[Mesa-dev] [PATCH 1/3] r600g: implement MSAA for Cayman
Ian Romanick
idr at freedesktop.org
Thu Aug 30 09:56:40 PDT 2012
On 08/30/2012 08:35 AM, Marek Olšák wrote:
> Everything works except for blitting MSAA colorbuffers, which isn't
Does this mean "from a MSAA color buffer to a MSAA color buffer" or does
this mean "from a MSAA color buffer to anything"?
> so trivial on Cayman. It's a rarely-used feature anyway.
> ---
> src/gallium/auxiliary/util/u_blitter.c | 3 +-
> src/gallium/auxiliary/util/u_blitter.h | 1 +
> src/gallium/drivers/r600/evergreen_hw_context.c | 1 +
> src/gallium/drivers/r600/evergreen_state.c | 196 +++++++++++++++--------
> src/gallium/drivers/r600/evergreend.h | 21 +++
> src/gallium/drivers/r600/r600_blit.c | 34 +++-
> src/gallium/drivers/r600/r600_pipe.h | 1 +
> src/gallium/drivers/r600/r600_state_common.c | 3 +-
> 8 files changed, 187 insertions(+), 73 deletions(-)
>
> diff --git a/src/gallium/auxiliary/util/u_blitter.c b/src/gallium/auxiliary/util/u_blitter.c
> index 03ed91f..44295c1 100644
> --- a/src/gallium/auxiliary/util/u_blitter.c
> +++ b/src/gallium/auxiliary/util/u_blitter.c
> @@ -1492,6 +1492,7 @@ void util_blitter_custom_resolve_color(struct blitter_context *blitter,
> unsigned dst_layer,
> struct pipe_resource *src,
> unsigned src_layer,
> + unsigned sample_mask,
> void *custom_blend)
> {
> struct blitter_context_priv *ctx = (struct blitter_context_priv*)blitter;
> @@ -1508,7 +1509,7 @@ void util_blitter_custom_resolve_color(struct blitter_context *blitter,
> pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_keep_depth_stencil);
> pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
> pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, 1, FALSE));
> - pipe->set_sample_mask(pipe, (1ull << MAX2(1, src->nr_samples)) - 1);
> + pipe->set_sample_mask(pipe, sample_mask);
>
> memset(&surf_tmpl, 0, sizeof(surf_tmpl));
> surf_tmpl.format = dst->format;
> diff --git a/src/gallium/auxiliary/util/u_blitter.h b/src/gallium/auxiliary/util/u_blitter.h
> index 2055fd6..6804073 100644
> --- a/src/gallium/auxiliary/util/u_blitter.h
> +++ b/src/gallium/auxiliary/util/u_blitter.h
> @@ -329,6 +329,7 @@ void util_blitter_custom_resolve_color(struct blitter_context *blitter,
> unsigned dst_layer,
> struct pipe_resource *src,
> unsigned src_layer,
> + unsigned sampled_mask,
> void *custom_blend);
>
> /* The functions below should be used to save currently bound constant state
> diff --git a/src/gallium/drivers/r600/evergreen_hw_context.c b/src/gallium/drivers/r600/evergreen_hw_context.c
> index d2f0949..b2ea7e2 100644
> --- a/src/gallium/drivers/r600/evergreen_hw_context.c
> +++ b/src/gallium/drivers/r600/evergreen_hw_context.c
> @@ -500,6 +500,7 @@ static const struct r600_reg cayman_context_reg_list[] = {
> {R_028798_CB_BLEND6_CONTROL, 0, 0},
> {R_02879C_CB_BLEND7_CONTROL, 0, 0},
> {R_028800_DB_DEPTH_CONTROL, 0, 0},
> + {CM_R_028804_DB_EQAA},
> {R_028808_CB_COLOR_CONTROL, 0, 0},
> {R_02880C_DB_SHADER_CONTROL, 0, 0},
> {R_028810_PA_CL_CLIP_CNTL, 0, 0},
> diff --git a/src/gallium/drivers/r600/evergreen_state.c b/src/gallium/drivers/r600/evergreen_state.c
> index 35bc391..28a83f2 100644
> --- a/src/gallium/drivers/r600/evergreen_state.c
> +++ b/src/gallium/drivers/r600/evergreen_state.c
