#include "util/u_format_s3tc.h"
#include "util/u_memory.h"
#include "util/u_pstipple.h"
+#include "util/u_resource.h"
/* Initialize an external atom (owned by ../radeon). */
static void
*list_elem = atom;
}
-unsigned si_array_mode(unsigned mode)
-{
- switch (mode) {
- case RADEON_SURF_MODE_LINEAR_ALIGNED:
- return V_009910_ARRAY_LINEAR_ALIGNED;
- case RADEON_SURF_MODE_1D:
- return V_009910_ARRAY_1D_TILED_THIN1;
- case RADEON_SURF_MODE_2D:
- return V_009910_ARRAY_2D_TILED_THIN1;
- default:
- case RADEON_SURF_MODE_LINEAR:
- return V_009910_ARRAY_LINEAR_GENERAL;
- }
-}
-
-uint32_t si_num_banks(struct si_screen *sscreen, struct r600_texture *tex)
-{
- if (sscreen->b.chip_class >= CIK &&
- sscreen->b.info.cik_macrotile_mode_array_valid) {
- unsigned index, tileb;
-
- tileb = 8 * 8 * tex->surface.bpe;
- tileb = MIN2(tex->surface.tile_split, tileb);
-
- for (index = 0; tileb > 64; index++) {
- tileb >>= 1;
- }
- assert(index < 16);
-
- return (sscreen->b.info.cik_macrotile_mode_array[index] >> 6) & 0x3;
- }
-
- if (sscreen->b.chip_class == SI &&
- sscreen->b.info.si_tile_mode_array_valid) {
- /* Don't use stencil_tiling_index, because num_banks is always
- * read from the depth mode. */
- unsigned tile_mode_index = tex->surface.tiling_index[0];
- assert(tile_mode_index < 32);
-
- return G_009910_NUM_BANKS(sscreen->b.info.si_tile_mode_array[tile_mode_index]);
- }
-
- /* The old way. */
- switch (sscreen->b.tiling_info.num_banks) {
- case 2:
- return V_02803C_ADDR_SURF_2_BANK;
- case 4:
- return V_02803C_ADDR_SURF_4_BANK;
- case 8:
- default:
- return V_02803C_ADDR_SURF_8_BANK;
- case 16:
- return V_02803C_ADDR_SURF_16_BANK;
- }
-}
-
-unsigned cik_tile_split(unsigned tile_split)
-{
- switch (tile_split) {
- case 64:
- tile_split = V_028040_ADDR_SURF_TILE_SPLIT_64B;
- break;
- case 128:
- tile_split = V_028040_ADDR_SURF_TILE_SPLIT_128B;
- break;
- case 256:
- tile_split = V_028040_ADDR_SURF_TILE_SPLIT_256B;
- break;
- case 512:
- tile_split = V_028040_ADDR_SURF_TILE_SPLIT_512B;
- break;
- default:
- case 1024:
- tile_split = V_028040_ADDR_SURF_TILE_SPLIT_1KB;
- break;
- case 2048:
- tile_split = V_028040_ADDR_SURF_TILE_SPLIT_2KB;
- break;
- case 4096:
- tile_split = V_028040_ADDR_SURF_TILE_SPLIT_4KB;
- break;
- }
- return tile_split;
-}
-
-unsigned cik_macro_tile_aspect(unsigned macro_tile_aspect)
-{
- switch (macro_tile_aspect) {
- default:
- case 1:
- macro_tile_aspect = V_02803C_ADDR_SURF_MACRO_ASPECT_1;
- break;
- case 2:
- macro_tile_aspect = V_02803C_ADDR_SURF_MACRO_ASPECT_2;
- break;
- case 4:
- macro_tile_aspect = V_02803C_ADDR_SURF_MACRO_ASPECT_4;
- break;
- case 8:
- macro_tile_aspect = V_02803C_ADDR_SURF_MACRO_ASPECT_8;
- break;
- }
- return macro_tile_aspect;
-}
-
-unsigned cik_bank_wh(unsigned bankwh)
-{
- switch (bankwh) {
- default:
- case 1:
- bankwh = V_02803C_ADDR_SURF_BANK_WIDTH_1;
- break;
- case 2:
- bankwh = V_02803C_ADDR_SURF_BANK_WIDTH_2;
- break;
- case 4:
- bankwh = V_02803C_ADDR_SURF_BANK_WIDTH_4;
- break;
- case 8:
- bankwh = V_02803C_ADDR_SURF_BANK_WIDTH_8;
- break;
- }
- return bankwh;
-}
-
-unsigned cik_db_pipe_config(struct si_screen *sscreen, unsigned tile_mode)
-{
- if (sscreen->b.info.si_tile_mode_array_valid) {
- uint32_t gb_tile_mode = sscreen->b.info.si_tile_mode_array[tile_mode];
-
- return G_009910_PIPE_CONFIG(gb_tile_mode);
- }
-
- /* This is probably broken for a lot of chips, but it's only used
- * if the kernel cannot return the tile mode array for CIK. */
- switch (sscreen->b.info.r600_num_tile_pipes) {
- case 16:
- return V_02803C_X_ADDR_SURF_P16_32X32_16X16;
- case 8:
- return V_02803C_X_ADDR_SURF_P8_32X32_16X16;
- case 4:
- default:
- if (sscreen->b.info.r600_num_backends == 4)
- return V_02803C_X_ADDR_SURF_P4_16X16;
- else
- return V_02803C_X_ADDR_SURF_P4_8X16;
- case 2:
- return V_02803C_ADDR_SURF_P2;
- }
-}
-
static unsigned si_map_swizzle(unsigned swizzle)
{
switch (swizzle) {
- case UTIL_FORMAT_SWIZZLE_Y:
+ case PIPE_SWIZZLE_Y:
return V_008F0C_SQ_SEL_Y;
- case UTIL_FORMAT_SWIZZLE_Z:
+ case PIPE_SWIZZLE_Z:
return V_008F0C_SQ_SEL_Z;
- case UTIL_FORMAT_SWIZZLE_W:
+ case PIPE_SWIZZLE_W:
return V_008F0C_SQ_SEL_W;
- case UTIL_FORMAT_SWIZZLE_0:
+ case PIPE_SWIZZLE_0:
return V_008F0C_SQ_SEL_0;
- case UTIL_FORMAT_SWIZZLE_1:
+ case PIPE_SWIZZLE_1:
return V_008F0C_SQ_SEL_1;
- default: /* UTIL_FORMAT_SWIZZLE_X */
+ default: /* PIPE_SWIZZLE_X */
return V_008F0C_SQ_SEL_X;
}
}
/*
* Inferred framebuffer and blender state.
*
- * One of the reasons this must be derived from the framebuffer state is that:
+ * One of the reasons CB_TARGET_MASK must be derived from the framebuffer state
+ * is that:
* - The blend state mask is 0xf most of the time.
* - The COLOR1 format isn't INVALID because of possible dual-source blending,
* so COLOR1 is enabled pretty much all the time.
*
* Another reason is to avoid a hang with dual source blending.
