRAY_SIZE(gs_output_semantic_name));
assert(attr >= VARYING_SLOT_VAR0);
gs_output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
gs_output_semantic_index[slot] =
st_get_generic_varying_index(st, attr);
break;
}
}
}
ureg_property(ureg, TGSI_PROPERTY_GS_INPUT_PRIM, stgp->Base.InputType);
ureg_property(ureg, TGSI_PROPERTY_GS_OUTPUT_PRIM, stgp->Base.OutputType);
ureg_property(ureg, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES,
stgp->Base.VerticesOut);
ureg_property(ureg, TGSI_PROPERTY_GS_INVOCATIONS, stgp->Base.Invocations);
st_translate_program(st->ctx,
TGSI_PROCESSOR_GEOMETRY,
ureg,
stgp->glsl_to_tgsi,
&stgp->Base.Base,
/* inputs */
gs_num_inputs,
inputMapping,
input_semantic_name,
input_semantic_index,
NULL,
NULL,
/* outputs */
gs_num_outputs,
outputMapping,
gs_output_semantic_name,
gs_output_semantic_index,
FALSE,
FALSE);
state.tokens = ureg_get_tokens(ureg, NULL);
ureg_destroy(ureg);
st_translate_stream_output_info(stgp->glsl_to_tgsi,
outputMapping,
&state.stream_output);
if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
_mesa_print_program(&stgp->Base.Base);
debug_printf("\n");
}
if (ST_DEBUG & DEBUG_TGSI) {
tgsi_dump(state.tokens, 0);
debug_printf("\n");
}
/* fill in new variant */
gpv->driver_shader = pipe->create_gs_state(pipe, &state);
gpv->key = *key;
ureg_free_tokens(state.tokens);
return gpv;
}
/**
* Translate a Mesa fragment shader into a TGSI shader using extra info in
* the key.
* \return new fragment program variant
*/
static struct st_fp_variant *
st_translate_fragment_program(struct st_context *st,
struct st_fragment_program *stfp,
const struct st_fp_variant_key *key)
{
struct pipe_context *pipe = st->pipe;
struct st_fp_variant *variant = CALLOC_STRUCT(st_fp_variant);
GLboolean deleteFP = GL_FALSE;
GLuint outputMapping[FRAG_RESULT_MAX];
GLuint inputMapping[VARYING_SLOT_MAX];
GLuint interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */
GLuint interpLocation[PIPE_MAX_SHADER_INPUTS];
GLuint attr;
GLbitfield64 inputsRead;
struct ureg_program *ureg;
GLboolean write_all = GL_FALSE;
ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
ubyte 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;
if (!variant)
return NULL;
assert(!(key->bitmap && key->drawpixels));
if (key->bitmap) {
/* glBitmap drawing */
struct gl_fragment_program *fp; /* we free this temp program below */
st_make_bitmap_fragment_program(st, &stfp->Base,
&fp, &variant->bitmap_sampler);
variant->parameters = _mesa_clone_parameter_list(fp->Base.Parameters);
stfp = st_fragment_program(fp);
deleteFP = GL_TRUE;
}
else if (key->drawpixels) {
/* glDrawPixels drawing */
struct gl_fragment_program *fp; /* we free this temp program below */
if (key->drawpixels_z || key->drawpixels_stencil) {
fp = st_make_drawpix_z_stencil_program(st, key->drawpixels_z,
key->drawpixels_stencil);
}
else {
/* RGBA */
st_make_drawpix_fragment_program(st, &stfp->Base, &fp);
variant->parameters = _mesa_clone_parameter_list(fp->Base.Parameters);
deleteFP = GL_TRUE;
}
stfp = st_fragment_program(fp);
}
if (!stfp->glsl_to_tgsi)
_mesa_remove_output_reads(&stfp->Base.Base, PROGRAM_OUTPUT);
/*
* Convert Mesa program inputs to TGSI input register semantics.
*/
inputsRead = stfp->Base.Base.InputsRead;
for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
if ((inputsRead & BITFIELD64_BIT(attr)) != 0) {
const GLuint slot = fs_num_inputs++;
inputMapping[attr] = slot;
if (stfp->Base.IsCentroid & BITFIELD64_BIT(attr))
interpLocation[slot] = TGSI_INTERPOLATE_LOC_CENTROID;
else if (stfp->Base.IsSample & BITFIELD64_BIT(attr))
interpLocation[slot] = TGSI_INTERPOLATE_LOC_SAMPLE;
else
interpLocation[slot] = TGSI_INTERPOLATE_LOC_CENTER;
if (key->persample_shading)
interpLocation[slot] = TGSI_INTERPOLATE_LOC_SAMPLE;
switch (attr) {
case VARYING_SLOT_POS:
input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
break;
case VARYING_SLOT_COL0:
input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
input_semantic_index[slot] = 0;
interpMode[slot] = st_translate_interp(stfp->Base.InterpQualifier[attr],
TRUE);
break;
case VARYING_SLOT_COL1:
input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
input_semantic_index[slot] = 1;
interpMode[slot] = st_translate_interp(stfp->Base.InterpQualifier[attr],
TRUE);
break;
case VARYING_SLOT_FOGC:
input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
break;
case VARYING_SLOT_FACE:
input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
break;
case VARYING_SLOT_PRIMITIVE_ID:
input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
break;
case VARYING_SLOT_LAYER:
input_semantic_name[slot] = TGSI_SEMANTIC_LAYER;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
break;
case VARYING_SLOT_VIEWPORT:
input_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
break;
case VARYING_SLOT_CLIP_DIST0:
input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
break;
case VARYING_SLOT_CLIP_DIST1:
input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
input_semantic_index[slot] = 1;
interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
break;
/* In most cases, there is nothing special about these
* inputs, so adopt a convention to use the generic
* semantic name and the mesa VARYING_SLOT_ number as the
* index.
