#include "util/u_inlines.h"
#include "pipe/p_defines.h"
+#include "util/format/u_format.h"
#include "util/u_math.h"
#include "util/u_memory.h"
#include "util/u_bitmask.h"
#include "svga_shader.h"
#include "svga_resource_texture.h"
#include "svga_tgsi.h"
+#include "svga_format.h"
#include "svga_hw_reg.h"
-static INLINE int
-compare_fs_keys(const struct svga_fs_compile_key *a,
- const struct svga_fs_compile_key *b)
-{
- unsigned keysize_a = svga_fs_key_size( a );
- unsigned keysize_b = svga_fs_key_size( b );
-
- if (keysize_a != keysize_b) {
- return (int)(keysize_a - keysize_b);
- }
- return memcmp( a, b, keysize_a );
-}
-
-
-/** Search for a fragment shader variant */
-static struct svga_shader_variant *
-search_fs_key(const struct svga_fragment_shader *fs,
- const struct svga_fs_compile_key *key)
-{
- struct svga_shader_variant *variant = fs->base.variants;
-
- assert(key);
-
- for ( ; variant; variant = variant->next) {
- if (compare_fs_keys( key, &variant->key.fkey ) == 0)
- return variant;
- }
-
- return NULL;
-}
-
-
/**
* 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.
+ * 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)
{
- static const float red[4] = { 1.0, 0.0, 0.0, 0.0 };
+#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);
+ ureg = ureg_create(PIPE_SHADER_FRAGMENT);
if (!ureg)
return NULL;
dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
- src = ureg_DECL_immediate(ureg, red, 4);
+ src = ureg_DECL_immediate(ureg, color, 4);
ureg_MOV(ureg, dst, src);
ureg_END(ureg);
- tokens = ureg_get_tokens(ureg, &num_tokens);
+ tokens = ureg_get_tokens(ureg, NULL);
ureg_destroy(ureg);
}
+static struct svga_shader_variant *
+translate_fragment_program(struct svga_context *svga,
+ const struct svga_fragment_shader *fs,
+ const struct svga_compile_key *key)
+{
+ if (svga_have_vgpu10(svga)) {
+ return svga_tgsi_vgpu10_translate(svga, &fs->base, key,
+ PIPE_SHADER_FRAGMENT);
+ }
+ else {
+ return svga_tgsi_vgpu9_translate(svga, &fs->base, key,
+ PIPE_SHADER_FRAGMENT);
+ }
+}
+
+
+/**
+ * Replace the given shader's instruction with a simple constant-color
+ * shader. We use this when normal shader translation fails.
+ */
+static struct svga_shader_variant *
+get_compiled_dummy_shader(struct svga_context *svga,
+ struct svga_fragment_shader *fs,
+ const struct svga_compile_key *key)
+{
+ const struct tgsi_token *dummy = get_dummy_fragment_shader();
+ struct svga_shader_variant *variant;
+
+ if (!dummy) {
+ return NULL;
+ }
+
+ FREE((void *) fs->base.tokens);
+ fs->base.tokens = dummy;
+
+ tgsi_scan_shader(fs->base.tokens, &fs->base.info);
+ fs->generic_inputs = svga_get_generic_inputs_mask(&fs->base.info);
+ svga_remap_generics(fs->generic_inputs, fs->generic_remap_table);
+
+ variant = translate_fragment_program(svga, fs, key);
+ return variant;
+}
+
+
/**
* Translate TGSI shader into an svga shader variant.
