static void
make_gs_key(struct svga_context *svga, struct svga_compile_key *key)
{
struct svga_geometry_shader *gs = svga->curr.gs;
memset(key, 0, sizeof *key);
/*
* SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER
*/
svga_init_shader_key_common(svga, PIPE_SHADER_GEOMETRY, key);
memcpy(key->generic_remap_table, gs->generic_remap_table,
sizeof(gs->generic_remap_table));
key->gs.vs_generic_outputs = svga->curr.vs->generic_outputs;
key->gs.need_prescale = svga->state.hw_clear.prescale.enabled;
key->gs.writes_psize = gs->base.info.writes_psize;
key->gs.wide_point = gs->wide_point;
key->sprite_coord_enable = svga->curr.rast->templ.sprite_coord_enable;
key->sprite_origin_lower_left = (svga->curr.rast->templ.sprite_coord_mode
== PIPE_SPRITE_COORD_LOWER_LEFT);
/* SVGA_NEW_RAST */
key->clip_plane_enable = svga->curr.rast->templ.clip_plane_enable;
}
/* SVGA_NEW_PRESCALE, SVGA_NEW_RAST, SVGA_NEW_FS
*/
static void
make_vs_key(struct svga_context *svga, struct svga_compile_key *key)
{
const unsigned shader = PIPE_SHADER_VERTEX;
memset(key, 0, sizeof *key);
if (svga->state.sw.need_swtnl && svga_have_vgpu10(svga)) {
/* Set both of these flags, to match compile_passthrough_vs() */
key->vs.passthrough = 1;
key->vs.undo_viewport = 1;
return;
}
/* SVGA_NEW_PRESCALE */
key->vs.need_prescale = svga->state.hw_clear.prescale.enabled &&
(svga->curr.gs == NULL);
/* SVGA_NEW_RAST */
key->vs.allow_psiz = svga->curr.rast->templ.point_size_per_vertex;
/* SVGA_NEW_FS */
key->vs.fs_generic_inputs = svga->curr.fs->generic_inputs;
svga_remap_generics(key->vs.fs_generic_inputs, key->generic_remap_table);
/* SVGA_NEW_VELEMENT */
key->vs.adjust_attrib_range = svga->curr.velems->adjust_attrib_range;
key->vs.adjust_attrib_w_1 = svga->curr.velems->adjust_attrib_w_1;
key->vs.attrib_is_pure_int = svga->curr.velems->attrib_is_pure_int;
key->vs.adjust_attrib_itof = svga->curr.velems->adjust_attrib_itof;
key->vs.adjust_attrib_utof = svga->curr.velems->adjust_attrib_utof;
key->vs.attrib_is_bgra = svga->curr.velems->attrib_is_bgra;
key->vs.attrib_puint_to_snorm = svga->curr.velems->attrib_puint_to_snorm;
key->vs.attrib_puint_to_uscaled = svga->curr.velems->attrib_puint_to_uscaled;
key->vs.attrib_puint_to_sscaled = svga->curr.velems->attrib_puint_to_sscaled;
/* SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER */
svga_init_shader_key_common(svga, shader, key);
/* SVGA_NEW_RAST */
key->clip_plane_enable = svga->curr.rast->templ.clip_plane_enable;
}
/* SVGA_NEW_TEXTURE_BINDING
* SVGA_NEW_RAST
* SVGA_NEW_NEED_SWTNL
* SVGA_NEW_SAMPLER
*/
static enum pipe_error
make_fs_key(const struct svga_context *svga,
struct svga_fragment_shader *fs,
struct svga_compile_key *key)
{
const unsigned shader = PIPE_SHADER_FRAGMENT;
unsigned i;
memset(key, 0, sizeof *key);
memcpy(key->generic_remap_table, fs->generic_remap_table,
sizeof(fs->generic_remap_table));
/* SVGA_NEW_GS, SVGA_NEW_VS
*/
if (svga->curr.gs) {
key->fs.gs_generic_outputs = svga->curr.gs->generic_outputs;
} else {
key->fs.vs_generic_outputs = svga->curr.vs->generic_outputs;
}
/* Only need fragment shader fixup for twoside lighting if doing
* hwtnl. Otherwise the draw module does the whole job for us.
*
* SVGA_NEW_SWTNL
*/
if (!svga->state.sw.need_swtnl) {
/* SVGA_NEW_RAST, SVGA_NEW_REDUCED_PRIMITIVE
*/
key->fs.light_twoside = svga->curr.rast->templ.light_twoside;
key->fs.front_ccw = svga->curr.rast->templ.front_ccw;
key->fs.pstipple = (svga->curr.rast->templ.poly_stipple_enable &&
svga->curr.reduced_prim == PIPE_PRIM_TRIANGLES);
key->fs.aa_point = (svga->curr.rast->templ.point_smooth &&
svga->curr.reduced_prim == PIPE_PRIM_POINTS &&
(svga->curr.rast->pointsize > 1.0 ||
svga->curr.vs->base.info.writes_psize));
if (key->fs.aa_point) {
assert(svga->curr.gs != NULL);
assert(svga->curr.gs->aa_point_coord_index != -1);
key->fs.aa_point_coord_index = svga->curr.gs->aa_point_coord_index;
}
}
/* The blend workaround for simulating logicop xor behaviour
* requires that the incoming fragment color be white. This change
* achieves that by creating a variant of the current fragment
* shader that overrides all output colors with 1,1,1,1
*
* This will work for most shaders, including those containing
* TEXKIL and/or depth-write. However, it will break on the
* combination of xor-logicop plus alphatest.