> @@ -635,9 +635,6 @@ boolean evergreen_is_format_supported(struct pipe_screen *screen,
> if (rscreen->info.drm_minor < 19)
> return FALSE;
>
> - if (rscreen->chip_class != EVERGREEN)
> - return FALSE;
> -
> switch (sample_count) {
> case 2:
> case 4:
> @@ -1102,8 +1099,12 @@ static struct pipe_sampler_view *evergreen_create_sampler_view(struct pipe_conte
> view->tex_resource_words[5] = S_030014_BASE_ARRAY(state->u.tex.first_layer) |
> S_030014_LAST_ARRAY(state->u.tex.last_layer);
> if (texture->nr_samples > 1) {
> + unsigned log_samples = util_logbase2(texture->nr_samples);
> + if (rscreen->chip_class == CAYMAN) {
> + view->tex_resource_words[4] |= S_030010_LOG2_NUM_FRAGMENTS(log_samples);
> + }
> /* LAST_LEVEL holds log2(nr_samples) for multisample textures */
> - view->tex_resource_words[5] |= S_030014_LAST_LEVEL(util_logbase2(texture->nr_samples));
> + view->tex_resource_words[5] |= S_030014_LAST_LEVEL(log_samples);
> } else {
> view->tex_resource_words[4] |= S_030010_BASE_LEVEL(state->u.tex.first_level);
> view->tex_resource_words[5] |= S_030014_LAST_LEVEL(state->u.tex.last_level);
> @@ -1324,6 +1325,12 @@ void evergreen_init_color_surface(struct r600_context *rctx,
> S_028C74_NON_DISP_TILING_ORDER(tile_type) |
> S_028C74_FMASK_BANK_HEIGHT(fmask_bankh);
>
> + if (rctx->chip_class == CAYMAN && rtex->resource.b.b.nr_samples > 1) {
> + unsigned log_samples = util_logbase2(rtex->resource.b.b.nr_samples);
> + color_attrib |= S_028C74_NUM_SAMPLES(log_samples) |
> + S_028C74_NUM_FRAGMENTS(log_samples);
> + }
> +
> ntype = V_028C70_NUMBER_UNORM;
> if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
> ntype = V_028C70_NUMBER_SRGB;
> @@ -1478,6 +1485,9 @@ static void evergreen_init_depth_surface(struct r600_context *rctx,
> S_028040_BANK_WIDTH(bankw) |
> S_028040_BANK_HEIGHT(bankh) |
> S_028040_MACRO_TILE_ASPECT(macro_aspect);
> + if (rscreen->chip_class == CAYMAN && rtex->resource.b.b.nr_samples > 1) {
> + surf->db_depth_info |= S_028040_NUM_SAMPLES(util_logbase2(rtex->resource.b.b.nr_samples));
> + }
> surf->db_depth_base = offset;
> surf->db_depth_view = S_028008_SLICE_START(surf->base.u.tex.first_layer) |
> S_028008_SLICE_MAX(surf->base.u.tex.last_layer);
> @@ -1530,7 +1540,7 @@ static uint32_t evergreen_set_ms_pos(struct pipe_context *ctx, struct r600_pipe_
> };
> static unsigned max_dist_4x = 6;
> /* 8xMSAA */
> - static uint32_t eg_sample_locs_8x[] = {
> + static uint32_t sample_locs_8x[] = {
> FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
> FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
> FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
> @@ -1540,7 +1550,57 @@ static uint32_t evergreen_set_ms_pos(struct pipe_context *ctx, struct r600_pipe_
> FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
> FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
> };
> - static uint32_t cm_sample_locs_8x[] = {
> + static unsigned max_dist_8x = 8;
> + struct r600_context *rctx = (struct r600_context *)ctx;
> + unsigned i;
> +
> + switch (nsample) {
> + case 2:
> + for (i = 0; i < Elements(sample_locs_2x); i++) {
> + r600_pipe_state_add_reg(rstate, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0 + i*4,
> + sample_locs_2x[i]);
> + }
> + return max_dist_2x;
> + case 4:
> + for (i = 0; i < Elements(sample_locs_4x); i++) {