*
* All that is required is that the vertex shader labels
* its own outputs similarly, and that the vertex shader
* generates at least every output required by the
* fragment shader plus fixed-function hardware (such as
* BFC).
*
* However, some drivers may need us to identify the PNTC and TEXi
* varyings if, for example, their capability to replace them with
* sprite coordinates is limited.
*/
case VARYING_SLOT_PNTC:
if (st->needs_texcoord_semantic) {
input_semantic_name[slot] = TGSI_SEMANTIC_PCOORD;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
break;
}
/* fall through */
case VARYING_SLOT_TEX0:
case VARYING_SLOT_TEX1:
case VARYING_SLOT_TEX2:
case VARYING_SLOT_TEX3:
case VARYING_SLOT_TEX4:
case VARYING_SLOT_TEX5:
case VARYING_SLOT_TEX6:
case VARYING_SLOT_TEX7:
if (st->needs_texcoord_semantic) {
input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
input_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
interpMode[slot] =
st_translate_interp(stfp->Base.InterpQualifier[attr], FALSE);
break;
}
/* fall through */
case VARYING_SLOT_VAR0:
default:
/* Semantic indices should be zero-based because drivers may choose
* to assign a fixed slot determined by that index.
* This is useful because ARB_separate_shader_objects uses location
* qualifiers for linkage, and if the semantic index corresponds to
* these locations, linkage passes in the driver become unecessary.
*
* If needs_texcoord_semantic is true, no semantic indices will be
* consumed for the TEXi varyings, and we can base the locations of
* the user varyings on VAR0. Otherwise, we use TEX0 as base index.
*/
assert(attr >= VARYING_SLOT_TEX0);
input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
input_semantic_index[slot] = st_get_generic_varying_index(st, attr);
if (attr == VARYING_SLOT_PNTC)
interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
else
interpMode[slot] = st_translate_interp(stfp->Base.InterpQualifier[attr],
FALSE);
break;
}
}
else {
inputMapping[attr] = -1;
}
}
/*
* Semantics and mapping for outputs
*/
{
uint numColors = 0;
GLbitfield64 outputsWritten = stfp->Base.Base.OutputsWritten;
/* if z is written, emit that first */
if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
fs_output_semantic_index[fs_num_outputs] = 0;
outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
fs_num_outputs++;
outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
}
if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) {
fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL;
fs_output_semantic_index[fs_num_outputs] = 0;
outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs;
fs_num_outputs++;
outputsWritten &= ~(1 << FRAG_RESULT_STENCIL);
}
if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)) {
fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_SAMPLEMASK;
fs_output_semantic_index[fs_num_outputs] = 0;
outputMapping[FRAG_RESULT_SAMPLE_MASK] = fs_num_outputs;
fs_num_outputs++;
outputsWritten &= ~(1 << FRAG_RESULT_SAMPLE_MASK);
}
/* handle remaining outputs (color) */
for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
if (outputsWritten & BITFIELD64_BIT(attr)) {
switch (attr) {
case FRAG_RESULT_DEPTH:
case FRAG_RESULT_STENCIL:
case FRAG_RESULT_SAMPLE_MASK:
/* handled above */
assert(0);
break;
case FRAG_RESULT_COLOR:
write_all = GL_TRUE; /* fallthrough */
default:
assert(attr == FRAG_RESULT_COLOR ||
(FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX));
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;
}
fs_num_outputs++;
}
}
}
ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT );
if (ureg == NULL) {
free(variant);
return NULL;
}
if (ST_DEBUG & DEBUG_MESA) {
_mesa_print_program(&stfp->Base.Base);
_mesa_print_program_parameters(st->ctx, &stfp->Base.Base);
debug_printf("\n");
}
if (write_all == GL_TRUE)
ureg_property(ureg, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS, 1);
