void *
util_make_geometry_passthrough_shader(struct pipe_context *pipe,
uint num_attribs,
const ubyte *semantic_names,
const ubyte *semantic_indexes)
{
static const unsigned zero[4] = {0, 0, 0, 0};
struct ureg_program *ureg;
struct ureg_dst dst[PIPE_MAX_SHADER_OUTPUTS];
struct ureg_src src[PIPE_MAX_SHADER_INPUTS];
struct ureg_src imm;
unsigned i;
ureg = ureg_create(PIPE_SHADER_GEOMETRY);
if (!ureg)
return NULL;
ureg_property(ureg, TGSI_PROPERTY_GS_INPUT_PRIM, PIPE_PRIM_POINTS);
ureg_property(ureg, TGSI_PROPERTY_GS_OUTPUT_PRIM, PIPE_PRIM_POINTS);
ureg_property(ureg, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES, 1);
ureg_property(ureg, TGSI_PROPERTY_GS_INVOCATIONS, 1);
imm = ureg_DECL_immediate_uint(ureg, zero, 4);
/**
* Loop over all the attribs and declare the corresponding
* declarations in the geometry shader
*/
for (i = 0; i < num_attribs; i++) {
src[i] = ureg_DECL_input(ureg, semantic_names[i],
semantic_indexes[i], 0, 1);
src[i] = ureg_src_dimension(src[i], 0);
dst[i] = ureg_DECL_output(ureg, semantic_names[i], semantic_indexes[i]);
}
/* MOV dst[i] src[i] */
for (i = 0; i < num_attribs; i++) {
ureg_MOV(ureg, dst[i], src[i]);
}
/* EMIT IMM[0] */
ureg_insn(ureg, TGSI_OPCODE_EMIT, NULL, 0, &imm, 1, 0);
/* END */
ureg_END(ureg);
return ureg_create_shader_and_destroy(ureg, pipe);
}
void *
util_make_vertex_passthrough_shader_with_so(struct pipe_context *pipe,
uint num_attribs,
const uint *semantic_names,
const uint *semantic_indexes,
bool window_space,
const struct pipe_stream_output_info *so)
{
struct ureg_program *ureg;
uint i;
ureg = ureg_create( PIPE_SHADER_VERTEX );
if (!ureg)
return NULL;
if (window_space)
ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, TRUE);
for (i = 0; i < num_attribs; i++) {
struct ureg_src src;
struct ureg_dst dst;
src = ureg_DECL_vs_input( ureg, i );
dst = ureg_DECL_output( ureg,
semantic_names[i],
semantic_indexes[i]);
ureg_MOV( ureg, dst, src );
}
ureg_END( ureg );
return ureg_create_shader_with_so_and_destroy( ureg, pipe, so );
}
void *
st_pbo_create_gs(struct st_context *st)
{
static const int zero = 0;
struct ureg_program *ureg;
struct ureg_dst out_pos;
struct ureg_dst out_layer;
struct ureg_src in_pos;
struct ureg_src imm;
unsigned i;
ureg = ureg_create(PIPE_SHADER_GEOMETRY);
if (!ureg)
return NULL;
ureg_property(ureg, TGSI_PROPERTY_GS_INPUT_PRIM, PIPE_PRIM_TRIANGLES);
ureg_property(ureg, TGSI_PROPERTY_GS_OUTPUT_PRIM, PIPE_PRIM_TRIANGLE_STRIP);
ureg_property(ureg, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES, 3);
out_pos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
out_layer = ureg_DECL_output(ureg, TGSI_SEMANTIC_LAYER, 0);
in_pos = ureg_DECL_input(ureg, TGSI_SEMANTIC_POSITION, 0, 0, 1);
imm = ureg_DECL_immediate_int(ureg, &zero, 1);
for (i = 0; i < 3; ++i) {
struct ureg_src in_pos_vertex = ureg_src_dimension(in_pos, i);
/* out_pos = in_pos[i] */
ureg_MOV(ureg, out_pos, in_pos_vertex);
/* out_layer.x = f2i(in_pos[i].z) */
ureg_F2I(ureg, ureg_writemask(out_layer, TGSI_WRITEMASK_X),
ureg_scalar(in_pos_vertex, TGSI_SWIZZLE_Z));
ureg_EMIT(ureg, ureg_scalar(imm, TGSI_SWIZZLE_X));
}
ureg_END(ureg);
return ureg_create_shader_and_destroy(ureg, st->pipe);
}
/**
* Emit fragment position/coordinate code.
*/
static void
emit_wpos(struct st_context *st,
struct st_translate *t,
const struct gl_program *program,
struct ureg_program *ureg)
{
const struct gl_fragment_program *fp =
(const struct gl_fragment_program *) program;
struct pipe_screen *pscreen = st->pipe->screen;
GLfloat adjX = 0.0f;
GLfloat adjY[2] = { 0.0f, 0.0f };
boolean invert = FALSE;
/* Query the pixel center conventions supported by the pipe driver and set
* adjX, adjY to help out if it cannot handle the requested one internally.
*
* The bias of the y-coordinate depends on whether y-inversion takes place
* (adjY[1]) or not (adjY[0]), which is in turn dependent on whether we are
* drawing to an FBO (causes additional inversion), and whether the the pipe
* driver origin and the requested origin differ (the latter condition is
* stored in the 'invert' variable).