> + r600_pipe_state_add_reg(rstate, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0 + i*4,
> + sample_locs_4x[i]);
> + }
> + return max_dist_4x;
> + case 8:
> + for (i = 0; i < Elements(sample_locs_8x); i++) {
> + r600_pipe_state_add_reg(rstate, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0 + i*4,
> + sample_locs_8x[i]);
> + }
> + return max_dist_8x;
> + default:
> + R600_ERR("Invalid nr_samples %i\n", nsample);
> + return 0;
> + }
> +}
> +
> +static uint32_t cayman_set_ms_pos(struct pipe_context *ctx, struct r600_pipe_state *rstate, int nsample)
> +{
> + /* 2xMSAA
> + * There are two locations (-4, 4), (4, -4). */
> + static uint32_t sample_locs_2x[] = {
> + FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
> + FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
> + FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
> + FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
> + };
> + static unsigned max_dist_2x = 4;
> + /* 4xMSAA
> + * There are 4 locations: (-2, -2), (2, 2), (-6, 6), (6, -6). */
> + static uint32_t sample_locs_4x[] = {
> + FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
> + FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
> + FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
> + FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
> + };
> + static unsigned max_dist_4x = 6;
> + /* 8xMSAA */
> + static uint32_t sample_locs_8x[] = {
> FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
> FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
> FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
> @@ -1552,7 +1612,7 @@ static uint32_t evergreen_set_ms_pos(struct pipe_context *ctx, struct r600_pipe_
> };
> static unsigned max_dist_8x = 8;
> /* 16xMSAA */
> - static uint32_t cm_sample_locs_16x[] = {
> + static uint32_t sample_locs_16x[] = {
> FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
> FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
> FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
> @@ -1572,7 +1632,7 @@ static uint32_t evergreen_set_ms_pos(struct pipe_context *ctx, struct r600_pipe_
> };
> static unsigned max_dist_16x = 8;
> struct r600_context *rctx = (struct r600_context *)ctx;
> - uint32_t max_dist, num_regs, *sample_locs, i;
> + uint32_t max_dist, num_regs, *sample_locs;
>
> switch (nsample) {
> case 2:
> @@ -1586,55 +1646,39 @@ static uint32_t evergreen_set_ms_pos(struct pipe_context *ctx, struct r600_pipe_
> max_dist = max_dist_4x;
> break;
> case 8:
> - if (rctx->chip_class == CAYMAN) {
> - sample_locs = cm_sample_locs_8x;
> - num_regs = Elements(cm_sample_locs_8x);
> - } else {
> - sample_locs = eg_sample_locs_8x;
> - num_regs = Elements(eg_sample_locs_8x);
> - }
> + sample_locs = sample_locs_8x;
> + num_regs = Elements(sample_locs_8x);
> max_dist = max_dist_8x;
> break;
> case 16:
> - if (rctx->chip_class == CAYMAN) {
> - sample_locs = cm_sample_locs_16x;
> - num_regs = Elements(cm_sample_locs_16x);
> - max_dist = max_dist_16x;
> - break;
> - }
> - /* fall through */
> + sample_locs = sample_locs_16x;
> + num_regs = Elements(sample_locs_16x);
> + max_dist = max_dist_16x;
> + break;
> default:
> R600_ERR("Invalid nr_samples %i\n", nsample);
> return 0;
> }
>
> - /* All the regs must be initialized. Otherwise weird rendering may occur. */
> - if (rctx->chip_class == CAYMAN) {