*/
-static void si_emit_cb_target_mask(struct si_context *sctx, struct r600_atom *atom)
+static void si_emit_cb_render_state(struct si_context *sctx, struct r600_atom *atom)
{
struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
struct si_state_blend *blend = sctx->queued.named.blend;
- uint32_t mask = 0, i;
+ uint32_t cb_target_mask = 0, i;
for (i = 0; i < sctx->framebuffer.state.nr_cbufs; i++)
if (sctx->framebuffer.state.cbufs[i])
- mask |= 0xf << (4*i);
+ cb_target_mask |= 0xf << (4*i);
if (blend)
- mask &= blend->cb_target_mask;
+ 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,
if (blend && blend->dual_src_blend &&
sctx->ps_shader.cso &&
(sctx->ps_shader.cso->info.colors_written & 0x3) != 0x3)
- mask = 0;
+ cb_target_mask = 0;
+
+ radeon_set_context_reg(cs, R_028238_CB_TARGET_MASK, cb_target_mask);
+
+ /* STONEY-specific register settings. */
+ if (sctx->b.family == CHIP_STONEY) {
+ unsigned spi_shader_col_format =
+ sctx->ps_shader.cso ?
+ sctx->ps_shader.current->key.ps.epilog.spi_shader_col_format : 0;
+ unsigned sx_ps_downconvert = 0;
+ unsigned sx_blend_opt_epsilon = 0;
+ unsigned sx_blend_opt_control = 0;
+
+ for (i = 0; i < sctx->framebuffer.state.nr_cbufs; i++) {
+ struct r600_surface *surf =
+ (struct r600_surface*)sctx->framebuffer.state.cbufs[i];
+ unsigned format, swap, spi_format, colormask;
+ bool has_alpha, has_rgb;
+
+ if (!surf)
+ continue;
+
+ format = G_028C70_FORMAT(surf->cb_color_info);
+ swap = G_028C70_COMP_SWAP(surf->cb_color_info);
+ spi_format = (spi_shader_col_format >> (i * 4)) & 0xf;
+ colormask = (cb_target_mask >> (i * 4)) & 0xf;
+
+ /* Set if RGB and A are present. */
+ has_alpha = !G_028C74_FORCE_DST_ALPHA_1(surf->cb_color_attrib);
+
+ if (format == V_028C70_COLOR_8 ||
+ format == V_028C70_COLOR_16 ||
+ format == V_028C70_COLOR_32)
+ has_rgb = !has_alpha;
+ else
+ has_rgb = true;
+
+ /* Check the colormask and export format. */
+ if (!(colormask & (PIPE_MASK_RGBA & ~PIPE_MASK_A)))
+ has_rgb = false;
+ if (!(colormask & PIPE_MASK_A))
+ has_alpha = false;
+
+ if (spi_format == V_028714_SPI_SHADER_ZERO) {
+ has_rgb = false;
+ has_alpha = false;
+ }
+
+ /* Disable value checking for disabled channels. */
+ if (!has_rgb)
+ sx_blend_opt_control |= S_02875C_MRT0_COLOR_OPT_DISABLE(1) << (i * 4);
+ if (!has_alpha)
+ sx_blend_opt_control |= S_02875C_MRT0_ALPHA_OPT_DISABLE(1) << (i * 4);
+
+ /* Enable down-conversion for 32bpp and smaller formats. */
+ switch (format) {
+ case V_028C70_COLOR_8:
+ case V_028C70_COLOR_8_8:
+ case V_028C70_COLOR_8_8_8_8:
+ /* For 1 and 2-channel formats, use the superset thereof. */
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR ||
+ spi_format == V_028714_SPI_SHADER_UINT16_ABGR ||
+ spi_format == V_028714_SPI_SHADER_SINT16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_8_8_8_8 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_8BIT_FORMAT << (i * 4);
+ }
+ break;
+
+ case V_028C70_COLOR_5_6_5:
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_5_6_5 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_6BIT_FORMAT << (i * 4);
+ }
+ break;
+
+ case V_028C70_COLOR_1_5_5_5:
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_1_5_5_5 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_5BIT_FORMAT << (i * 4);
+ }
+ break;
+
+ case V_028C70_COLOR_4_4_4_4:
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_4_4_4_4 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_4BIT_FORMAT << (i * 4);
+ }
+ break;
+
+ case V_028C70_COLOR_32:
+ if (swap == V_0280A0_SWAP_STD &&
+ spi_format == V_028714_SPI_SHADER_32_R)
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_32_R << (i * 4);
+ else if (swap == V_0280A0_SWAP_ALT_REV &&
+ spi_format == V_028714_SPI_SHADER_32_AR)
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_32_A << (i * 4);
+ break;
+
+ case V_028C70_COLOR_16:
+ case V_028C70_COLOR_16_16:
+ /* For 1-channel formats, use the superset thereof. */
+ if (spi_format == V_028714_SPI_SHADER_UNORM16_ABGR ||
+ spi_format == V_028714_SPI_SHADER_SNORM16_ABGR ||
+ spi_format == V_028714_SPI_SHADER_UINT16_ABGR ||
+ spi_format == V_028714_SPI_SHADER_SINT16_ABGR) {
+ if (swap == V_0280A0_SWAP_STD ||
+ swap == V_0280A0_SWAP_STD_REV)
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_16_16_GR << (i * 4);
+ else
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_16_16_AR << (i * 4);
+ }
+ break;
+
+ case V_028C70_COLOR_10_11_11:
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_10_11_11 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_11BIT_FORMAT << (i * 4);
+ }
+ break;
+
+ case V_028C70_COLOR_2_10_10_10:
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_2_10_10_10 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_10BIT_FORMAT << (i * 4);
+ }
+ break;
+ }
+ }
+
+ if (sctx->screen->b.debug_flags & DBG_NO_RB_PLUS) {
+ sx_ps_downconvert = 0;
+ sx_blend_opt_epsilon = 0;
+ sx_blend_opt_control = 0;
+ }
- radeon_set_context_reg(cs, R_028238_CB_TARGET_MASK, mask);
+ radeon_set_context_reg_seq(cs, R_028754_SX_PS_DOWNCONVERT, 3);
+ radeon_emit(cs, sx_ps_downconvert); /* R_028754_SX_PS_DOWNCONVERT */
