radeonsi: initialize SX_PS_DOWNCONVERT to 0 on Stoney
[mesa.git] / src / gallium / drivers / radeonsi / si_state.c
index 1e2f32a87991a852a3bd3ffae6627560d228f5eb..6d97049c0f3a1e91bbca60946b4bdf216c05a7a2 100644 (file)
@@ -50,7 +50,6 @@ void si_init_atom(struct si_context *sctx, struct r600_atom *atom,
                  void (*emit_func)(struct si_context *ctx, struct r600_atom *state))
 {
        atom->emit = (void*)emit_func;
-       atom->dirty = false;
        atom->id = list_elem - sctx->atoms.array + 1; /* index+1 in the atom array */
        *list_elem = atom;
 }
@@ -266,8 +265,9 @@ static void si_emit_cb_target_mask(struct si_context *sctx, struct r600_atom *at
         *
         * Reproducible with Unigine Heaven 4.0 and drirc missing.
         */
-       if (blend->dual_src_blend &&
-           (sctx->ps_shader->ps_colors_written & 0x3) != 0x3)
+       if (blend && blend->dual_src_blend &&
+           sctx->ps_shader.cso &&
+           (sctx->ps_shader.cso->ps_colors_written & 0x3) != 0x3)
                mask = 0;
 
        radeon_set_context_reg(cs, R_028238_CB_TARGET_MASK, mask);
@@ -637,7 +637,7 @@ static void si_update_poly_offset_state(struct si_context *sctx)
 {
        struct si_state_rasterizer *rs = sctx->queued.named.rasterizer;
 
-       if (!rs || !sctx->framebuffer.state.zsbuf)
+       if (!rs || !rs->uses_poly_offset || !sctx->framebuffer.state.zsbuf)
                return;
 
        switch (sctx->framebuffer.state.zsbuf->texture->format) {
@@ -687,14 +687,18 @@ static void *si_create_rs_state(struct pipe_context *ctx,
 
        rs->two_side = state->light_twoside;
        rs->multisample_enable = state->multisample;
+       rs->force_persample_interp = state->force_persample_interp;
        rs->clip_plane_enable = state->clip_plane_enable;
        rs->line_stipple_enable = state->line_stipple_enable;
        rs->poly_stipple_enable = state->poly_stipple_enable;
        rs->line_smooth = state->line_smooth;
        rs->poly_smooth = state->poly_smooth;
-
+       rs->uses_poly_offset = state->offset_point || state->offset_line ||
+                              state->offset_tri;
+       rs->clamp_fragment_color = state->clamp_fragment_color;
        rs->flatshade = state->flatshade;
        rs->sprite_coord_enable = state->sprite_coord_enable;
+       rs->rasterizer_discard = state->rasterizer_discard;
        rs->pa_sc_line_stipple = state->line_stipple_enable ?
                                S_028A0C_LINE_PATTERN(state->line_stipple_pattern) |
                                S_028A0C_REPEAT_COUNT(state->line_stipple_factor) : 0;
@@ -758,6 +762,8 @@ static void *si_create_rs_state(struct pipe_context *ctx,
                                   state->fill_back != PIPE_POLYGON_MODE_FILL) |
                S_028814_POLYMODE_FRONT_PTYPE(si_translate_fill(state->fill_front)) |
                S_028814_POLYMODE_BACK_PTYPE(si_translate_fill(state->fill_back)));
+       si_pm4_set_reg(pm4, R_00B130_SPI_SHADER_USER_DATA_VS_0 +
+                      SI_SGPR_VS_STATE_BITS * 4, state->clamp_vertex_color);
 