> - r600_pipe_state_add_reg(rstate, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs[0]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs[1]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs[2]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs[3]);
> - if (num_regs <= 8) {
> - r600_pipe_state_add_reg(rstate, CM_R_028BFC_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_1, sample_locs[4]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C0C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_1, sample_locs[5]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C1C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_1, sample_locs[6]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C2C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_1, sample_locs[7]);
> - }
> - if (num_regs <= 16) {
> - r600_pipe_state_add_reg(rstate, CM_R_028C00_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_2, sample_locs[8]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C10_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_2, sample_locs[9]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C20_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_2, sample_locs[10]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C30_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_2, sample_locs[11]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C04_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_3, sample_locs[12]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C14_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_3, sample_locs[13]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C24_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_3, sample_locs[14]);
> - r600_pipe_state_add_reg(rstate, CM_R_028C34_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_3, sample_locs[15]);
> - }
> - } else {
> - for (i = 0; i < num_regs; i++) {
> - r600_pipe_state_add_reg(rstate, R_028C1C_PA_SC_AA_SAMPLE_LOCS_0 + i*4,
> - sample_locs[i]);
> - }
> + r600_pipe_state_add_reg(rstate, CM_R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs[0]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs[1]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs[2]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs[3]);
> + if (num_regs <= 8) {
> + r600_pipe_state_add_reg(rstate, CM_R_028BFC_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_1, sample_locs[4]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C0C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_1, sample_locs[5]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C1C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_1, sample_locs[6]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C2C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_1, sample_locs[7]);
> + }
> + if (num_regs <= 16) {