+ radeon_emit(cs, sx_blend_opt_epsilon); /* R_028758_SX_BLEND_OPT_EPSILON */
+ radeon_emit(cs, sx_blend_opt_control); /* R_02875C_SX_BLEND_OPT_CONTROL */
+ }
}
/*
}
}
+/**
+ * Get rid of DST in the blend factors by commuting the operands:
+ * func(src * DST, dst * 0) ---> func(src * 0, dst * SRC)
+ */
+static void si_blend_remove_dst(unsigned *func, unsigned *src_factor,
+ unsigned *dst_factor, unsigned expected_dst,
+ unsigned replacement_src)
+{
+ if (*src_factor == expected_dst &&
+ *dst_factor == PIPE_BLENDFACTOR_ZERO) {
+ *src_factor = PIPE_BLENDFACTOR_ZERO;
+ *dst_factor = replacement_src;
+
+ /* Commuting the operands requires reversing subtractions. */
+ if (*func == PIPE_BLEND_SUBTRACT)
+ *func = PIPE_BLEND_REVERSE_SUBTRACT;
+ else if (*func == PIPE_BLEND_REVERSE_SUBTRACT)
+ *func = PIPE_BLEND_SUBTRACT;
+ }
+}
+
+static bool si_blend_factor_uses_dst(unsigned factor)
+{
+ return factor == PIPE_BLENDFACTOR_DST_COLOR ||
+ factor == PIPE_BLENDFACTOR_DST_ALPHA ||
+ factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
+ factor == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
+ factor == PIPE_BLENDFACTOR_INV_DST_COLOR;
+}
+
static void *si_create_blend_state_mode(struct pipe_context *ctx,
const struct pipe_blend_state *state,
unsigned mode)
struct si_context *sctx = (struct si_context*)ctx;
struct si_state_blend *blend = CALLOC_STRUCT(si_state_blend);
struct si_pm4_state *pm4 = &blend->pm4;
-
+ uint32_t sx_mrt_blend_opt[8] = {0};
uint32_t color_control = 0;
if (!blend)
return NULL;
+ blend->alpha_to_coverage = state->alpha_to_coverage;
blend->alpha_to_one = state->alpha_to_one;
blend->dual_src_blend = util_blend_state_is_dual(state, 0);
S_028B70_ALPHA_TO_MASK_OFFSET2(2) |
S_028B70_ALPHA_TO_MASK_OFFSET3(2));
+ if (state->alpha_to_coverage)
+ blend->need_src_alpha_4bit |= 0xf;
+
blend->cb_target_mask = 0;
for (int i = 0; i < 8; i++) {
/* state->rt entries > 0 only written if independent blending */
unsigned srcA = state->rt[j].alpha_src_factor;
unsigned dstA = state->rt[j].alpha_dst_factor;
+ unsigned srcRGB_opt, dstRGB_opt, srcA_opt, dstA_opt;
unsigned blend_cntl = 0;
- /* we pretend 8 buffer are used, CB_SHADER_MASK will disable unused one */
- blend->cb_target_mask |= state->rt[j].colormask << (4 * i);
+ sx_mrt_blend_opt[i] =
+ S_028760_COLOR_COMB_FCN(V_028760_OPT_COMB_BLEND_DISABLED) |
+ S_028760_ALPHA_COMB_FCN(V_028760_OPT_COMB_BLEND_DISABLED);
+
+ 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) {
si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
continue;
}
+ /* Blending optimizations for Stoney.
+ * These transformations don't change the behavior.
+ *
+ * First, get rid of DST in the blend factors:
+ * func(src * DST, dst * 0) ---> func(src * 0, dst * SRC)
+ */
+ si_blend_remove_dst(&eqRGB, &srcRGB, &dstRGB,
+ PIPE_BLENDFACTOR_DST_COLOR,
+ PIPE_BLENDFACTOR_SRC_COLOR);
+ si_blend_remove_dst(&eqA, &srcA, &dstA,
+ PIPE_BLENDFACTOR_DST_COLOR,
+ PIPE_BLENDFACTOR_SRC_COLOR);
+ si_blend_remove_dst(&eqA, &srcA, &dstA,
+ PIPE_BLENDFACTOR_DST_ALPHA,
+ PIPE_BLENDFACTOR_SRC_ALPHA);
+
+ /* Look up the ideal settings from tables. */
+ srcRGB_opt = si_translate_blend_opt_factor(srcRGB, false);
+ dstRGB_opt = si_translate_blend_opt_factor(dstRGB, false);
+ srcA_opt = si_translate_blend_opt_factor(srcA, true);
+ dstA_opt = si_translate_blend_opt_factor(dstA, true);
+
+ /* Handle interdependencies. */
+ if (si_blend_factor_uses_dst(srcRGB))
+ dstRGB_opt = V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_NONE;
+ if (si_blend_factor_uses_dst(srcA))
+ dstA_opt = V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_NONE;
+
+ if (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE &&
+ (dstRGB == PIPE_BLENDFACTOR_ZERO ||
+ dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE))
+ dstRGB_opt = V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_A0;
+
+ /* Set the final value. */
+ sx_mrt_blend_opt[i] =
+ S_028760_COLOR_SRC_OPT(srcRGB_opt) |
+ S_028760_COLOR_DST_OPT(dstRGB_opt) |
+ S_028760_COLOR_COMB_FCN(si_translate_blend_opt_function(eqRGB)) |
+ S_028760_ALPHA_SRC_OPT(srcA_opt) |
+ S_028760_ALPHA_DST_OPT(dstA_opt) |
+ S_028760_ALPHA_COMB_FCN(si_translate_blend_opt_function(eqA));
+
+ /* Set blend state. */
blend_cntl |= S_028780_ENABLE(1);
blend_cntl |= S_028780_COLOR_COMB_FCN(si_translate_blend_function(eqRGB));
blend_cntl |= S_028780_COLOR_SRCBLEND(si_translate_blend_factor(srcRGB));
blend_cntl |= S_028780_ALPHA_DESTBLEND(si_translate_blend_factor(dstA));
}
si_pm4_set_reg(pm4, R_028780_CB_BLEND0_CONTROL + i * 4, blend_cntl);
+
+ blend->blend_enable_4bit |= 0xfu << (i * 4);
+
+ /* This is only important for formats without alpha. */
+ if (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
+ dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
+ srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
+ dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA)
+ blend->need_src_alpha_4bit |= 0xfu << (i * 4);
}
if (blend->cb_target_mask) {
}
if (sctx->b.family == CHIP_STONEY) {
- uint32_t sx_blend_opt_control = 0;
-
- for (int i = 0; i < 8; i++) {
- const int j = state->independent_blend_enable ? i : 0;