if (stfp->Base.FragDepthLayout != FRAG_DEPTH_LAYOUT_NONE) {
switch (stfp->Base.FragDepthLayout) {
case FRAG_DEPTH_LAYOUT_ANY:
ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
TGSI_FS_DEPTH_LAYOUT_ANY);
break;
case FRAG_DEPTH_LAYOUT_GREATER:
ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
TGSI_FS_DEPTH_LAYOUT_GREATER);
break;
case FRAG_DEPTH_LAYOUT_LESS:
ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
TGSI_FS_DEPTH_LAYOUT_LESS);
break;
case FRAG_DEPTH_LAYOUT_UNCHANGED:
ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
TGSI_FS_DEPTH_LAYOUT_UNCHANGED);
break;
default:
assert(0);
}
}
if (stfp->glsl_to_tgsi)
st_translate_program(st->ctx,
TGSI_PROCESSOR_FRAGMENT,
ureg,
stfp->glsl_to_tgsi,
&stfp->Base.Base,
/* inputs */
fs_num_inputs,
inputMapping,
input_semantic_name,
input_semantic_index,
interpMode,
interpLocation,
/* outputs */
fs_num_outputs,
outputMapping,
fs_output_semantic_name,
fs_output_semantic_index, FALSE,
key->clamp_color );
else
st_translate_mesa_program(st->ctx,
TGSI_PROCESSOR_FRAGMENT,
ureg,
&stfp->Base.Base,
/* inputs */
fs_num_inputs,
inputMapping,
input_semantic_name,
input_semantic_index,
interpMode,
/* outputs */
fs_num_outputs,
outputMapping,
fs_output_semantic_name,
fs_output_semantic_index, FALSE,
key->clamp_color);
variant->tgsi.tokens = ureg_get_tokens( ureg, NULL );
ureg_destroy( ureg );
if (ST_DEBUG & DEBUG_TGSI) {
tgsi_dump(variant->tgsi.tokens, 0/*TGSI_DUMP_VERBOSE*/);
debug_printf("\n");
}
/* fill in variant */
variant->driver_shader = pipe->create_fs_state(pipe, &variant->tgsi);
variant->key = *key;
if (deleteFP) {
/* Free the temporary program made above */
struct gl_fragment_program *fp = &stfp->Base;
_mesa_reference_fragprog(st->ctx, &fp, NULL);
}
return variant;
}
/**
* Translate a vertex program to create a new variant.
*/
static struct st_vp_variant *
st_translate_vertex_program(struct st_context *st,
struct st_vertex_program *stvp,
const struct st_vp_variant_key *key)
{
struct st_vp_variant *vpv = CALLOC_STRUCT(st_vp_variant);
struct pipe_context *pipe = st->pipe;
struct ureg_program *ureg;
enum pipe_error error;
unsigned num_outputs;
st_prepare_vertex_program(st->ctx, stvp);
if (!stvp->glsl_to_tgsi)
{
_mesa_remove_output_reads(&stvp->Base.Base, PROGRAM_OUTPUT);
}
ureg = ureg_create( TGSI_PROCESSOR_VERTEX );
if (ureg == NULL) {
free(vpv);
return NULL;
}
vpv->key = *key;
vpv->num_inputs = stvp->num_inputs;
num_outputs = stvp->num_outputs;
if (key->passthrough_edgeflags) {
vpv->num_inputs++;
num_outputs++;
}
if (ST_DEBUG & DEBUG_MESA) {
_mesa_print_program(&stvp->Base.Base);
_mesa_print_program_parameters(st->ctx, &stvp->Base.Base);
debug_printf("\n");
}
if (stvp->glsl_to_tgsi)
error = st_translate_program(st->ctx,
TGSI_PROCESSOR_VERTEX,
ureg,
stvp->glsl_to_tgsi,
&stvp->Base.Base,
/* inputs */
vpv->num_inputs,
stvp->input_to_index,
NULL, /* input semantic name */
NULL, /* input semantic index */
NULL, /* interp mode */
NULL, /* interp location */
/* outputs */
num_outputs,
stvp->result_to_output,
stvp->output_semantic_name,
stvp->output_semantic_index,
key->passthrough_edgeflags,
key->clamp_color);
else
error = st_translate_mesa_program(st->ctx,
TGSI_PROCESSOR_VERTEX,
ureg,
&stvp->Base.Base,
/* inputs */
vpv->num_inputs,
stvp->input_to_index,
NULL, /* input semantic name */
NULL, /* input semantic index */
NULL,
/* outputs */
num_outputs,
stvp->result_to_output,
stvp->output_semantic_name,
stvp->output_semantic_index,
key->passthrough_edgeflags,
key->clamp_color);
if (error)
goto fail;
vpv->tgsi.tokens = ureg_get_tokens( ureg, NULL );
if (!vpv->tgsi.tokens)
goto fail;
ureg_destroy( ureg );
if (stvp->glsl_to_tgsi) {
st_translate_stream_output_info(stvp->glsl_to_tgsi,
stvp->result_to_output,
&vpv->tgsi.stream_output);
}
if (ST_DEBUG & DEBUG_TGSI) {
tgsi_dump(vpv->tgsi.tokens, 0);
debug_printf("\n");
}