> + r600_pipe_state_add_reg(rstate, CM_R_028C00_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_2, sample_locs[8]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C10_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_2, sample_locs[9]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C20_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_2, sample_locs[10]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C30_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_2, sample_locs[11]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C04_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_3, sample_locs[12]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C14_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_3, sample_locs[13]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C24_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_3, sample_locs[14]);
> + r600_pipe_state_add_reg(rstate, CM_R_028C34_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_3, sample_locs[15]);
> }
> return max_dist;
> }
> @@ -1647,7 +1691,7 @@ static void evergreen_set_framebuffer_state(struct pipe_context *ctx,
> struct r600_surface *surf;
> struct r600_resource *res;
> struct r600_texture *rtex;
> - uint32_t tl, br, i, nr_samples;
> + uint32_t tl, br, i, nr_samples, log_samples;
>
> if (rstate == NULL)
> return;
> @@ -1702,7 +1746,9 @@ static void evergreen_set_framebuffer_state(struct pipe_context *ctx,
> r600_pipe_state_add_reg(rstate, R_028C88_CB_COLOR0_FMASK_SLICE + i * 0x3c,
> surf->cb_color_fmask_slice);
>
> - if (rtex->fmask_size && rtex->cmask_size) {
> + /* Cayman can fetch from a compressed MSAA colorbuffer,
> + * so it's pointless to track them. */
> + if (rctx->chip_class != CAYMAN && rtex->fmask_size && rtex->cmask_size) {
> rctx->compressed_cb_mask |= 1 << i;
> }
> }
> @@ -1775,27 +1821,43 @@ static void evergreen_set_framebuffer_state(struct pipe_context *ctx,
> nr_samples = 0;
>
> if (nr_samples > 1) {
> - unsigned log_samples = util_logbase2(nr_samples);
> - unsigned max_dist, line_cntl, aa_config;
> -
> - max_dist = evergreen_set_ms_pos(ctx, rstate, nr_samples);
> -
> - line_cntl = S_028C00_LAST_PIXEL(1) |
> - S_028C00_EXPAND_LINE_WIDTH(1);
> - aa_config = S_028C04_MSAA_NUM_SAMPLES(log_samples) |
> - S_028C04_MAX_SAMPLE_DIST(max_dist);
> + unsigned line_cntl = S_028C00_LAST_PIXEL(1) |
> + S_028C00_EXPAND_LINE_WIDTH(1);
> + log_samples = util_logbase2(nr_samples);
>
> if (rctx->chip_class == CAYMAN) {
> + unsigned max_dist = cayman_set_ms_pos(ctx, rstate, nr_samples);
> +
> r600_pipe_state_add_reg(rstate, CM_R_028BDC_PA_SC_LINE_CNTL, line_cntl);
> - r600_pipe_state_add_reg(rstate, CM_R_028BE0_PA_SC_AA_CONFIG, aa_config);
> + r600_pipe_state_add_reg(rstate, CM_R_028BE0_PA_SC_AA_CONFIG,
> + S_028BE0_MSAA_NUM_SAMPLES(log_samples) |
> + S_028BE0_MAX_SAMPLE_DIST(max_dist) |
> + S_028BE0_MSAA_EXPOSED_SAMPLES(log_samples));
> + r600_pipe_state_add_reg(rstate, CM_R_028804_DB_EQAA,
> + S_028804_MAX_ANCHOR_SAMPLES(log_samples) |
> + S_028804_PS_ITER_SAMPLES(log_samples) |
> + S_028804_MASK_EXPORT_NUM_SAMPLES(log_samples) |
> + S_028804_ALPHA_TO_MASK_NUM_SAMPLES(log_samples) |
> + S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
> + S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
> } else {
> + unsigned max_dist = evergreen_set_ms_pos(ctx, rstate, nr_samples);
> +
> r600_pipe_state_add_reg(rstate, R_028C00_PA_SC_LINE_CNTL, line_cntl);
> - r600_pipe_state_add_reg(rstate, R_028C04_PA_SC_AA_CONFIG, aa_config);
> + r600_pipe_state_add_reg(rstate, R_028C04_PA_SC_AA_CONFIG,
> + S_028C04_MSAA_NUM_SAMPLES(log_samples) |
> + S_028C04_MAX_SAMPLE_DIST(max_dist));
> }
> } else {
> + log_samples = 0;
> +
> if (rctx->chip_class == CAYMAN) {