*/
static enum pipe_error
compile_fs(struct svga_context *svga,
struct svga_fragment_shader *fs,
- const struct svga_fs_compile_key *key,
+ const struct svga_compile_key *key,
struct svga_shader_variant **out_variant)
{
struct svga_shader_variant *variant;
enum pipe_error ret = PIPE_ERROR;
- variant = svga_translate_fragment_program( fs, key );
+ variant = translate_fragment_program(svga, fs, key);
if (variant == NULL) {
- /* some problem during translation, try the dummy shader */
- const struct tgsi_token *dummy = get_dummy_fragment_shader();
- if (!dummy) {
- ret = PIPE_ERROR_OUT_OF_MEMORY;
- goto fail;
- }
- debug_printf("Failed to compile fragment shader, using dummy shader instead.\n");
- FREE((void *) fs->base.tokens);
- fs->base.tokens = dummy;
- variant = svga_translate_fragment_program(fs, key);
- if (variant == NULL) {
- ret = PIPE_ERROR;
- goto fail;
- }
+ debug_printf("Failed to compile fragment shader,"
+ " using dummy shader instead.\n");
+ variant = get_compiled_dummy_shader(svga, fs, key);
+ }
+ else if (svga_shader_too_large(svga, variant)) {
+ /* too big, use dummy shader */
+ debug_printf("Shader too large (%u bytes),"
+ " using dummy shader instead.\n",
+ (unsigned) (variant->nr_tokens
+ * sizeof(variant->tokens[0])));
+ /* Free the too-large variant */
+ svga_destroy_shader_variant(svga, variant);
+ /* Use simple pass-through shader instead */
+ variant = get_compiled_dummy_shader(svga, fs, key);
}
- ret = svga_define_shader(svga, SVGA3D_SHADERTYPE_PS, variant);
- if (ret != PIPE_OK)
- goto fail;
+ if (!variant) {
+ return PIPE_ERROR;
+ }
+
+ ret = svga_define_shader(svga, variant);
+ if (ret != PIPE_OK) {
+ svga_destroy_shader_variant(svga, variant);
+ return ret;
+ }
*out_variant = variant;
- /* insert variants at head of linked list */
+ /* insert variant at head of linked list */
variant->next = fs->base.variants;
fs->base.variants = variant;
return PIPE_OK;
-
-fail:
- if (variant) {
- svga_destroy_shader_variant(svga, SVGA3D_SHADERTYPE_PS, variant);
- }
- return ret;
}
static enum pipe_error
make_fs_key(const struct svga_context *svga,
struct svga_fragment_shader *fs,
- struct svga_fs_compile_key *key)
+ struct svga_compile_key *key)
{
+ const enum pipe_shader_type shader = PIPE_SHADER_FRAGMENT;
unsigned i;
- int idx = 0;
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;
+ key->fs.layer_to_zero = !svga->curr.gs->base.info.writes_layer;
+ } else {
+ key->fs.vs_generic_outputs = svga->curr.vs->generic_outputs;
+ key->fs.layer_to_zero = 1;
+ }
+
/* 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_RAST, SVGA_NEW_REDUCED_PRIMITIVE
+ */
+ enum pipe_prim_type prim_mode;
+ struct svga_shader *shader;
+
+ /* Find the last shader in the vertex pipeline and the output primitive mode
+ * from that shader.
*/
- key->light_twoside = svga->curr.rast->templ.light_twoside;
- key->front_ccw = svga->curr.rast->templ.front_ccw;
+ if (svga->curr.tes) {
+ shader = &svga->curr.tes->base;
+ prim_mode = shader->info.properties[TGSI_PROPERTY_TES_PRIM_MODE];
+ } else if (svga->curr.gs) {
+ shader = &svga->curr.gs->base;
+ prim_mode = shader->info.properties[TGSI_PROPERTY_GS_OUTPUT_PRIM];
+ } else {
+ shader = &svga->curr.vs->base;
+ prim_mode = svga->curr.reduced_prim;
+ }
+
+ 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 &&
+ prim_mode == PIPE_PRIM_TRIANGLES);
+
+ key->fs.aa_point = (svga->curr.rast->templ.point_smooth &&
+ prim_mode == PIPE_PRIM_POINTS &&
+ (svga->curr.rast->pointsize > 1.0 ||
+ shader->info.writes_psize));
+
+ if (key->fs.aa_point && svga->curr.gs) {
+ 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
*
* SVGA_NEW_BLEND
*/
- if (svga->curr.blend->need_white_fragments) {
- key->white_fragments = 1;
+ key->fs.white_fragments = svga->curr.blend->need_white_fragments;
+
+ key->fs.alpha_to_one = svga->curr.blend->alpha_to_one;
