si_init_external_atom(struct si_context *sctx, struct r600_atom *atom,
struct r600_atom **list_elem)
{
- atom->id = list_elem - sctx->atoms.array + 1;
+ atom->id = list_elem - sctx->atoms.array;
*list_elem = atom;
}
void (*emit_func)(struct si_context *ctx, struct r600_atom *state))
{
atom->emit = (void*)emit_func;
- atom->id = list_elem - sctx->atoms.array + 1; /* index+1 in the atom array */
+ atom->id = list_elem - sctx->atoms.array;
*list_elem = atom;
}
{
struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
struct si_state_blend *blend = sctx->queued.named.blend;
- uint32_t cb_target_mask, i;
+ /* CB_COLORn_INFO.FORMAT=INVALID should disable unbound colorbuffers,
+ * but you never know. */
+ uint32_t cb_target_mask = sctx->framebuffer.colorbuf_enabled_4bit;
+ unsigned i;
- /* CB_COLORn_INFO.FORMAT=INVALID disables empty colorbuffer slots. */
if (blend)
- cb_target_mask = blend->cb_target_mask;
- else
- cb_target_mask = 0xffffffff;
+ cb_target_mask &= blend->cb_target_mask;
/* Avoid a hang that happens when dual source blending is enabled
* but there is not enough color outputs. This is undefined behavior,
S_028760_ALPHA_COMB_FCN(V_028760_OPT_COMB_BLEND_DISABLED);
/* Only set dual source blending for MRT0 to avoid a hang. */
- if (i >= 1 && blend->dual_src_blend)
+ if (i >= 1 && blend->dual_src_blend) {
+ /* Vulkan does this for dual source blending. */
+ if (i == 1)
+ blend_cntl |= S_028780_ENABLE(1);
+
+ si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
continue;
+ }
/* Only addition and subtraction equations are supported with
* dual source blending.
(eqRGB == PIPE_BLEND_MIN || eqRGB == PIPE_BLEND_MAX ||
eqA == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MAX)) {
assert(!"Unsupported equation for dual source blending");
+ si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
continue;
}
- if (!state->rt[j].colormask)
- continue;
-
/* cb_render_state will disable unused ones */
blend->cb_target_mask |= (unsigned)state->rt[j].colormask << (4 * i);
- if (!state->rt[j].blend_enable) {
+ if (!state->rt[j].colormask || !state->rt[j].blend_enable) {
si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
continue;
}
}
if (sctx->b.family == CHIP_STONEY) {
+ /* Disable RB+ blend optimizations for dual source blending.
+ * Vulkan does this.
+ */
+ if (blend->dual_src_blend) {
+ for (int i = 0; i < 8; i++) {
+ sx_mrt_blend_opt[i] =
+ S_028760_COLOR_COMB_FCN(V_028760_OPT_COMB_NONE) |
+ S_028760_ALPHA_COMB_FCN(V_028760_OPT_COMB_NONE);
+ }
+ }
+
for (int i = 0; i < 8; i++)
si_pm4_set_reg(pm4, R_028760_SX_MRT0_BLEND_OPT + i * 4,
sx_mrt_blend_opt[i]);
unsigned ucp_mask = clipdist_mask ? 0 : rs->clip_plane_enable & SIX_BITS;
unsigned culldist_mask = info->culldist_writemask << info->num_written_clipdistance;
unsigned total_mask;
+ bool misc_vec_ena;
if (vs->key.opt.hw_vs.clip_disable) {
assert(!info->culldist_writemask);
}
total_mask = clipdist_mask | culldist_mask;
+ /* Clip distances on points have no effect, so need to be implemented
+ * as cull distances. This applies for the clipvertex case as well.
+ *
+ * Setting this for primitives other than points should have no adverse
+ * effects.