> r600_pipe_state_add_reg(rstate, CM_R_028BDC_PA_SC_LINE_CNTL, S_028C00_LAST_PIXEL(1));
> r600_pipe_state_add_reg(rstate, CM_R_028BE0_PA_SC_AA_CONFIG, 0);
> + r600_pipe_state_add_reg(rstate, CM_R_028804_DB_EQAA,
> + S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
> + S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
> +
> } else {
> r600_pipe_state_add_reg(rstate, R_028C00_PA_SC_LINE_CNTL, S_028C00_LAST_PIXEL(1));
> r600_pipe_state_add_reg(rstate, R_028C04_PA_SC_AA_CONFIG, 0);
> @@ -1819,6 +1881,11 @@ static void evergreen_set_framebuffer_state(struct pipe_context *ctx,
> rctx->alphatest_state.bypass = false;
> r600_atom_dirty(rctx, &rctx->alphatest_state.atom);
> }
> +
> + if (rctx->chip_class == CAYMAN && rctx->db_misc_state.log_samples != log_samples) {
> + rctx->db_misc_state.log_samples = log_samples;
> + r600_atom_dirty(rctx, &rctx->db_misc_state.atom);
> + }
> }
>
> static void evergreen_emit_cb_misc_state(struct r600_context *rctx, struct r600_atom *atom)
> @@ -1849,6 +1916,9 @@ static void evergreen_emit_db_misc_state(struct r600_context *rctx, struct r600_
>
> if (a->occlusion_query_enabled) {
> db_count_control |= S_028004_PERFECT_ZPASS_COUNTS(1);
> + if (rctx->chip_class == CAYMAN) {
> + db_count_control |= S_028004_SAMPLE_RATE(a->log_samples);
> + }
> db_render_override |= S_02800C_NOOP_CULL_DISABLE(1);
> }
>
> @@ -2214,8 +2284,6 @@ static void cayman_init_atom_start_cs(struct r600_context *rctx)
> r600_store_value(cb, 0); /* CM_R_0288E8_SQ_LDS_ALLOC */
> r600_store_value(cb, 0); /* R_0288EC_SQ_LDS_ALLOC_PS */
>
> - r600_store_context_reg(cb, CM_R_028804_DB_EQAA, 0x110000);
> -
> r600_store_context_reg_seq(cb, R_028380_SQ_VTX_SEMANTIC_0, 34);
> r600_store_value(cb, 0); /* R_028380_SQ_VTX_SEMANTIC_0 */
> r600_store_value(cb, 0);
> diff --git a/src/gallium/drivers/r600/evergreend.h b/src/gallium/drivers/r600/evergreend.h
> index 0c56aa6..91d78f8 100644
> --- a/src/gallium/drivers/r600/evergreend.h
> +++ b/src/gallium/drivers/r600/evergreend.h
> @@ -378,6 +378,9 @@
> #define S_028C74_BANK_HEIGHT(x) (((x) & 0x3) << 16)
> #define S_028C74_MACRO_TILE_ASPECT(x) (((x) & 0x3) << 19)
> #define S_028C74_FMASK_BANK_HEIGHT(x) (((x) & 0x3) << 22)
> +#define S_028C74_NUM_SAMPLES(x) (((x) & 0x7) << 24) /* cayman only */
> +#define S_028C74_NUM_FRAGMENTS(x) (((x) & 0x3) << 27) /* cayman only */
> +#define S_028C74_FORCE_DST_ALPHA_1(x) (((x) & 0x1) << 31) /* cayman only */
>
> #define R_028C78_CB_COLOR0_DIM 0x028C78
> #define S_028C78_WIDTH_MAX(x) (((x) & 0xFFFF) << 0)
> @@ -542,6 +545,7 @@
> #define V_028040_Z_16 0x00000001
> #define V_028040_Z_24 0x00000002
> #define V_028040_Z_32_FLOAT 0x00000003
> +#define S_028040_NUM_SAMPLES(x) (((x) & 0x3) << 2) /* cayman only */
> #define S_028040_ARRAY_MODE(x) (((x) & 0xF) << 4)
> #define G_028040_ARRAY_MODE(x) (((x) >> 4) & 0xF)
> #define C_028040_ARRAY_MODE 0xFFFFFF0F
> @@ -1050,6 +1054,7 @@
> #define S_030010_ENDIAN_SWAP(x) (((x) & 0x3) << 12)
> #define G_030010_ENDIAN_SWAP(x) (((x) >> 12) & 0x3)
> #define C_030010_ENDIAN_SWAP 0xFFFFCFFF
> +#define S_030010_LOG2_NUM_FRAGMENTS(x) (((x) & 0x3) << 14) /* cayman only */
> #define S_030010_DST_SEL_X(x) (((x) & 0x7) << 16)
> #define G_030010_DST_SEL_X(x) (((x) >> 16) & 0x7)
> #define C_030010_DST_SEL_X 0xFFF8FFFF
> @@ -1574,6 +1579,7 @@