vpv->driver_shader = pipe->create_vs_state(pipe, &vpv->tgsi);
return vpv;
fail:
debug_printf("%s: failed to translate Mesa program:\n", __func__);
_mesa_print_program(&stvp->Base.Base);
debug_assert(0);
ureg_destroy( ureg );
return NULL;
}
/* Create a compute shader implementing clear_buffer or copy_buffer. */
void *si_create_dma_compute_shader(struct pipe_context *ctx,
unsigned num_dwords_per_thread,
bool dst_stream_cache_policy, bool is_copy)
{
assert(util_is_power_of_two_nonzero(num_dwords_per_thread));
unsigned store_qualifier = TGSI_MEMORY_COHERENT | TGSI_MEMORY_RESTRICT;
if (dst_stream_cache_policy)
store_qualifier |= TGSI_MEMORY_STREAM_CACHE_POLICY;
/* Don't cache loads, because there is no reuse. */
unsigned load_qualifier = store_qualifier | TGSI_MEMORY_STREAM_CACHE_POLICY;
unsigned num_mem_ops = MAX2(1, num_dwords_per_thread / 4);
unsigned *inst_dwords = alloca(num_mem_ops * sizeof(unsigned));
for (unsigned i = 0; i < num_mem_ops; i++) {
if (i*4 < num_dwords_per_thread)
inst_dwords[i] = MIN2(4, num_dwords_per_thread - i*4);
}
struct ureg_program *ureg = ureg_create(PIPE_SHADER_COMPUTE);
if (!ureg)
return NULL;
ureg_property(ureg, TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH, 64);
ureg_property(ureg, TGSI_PROPERTY_CS_FIXED_BLOCK_HEIGHT, 1);
ureg_property(ureg, TGSI_PROPERTY_CS_FIXED_BLOCK_DEPTH, 1);
struct ureg_src value;
if (!is_copy) {
ureg_property(ureg, TGSI_PROPERTY_CS_USER_DATA_DWORDS, inst_dwords[0]);
value = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_CS_USER_DATA, 0);
}
struct ureg_src tid = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_THREAD_ID, 0);
struct ureg_src blk = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_BLOCK_ID, 0);
struct ureg_dst store_addr = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_X);
struct ureg_dst load_addr = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_X);
struct ureg_dst dstbuf = ureg_dst(ureg_DECL_buffer(ureg, 0, false));
struct ureg_src srcbuf;
struct ureg_src *values = NULL;
if (is_copy) {
srcbuf = ureg_DECL_buffer(ureg, 1, false);
values = malloc(num_mem_ops * sizeof(struct ureg_src));
}
/* If there are multiple stores, the first store writes into 0+tid,
* the 2nd store writes into 64+tid, the 3rd store writes into 128+tid, etc.
*/
ureg_UMAD(ureg, store_addr, blk, ureg_imm1u(ureg, 64 * num_mem_ops), tid);
/* Convert from a "store size unit" into bytes. */
ureg_UMUL(ureg, store_addr, ureg_src(store_addr),
ureg_imm1u(ureg, 4 * inst_dwords[0]));
ureg_MOV(ureg, load_addr, ureg_src(store_addr));
/* Distance between a load and a store for latency hiding. */
unsigned load_store_distance = is_copy ? 8 : 0;
for (unsigned i = 0; i < num_mem_ops + load_store_distance; i++) {
int d = i - load_store_distance;
if (is_copy && i < num_mem_ops) {
if (i) {
ureg_UADD(ureg, load_addr, ureg_src(load_addr),
ureg_imm1u(ureg, 4 * inst_dwords[i] * 64));
}
values[i] = ureg_src(ureg_DECL_temporary(ureg));
struct ureg_dst dst =
ureg_writemask(ureg_dst(values[i]),
u_bit_consecutive(0, inst_dwords[i]));
struct ureg_src srcs[] = {srcbuf, ureg_src(load_addr)};
ureg_memory_insn(ureg, TGSI_OPCODE_LOAD, &dst, 1, srcs, 2,
load_qualifier, TGSI_TEXTURE_BUFFER, 0);
}
if (d >= 0) {
if (d) {
ureg_UADD(ureg, store_addr, ureg_src(store_addr),
ureg_imm1u(ureg, 4 * inst_dwords[d] * 64));
}
struct ureg_dst dst =
ureg_writemask(dstbuf, u_bit_consecutive(0, inst_dwords[d]));
struct ureg_src srcs[] =
{ureg_src(store_addr), is_copy ? values[d] : value};
ureg_memory_insn(ureg, TGSI_OPCODE_STORE, &dst, 1, srcs, 2,
store_qualifier, TGSI_TEXTURE_BUFFER, 0);
}
}
ureg_END(ureg);
struct pipe_compute_state state = {};
state.ir_type = PIPE_SHADER_IR_TGSI;
state.prog = ureg_get_tokens(ureg, NULL);
void *cs = ctx->create_compute_state(ctx, &state);
ureg_destroy(ureg);
free(values);
return cs;
}
/**
* Translate a geometry program to create a new variant.