        /* Precalculate polygon offset states for 16-bit, 24-bit, and 32-bit zbuffers. */
        for (i = 0; i < 3; i++) {
@@ -997,10 +1003,10 @@ static void si_emit_db_render_state(struct si_context *sctx, struct r600_atom *s
                            S_028000_STENCIL_COPY(sctx->dbcb_stencil_copy_enabled) |
                            S_028000_COPY_CENTROID(1) |
                            S_028000_COPY_SAMPLE(sctx->dbcb_copy_sample));
-       } else if (sctx->db_inplace_flush_enabled) {
+       } else if (sctx->db_flush_depth_inplace || sctx->db_flush_stencil_inplace) {
                radeon_emit(cs,
-                           S_028000_DEPTH_COMPRESS_DISABLE(1) |
-                           S_028000_STENCIL_COMPRESS_DISABLE(1));
+                           S_028000_DEPTH_COMPRESS_DISABLE(sctx->db_flush_depth_inplace) |
+                           S_028000_STENCIL_COMPRESS_DISABLE(sctx->db_flush_stencil_inplace));
        } else if (sctx->db_depth_clear) {
                radeon_emit(cs, S_028000_DEPTH_CLEAR_ENABLE(1));
        } else {
@@ -1531,9 +1537,14 @@ static unsigned si_tex_compare(unsigned compare)
        }
 }
 
-static unsigned si_tex_dim(unsigned dim, unsigned nr_samples)
+static unsigned si_tex_dim(unsigned res_target, unsigned view_target,
+                          unsigned nr_samples)
 {
-       switch (dim) {
+       if (view_target == PIPE_TEXTURE_CUBE ||
+           view_target == PIPE_TEXTURE_CUBE_ARRAY)
+               res_target = view_target;
+
+       switch (res_target) {
        default:
        case PIPE_TEXTURE_1D:
                return V_008F1C_SQ_RSRC_IMG_1D;
@@ -1915,8 +1926,21 @@ static void si_initialize_color_surface(struct si_context *sctx,
        surf->cb_color_info = color_info;
        surf->cb_color_attrib = color_attrib;
 
-       if (sctx->b.chip_class >= VI)
-               surf->cb_dcc_control = S_028C78_OVERWRITE_COMBINER_DISABLE(1);
+       if (sctx->b.chip_class >= VI && rtex->dcc_buffer) {
+               unsigned max_uncompressed_block_size = 2;
+               uint64_t dcc_offset = rtex->surface.level[level].dcc_offset;
+
+               if (rtex->surface.nsamples > 1) {
+                       if (rtex->surface.bpe == 1)
+                               max_uncompressed_block_size = 0;
+                       else if (rtex->surface.bpe == 2)
+                               max_uncompressed_block_size = 1;
+               }
+
+               surf->cb_dcc_control = S_028C78_MAX_UNCOMPRESSED_BLOCK_SIZE(max_uncompressed_block_size) |
+                                      S_028C78_INDEPENDENT_64B_BLOCKS(1);
+               surf->cb_dcc_base = (rtex->dcc_buffer->gpu_address + dcc_offset) >> 8;
+       }
 
        if (rtex->fmask.size) {
                surf->cb_color_fmask = (offset + rtex->fmask.offset) >> 8;
@@ -2237,7 +2261,13 @@ static void si_emit_framebuffer_state(struct si_context *sctx, struct r600_atom
                if (tex->cmask_buffer && tex->cmask_buffer != &tex->resource) {
                        radeon_add_to_buffer_list(&sctx->b, &sctx->b.rings.gfx,
                                tex->cmask_buffer, RADEON_USAGE_READWRITE,
-                               RADEON_PRIO_COLOR_META);
+                               RADEON_PRIO_CMASK);
+               }
+
+               if (tex->dcc_buffer && tex->dcc_buffer != &tex->resource) {
+                       radeon_add_to_buffer_list(&sctx->b, &sctx->b.rings.gfx,
+                               tex->dcc_buffer, RADEON_USAGE_READWRITE,
+                               RADEON_PRIO_DCC);
                }
 
                radeon_set_context_reg_seq(cs, R_028C60_CB_COLOR0_BASE + i * 0x3C,
@@ -2257,7 +2287,7 @@ static void si_emit_framebuffer_state(struct si_context *sctx, struct r600_atom
                radeon_emit(cs, tex->color_clear_value[1]);     /* R_028C90_CB_COLOR0_CLEAR_WORD1 */
 