+ */
+ clipdist_mask &= rs->clip_plane_enable;
+ culldist_mask |= clipdist_mask;
+
+ misc_vec_ena = info->writes_psize || info->writes_edgeflag ||
+ info->writes_layer || info->writes_viewport_index;
+
radeon_set_context_reg(cs, R_02881C_PA_CL_VS_OUT_CNTL,
S_02881C_USE_VTX_POINT_SIZE(info->writes_psize) |
S_02881C_USE_VTX_EDGE_FLAG(info->writes_edgeflag) |
S_02881C_USE_VTX_VIEWPORT_INDX(info->writes_viewport_index) |
S_02881C_VS_OUT_CCDIST0_VEC_ENA((total_mask & 0x0F) != 0) |
S_02881C_VS_OUT_CCDIST1_VEC_ENA((total_mask & 0xF0) != 0) |
- S_02881C_VS_OUT_MISC_VEC_ENA(info->writes_psize ||
- info->writes_edgeflag ||
- info->writes_layer ||
- info->writes_viewport_index) |
- S_02881C_VS_OUT_MISC_SIDE_BUS_ENA(1) |
- (rs->clip_plane_enable &
- clipdist_mask) | (culldist_mask << 8));
+ S_02881C_VS_OUT_MISC_VEC_ENA(misc_vec_ena) |
+ S_02881C_VS_OUT_MISC_SIDE_BUS_ENA(misc_vec_ena) |
+ clipdist_mask | (culldist_mask << 8));
radeon_set_context_reg(cs, R_028810_PA_CL_CLIP_CNTL,
rs->pa_cl_clip_cntl |
ucp_mask |
{
struct si_state_rasterizer *rs = sctx->queued.named.rasterizer;
- if (!rs || !rs->uses_poly_offset || !sctx->framebuffer.state.zsbuf)
+ if (!rs || !rs->uses_poly_offset || !sctx->framebuffer.state.zsbuf) {
+ si_pm4_bind_state(sctx, poly_offset, NULL);
return;
+ }
/* Use the user format, not db_render_format, so that the polygon
* offset behaves as expected by applications.
si_update_poly_offset_state(sctx);
si_mark_atom_dirty(sctx, &sctx->clip_regs);
+ sctx->ia_multi_vgt_param_key.u.line_stipple_enabled =
+ rs->line_stipple_enable;
sctx->do_update_shaders = true;
}
case PIPE_FORMAT_Z16_UNORM:
return V_008F14_IMG_DATA_FORMAT_16;
case PIPE_FORMAT_X24S8_UINT:
+ case PIPE_FORMAT_S8X24_UINT:
+ /*
+ * Implemented as an 8_8_8_8 data format to fix texture
+ * gathers in stencil sampling. This affects at least
+ * GL45-CTS.texture_cube_map_array.sampling on VI.
+ */
+ return V_008F14_IMG_DATA_FORMAT_8_8_8_8;
case PIPE_FORMAT_Z24X8_UNORM:
case PIPE_FORMAT_Z24_UNORM_S8_UINT:
return V_008F14_IMG_DATA_FORMAT_8_24;
case PIPE_FORMAT_X8Z24_UNORM:
- case PIPE_FORMAT_S8X24_UINT:
case PIPE_FORMAT_S8_UINT_Z24_UNORM:
return V_008F14_IMG_DATA_FORMAT_24_8;
case PIPE_FORMAT_S8_UINT:
const struct util_format_description *desc,
int first_non_void)
{
- unsigned type;
int i;
if (desc->format == PIPE_FORMAT_R11G11B10_FLOAT)
return V_008F0C_BUF_DATA_FORMAT_10_11_11;
assert(first_non_void >= 0);
- type = desc->channel[first_non_void].type;
-
- if (type == UTIL_FORMAT_TYPE_FIXED)
- return V_008F0C_BUF_DATA_FORMAT_INVALID;
if (desc->nr_channels == 4 &&
desc->channel[0].size == 10 &&
}
break;
case 32:
- /* From the Southern Islands ISA documentation about MTBUF:
- * 'Memory reads of data in memory that is 32 or 64 bits do not
- * undergo any format conversion.'