*/
static struct st_gp_variant *
st_translate_geometry_program(struct st_context *st,
struct st_geometry_program *stgp,
const struct st_gp_variant_key *key)
{
GLuint inputMapping[GEOM_ATTRIB_MAX];
GLuint outputMapping[GEOM_RESULT_MAX];
struct pipe_context *pipe = st->pipe;
enum pipe_error error;
GLuint attr;
const GLbitfield inputsRead = stgp->Base.Base.InputsRead;
GLuint vslot = 0;
GLuint num_generic = 0;
uint gs_num_inputs = 0;
uint gs_builtin_inputs = 0;
uint gs_array_offset = 0;
ubyte gs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
ubyte gs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
uint gs_num_outputs = 0;
GLint i;
GLuint maxSlot = 0;
struct ureg_program *ureg;
struct st_gp_variant *gpv;
gpv = CALLOC_STRUCT(st_gp_variant);
if (!gpv)
return NULL;
_mesa_remove_output_reads(&stgp->Base.Base, PROGRAM_OUTPUT);
_mesa_remove_output_reads(&stgp->Base.Base, PROGRAM_VARYING);
ureg = ureg_create( TGSI_PROCESSOR_GEOMETRY );
if (ureg == NULL) {
FREE(gpv);
return NULL;
}
/* which vertex output goes to the first geometry input */
vslot = 0;
memset(inputMapping, 0, sizeof(inputMapping));
memset(outputMapping, 0, sizeof(outputMapping));
/*
* Convert Mesa program inputs to TGSI input register semantics.
*/
for (attr = 0; attr < GEOM_ATTRIB_MAX; attr++) {
if (inputsRead & (1 << attr)) {
const GLuint slot = gs_num_inputs;
gs_num_inputs++;
inputMapping[attr] = slot;
stgp->input_map[slot + gs_array_offset] = vslot - gs_builtin_inputs;
stgp->input_to_index[attr] = vslot;
stgp->index_to_input[vslot] = attr;
++vslot;
if (attr != GEOM_ATTRIB_PRIMITIVE_ID) {
gs_array_offset += 2;
} else
++gs_builtin_inputs;
#if 0
debug_printf("input map at %d = %d\n",
slot + gs_array_offset, stgp->input_map[slot + gs_array_offset]);
#endif
switch (attr) {
case GEOM_ATTRIB_PRIMITIVE_ID:
stgp->input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
stgp->input_semantic_index[slot] = 0;
break;
case GEOM_ATTRIB_POSITION:
stgp->input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
stgp->input_semantic_index[slot] = 0;
break;
case GEOM_ATTRIB_COLOR0:
stgp->input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
stgp->input_semantic_index[slot] = 0;
break;
case GEOM_ATTRIB_COLOR1:
stgp->input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
stgp->input_semantic_index[slot] = 1;
break;
case GEOM_ATTRIB_FOG_FRAG_COORD:
stgp->input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
stgp->input_semantic_index[slot] = 0;
break;
case GEOM_ATTRIB_TEX_COORD:
stgp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
stgp->input_semantic_index[slot] = num_generic++;
break;
case GEOM_ATTRIB_VAR0:
/* fall-through */
default:
stgp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
stgp->input_semantic_index[slot] = num_generic++;
}
}
}
/* initialize output semantics to defaults */
for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
gs_output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
gs_output_semantic_index[i] = 0;
}
num_generic = 0;
/*
* Determine number of outputs, the (default) output register
* mapping and the semantic information for each output.
*/
for (attr = 0; attr < GEOM_RESULT_MAX; attr++) {
if (stgp->Base.Base.OutputsWritten & BITFIELD64_BIT(attr)) {
GLuint slot;
slot = gs_num_outputs;
gs_num_outputs++;
outputMapping[attr] = slot;
switch (attr) {
case GEOM_RESULT_POS:
assert(slot == 0);
gs_output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
gs_output_semantic_index[slot] = 0;
break;
case GEOM_RESULT_COL0:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
gs_output_semantic_index[slot] = 0;
break;
case GEOM_RESULT_COL1:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
gs_output_semantic_index[slot] = 1;
break;
case GEOM_RESULT_SCOL0:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
gs_output_semantic_index[slot] = 0;
break;
case GEOM_RESULT_SCOL1:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
gs_output_semantic_index[slot] = 1;
break;
case GEOM_RESULT_FOGC:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
gs_output_semantic_index[slot] = 0;
break;
case GEOM_RESULT_PSIZ:
gs_output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
gs_output_semantic_index[slot] = 0;
break;
case GEOM_RESULT_TEX0:
case GEOM_RESULT_TEX1:
case GEOM_RESULT_TEX2:
case GEOM_RESULT_TEX3:
case GEOM_RESULT_TEX4:
case GEOM_RESULT_TEX5:
case GEOM_RESULT_TEX6:
case GEOM_RESULT_TEX7:
/* fall-through */
case GEOM_RESULT_VAR0:
/* fall-through */
default:
assert(slot < Elements(gs_output_semantic_name));
/* use default semantic info */
gs_output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
gs_output_semantic_index[slot] = num_generic++;
}
}
}