*
* Ultimately, we could implement alphatest in the shader using
* texkil prior to overriding the outgoing fragment color.
*
* SVGA_NEW_BLEND
*/
if (svga->curr.blend->need_white_fragments) {
key->fs.white_fragments = 1;
}
#ifdef DEBUG
/*
* We expect a consistent set of samplers and sampler views.
* Do some debug checks/warnings here.
*/
{
static boolean warned = FALSE;
unsigned i, n = MAX2(svga->curr.num_sampler_views[shader],
svga->curr.num_samplers[shader]);
/* Only warn once to prevent too much debug output */
if (!warned) {
if (svga->curr.num_sampler_views[shader] !=
svga->curr.num_samplers[shader]) {
debug_printf("svga: mismatched number of sampler views (%u) "
"vs. samplers (%u)\n",
svga->curr.num_sampler_views[shader],
svga->curr.num_samplers[shader]);
}
for (i = 0; i < n; i++) {
if ((svga->curr.sampler_views[shader][i] == NULL) !=
(svga->curr.sampler[shader][i] == NULL))
debug_printf("sampler_view[%u] = %p but sampler[%u] = %p\n",
i, svga->curr.sampler_views[shader][i],
i, svga->curr.sampler[shader][i]);
}
warned = TRUE;
}
}
#endif
/* XXX: want to limit this to the textures that the shader actually
* refers to.
*
* SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER
*/
svga_init_shader_key_common(svga, shader, key);
for (i = 0; i < svga->curr.num_samplers[shader]; ++i) {
struct pipe_sampler_view *view = svga->curr.sampler_views[shader][i];
const struct svga_sampler_state *sampler = svga->curr.sampler[shader][i];
if (view) {
struct pipe_resource *tex = view->texture;
if (tex->target != PIPE_BUFFER) {
struct svga_texture *stex = svga_texture(tex);
SVGA3dSurfaceFormat format = stex->key.format;
if (!svga_have_vgpu10(svga) &&
(format == SVGA3D_Z_D16 ||
format == SVGA3D_Z_D24X8 ||
format == SVGA3D_Z_D24S8)) {
/* If we're sampling from a SVGA3D_Z_D16, SVGA3D_Z_D24X8,
* or SVGA3D_Z_D24S8 surface, we'll automatically get
* shadow comparison. But we only get LEQUAL mode.
* Set TEX_COMPARE_NONE here so we don't emit the extra FS
* code for shadow comparison.
*/
key->tex[i].compare_mode = PIPE_TEX_COMPARE_NONE;
key->tex[i].compare_func = PIPE_FUNC_NEVER;
/* These depth formats _only_ support comparison mode and
* not ordinary sampling so warn if the later is expected.
*/
if (sampler->compare_mode != PIPE_TEX_COMPARE_R_TO_TEXTURE) {
debug_warn_once("Unsupported shadow compare mode");
}
/* The shader translation code can emit code to
* handle ALWAYS and NEVER compare functions
*/
else if (sampler->compare_func == PIPE_FUNC_ALWAYS ||
sampler->compare_func == PIPE_FUNC_NEVER) {
key->tex[i].compare_mode = sampler->compare_mode;
key->tex[i].compare_func = sampler->compare_func;
}
else if (sampler->compare_func != PIPE_FUNC_LEQUAL) {
debug_warn_once("Unsupported shadow compare function");
}
}
else {
/* For other texture formats, just use the compare func/mode
* as-is. Should be no-ops for color textures. For depth
* textures, we do not get automatic depth compare. We have
* to do it ourselves in the shader. And we don't get PCF.
*/
key->tex[i].compare_mode = sampler->compare_mode;
key->tex[i].compare_func = sampler->compare_func;
}
}
}
}
/* sprite coord gen state */
for (i = 0; i < svga->curr.num_samplers[shader]; ++i) {
key->tex[i].sprite_texgen =
svga->curr.rast->templ.sprite_coord_enable & (1 << i);
}
key->sprite_origin_lower_left = (svga->curr.rast->templ.sprite_coord_mode
== PIPE_SPRITE_COORD_LOWER_LEFT);
key->fs.flatshade = svga->curr.rast->templ.flatshade;
/* SVGA_NEW_DEPTH_STENCIL_ALPHA */
if (svga_have_vgpu10(svga)) {
/* Alpha testing is not supported in integer-valued render targets. */
if (svga_has_any_integer_cbufs(svga)) {
key->fs.alpha_func = SVGA3D_CMP_ALWAYS;
key->fs.alpha_ref = 0;
}
else {
key->fs.alpha_func = svga->curr.depth->alphafunc;
key->fs.alpha_ref = svga->curr.depth->alpharef;
}
}
/* SVGA_NEW_FRAME_BUFFER */
if (fs->base.info.properties[TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS]) {
/* Replicate color0 output to N colorbuffers */
key->fs.write_color0_to_n_cbufs = svga->curr.framebuffer.nr_cbufs;
}
return PIPE_OK;
}