> #define R_028004_DB_COUNT_CONTROL 0x00028004
> #define S_028004_ZPASS_INCREMENT_DISABLE (((x) & 0x1) << 0)
> #define S_028004_PERFECT_ZPASS_COUNTS(x) (((x) & 0x1) << 1)
> +#define S_028004_SAMPLE_RATE(x) (((x) & 0x7) << 4) /* cayman only */
> #define R_028008_DB_DEPTH_VIEW 0x00028008
> #define S_028008_SLICE_START(x) (((x) & 0x7FF) << 0)
> #define G_028008_SLICE_START(x) (((x) >> 0) & 0x7FF)
> @@ -2224,11 +2230,26 @@
> #define CM_R_0288E8_SQ_LDS_ALLOC 0x000288E8
>
> #define CM_R_028804_DB_EQAA 0x00028804
> +#define S_028804_MAX_ANCHOR_SAMPLES(x) (((x) & 0x7) << 0)
> +#define S_028804_PS_ITER_SAMPLES(x) (((x) & 0x7) << 4)
> +#define S_028804_MASK_EXPORT_NUM_SAMPLES(x) (((x) & 0x7) << 8)
> +#define S_028804_ALPHA_TO_MASK_NUM_SAMPLES(x) (((x) & 0x7) << 12)
> +#define S_028804_HIGH_QUALITY_INTERSECTIONS(x) (((x) & 0x1) << 16)
> +#define S_028804_INCOHERENT_EQAA_READS(x) (((x) & 0x1) << 17)
> +#define S_028804_INTERPOLATE_COMP_Z(x) (((x) & 0x1) << 18)
> +#define S_028804_INTERPOLATE_SRC_Z(x) (((x) & 0x1) << 19)
> +#define S_028804_STATIC_ANCHOR_ASSOCIATIONS(x) (((x) & 0x1) << 20)
> +#define S_028804_ALPHA_TO_MASK_EQAA_DISABLE(x) (((x) & 0x1) << 21)
>
> #define CM_R_028BD4_PA_SC_CENTROID_PRIORITY_0 0x00028BD4
> #define CM_R_028BD8_PA_SC_CENTROID_PRIORITY_1 0x00028BD8
> #define CM_R_028BDC_PA_SC_LINE_CNTL 0x28bdc
> #define CM_R_028BE0_PA_SC_AA_CONFIG 0x28be0
> +#define S_028BE0_MSAA_NUM_SAMPLES(x) (((x) & 0x7) << 0)
> +#define S_028BE0_AA_MASK_CENTROID_DTMN(x) (((x) & 0x1) << 4)
> +#define S_028BE0_MAX_SAMPLE_DIST(x) (((x) & 0xf) << 13)
> +#define S_028BE0_MSAA_EXPOSED_SAMPLES(x) (((x) & 0x7) << 20)
> +#define S_028BE0_DETAIL_TO_EXPOSED_MODE(x) (((x) & 0x3) << 24)
> #define CM_R_028BE4_PA_SU_VTX_CNTL 0x28be4
> #define CM_R_028BE8_PA_CL_GB_VERT_CLIP_ADJ 0x28be8
> #define CM_R_028BEC_PA_CL_GB_VERT_DISC_ADJ 0x28bec
> diff --git a/src/gallium/drivers/r600/r600_blit.c b/src/gallium/drivers/r600/r600_blit.c
> index 5f39e0d..072df14 100644
> --- a/src/gallium/drivers/r600/r600_blit.c
> +++ b/src/gallium/drivers/r600/r600_blit.c
> @@ -260,6 +260,8 @@ static void r600_blit_decompress_color(struct pipe_context *ctx,
> struct r600_context *rctx = (struct r600_context *)ctx;
> unsigned layer, level, checked_last_layer, max_layer;
>
> + assert(rctx->chip_class != CAYMAN);
> +
> if (!rtex->dirty_level_mask)
> return;
>
> @@ -304,6 +306,13 @@ void r600_decompress_color_textures(struct r600_context *rctx,
> unsigned i;
> unsigned mask = textures->compressed_colortex_mask;
>
> + /* Cayman cannot decompress an MSAA colorbuffer,
> + * but it can read it compressed, so skip this. */
> + assert(rctx->chip_class != CAYMAN);
> + if (rctx->chip_class == CAYMAN) {
> + return;
> + }
> +
> while (mask) {
> struct pipe_sampler_view *view;
> struct r600_texture *tex;
> @@ -341,7 +350,7 @@ static void r600_copy_first_sample(struct pipe_context *ctx,
> info->src.layer, info->src.layer,
> 0, 0);
> }
> - if (rsrc->fmask_size && rsrc->cmask_size) {
> + if (rctx->chip_class != CAYMAN && rsrc->fmask_size && rsrc->cmask_size) {
> r600_blit_decompress_color(ctx, rsrc,
> 0, 0,
> info->src.layer, info->src.layer);
> @@ -399,6 +408,8 @@ static void r600_color_resolve(struct pipe_context *ctx,
> struct pipe_screen *screen = ctx->screen;
> struct pipe_resource *tmp, templ;
> struct pipe_box box;