assert(gs_output_semantic_name[0] == TGSI_SEMANTIC_POSITION);
/* find max output slot referenced to compute gs_num_outputs */
for (attr = 0; attr < GEOM_RESULT_MAX; attr++) {
if (outputMapping[attr] != ~0 && outputMapping[attr] > maxSlot)
maxSlot = outputMapping[attr];
}
gs_num_outputs = maxSlot + 1;
#if 0 /* debug */
{
GLuint i;
printf("outputMapping? %d\n", outputMapping ? 1 : 0);
if (outputMapping) {
printf("attr -> slot\n");
for (i = 0; i < 16; i++) {
printf(" %2d %3d\n", i, outputMapping[i]);
}
}
printf("slot sem_name sem_index\n");
for (i = 0; i < gs_num_outputs; i++) {
printf(" %2d %d %d\n",
i,
gs_output_semantic_name[i],
gs_output_semantic_index[i]);
}
}
#endif
/* free old shader state, if any */
if (stgp->tgsi.tokens) {
st_free_tokens(stgp->tgsi.tokens);
stgp->tgsi.tokens = NULL;
}
ureg_property_gs_input_prim(ureg, stgp->Base.InputType);
ureg_property_gs_output_prim(ureg, stgp->Base.OutputType);
ureg_property_gs_max_vertices(ureg, stgp->Base.VerticesOut);
error = st_translate_mesa_program(st->ctx,
TGSI_PROCESSOR_GEOMETRY,
ureg,
&stgp->Base.Base,
/* inputs */
gs_num_inputs,
inputMapping,
stgp->input_semantic_name,
stgp->input_semantic_index,
NULL,
/* outputs */
gs_num_outputs,
outputMapping,
gs_output_semantic_name,
gs_output_semantic_index,
FALSE);
stgp->num_inputs = gs_num_inputs;
stgp->tgsi.tokens = ureg_get_tokens( ureg, NULL );
ureg_destroy( ureg );
/* fill in new variant */
gpv->driver_shader = pipe->create_gs_state(pipe, &stgp->tgsi);
gpv->key = *key;
if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
_mesa_print_program(&stgp->Base.Base);
debug_printf("\n");
}
if (ST_DEBUG & DEBUG_TGSI) {
tgsi_dump(stgp->tgsi.tokens, 0);
debug_printf("\n");
}
return gpv;
}
/**
* Translate a Mesa fragment shader into a TGSI shader using extra info in
* the key.
* \return new fragment program variant
*/
static struct st_fp_variant *
st_translate_fragment_program(struct st_context *st,
struct st_fragment_program *stfp,
const struct st_fp_variant_key *key)
{
struct pipe_context *pipe = st->pipe;
struct st_fp_variant *variant = CALLOC_STRUCT(st_fp_variant);
GLboolean deleteFP = GL_FALSE;
if (!variant)
return NULL;
assert(!(key->bitmap && key->drawpixels));
#if FEATURE_drawpix
if (key->bitmap) {
/* glBitmap drawing */
struct gl_fragment_program *fp; /* we free this temp program below */
st_make_bitmap_fragment_program(st, &stfp->Base,
&fp, &variant->bitmap_sampler);
variant->parameters = _mesa_clone_parameter_list(fp->Base.Parameters);
stfp = st_fragment_program(fp);
deleteFP = GL_TRUE;
}
else if (key->drawpixels) {
/* glDrawPixels drawing */
struct gl_fragment_program *fp; /* we free this temp program below */
if (key->drawpixels_z || key->drawpixels_stencil) {
fp = st_make_drawpix_z_stencil_program(st, key->drawpixels_z,
key->drawpixels_stencil);
}
else {
/* RGBA */
st_make_drawpix_fragment_program(st, &stfp->Base, &fp);
variant->parameters = _mesa_clone_parameter_list(fp->Base.Parameters);
deleteFP = GL_TRUE;
}
stfp = st_fragment_program(fp);
}
#endif
if (!stfp->tgsi.tokens) {
/* need to translate Mesa instructions to TGSI now */
GLuint outputMapping[FRAG_RESULT_MAX];
GLuint inputMapping[FRAG_ATTRIB_MAX];
GLuint interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */
GLuint attr;
enum pipe_error error;
const GLbitfield inputsRead = stfp->Base.Base.InputsRead;
struct ureg_program *ureg;
GLboolean write_all = GL_FALSE;
ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
ubyte 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;
_mesa_remove_output_reads(&stfp->Base.Base, PROGRAM_OUTPUT);
/*
* 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++;
inputMapping[attr] = slot;
switch (attr) {
case FRAG_ATTRIB_WPOS:
input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
break;
case FRAG_ATTRIB_COL0:
input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
break;
case FRAG_ATTRIB_COL1:
input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
input_semantic_index[slot] = 1;
interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
break;
case FRAG_ATTRIB_FOGC:
input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
break;
case FRAG_ATTRIB_FACE:
input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
input_semantic_index[slot] = 0;
interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
break;
/* In most cases, there is nothing special about these
* inputs, so adopt a convention to use the generic
* semantic name and the mesa FRAG_ATTRIB_ number as the
* index.
*
* All that is required is that the vertex shader labels
* its own outputs similarly, and that the vertex shader
* generates at least every output required by the
* fragment shader plus fixed-function hardware (such as
* BFC).