-
- /* TODO: We can also set this if the surface doesn't contain RGB. */
- if (!state->rt[j].blend_enable ||
- !(state->rt[j].colormask & (PIPE_MASK_R | PIPE_MASK_G | PIPE_MASK_B)))
- sx_blend_opt_control |= S_02875C_MRT0_COLOR_OPT_DISABLE(1) << (4 * i);
-
- /* TODO: We can also set this if the surface doesn't contain alpha. */
- if (!state->rt[j].blend_enable ||
- !(state->rt[j].colormask & PIPE_MASK_A))
- sx_blend_opt_control |= S_02875C_MRT0_ALPHA_OPT_DISABLE(1) << (4 * i);
-
- if (!state->rt[j].blend_enable) {
- si_pm4_set_reg(pm4, R_028760_SX_MRT0_BLEND_OPT + i * 4,
- S_028760_COLOR_COMB_FCN(V_028760_OPT_COMB_BLEND_DISABLED) |
- S_028760_ALPHA_COMB_FCN(V_028760_OPT_COMB_BLEND_DISABLED));
- continue;
- }
-
+ for (int i = 0; i < 8; i++)
si_pm4_set_reg(pm4, R_028760_SX_MRT0_BLEND_OPT + i * 4,
- S_028760_COLOR_SRC_OPT(si_translate_blend_opt_factor(state->rt[j].rgb_src_factor, false)) |
- S_028760_COLOR_DST_OPT(si_translate_blend_opt_factor(state->rt[j].rgb_dst_factor, false)) |
- S_028760_COLOR_COMB_FCN(si_translate_blend_opt_function(state->rt[j].rgb_func)) |
- S_028760_ALPHA_SRC_OPT(si_translate_blend_opt_factor(state->rt[j].alpha_src_factor, true)) |
- S_028760_ALPHA_DST_OPT(si_translate_blend_opt_factor(state->rt[j].alpha_dst_factor, true)) |
- S_028760_ALPHA_COMB_FCN(si_translate_blend_opt_function(state->rt[j].alpha_func)));
- }
-
- si_pm4_set_reg(pm4, R_02875C_SX_BLEND_OPT_CONTROL, sx_blend_opt_control);
+ sx_mrt_blend_opt[i]);
- /* RB+ doesn't work with dual source blending */
- if (blend->dual_src_blend)
+ /* RB+ doesn't work with dual source blending, logic op, and RESOLVE. */
+ if (blend->dual_src_blend || state->logicop_enable ||
+ mode == V_028808_CB_RESOLVE)
color_control |= S_028808_DISABLE_DUAL_QUAD(1);
}
{
struct si_context *sctx = (struct si_context *)ctx;
si_pm4_bind_state(sctx, blend, (struct si_state_blend *)state);
- si_mark_atom_dirty(sctx, &sctx->cb_target_mask);
+ si_mark_atom_dirty(sctx, &sctx->cb_render_state);
}
static void si_delete_blend_state(struct pipe_context *ctx, void *state)
}
/*
- * Clipping, scissors and viewport
+ * Clipping
*/
static void si_set_clip_state(struct pipe_context *ctx,
cb.user_buffer = state->ucp;
cb.buffer_offset = 0;
cb.buffer_size = 4*4*8;
- ctx->set_constant_buffer(ctx, PIPE_SHADER_VERTEX, SI_DRIVER_STATE_CONST_BUF, &cb);
+ si_set_rw_buffer(sctx, SI_VS_CONST_CLIP_PLANES, &cb);
pipe_resource_reference(&cb.buffer, NULL);
}
S_028AB4_REUSE_OFF(info->writes_viewport_index));
}
-static void si_set_scissor_states(struct pipe_context *ctx,
- unsigned start_slot,
- unsigned num_scissors,
- const struct pipe_scissor_state *state)
-{
- struct si_context *sctx = (struct si_context *)ctx;
- int i;
-
- for (i = 0; i < num_scissors; i++)
- sctx->scissors.states[start_slot + i] = state[i];
-
- sctx->scissors.dirty_mask |= ((1 << num_scissors) - 1) << start_slot;
- si_mark_atom_dirty(sctx, &sctx->scissors.atom);
-}
-
-static void si_emit_scissors(struct si_context *sctx, struct r600_atom *atom)
-{
- struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
- struct pipe_scissor_state *states = sctx->scissors.states;
- unsigned mask = sctx->scissors.dirty_mask;
-
- /* The simple case: Only 1 viewport is active. */
- if (mask & 1 &&
- !si_get_vs_info(sctx)->writes_viewport_index) {
- radeon_set_context_reg_seq(cs, R_028250_PA_SC_VPORT_SCISSOR_0_TL, 2);
- radeon_emit(cs, S_028250_TL_X(states[0].minx) |
- S_028250_TL_Y(states[0].miny) |
- S_028250_WINDOW_OFFSET_DISABLE(1));
- radeon_emit(cs, S_028254_BR_X(states[0].maxx) |
- S_028254_BR_Y(states[0].maxy));
- sctx->scissors.dirty_mask &= ~1; /* clear one bit */
- return;
- }
-
- while (mask) {
- int start, count, i;
-
- u_bit_scan_consecutive_range(&mask, &start, &count);
-
- radeon_set_context_reg_seq(cs, R_028250_PA_SC_VPORT_SCISSOR_0_TL +
- start * 4 * 2, count * 2);
- for (i = start; i < start+count; i++) {
- radeon_emit(cs, S_028250_TL_X(states[i].minx) |
- S_028250_TL_Y(states[i].miny) |
- S_028250_WINDOW_OFFSET_DISABLE(1));
- radeon_emit(cs, S_028254_BR_X(states[i].maxx) |
- S_028254_BR_Y(states[i].maxy));
- }
- }
- sctx->scissors.dirty_mask = 0;
-}
-
-static void si_set_viewport_states(struct pipe_context *ctx,
- unsigned start_slot,
- unsigned num_viewports,
- const struct pipe_viewport_state *state)
-{
- struct si_context *sctx = (struct si_context *)ctx;
- int i;
-
- for (i = 0; i < num_viewports; i++)
- sctx->viewports.states[start_slot + i] = state[i];
-
- sctx->viewports.dirty_mask |= ((1 << num_viewports) - 1) << start_slot;
- si_mark_atom_dirty(sctx, &sctx->viewports.atom);
-}
-
-static void si_emit_viewports(struct si_context *sctx, struct r600_atom *atom)
-{
- struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
- struct pipe_viewport_state *states = sctx->viewports.states;
- unsigned mask = sctx->viewports.dirty_mask;
-
- /* The simple case: Only 1 viewport is active. */
- if (mask & 1 &&
- !si_get_vs_info(sctx)->writes_viewport_index) {
- radeon_set_context_reg_seq(cs, R_02843C_PA_CL_VPORT_XSCALE, 6);
- radeon_emit(cs, fui(states[0].scale[0]));
- radeon_emit(cs, fui(states[0].translate[0]));