> + unsigned sample_mask =
> + rctx->chip_class == CAYMAN ? ~0 : ((1ull << MAX2(1, info->src.res->nr_samples)) - 1);
>
> assert((info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA);
>
> @@ -407,7 +418,7 @@ static void r600_color_resolve(struct pipe_context *ctx,
> util_blitter_custom_resolve_color(rctx->blitter,
> info->dst.res, info->dst.level, info->dst.layer,
> info->src.res, info->src.layer,
> - rctx->custom_blend_resolve);
> + sample_mask, rctx->custom_blend_resolve);
> r600_blitter_end(ctx);
> return;
> }
> @@ -432,7 +443,7 @@ static void r600_color_resolve(struct pipe_context *ctx,
> util_blitter_custom_resolve_color(rctx->blitter,
> tmp, 0, 0,
> info->src.res, info->src.layer,
> - rctx->custom_blend_resolve);
> + sample_mask, rctx->custom_blend_resolve);
> r600_blitter_end(ctx);
>
> /* this is correct for upside-down blits too */
> @@ -685,7 +696,7 @@ static void r600_resource_copy_region(struct pipe_context *ctx,
> src_box->z, src_box->z + src_box->depth - 1,
> 0, u_max_sample(src));
> }
> - if (rsrc->fmask_size && rsrc->cmask_size) {
> + if (rctx->chip_class != CAYMAN && rsrc->fmask_size && rsrc->cmask_size) {
> r600_blit_decompress_color(ctx, rsrc, src_level, src_level,
> src_box->z, src_box->z + src_box->depth - 1);
> }
> @@ -748,11 +759,20 @@ static void r600_resource_copy_region(struct pipe_context *ctx,
> restore_orig[1] = TRUE;
> }
>
> - for (i = 0; i <= last_sample; i++) {
> + /* XXX Properly implement multisample textures on Cayman. In the meantime,
> + * copy only the first sample (which is the only one that doesn't return garbage). */
> + if (rctx->chip_class == CAYMAN) {
> r600_blitter_begin(ctx, R600_COPY_TEXTURE);
> - util_blitter_copy_texture(rctx->blitter, dst, dst_level, 1 << i, dstx, dsty, dstz,
> - src, src_level, i, psbox);
> + util_blitter_copy_texture(rctx->blitter, dst, dst_level, ~0, dstx, dsty, dstz,
> + src, src_level, 0, psbox);
> r600_blitter_end(ctx);
> + } else {
> + for (i = 0; i <= last_sample; i++) {
> + r600_blitter_begin(ctx, R600_COPY_TEXTURE);
> + util_blitter_copy_texture(rctx->blitter, dst, dst_level, 1 << i, dstx, dsty, dstz,
> + src, src_level, i, psbox);
> + r600_blitter_end(ctx);
> + }
> }
>
> if (restore_orig[0])
> diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
> index 8c718a7..721334d 100644
> --- a/src/gallium/drivers/r600/r600_pipe.h
> +++ b/src/gallium/drivers/r600/r600_pipe.h
> @@ -83,6 +83,7 @@ struct r600_db_misc_state {
> bool flush_depthstencil_through_cb;
> bool copy_depth, copy_stencil;
> unsigned copy_sample;
> + unsigned log_samples;
> };
>
> struct r600_cb_misc_state {
> diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
> index e880132..68f53c6 100644
> --- a/src/gallium/drivers/r600/r600_state_common.c
> +++ b/src/gallium/drivers/r600/r600_state_common.c
> @@ -631,7 +631,8 @@ void r600_set_sampler_views(struct pipe_context *pipe,
> dst->views.compressed_depthtex_mask &= ~(1 << i);
> }
>
> - if (rtex->cmask_size && rtex->fmask_size) {
> + /* Track compressed colorbuffers for Evergreen (Cayman doesn't need this). */
> + if (rctx->chip_class != CAYMAN && rtex->cmask_size && rtex->fmask_size) {
> dst->views.compressed_colortex_mask |= 1 << i;
> } else {
> dst->views.compressed_colortex_mask &= ~(1 << i);
>
More information about the mesa-dev
mailing list