*
* There is no requirement that semantic indexes start at
* zero or be restricted to a particular range -- nobody
* should be building tables based on semantic index.
*/
case FRAG_ATTRIB_PNTC:
case FRAG_ATTRIB_TEX0:
case FRAG_ATTRIB_TEX1:
case FRAG_ATTRIB_TEX2:
case FRAG_ATTRIB_TEX3:
case FRAG_ATTRIB_TEX4:
case FRAG_ATTRIB_TEX5:
case FRAG_ATTRIB_TEX6:
case FRAG_ATTRIB_TEX7:
case FRAG_ATTRIB_VAR0:
default:
/* Actually, let's try and zero-base this just for
* readability of the generated TGSI.
*/
assert(attr >= FRAG_ATTRIB_TEX0);
input_semantic_index[slot] = (attr - FRAG_ATTRIB_TEX0);
input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
if (attr == FRAG_ATTRIB_PNTC)
interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
else
interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
break;
}
}
else {
inputMapping[attr] = -1;
}
}
/*
* Semantics and mapping for outputs
*/
{
uint numColors = 0;
GLbitfield64 outputsWritten = stfp->Base.Base.OutputsWritten;
/* if z is written, emit that first */
if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
fs_output_semantic_index[fs_num_outputs] = 0;
outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
fs_num_outputs++;
outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
}
if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) {
fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL;
fs_output_semantic_index[fs_num_outputs] = 0;
outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs;
fs_num_outputs++;
outputsWritten &= ~(1 << FRAG_RESULT_STENCIL);
}
/* handle remaning outputs (color) */
for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
if (outputsWritten & BITFIELD64_BIT(attr)) {
switch (attr) {
case FRAG_RESULT_DEPTH:
case FRAG_RESULT_STENCIL:
/* handled above */
assert(0);
break;
case FRAG_RESULT_COLOR:
write_all = GL_TRUE; /* fallthrough */
default:
assert(attr == FRAG_RESULT_COLOR ||
(FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX));
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;
}
fs_num_outputs++;
}
}
}
ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT );
if (ureg == NULL)
return NULL;
if (ST_DEBUG & DEBUG_MESA) {
_mesa_print_program(&stfp->Base.Base);
_mesa_print_program_parameters(st->ctx, &stfp->Base.Base);
debug_printf("\n");
}
if (write_all == GL_TRUE)
ureg_property_fs_color0_writes_all_cbufs(ureg, 1);
error = st_translate_mesa_program(st->ctx,
TGSI_PROCESSOR_FRAGMENT,
ureg,
&stfp->Base.Base,
/* inputs */
fs_num_inputs,
inputMapping,
input_semantic_name,
input_semantic_index,
interpMode,
/* outputs */
fs_num_outputs,
outputMapping,
fs_output_semantic_name,
fs_output_semantic_index, FALSE );
stfp->tgsi.tokens = ureg_get_tokens( ureg, NULL );
ureg_destroy( ureg );
}
/* fill in variant */
variant->driver_shader = pipe->create_fs_state(pipe, &stfp->tgsi);
variant->key = *key;
if (ST_DEBUG & DEBUG_TGSI) {
tgsi_dump( stfp->tgsi.tokens, 0/*TGSI_DUMP_VERBOSE*/ );
debug_printf("\n");
}
if (deleteFP) {
/* Free the temporary program made above */
struct gl_fragment_program *fp = &stfp->Base;
_mesa_reference_fragprog(st->ctx, &fp, NULL);
}
return variant;
}
/**
* The current vertex shader is already executed by the 'draw'
* module, so we just need to generate a simple vertex shader
* to pass through all those VS outputs that will
* be consumed by the fragment shader.
* Used when we employ the 'draw' module.
*/
static enum pipe_error
compile_passthrough_vs(struct svga_context *svga,
struct svga_vertex_shader *vs,
struct svga_fragment_shader *fs,
struct svga_shader_variant **out_variant)
{
struct svga_shader_variant *variant = NULL;
unsigned num_inputs;
unsigned i;
unsigned num_elements;
struct svga_vertex_shader new_vs;
struct ureg_src src[PIPE_MAX_SHADER_INPUTS];
struct ureg_dst dst[PIPE_MAX_SHADER_OUTPUTS];
struct ureg_program *ureg;
unsigned num_tokens;
struct svga_compile_key key;
enum pipe_error ret;
assert(svga_have_vgpu10(svga));
assert(fs);
num_inputs = fs->base.info.num_inputs;
ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
if (!ureg)
return PIPE_ERROR_OUT_OF_MEMORY;
/* draw will always add position */
dst[0] = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
src[0] = ureg_DECL_vs_input(ureg, 0);
num_elements = 1;
/**
* swtnl backend redefines the input layout based on the
* fragment shader's inputs. So we only need to passthrough
* those inputs that will be consumed by the fragment shader.
* Note: DX10 requires the number of vertex elements
* specified in the input layout to be no less than the
* number of inputs to the vertex shader.