+
+#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
*/
- for (i = 0; i < svga->curr.num_sampler_views; i++) {
- if (svga->curr.sampler_views[i]) {
- assert(svga->curr.sampler[i]);
- assert(svga->curr.sampler_views[i]->texture);
- key->tex[i].texture_target = svga->curr.sampler_views[i]->texture->target;
- if (!svga->curr.sampler[i]->normalized_coords) {
- key->tex[i].width_height_idx = idx++;
- key->tex[i].unnormalized = TRUE;
- ++key->num_unnormalized_coords;
- }
-
- key->tex[i].swizzle_r = svga->curr.sampler_views[i]->swizzle_r;
- key->tex[i].swizzle_g = svga->curr.sampler_views[i]->swizzle_g;
- key->tex[i].swizzle_b = svga->curr.sampler_views[i]->swizzle_b;
- key->tex[i].swizzle_a = svga->curr.sampler_views[i]->swizzle_a;
- }
- }
- key->num_textures = svga->curr.num_sampler_views;
-
- idx = 0;
- for (i = 0; i < svga->curr.num_samplers; ++i) {
- if (svga->curr.sampler_views[i]) {
- struct pipe_resource *tex = svga->curr.sampler_views[i]->texture;
- struct svga_texture *stex = svga_texture(tex);
- SVGA3dSurfaceFormat format = stex->key.format;
-
- if (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 (svga->curr.sampler[i]->compare_mode !=
- PIPE_TEX_COMPARE_R_TO_TEXTURE) {
- debug_warn_once("Unsupported shadow compare mode");
- }
- /* The only supported comparison mode is LEQUAL */
- if (svga->curr.sampler[i]->compare_func != PIPE_FUNC_LEQUAL) {
- debug_warn_once("Unsupported shadow compare function");
+ svga_init_shader_key_common(svga, shader, &fs->base, 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 = svga->curr.sampler[i]->compare_mode;
- key->tex[i].compare_func = svga->curr.sampler[i]->compare_func;
- }
}
}
/* sprite coord gen state */
- for (i = 0; i < svga->curr.num_samplers; ++i) {
- key->tex[i].sprite_texgen =
- svga->curr.rast->templ.sprite_coord_enable & (1 << i);
- }
+ 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_FRAME_BUFFER */
- if (fs->base.info.color0_writes_all_cbufs) {
- /* Replicate color0 output to N colorbuffers */
- key->write_color0_to_n_cbufs = svga->curr.framebuffer.nr_cbufs;
+ 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 | SVGA_NEW_BLEND */
+ if (fs->base.info.properties[TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS] ||
+ svga->curr.blend->need_white_fragments) {
+ /* Replicate color0 output (or white) to N colorbuffers */
+ key->fs.write_color0_to_n_cbufs = svga->curr.framebuffer.nr_cbufs;
+ }
+
+ return PIPE_OK;
+}
+
+
+/**
+ * svga_reemit_fs_bindings - Reemit the fragment shader bindings
+ */
+enum pipe_error
+svga_reemit_fs_bindings(struct svga_context *svga)
+{
+ enum pipe_error ret;
+
+ assert(svga->rebind.flags.fs);
+ assert(svga_have_gb_objects(svga));
+
+ if (!svga->state.hw_draw.fs)
+ return PIPE_OK;
+
+ if (!svga_need_to_rebind_resources(svga)) {
+ ret = svga->swc->resource_rebind(svga->swc, NULL,
+ svga->state.hw_draw.fs->gb_shader,
+ SVGA_RELOC_READ);
+ }
+ else {
+ if (svga_have_vgpu10(svga))
+ ret = SVGA3D_vgpu10_SetShader(svga->swc, SVGA3D_SHADERTYPE_PS,
+ svga->state.hw_draw.fs->gb_shader,
+ svga->state.hw_draw.fs->id);
+ else
+ ret = SVGA3D_SetGBShader(svga->swc, SVGA3D_SHADERTYPE_PS,
+ svga->state.hw_draw.fs->gb_shader);
}
+ if (ret != PIPE_OK)
+ return ret;
+
+ svga->rebind.flags.fs = FALSE;
return PIPE_OK;
}
+
static enum pipe_error
-emit_hw_fs(struct svga_context *svga, unsigned dirty)
+emit_hw_fs(struct svga_context *svga, uint64_t dirty)
{
struct svga_shader_variant *variant = NULL;
- unsigned id = SVGA3D_INVALID_ID;
enum pipe_error ret = PIPE_OK;
-
struct svga_fragment_shader *fs = svga->curr.fs;
- struct svga_fs_compile_key key;
+ struct svga_compile_key key;
+ struct svga_shader *prevShader = NULL; /* shader in the previous stage */
+
+ SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_EMITFS);
+
+ prevShader = svga->curr.gs ?