- radeon_emit(cs, fui(states[0].scale[1]));
- radeon_emit(cs, fui(states[0].translate[1]));
- radeon_emit(cs, fui(states[0].scale[2]));
- radeon_emit(cs, fui(states[0].translate[2]));
- sctx->viewports.dirty_mask &= ~1; /* clear one bit */
- return;
- }
-
- while (mask) {
- int start, count, i;
-
- u_bit_scan_consecutive_range(&mask, &start, &count);
-
- radeon_set_context_reg_seq(cs, R_02843C_PA_CL_VPORT_XSCALE +
- start * 4 * 6, count * 6);
- for (i = start; i < start+count; i++) {
- radeon_emit(cs, fui(states[i].scale[0]));
- radeon_emit(cs, fui(states[i].translate[0]));
- radeon_emit(cs, fui(states[i].scale[1]));
- radeon_emit(cs, fui(states[i].translate[1]));
- radeon_emit(cs, fui(states[i].scale[2]));
- radeon_emit(cs, fui(states[i].translate[2]));
- }
- }
- sctx->viewports.dirty_mask = 0;
-}
-
/*
* inferred state between framebuffer and rasterizer
*/
return NULL;
}
+ rs->scissor_enable = state->scissor;
rs->two_side = state->light_twoside;
rs->multisample_enable = state->multisample;
rs->force_persample_interp = state->force_persample_interp;
S_028A0C_LINE_PATTERN(state->line_stipple_pattern) |
S_028A0C_REPEAT_COUNT(state->line_stipple_factor) : 0;
rs->pa_cl_clip_cntl =
- S_028810_PS_UCP_MODE(3) |
S_028810_DX_CLIP_SPACE_DEF(state->clip_halfz) |
S_028810_ZCLIP_NEAR_DISABLE(!state->depth_clip) |
S_028810_ZCLIP_FAR_DISABLE(!state->depth_clip) |
S_028A48_MSAA_ENABLE(state->multisample ||
state->poly_smooth ||
state->line_smooth) |
- S_028A48_VPORT_SCISSOR_ENABLE(state->scissor));
+ S_028A48_VPORT_SCISSOR_ENABLE(1));
si_pm4_set_reg(pm4, R_028BE4_PA_SU_VTX_CNTL,
S_028BE4_PIX_CENTER(state->half_pixel_center) |
(!old_rs || old_rs->multisample_enable != rs->multisample_enable))
si_mark_atom_dirty(sctx, &sctx->db_render_state);
+ r600_set_scissor_enable(&sctx->b, rs->scissor_enable);
+
si_pm4_bind_state(sctx, rasterizer, rs);
si_update_poly_offset_state(sctx);
/* DB RENDER STATE */
+static void si_set_active_query_state(struct pipe_context *ctx, boolean enable)
+{
+ struct si_context *sctx = (struct si_context*)ctx;
+
+ /* Pipeline stat & streamout queries. */
+ if (enable) {
+ sctx->b.flags &= ~R600_CONTEXT_STOP_PIPELINE_STATS;
+ sctx->b.flags |= R600_CONTEXT_START_PIPELINE_STATS;
+ } else {
+ sctx->b.flags &= ~R600_CONTEXT_START_PIPELINE_STATS;
+ sctx->b.flags |= R600_CONTEXT_STOP_PIPELINE_STATS;
+ }
+
+ /* Occlusion queries. */
+ if (sctx->occlusion_queries_disabled != !enable) {
+ sctx->occlusion_queries_disabled = !enable;
+ si_mark_atom_dirty(sctx, &sctx->db_render_state);
+ }
+}
+
static void si_set_occlusion_query_state(struct pipe_context *ctx, bool enable)
{
struct si_context *sctx = (struct si_context*)ctx;
}
/* DB_COUNT_CONTROL (occlusion queries) */
- if (sctx->b.num_occlusion_queries > 0) {
+ if (sctx->b.num_occlusion_queries > 0 &&
+ !sctx->occlusion_queries_disabled) {
+ bool perfect = sctx->b.num_perfect_occlusion_queries > 0;
+
if (sctx->b.chip_class >= CIK) {
radeon_emit(cs,
- S_028004_PERFECT_ZPASS_COUNTS(1) |
+ S_028004_PERFECT_ZPASS_COUNTS(perfect) |
S_028004_SAMPLE_RATE(sctx->framebuffer.log_samples) |
S_028004_ZPASS_ENABLE(1) |
S_028004_SLICE_EVEN_ENABLE(1) |
S_028004_SLICE_ODD_ENABLE(1));
} else {
radeon_emit(cs,
- S_028004_PERFECT_ZPASS_COUNTS(1) |
+ S_028004_PERFECT_ZPASS_COUNTS(perfect) |
S_028004_SAMPLE_RATE(sctx->framebuffer.log_samples));
}
} else {
/* DB_RENDER_OVERRIDE2 */
radeon_set_context_reg(cs, R_028010_DB_RENDER_OVERRIDE2,
S_028010_DISABLE_ZMASK_EXPCLEAR_OPTIMIZATION(sctx->db_depth_disable_expclear) |
- S_028010_DISABLE_SMEM_EXPCLEAR_OPTIMIZATION(sctx->db_stencil_disable_expclear));
+ S_028010_DISABLE_SMEM_EXPCLEAR_OPTIMIZATION(sctx->db_stencil_disable_expclear) |
+ S_028010_DECOMPRESS_Z_ON_FLUSH(sctx->framebuffer.nr_samples >= 4));
db_shader_control = S_02880C_ALPHA_TO_MASK_DISABLE(sctx->framebuffer.cb0_is_integer) |
sctx->ps_db_shader_control;
/* Bug workaround for smoothing (overrasterization) on SI. */
- if (sctx->b.chip_class == SI && sctx->smoothing_enabled)
+ if (sctx->b.chip_class == SI && sctx->smoothing_enabled) {
+ db_shader_control &= C_02880C_Z_ORDER;
db_shader_control |= S_02880C_Z_ORDER(V_02880C_LATE_Z);
- else
- db_shader_control |= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
+ }
/* Disable the gl_SampleMask fragment shader output if MSAA is disabled. */
if (sctx->framebuffer.nr_samples <= 1 || (rs && !rs->multisample_enable))
if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
return V_028C70_COLOR_INVALID;
+ /* hw cannot support mixed formats (except depth/stencil, since
+ * stencil is not written to). */
+ if (desc->is_mixed && desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
+ return V_028C70_COLOR_INVALID;
+
switch (desc->nr_channels) {
case 1:
switch (desc->channel[0].size) {
}
}
-/* Returns the size in bits of the widest component of a CB format */
-static unsigned si_colorformat_max_comp_size(uint32_t colorformat)
-{
- switch(colorformat) {
- case V_028C70_COLOR_4_4_4_4:
- return 4;
-
- case V_028C70_COLOR_1_5_5_5:
- case V_028C70_COLOR_5_5_5_1:
- return 5;
-
- case V_028C70_COLOR_5_6_5:
- return 6;
-
- case V_028C70_COLOR_8:
- case V_028C70_COLOR_8_8:
- case V_028C70_COLOR_8_8_8_8:
- return 8;
-
- case V_028C70_COLOR_10_10_10_2:
- case V_028C70_COLOR_2_10_10_10:
- return 10;
-
- case V_028C70_COLOR_10_11_11:
- case V_028C70_COLOR_11_11_10:
- return 11;
-
- case V_028C70_COLOR_16:
- case V_028C70_COLOR_16_16:
- case V_028C70_COLOR_16_16_16_16:
- return 16;
-
- case V_028C70_COLOR_8_24:
- case V_028C70_COLOR_24_8:
- return 24;
-
- case V_028C70_COLOR_32:
- case V_028C70_COLOR_32_32:
- case V_028C70_COLOR_32_32_32_32:
- case V_028C70_COLOR_X24_8_32_FLOAT:
- return 32;
- }
-
- assert(!"Unknown maximum component size");
- return 0;
-}
-
static uint32_t si_translate_dbformat(enum pipe_format format)
{
switch (format) {
}
}
+ if (desc->layout == UTIL_FORMAT_LAYOUT_ETC &&
+ sscreen->b.family == CHIP_STONEY) {
+ switch (format) {
+ case PIPE_FORMAT_ETC1_RGB8:
+ case PIPE_FORMAT_ETC2_RGB8:
+ case PIPE_FORMAT_ETC2_SRGB8:
+ return V_008F14_IMG_DATA_FORMAT_ETC2_RGB;
+ case PIPE_FORMAT_ETC2_RGB8A1:
+ case PIPE_FORMAT_ETC2_SRGB8A1:
+ return V_008F14_IMG_DATA_FORMAT_ETC2_RGBA1;
+ case PIPE_FORMAT_ETC2_RGBA8:
+ case PIPE_FORMAT_ETC2_SRGBA8:
+ return V_008F14_IMG_DATA_FORMAT_ETC2_RGBA;
+ case PIPE_FORMAT_ETC2_R11_UNORM:
+ case PIPE_FORMAT_ETC2_R11_SNORM:
+ return V_008F14_IMG_DATA_FORMAT_ETC2_R;
+ case PIPE_FORMAT_ETC2_RG11_UNORM:
+ case PIPE_FORMAT_ETC2_RG11_SNORM:
+ return V_008F14_IMG_DATA_FORMAT_ETC2_RG;
+ default:
+ goto out_unknown;
+ }
+ }
+
if (desc->layout == UTIL_FORMAT_LAYOUT_BPTC) {
if (!enable_compressed_formats)
goto out_unknown;
/* R8G8Bx_SNORM - TODO CxV8U8 */
+ /* hw cannot support mixed formats (except depth/stencil, since only
+ * depth is read).*/
+ if (desc->is_mixed && desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
+ goto out_unknown;
+
/* See whether the components are of the same size. */
for (i = 1; i < desc->nr_channels; i++) {
uniform = uniform && desc->channel[0].size == desc->channel[i].size;
}
}
-static unsigned si_tex_filter(unsigned filter)
-{
- switch (filter) {
- default:
- case PIPE_TEX_FILTER_NEAREST:
- return V_008F38_SQ_TEX_XY_FILTER_POINT;
- case PIPE_TEX_FILTER_LINEAR:
- return V_008F38_SQ_TEX_XY_FILTER_BILINEAR;
- }
-}
-
static unsigned si_tex_mipfilter(unsigned filter)
{
switch (filter) {
const struct util_format_description *desc,
int first_non_void)
{
- unsigned type = desc->channel[first_non_void].type;
+ unsigned type;
int i;
- if (type == UTIL_FORMAT_TYPE_FIXED)
- return V_008F0C_BUF_DATA_FORMAT_INVALID;
-
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 &&
desc->channel[1].size == 10 &&
if (desc->format == PIPE_FORMAT_R11G11B10_FLOAT)
return V_008F0C_BUF_NUM_FORMAT_FLOAT;
+ assert(first_non_void >= 0);
+
switch (desc->channel[first_non_void].type) {
case UTIL_FORMAT_TYPE_SIGNED:
if (desc->channel[first_non_void].normalized)
static bool si_is_colorbuffer_format_supported(enum pipe_format format)
{
return si_translate_colorformat(format) != V_028C70_COLOR_INVALID &&
- r600_translate_colorswap(format) != ~0U;
+ r600_translate_colorswap(format, FALSE) != ~0U;
}
static bool si_is_zs_format_supported(enum pipe_format format)
case 4:
case 8:
break;
+ case 16:
+ if (format == PIPE_FORMAT_NONE)
+ return TRUE;
+ else
+ return FALSE;
default:
return FALSE;
}
}
- if (usage & PIPE_BIND_SAMPLER_VIEW) {
+ if (usage & (PIPE_BIND_SAMPLER_VIEW |
+ PIPE_BIND_SHADER_IMAGE)) {
if (target == PIPE_BUFFER) {
if (si_is_vertex_format_supported(screen, format))
- retval |= PIPE_BIND_SAMPLER_VIEW;
+ retval |= usage & (PIPE_BIND_SAMPLER_VIEW |
+ PIPE_BIND_SHADER_IMAGE);
} else {
if (si_is_sampler_format_supported(screen, format))
- retval |= PIPE_BIND_SAMPLER_VIEW;
+ retval |= usage & (PIPE_BIND_SAMPLER_VIEW |
+ PIPE_BIND_SHADER_IMAGE);
}
}
if (usage & PIPE_BIND_TRANSFER_WRITE)
retval |= PIPE_BIND_TRANSFER_WRITE;
- return retval == usage;
-}
-
-unsigned si_tile_mode_index(struct r600_texture *rtex, unsigned level, bool stencil)
-{
- unsigned tile_mode_index = 0;
+ if ((usage & PIPE_BIND_LINEAR) &&
+ !util_format_is_compressed(format) &&
+ !(usage & PIPE_BIND_DEPTH_STENCIL))
+ retval |= PIPE_BIND_LINEAR;
- if (stencil) {
- tile_mode_index = rtex->surface.stencil_tiling_index[level];
- } else {
- tile_mode_index = rtex->surface.tiling_index[level];
- }
- return tile_mode_index;
+ return retval == usage;
}
/*
* framebuffer handling
*/
-static void si_initialize_color_surface(struct si_context *sctx,
- struct r600_surface *surf)
+static void si_choose_spi_color_formats(struct r600_surface *surf,
+ unsigned format, unsigned swap,
+ unsigned ntype, bool is_depth)
+{
+ /* Alpha is needed for alpha-to-coverage.
+ * Blending may be with or without alpha.
+ */
+ unsigned normal = 0; /* most optimal, may not support blending or export alpha */
+ unsigned alpha = 0; /* exports alpha, but may not support blending */
+ unsigned blend = 0; /* supports blending, but may not export alpha */
+ unsigned blend_alpha = 0; /* least optimal, supports blending and exports alpha */
+
+ /* Choose the SPI color formats. These are required values for Stoney/RB+.
+ * Other chips have multiple choices, though they are not necessarily better.