*
* For height = 100 (i = integer, h = half-integer, l = lower, u = upper):
*
* center shift only:
* i -> h: +0.5
* h -> i: -0.5
*
* inversion only:
* l,i -> u,i: ( 0.0 + 1.0) * -1 + 100 = 99
* l,h -> u,h: ( 0.5 + 0.0) * -1 + 100 = 99.5
* u,i -> l,i: (99.0 + 1.0) * -1 + 100 = 0
* u,h -> l,h: (99.5 + 0.0) * -1 + 100 = 0.5
*
* inversion and center shift:
* l,i -> u,h: ( 0.0 + 0.5) * -1 + 100 = 99.5
* l,h -> u,i: ( 0.5 + 0.5) * -1 + 100 = 99
* u,i -> l,h: (99.0 + 0.5) * -1 + 100 = 0.5
* u,h -> l,i: (99.5 + 0.5) * -1 + 100 = 0
*/
if (fp->OriginUpperLeft) {
/* Fragment shader wants origin in upper-left */
if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT)) {
/* the driver supports upper-left origin */
}
else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT)) {
/* the driver supports lower-left origin, need to invert Y */
ureg_property(ureg, TGSI_PROPERTY_FS_COORD_ORIGIN,
TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
invert = TRUE;
}
else
assert(0);
}
else {
/* Fragment shader wants origin in lower-left */
if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT))
/* the driver supports lower-left origin */
ureg_property(ureg, TGSI_PROPERTY_FS_COORD_ORIGIN,
TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT))
/* the driver supports upper-left origin, need to invert Y */
invert = TRUE;
else
assert(0);
}
if (fp->PixelCenterInteger) {
/* Fragment shader wants pixel center integer */
if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) {
/* the driver supports pixel center integer */
adjY[1] = 1.0f;
ureg_property(ureg, TGSI_PROPERTY_FS_COORD_PIXEL_CENTER,
TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
}
else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) {
/* the driver supports pixel center half integer, need to bias X,Y */
adjX = -0.5f;
adjY[0] = -0.5f;
adjY[1] = 0.5f;
}
else
assert(0);
}
else {
/* Fragment shader wants pixel center half integer */
if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) {
/* the driver supports pixel center half integer */
}
else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) {
/* the driver supports pixel center integer, need to bias X,Y */
adjX = adjY[0] = adjY[1] = 0.5f;
ureg_property(ureg, TGSI_PROPERTY_FS_COORD_PIXEL_CENTER,
TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
}
else
assert(0);
}
/* we invert after adjustment so that we avoid the MOV to temporary,
* and reuse the adjustment ADD instead */
emit_wpos_adjustment(t, program, invert, adjX, adjY);
}
/**
* 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;
}
void *si_get_blitter_vs(struct si_context *sctx, enum blitter_attrib_type type,
unsigned num_layers)
{
unsigned vs_blit_property;
void **vs;
switch (type) {
case UTIL_BLITTER_ATTRIB_NONE:
vs = num_layers > 1 ? &sctx->vs_blit_pos_layered :
&sctx->vs_blit_pos;
vs_blit_property = SI_VS_BLIT_SGPRS_POS;
break;
case UTIL_BLITTER_ATTRIB_COLOR:
vs = num_layers > 1 ? &sctx->vs_blit_color_layered :
&sctx->vs_blit_color;
vs_blit_property = SI_VS_BLIT_SGPRS_POS_COLOR;
break;
case UTIL_BLITTER_ATTRIB_TEXCOORD_XY:
case UTIL_BLITTER_ATTRIB_TEXCOORD_XYZW:
assert(num_layers == 1);
vs = &sctx->vs_blit_texcoord;
vs_blit_property = SI_VS_BLIT_SGPRS_POS_TEXCOORD;
break;
default:
assert(0);
return NULL;
}
if (*vs)
return *vs;
struct ureg_program *ureg = ureg_create(PIPE_SHADER_VERTEX);
if (!ureg)
return NULL;
/* Tell the shader to load VS inputs from SGPRs: */
ureg_property(ureg, TGSI_PROPERTY_VS_BLIT_SGPRS, vs_blit_property);
ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, true);
/* This is just a pass-through shader with 1-3 MOV instructions. */
ureg_MOV(ureg,
ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0),
ureg_DECL_vs_input(ureg, 0));
if (type != UTIL_BLITTER_ATTRIB_NONE) {
ureg_MOV(ureg,
ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 0),
ureg_DECL_vs_input(ureg, 1));
}
if (num_layers > 1) {
struct ureg_src instance_id =
ureg_DECL_system_value(ureg, TGSI_SEMANTIC_INSTANCEID, 0);
struct ureg_dst layer =
ureg_DECL_output(ureg, TGSI_SEMANTIC_LAYER, 0);
ureg_MOV(ureg, ureg_writemask(layer, TGSI_WRITEMASK_X),
ureg_scalar(instance_id, TGSI_SWIZZLE_X));
}
ureg_END(ureg);
*vs = ureg_create_shader_and_destroy(ureg, &sctx->b);
return *vs;
}
/* 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;
}