*/
for (i = 0; i < num_inputs; i++) {
switch (fs->base.info.input_semantic_name[i]) {
case TGSI_SEMANTIC_COLOR:
case TGSI_SEMANTIC_GENERIC:
case TGSI_SEMANTIC_FOG:
dst[num_elements] = ureg_DECL_output(ureg,
fs->base.info.input_semantic_name[i],
fs->base.info.input_semantic_index[i]);
src[num_elements] = ureg_DECL_vs_input(ureg, num_elements);
num_elements++;
break;
default:
break;
}
}
for (i = 0; i < num_elements; i++) {
ureg_MOV(ureg, dst[i], src[i]);
}
ureg_END(ureg);
memset(&new_vs, 0, sizeof(new_vs));
new_vs.base.tokens = ureg_get_tokens(ureg, &num_tokens);
tgsi_scan_shader(new_vs.base.tokens, &new_vs.base.info);
memset(&key, 0, sizeof(key));
key.vs.undo_viewport = 1;
ret = compile_vs(svga, &new_vs, &key, &variant);
if (ret != PIPE_OK)
return ret;
ureg_free_tokens(new_vs.base.tokens);
ureg_destroy(ureg);
/* Overwrite the variant key to indicate it's a pass-through VS */
memset(&variant->key, 0, sizeof(variant->key));
variant->key.vs.passthrough = 1;
variant->key.vs.undo_viewport = 1;
*out_variant = variant;
return PIPE_OK;
}
static void *
combine_shaders(const struct shader_asm_info *shaders[SHADER_STAGES], int num_shaders,
struct pipe_context *pipe,
struct pipe_shader_state *shader)
{
VGboolean declare_input = VG_FALSE;
VGint start_const = -1, end_const = 0;
VGint start_temp = -1, end_temp = 0;
VGint start_sampler = -1, end_sampler = 0;
VGint i, current_shader = 0;
VGint num_consts, num_temps, num_samplers;
struct ureg_program *ureg;
struct ureg_src in[2];
struct ureg_src *sampler = NULL;
struct ureg_src *constant = NULL;
struct ureg_dst out, *temp = NULL;
void *p = NULL;
for (i = 0; i < num_shaders; ++i) {
if (shaders[i]->num_consts)
start_const = range_min(start_const, shaders[i]->start_const);
if (shaders[i]->num_temps)
start_temp = range_min(start_temp, shaders[i]->start_temp);
if (shaders[i]->num_samplers)
start_sampler = range_min(start_sampler, shaders[i]->start_sampler);
end_const = range_max(end_const, shaders[i]->start_const +
shaders[i]->num_consts);
end_temp = range_max(end_temp, shaders[i]->start_temp +
shaders[i]->num_temps);
end_sampler = range_max(end_sampler, shaders[i]->start_sampler +
shaders[i]->num_samplers);
if (shaders[i]->needs_position)
declare_input = VG_TRUE;
}
/* if they're still unitialized, initialize them */
if (start_const < 0)
start_const = 0;
if (start_temp < 0)
start_temp = 0;
if (start_sampler < 0)
start_sampler = 0;
num_consts = end_const - start_const;
num_temps = end_temp - start_temp;
num_samplers = end_sampler - start_sampler;
ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
if (!ureg)
return NULL;
if (declare_input) {
in[0] = ureg_DECL_fs_input(ureg,
TGSI_SEMANTIC_POSITION,
0,
TGSI_INTERPOLATE_LINEAR);
in[1] = ureg_DECL_fs_input(ureg,
TGSI_SEMANTIC_GENERIC,
0,
TGSI_INTERPOLATE_PERSPECTIVE);
}
/* we always have a color output */
out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
if (num_consts >= 1) {
constant = (struct ureg_src *) malloc(sizeof(struct ureg_src) * end_const);
for (i = start_const; i < end_const; i++) {
constant[i] = ureg_DECL_constant(ureg, i);
}
}
if (num_temps >= 1) {
temp = (struct ureg_dst *) malloc(sizeof(struct ureg_dst) * end_temp);
for (i = start_temp; i < end_temp; i++) {
temp[i] = ureg_DECL_temporary(ureg);
}
}
if (num_samplers >= 1) {
sampler = (struct ureg_src *) malloc(sizeof(struct ureg_src) * end_sampler);
for (i = start_sampler; i < end_sampler; i++) {
sampler[i] = ureg_DECL_sampler(ureg, i);
}
}
while (current_shader < num_shaders) {
if ((current_shader + 1) == num_shaders) {
shaders[current_shader]->func(ureg,
&out,
in,
sampler,
temp,
constant);
} else {
shaders[current_shader]->func(ureg,
&temp[0],
in,
sampler,
temp,
constant);
}
current_shader++;
}
ureg_END(ureg);
shader->tokens = ureg_finalize(ureg);
if(!shader->tokens)
return NULL;
p = pipe->create_fs_state(pipe, shader);
ureg_destroy(ureg);
if (num_temps >= 1) {
for (i = start_temp; i < end_temp; i++) {
ureg_release_temporary(ureg, temp[i]);
}
}
if (temp)
free(temp);
if (constant)
free(constant);
if (sampler)
free(sampler);
return p;
}