+ */
+ switch (format) {
+ case V_028C70_COLOR_5_6_5:
+ case V_028C70_COLOR_1_5_5_5:
+ case V_028C70_COLOR_5_5_5_1:
+ case V_028C70_COLOR_4_4_4_4:
+ case V_028C70_COLOR_10_11_11:
+ case V_028C70_COLOR_11_11_10:
+ case V_028C70_COLOR_8:
+ case V_028C70_COLOR_8_8:
+ case V_028C70_COLOR_8_8_8_8:
+ case V_028C70_COLOR_10_10_10_2:
+ case V_028C70_COLOR_2_10_10_10:
+ if (ntype == V_028C70_NUMBER_UINT)
+ alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_UINT16_ABGR;
+ else if (ntype == V_028C70_NUMBER_SINT)
+ alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_SINT16_ABGR;
+ else
+ alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_FP16_ABGR;
+ break;
+
+ case V_028C70_COLOR_16:
+ case V_028C70_COLOR_16_16:
+ case V_028C70_COLOR_16_16_16_16:
+ if (ntype == V_028C70_NUMBER_UNORM ||
+ ntype == V_028C70_NUMBER_SNORM) {
+ /* UNORM16 and SNORM16 don't support blending */
+ if (ntype == V_028C70_NUMBER_UNORM)
+ normal = alpha = V_028714_SPI_SHADER_UNORM16_ABGR;
+ else
+ normal = alpha = V_028714_SPI_SHADER_SNORM16_ABGR;
+
+ /* Use 32 bits per channel for blending. */
+ if (format == V_028C70_COLOR_16) {
+ if (swap == V_028C70_SWAP_STD) { /* R */
+ blend = V_028714_SPI_SHADER_32_R;
+ blend_alpha = V_028714_SPI_SHADER_32_AR;
+ } else if (swap == V_028C70_SWAP_ALT_REV) /* A */
+ blend = blend_alpha = V_028714_SPI_SHADER_32_AR;
+ else
+ assert(0);
+ } else if (format == V_028C70_COLOR_16_16) {
+ if (swap == V_028C70_SWAP_STD) { /* RG */
+ blend = V_028714_SPI_SHADER_32_GR;
+ blend_alpha = V_028714_SPI_SHADER_32_ABGR;
+ } else if (swap == V_028C70_SWAP_ALT) /* RA */
+ blend = blend_alpha = V_028714_SPI_SHADER_32_AR;
+ else
+ assert(0);
+ } else /* 16_16_16_16 */
+ blend = blend_alpha = V_028714_SPI_SHADER_32_ABGR;
+ } else if (ntype == V_028C70_NUMBER_UINT)
+ alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_UINT16_ABGR;
+ else if (ntype == V_028C70_NUMBER_SINT)
+ alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_SINT16_ABGR;
+ else if (ntype == V_028C70_NUMBER_FLOAT)
+ alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_FP16_ABGR;
+ else
+ assert(0);
+ break;
+
+ case V_028C70_COLOR_32:
+ if (swap == V_028C70_SWAP_STD) { /* R */
+ blend = normal = V_028714_SPI_SHADER_32_R;
+ alpha = blend_alpha = V_028714_SPI_SHADER_32_AR;
+ } else if (swap == V_028C70_SWAP_ALT_REV) /* A */
+ alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_32_AR;
+ else
+ assert(0);
+ break;
+
+ case V_028C70_COLOR_32_32:
+ if (swap == V_028C70_SWAP_STD) { /* RG */
+ blend = normal = V_028714_SPI_SHADER_32_GR;
+ alpha = blend_alpha = V_028714_SPI_SHADER_32_ABGR;
+ } else if (swap == V_028C70_SWAP_ALT) /* RA */
+ alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_32_AR;
+ else
+ assert(0);
+ break;
+
+ case V_028C70_COLOR_32_32_32_32:
+ case V_028C70_COLOR_8_24:
+ case V_028C70_COLOR_24_8:
+ case V_028C70_COLOR_X24_8_32_FLOAT:
+ alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_32_ABGR;
+ break;
+
+ default:
+ assert(0);
+ return;
+ }
+
+ /* The DB->CB copy needs 32_ABGR. */
+ if (is_depth)
+ alpha = blend = blend_alpha = normal = V_028714_SPI_SHADER_32_ABGR;
+
+ surf->spi_shader_col_format = normal;
+ surf->spi_shader_col_format_alpha = alpha;
+ surf->spi_shader_col_format_blend = blend;
+ surf->spi_shader_col_format_blend_alpha = blend_alpha;
+}
+
+static void si_initialize_color_surface(struct si_context *sctx,
+ struct r600_surface *surf)
{
struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
- unsigned level = surf->base.u.tex.level;
- uint64_t offset = rtex->surface.level[level].offset;
- unsigned pitch, slice;
- unsigned color_info, color_attrib, color_pitch, color_view;
- unsigned tile_mode_index;
+ unsigned color_info, color_attrib, color_view;
unsigned format, swap, ntype, endian;
const struct util_format_description *desc;
int i;
unsigned blend_clamp = 0, blend_bypass = 0;
- unsigned max_comp_size;
-
- /* Layered rendering doesn't work with LINEAR_GENERAL.
- * (LINEAR_ALIGNED and others work) */
- if (rtex->surface.level[level].mode == RADEON_SURF_MODE_LINEAR) {
- assert(surf->base.u.tex.first_layer == surf->base.u.tex.last_layer);
- offset += rtex->surface.level[level].slice_size *
- surf->base.u.tex.first_layer;
- color_view = 0;
- } else {
- color_view = S_028C6C_SLICE_START(surf->base.u.tex.first_layer) |
- S_028C6C_SLICE_MAX(surf->base.u.tex.last_layer);
- }
- pitch = (rtex->surface.level[level].nblk_x) / 8 - 1;
- slice = (rtex->surface.level[level].nblk_x * rtex->surface.level[level].nblk_y) / 64;
- if (slice) {
- slice = slice - 1;
- }
-
- tile_mode_index = si_tile_mode_index(rtex, level, false);
+ color_view = S_028C6C_SLICE_START(surf->base.u.tex.first_layer) |
+ S_028C6C_SLICE_MAX(surf->base.u.tex.last_layer);
desc = util_format_description(surf->base.format);
for (i = 0; i < 4; i++) {
R600_ERR("Invalid CB format: %d, disabling CB.\n", surf->base.format);
}
assert(format != V_028C70_COLOR_INVALID);
- swap = r600_translate_colorswap(surf->base.format);
- if (rtex->resource.b.b.usage == PIPE_USAGE_STAGING) {
- endian = V_028C70_ENDIAN_NONE;
- } else {
- endian = si_colorformat_endian_swap(format);
- }
+ swap = r600_translate_colorswap(surf->base.format, FALSE);
+ endian = si_colorformat_endian_swap(format);
/* blend clamp should be set for all NORM/SRGB types */
if (ntype == V_028C70_NUMBER_UNORM ||
blend_bypass = 1;
}
+ if ((ntype == V_028C70_NUMBER_UINT || ntype == V_028C70_NUMBER_SINT) &&
+ (format == V_028C70_COLOR_8 ||
+ format == V_028C70_COLOR_8_8 ||
+ format == V_028C70_COLOR_8_8_8_8))
+ surf->color_is_int8 = true;
+
color_info = S_028C70_FORMAT(format) |
S_028C70_COMP_SWAP(swap) |
S_028C70_BLEND_CLAMP(blend_clamp) |
S_028C70_NUMBER_TYPE(ntype) |
S_028C70_ENDIAN(endian);
- color_pitch = S_028C64_TILE_MAX(pitch);
-
- color_attrib = S_028C74_TILE_MODE_INDEX(tile_mode_index) |