                if (sctx->b.chip_class >= VI)
-                       radeon_emit(cs, 0);     /* R_028C94_CB_COLOR0_DCC_BASE */
+                       radeon_emit(cs, cb->cb_dcc_base);       /* R_028C94_CB_COLOR0_DCC_BASE */
        }
        /* set CB_COLOR1_INFO for possible dual-src blending */
        if (i == 1 && state->cbufs[0] &&
@@ -2284,7 +2314,7 @@ static void si_emit_framebuffer_state(struct si_context *sctx, struct r600_atom
                if (zb->db_htile_data_base) {
                        radeon_add_to_buffer_list(&sctx->b, &sctx->b.rings.gfx,
                                              rtex->htile_buffer, RADEON_USAGE_READWRITE,
-                                             RADEON_PRIO_DEPTH_META);
+                                             RADEON_PRIO_HTILE);
                }
 
                radeon_set_context_reg(cs, R_028008_DB_DEPTH_VIEW, zb->db_depth_view);
@@ -2387,6 +2417,7 @@ si_create_sampler_view_custom(struct pipe_context *ctx,
        struct radeon_surf_level *surflevel;
        int first_non_void;
        uint64_t va;
+       unsigned last_layer = state->u.tex.last_layer;
 
        if (view == NULL)
                return NULL;
@@ -2410,6 +2441,12 @@ si_create_sampler_view_custom(struct pipe_context *ctx,
        pipe_resource_reference(&view->base.texture, texture);
        view->resource = &tmp->resource;
 
+       if (state->format == PIPE_FORMAT_X24S8_UINT ||
+           state->format == PIPE_FORMAT_S8X24_UINT ||
+           state->format == PIPE_FORMAT_X32_S8X24_UINT ||
+           state->format == PIPE_FORMAT_S8_UINT)
+               view->is_stencil_sampler = true;
+
        /* Buffer resource. */
        if (texture->target == PIPE_BUFFER) {
                unsigned stride, num_records;
@@ -2586,6 +2623,13 @@ si_create_sampler_view_custom(struct pipe_context *ctx,
        } else if (texture->target == PIPE_TEXTURE_CUBE_ARRAY)
                depth = texture->array_size / 6;
 
+       /* This is not needed if state trackers set last_layer correctly. */
+       if (state->target == PIPE_TEXTURE_1D ||
+           state->target == PIPE_TEXTURE_2D ||
+           state->target == PIPE_TEXTURE_RECT ||
+           state->target == PIPE_TEXTURE_CUBE)
+               last_layer = state->u.tex.first_layer;
+
        va = tmp->resource.gpu_address + surflevel[base_level].offset;
 
        view->state[0] = va >> 8;
@@ -2605,12 +2649,23 @@ si_create_sampler_view_custom(struct pipe_context *ctx,
                                                      last_level) |
                          S_008F1C_TILING_INDEX(si_tile_mode_index(tmp, base_level, false)) |
                          S_008F1C_POW2_PAD(texture->last_level > 0) |
-                         S_008F1C_TYPE(si_tex_dim(texture->target, texture->nr_samples)));
+                         S_008F1C_TYPE(si_tex_dim(texture->target, state->target,
+                                                  texture->nr_samples)));
        view->state[4] = (S_008F20_DEPTH(depth - 1) | S_008F20_PITCH(pitch - 1));
        view->state[5] = (S_008F24_BASE_ARRAY(state->u.tex.first_layer) |
-                         S_008F24_LAST_ARRAY(state->u.tex.last_layer));
-       view->state[6] = 0;
-       view->state[7] = 0;
+                         S_008F24_LAST_ARRAY(last_layer));
+
+       if (tmp->dcc_buffer) {
+               uint64_t dcc_offset = surflevel[base_level].dcc_offset;
+               unsigned swap = r600_translate_colorswap(pipe_format);
+
+               view->state[6] = S_008F28_COMPRESSION_EN(1) | S_008F28_ALPHA_IS_ON_MSB(swap <= 1);
+               view->state[7] = (tmp->dcc_buffer->gpu_address + dcc_offset) >> 8;
+               view->dcc_buffer = tmp->dcc_buffer;
+       } else {
+               view->state[6] = 0;
+               view->state[7] = 0;
+       }
 