+ &svga->curr.gs->base : (svga->curr.tes ?
+ &svga->curr.tes->base : &svga->curr.vs->base);
+
+ /* Disable rasterization if rasterizer_discard flag is set or
+ * vs/gs does not output position.
+ */
+ svga->disable_rasterizer =
+ svga->curr.rast->templ.rasterizer_discard ||
+ !prevShader->info.writes_position;
+
+ /* Set FS to NULL when rasterization is to be disabled */
+ if (svga->disable_rasterizer) {
+ /* Set FS to NULL if it has not been done */
+ if (svga->state.hw_draw.fs) {
+ ret = svga_set_shader(svga, SVGA3D_SHADERTYPE_PS, NULL);
+ if (ret != PIPE_OK)
+ goto done;
+ }
+ svga->rebind.flags.fs = FALSE;
+ svga->state.hw_draw.fs = NULL;
+ goto done;
+ }
/* SVGA_NEW_BLEND
* SVGA_NEW_TEXTURE_BINDING
* SVGA_NEW_NEED_SWTNL
* SVGA_NEW_SAMPLER
* SVGA_NEW_FRAME_BUFFER
+ * SVGA_NEW_DEPTH_STENCIL_ALPHA
+ * SVGA_NEW_VS
*/
- ret = make_fs_key( svga, fs, &key );
+ ret = make_fs_key(svga, fs, &key);
if (ret != PIPE_OK)
- return ret;
+ goto done;
- variant = search_fs_key( fs, &key );
+ variant = svga_search_shader_key(&fs->base, &key);
if (!variant) {
- ret = compile_fs( svga, fs, &key, &variant );
+ ret = compile_fs(svga, fs, &key, &variant);
if (ret != PIPE_OK)
- return ret;
+ goto done;
}
- assert (variant);
- id = variant->id;
-
- assert(id != SVGA3D_INVALID_ID);
+ assert(variant);
if (variant != svga->state.hw_draw.fs) {
- ret = SVGA3D_SetShader(svga->swc,
- SVGA3D_SHADERTYPE_PS,
- id );
+ ret = svga_set_shader(svga, SVGA3D_SHADERTYPE_PS, variant);
if (ret != PIPE_OK)
- return ret;
+ goto done;
+
+ svga->rebind.flags.fs = FALSE;
svga->dirty |= SVGA_NEW_FS_VARIANT;
- svga->state.hw_draw.fs = variant;
+ svga->state.hw_draw.fs = variant;
}
- return PIPE_OK;
+done:
+ SVGA_STATS_TIME_POP(svga_sws(svga));
+ return ret;
}
struct svga_tracked_state svga_hw_fs =
{
"fragment shader (hwtnl)",
(SVGA_NEW_FS |
+ SVGA_NEW_GS |
+ SVGA_NEW_VS |
SVGA_NEW_TEXTURE_BINDING |
SVGA_NEW_NEED_SWTNL |
SVGA_NEW_RAST |
+ SVGA_NEW_STIPPLE |
+ SVGA_NEW_REDUCED_PRIMITIVE |
SVGA_NEW_SAMPLER |
SVGA_NEW_FRAME_BUFFER |
+ SVGA_NEW_DEPTH_STENCIL_ALPHA |
SVGA_NEW_BLEND),
emit_hw_fs
};
projects / mesa.git / blobdiff