- */
- if (type != UTIL_FORMAT_TYPE_FLOAT &&
- !desc->channel[first_non_void].pure_integer)
- return V_008F0C_BUF_DATA_FORMAT_INVALID;
-
switch (desc->nr_channels) {
case 1:
return V_008F0C_BUF_DATA_FORMAT_32;
switch (desc->channel[first_non_void].type) {
case UTIL_FORMAT_TYPE_SIGNED:
- if (desc->channel[first_non_void].normalized)
- return V_008F0C_BUF_NUM_FORMAT_SNORM;
- else if (desc->channel[first_non_void].pure_integer)
+ case UTIL_FORMAT_TYPE_FIXED:
+ if (desc->channel[first_non_void].size >= 32 ||
+ desc->channel[first_non_void].pure_integer)
return V_008F0C_BUF_NUM_FORMAT_SINT;
+ else if (desc->channel[first_non_void].normalized)
+ return V_008F0C_BUF_NUM_FORMAT_SNORM;
else
return V_008F0C_BUF_NUM_FORMAT_SSCALED;
break;
case UTIL_FORMAT_TYPE_UNSIGNED:
- if (desc->channel[first_non_void].normalized)
- return V_008F0C_BUF_NUM_FORMAT_UNORM;
- else if (desc->channel[first_non_void].pure_integer)
+ if (desc->channel[first_non_void].size >= 32 ||
+ desc->channel[first_non_void].pure_integer)
return V_008F0C_BUF_NUM_FORMAT_UINT;
+ else if (desc->channel[first_non_void].normalized)
+ return V_008F0C_BUF_NUM_FORMAT_UNORM;
else
return V_008F0C_BUF_NUM_FORMAT_USCALED;
break;
si_dec_framebuffer_counters(&sctx->framebuffer.state);
util_copy_framebuffer_state(&sctx->framebuffer.state, state);
+ sctx->framebuffer.colorbuf_enabled_4bit = 0;
sctx->framebuffer.spi_shader_col_format = 0;
sctx->framebuffer.spi_shader_col_format_alpha = 0;
sctx->framebuffer.spi_shader_col_format_blend = 0;
si_initialize_color_surface(sctx, surf);
}
+ sctx->framebuffer.colorbuf_enabled_4bit |= 0xf << (i * 4);
sctx->framebuffer.spi_shader_col_format |=
surf->spi_shader_col_format << (i * 4);
sctx->framebuffer.spi_shader_col_format_alpha |=
sctx->need_check_render_feedback = true;
sctx->do_update_shaders = true;
+ sctx->framebuffer.do_update_surf_dirtiness = true;
}
static void si_emit_framebuffer_state(struct si_context *sctx, struct r600_atom *atom)
if (desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) {
const unsigned char swizzle_xxxx[4] = {0, 0, 0, 0};
const unsigned char swizzle_yyyy[4] = {1, 1, 1, 1};
+ const unsigned char swizzle_wwww[4] = {3, 3, 3, 3};
switch (pipe_format) {
case PIPE_FORMAT_S8_UINT_Z24_UNORM:
- case PIPE_FORMAT_X24S8_UINT:
case PIPE_FORMAT_X32_S8X24_UINT:
case PIPE_FORMAT_X8Z24_UNORM:
util_format_compose_swizzles(swizzle_yyyy, state_swizzle, swizzle);
break;
+ case PIPE_FORMAT_X24S8_UINT:
+ /*
+ * X24S8 is implemented as an 8_8_8_8 data format, to
+ * fix texture gathers. This affects at least
+ * GL45-CTS.texture_cube_map_array.sampling on VI.