        /* Initialize the sampler view for FMASK. */
        if (tmp->fmask.size) {
@@ -2643,11 +2698,12 @@ si_create_sampler_view_custom(struct pipe_context *ctx,
                                       S_008F1C_DST_SEL_Z(V_008F1C_SQ_SEL_X) |
                                       S_008F1C_DST_SEL_W(V_008F1C_SQ_SEL_X) |
                                       S_008F1C_TILING_INDEX(tmp->fmask.tile_mode_index) |
-                                      S_008F1C_TYPE(si_tex_dim(texture->target, 0));
+                                      S_008F1C_TYPE(si_tex_dim(texture->target,
+                                                               state->target, 0));
                view->fmask_state[4] = S_008F20_DEPTH(depth - 1) |
                                       S_008F20_PITCH(tmp->fmask.pitch - 1);
                view->fmask_state[5] = S_008F24_BASE_ARRAY(state->u.tex.first_layer) |
-                                      S_008F24_LAST_ARRAY(state->u.tex.last_layer);
+                                      S_008F24_LAST_ARRAY(last_layer);
                view->fmask_state[6] = 0;
                view->fmask_state[7] = 0;
        }
@@ -3237,7 +3293,7 @@ static void si_init_config(struct si_context *sctx)
        si_pm4_set_reg(pm4, R_028A1C_VGT_HOS_MIN_TESS_LEVEL, fui(0));
 
        /* FIXME calculate these values somehow ??? */
-       si_pm4_set_reg(pm4, R_028A54_VGT_GS_PER_ES, 0x80);
+       si_pm4_set_reg(pm4, R_028A54_VGT_GS_PER_ES, SI_GS_PER_ES);
        si_pm4_set_reg(pm4, R_028A58_VGT_ES_PER_GS, 0x40);
        si_pm4_set_reg(pm4, R_028A5C_VGT_GS_PER_VS, 0x2);
 
@@ -3311,6 +3367,7 @@ static void si_init_config(struct si_context *sctx)
                break;
        case CHIP_KABINI:
        case CHIP_MULLINS:
+       case CHIP_STONEY:
                raster_config = 0x00000000;
                raster_config_1 = 0x00000000;
                break;
@@ -3381,16 +3438,20 @@ static void si_init_config(struct si_context *sctx)
 
        if (sctx->b.chip_class >= VI) {
                si_pm4_set_reg(pm4, R_028424_CB_DCC_CONTROL,
-                              S_028424_OVERWRITE_COMBINER_MRT_SHARING_DISABLE(1));
+                              S_028424_OVERWRITE_COMBINER_MRT_SHARING_DISABLE(1) |
+                              S_028424_OVERWRITE_COMBINER_WATERMARK(4));
                si_pm4_set_reg(pm4, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL, 30);
                si_pm4_set_reg(pm4, R_028C5C_VGT_OUT_DEALLOC_CNTL, 32);
        }
 
+       if (sctx->b.family == CHIP_STONEY)
+               si_pm4_set_reg(pm4, R_028754_SX_PS_DOWNCONVERT, 0);
+
        si_pm4_set_reg(pm4, R_028080_TA_BC_BASE_ADDR, border_color_va >> 8);
        if (sctx->b.chip_class >= CIK)
                si_pm4_set_reg(pm4, R_028084_TA_BC_BASE_ADDR_HI, border_color_va >> 40);
        si_pm4_add_bo(pm4, sctx->border_color_buffer, RADEON_USAGE_READ,
-                     RADEON_PRIO_SHADER_DATA);
+                     RADEON_PRIO_BORDER_COLORS);
 
        si_pm4_upload_indirect_buffer(sctx, pm4);
        sctx->init_config = pm4;