+ */
+ util_format_compose_swizzles(swizzle_wwww, state_swizzle, swizzle);
+ break;
default:
util_format_compose_swizzles(swizzle_xxxx, state_swizzle, swizzle);
}
}
}
+#ifdef DEBUG
+ rstate->magic = SI_SAMPLER_STATE_MAGIC;
+#endif
rstate->val[0] = (S_008F30_CLAMP_X(si_tex_wrap(state->wrap_s)) |
S_008F30_CLAMP_Y(si_tex_wrap(state->wrap_t)) |
S_008F30_CLAMP_Z(si_tex_wrap(state->wrap_r)) |
static void si_delete_sampler_state(struct pipe_context *ctx, void *state)
{
+#ifdef DEBUG
+ struct si_sampler_state *s = state;
+
+ assert(s->magic == SI_SAMPLER_STATE_MAGIC);
+ s->magic = 0;
+#endif
free(state);
}
const struct pipe_vertex_element *elements)
{
struct si_vertex_element *v = CALLOC_STRUCT(si_vertex_element);
+ bool used[SI_NUM_VERTEX_BUFFERS] = {};
int i;
assert(count <= SI_MAX_ATTRIBS);
v->count = count;
for (i = 0; i < count; ++i) {
const struct util_format_description *desc;
+ const struct util_format_channel_description *channel;
unsigned data_format, num_format;
int first_non_void;
+ unsigned vbo_index = elements[i].vertex_buffer_index;
+
+ if (vbo_index >= SI_NUM_VERTEX_BUFFERS) {
+ FREE(v);
+ return NULL;
+ }
+
+ if (!used[vbo_index]) {
+ v->first_vb_use_mask |= 1 << i;
+ used[vbo_index] = true;
+ }
desc = util_format_description(elements[i].src_format);
first_non_void = util_format_get_first_non_void_channel(elements[i].src_format);
data_format = si_translate_buffer_dataformat(ctx->screen, desc, first_non_void);
num_format = si_translate_buffer_numformat(ctx->screen, desc, first_non_void);
+ channel = first_non_void >= 0 ? &desc->channel[first_non_void] : NULL;
v->rsrc_word3[i] = S_008F0C_DST_SEL_X(si_map_swizzle(desc->swizzle[0])) |
S_008F0C_DST_SEL_Y(si_map_swizzle(desc->swizzle[1])) |
*/
if (data_format == V_008F0C_BUF_DATA_FORMAT_2_10_10_10) {
if (num_format == V_008F0C_BUF_NUM_FORMAT_SNORM) {
- v->fix_fetch |= SI_FIX_FETCH_A2_SNORM << (2 * i);
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_A2_SNORM << (4 * i);
} else if (num_format == V_008F0C_BUF_NUM_FORMAT_SSCALED) {
- v->fix_fetch |= SI_FIX_FETCH_A2_SSCALED << (2 * i);
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_A2_SSCALED << (4 * i);
} else if (num_format == V_008F0C_BUF_NUM_FORMAT_SINT) {
/* This isn't actually used in OpenGL. */
- v->fix_fetch |= SI_FIX_FETCH_A2_SINT << (2 * i);
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_A2_SINT << (4 * i);
+ }
+ } else if (channel && channel->type == UTIL_FORMAT_TYPE_FIXED) {
+ if (desc->swizzle[3] == PIPE_SWIZZLE_1)
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_RGBX_32_FIXED << (4 * i);
+ else
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_RGBA_32_FIXED << (4 * i);
+ } else if (channel && channel->size == 32 && !channel->pure_integer) {
+ if (channel->type == UTIL_FORMAT_TYPE_SIGNED) {
+ if (channel->normalized) {
+ if (desc->swizzle[3] == PIPE_SWIZZLE_1)
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_RGBX_32_SNORM << (4 * i);
+ else
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_RGBA_32_SNORM << (4 * i);
+ } else {
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_RGBA_32_SSCALED << (4 * i);
+ }
+ } else if (channel->type == UTIL_FORMAT_TYPE_UNSIGNED) {
+ if (channel->normalized) {
+ if (desc->swizzle[3] == PIPE_SWIZZLE_1)
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_RGBX_32_UNORM << (4 * i);
+ else
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_RGBA_32_UNORM << (4 * i);
+ } else {
+ v->fix_fetch |= (uint64_t)SI_FIX_FETCH_RGBA_32_USCALED << (4 * i);
+ }
}
}
pipe_resource_reference(&cb.buffer, NULL);
}
-static void si_texture_barrier(struct pipe_context *ctx)
+static void si_texture_barrier(struct pipe_context *ctx, unsigned flags)
{
struct si_context *sctx = (struct si_context *)ctx;
sctx->b.flags |= SI_CONTEXT_INV_VMEM_L1 |
SI_CONTEXT_INV_GLOBAL_L2 |
- SI_CONTEXT_FLUSH_AND_INV_CB |
- SI_CONTEXT_CS_PARTIAL_FLUSH;
+ SI_CONTEXT_FLUSH_AND_INV_CB;
+ sctx->framebuffer.do_update_surf_dirtiness = true;
}
/* This only ensures coherency for shader image/buffer stores. */
raster_config_1 = 0x0000002a;
break;
case CHIP_POLARIS11:
+ case CHIP_POLARIS12:
raster_config = 0x16000012;
raster_config_1 = 0x00000000;
break;