- S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] == UTIL_FORMAT_SWIZZLE_1);
+ /* Intensity is implemented as Red, so treat it that way. */
+ color_attrib = S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] == PIPE_SWIZZLE_1 ||
+ util_format_is_intensity(surf->base.format));
if (rtex->resource.b.b.nr_samples > 1) {
unsigned log_samples = util_logbase2(rtex->resource.b.b.nr_samples);
color_info |= S_028C70_COMPRESSION(1);
unsigned fmask_bankh = util_logbase2(rtex->fmask.bank_height);
- color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(rtex->fmask.tile_mode_index);
-
if (sctx->b.chip_class == SI) {
/* due to a hw bug, FMASK_BANK_HEIGHT must be set on SI too */
color_attrib |= S_028C74_FMASK_BANK_HEIGHT(fmask_bankh);
}
- if (sctx->b.chip_class >= CIK) {
- color_pitch |= S_028C64_FMASK_TILE_MAX(rtex->fmask.pitch_in_pixels / 8 - 1);
- }
}
}
- offset += rtex->resource.gpu_address;
-
- surf->cb_color_base = offset >> 8;
- surf->cb_color_pitch = color_pitch;
- surf->cb_color_slice = S_028C68_TILE_MAX(slice);
surf->cb_color_view = color_view;
surf->cb_color_info = color_info;
surf->cb_color_attrib = color_attrib;
- if (sctx->b.chip_class >= VI && rtex->dcc_buffer) {
+ if (sctx->b.chip_class >= VI) {
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)
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;
- surf->cb_color_fmask_slice = S_028C88_TILE_MAX(rtex->fmask.slice_tile_max);
- } else {
- /* This must be set for fast clear to work without FMASK. */
- surf->cb_color_fmask = surf->cb_color_base;
- surf->cb_color_fmask_slice = surf->cb_color_slice;
- surf->cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index);
-
- if (sctx->b.chip_class == SI) {
- unsigned bankh = util_logbase2(rtex->surface.bankh);
- surf->cb_color_attrib |= S_028C74_FMASK_BANK_HEIGHT(bankh);
- }
-
- if (sctx->b.chip_class >= CIK) {
- surf->cb_color_pitch |= S_028C64_FMASK_TILE_MAX(pitch);
- }
+ /* This must be set for fast clear to work without FMASK. */
+ if (!rtex->fmask.size && sctx->b.chip_class == SI) {
+ unsigned bankh = util_logbase2(rtex->surface.bankh);
+ surf->cb_color_attrib |= S_028C74_FMASK_BANK_HEIGHT(bankh);
}
/* Determine pixel shader export format */
- max_comp_size = si_colorformat_max_comp_size(format);
- if (ntype == V_028C70_NUMBER_SRGB ||
- ((ntype == V_028C70_NUMBER_UNORM || ntype == V_028C70_NUMBER_SNORM) &&
- max_comp_size <= 10) ||
- (ntype == V_028C70_NUMBER_FLOAT && max_comp_size <= 16)) {
- surf->export_16bpc = true;
- }
-
- if (sctx->b.family == CHIP_STONEY &&
- !(sctx->screen->b.debug_flags & DBG_NO_RB_PLUS)) {
- switch (desc->channel[0].size) {
- case 32:
- if (desc->nr_channels == 1) {
- if (swap == V_0280A0_SWAP_STD)
- surf->sx_ps_downconvert = V_028754_SX_RT_EXPORT_32_R;
- else if (swap == V_0280A0_SWAP_ALT_REV)
- surf->sx_ps_downconvert = V_028754_SX_RT_EXPORT_32_A;
- }
- break;
- case 16:
- /* For 1-channel formats, use the superset thereof. */
- if (desc->nr_channels <= 2) {
- if (swap == V_0280A0_SWAP_STD ||
- swap == V_0280A0_SWAP_STD_REV)
- surf->sx_ps_downconvert = V_028754_SX_RT_EXPORT_16_16_GR;
- else
- surf->sx_ps_downconvert = V_028754_SX_RT_EXPORT_16_16_AR;
- }
- break;
- case 11:
- if (desc->nr_channels == 3) {
- surf->sx_ps_downconvert = V_028754_SX_RT_EXPORT_10_11_11;
- surf->sx_blend_opt_epsilon = V_028758_11BIT_FORMAT;
- }
- break;
- case 10:
- if (desc->nr_channels == 4) {
- surf->sx_ps_downconvert = V_028754_SX_RT_EXPORT_2_10_10_10;
- surf->sx_blend_opt_epsilon = V_028758_10BIT_FORMAT;
- }
- break;
- case 8:
- /* For 1 and 2-channel formats, use the superset thereof. */
- surf->sx_ps_downconvert = V_028754_SX_RT_EXPORT_8_8_8_8;
- surf->sx_blend_opt_epsilon = V_028758_8BIT_FORMAT;
- break;
- case 5:
- if (desc->nr_channels == 3) {
- surf->sx_ps_downconvert = V_028754_SX_RT_EXPORT_5_6_5;
- surf->sx_blend_opt_epsilon = V_028758_6BIT_FORMAT;
- } else if (desc->nr_channels == 4) {
- surf->sx_ps_downconvert = V_028754_SX_RT_EXPORT_1_5_5_5;
- surf->sx_blend_opt_epsilon = V_028758_5BIT_FORMAT;
- }
- break;
- case 4:
- /* For 1 nad 2-channel formats, use the superset thereof. */
- surf->sx_ps_downconvert = V_028754_SX_RT_EXPORT_4_4_4_4;
- surf->sx_blend_opt_epsilon = V_028758_4BIT_FORMAT;
- break;
- }
- }
+ si_choose_spi_color_formats(surf, format, swap, ntype, rtex->is_depth);
surf->color_initialized = true;
}
static void si_init_depth_surface(struct si_context *sctx,
struct r600_surface *surf)
{
- struct si_screen *sscreen = sctx->screen;
struct r600_texture *rtex = (struct r600_texture*)surf->base.texture;
unsigned level = surf->base.u.tex.level;
struct radeon_surf_level *levelinfo = &rtex->surface.level[level];
- unsigned format, tile_mode_index, array_mode;
- unsigned macro_aspect, tile_split, stile_split, bankh, bankw, nbanks, pipe_config;
+ unsigned format;
uint32_t z_info, s_info, db_depth_info;
uint64_t z_offs, s_offs;
uint32_t db_htile_data_base, db_htile_surface, pa_su_poly_offset_db_fmt_cntl = 0;
s_info = S_028044_FORMAT(V_028044_STENCIL_INVALID);
if (sctx->b.chip_class >= CIK) {
- switch (rtex->surface.level[level].mode) {
- case RADEON_SURF_MODE_2D:
- array_mode = V_02803C_ARRAY_2D_TILED_THIN1;
- break;
- case RADEON_SURF_MODE_1D:
- case RADEON_SURF_MODE_LINEAR_ALIGNED:
- case RADEON_SURF_MODE_LINEAR:
- default:
- array_mode = V_02803C_ARRAY_1D_TILED_THIN1;
- break;
- }
- tile_split = rtex->surface.tile_split;
- stile_split = rtex->surface.stencil_tile_split;
- macro_aspect = rtex->surface.mtilea;
- bankw = rtex->surface.bankw;
- bankh = rtex->surface.bankh;
- tile_split = cik_tile_split(tile_split);
- stile_split = cik_tile_split(stile_split);
- macro_aspect = cik_macro_tile_aspect(macro_aspect);
- bankw = cik_bank_wh(bankw);
- bankh = cik_bank_wh(bankh);
- nbanks = si_num_banks(sscreen, rtex);
- tile_mode_index = si_tile_mode_index(rtex, level, false);
- pipe_config = cik_db_pipe_config(sscreen, tile_mode_index);
-
- db_depth_info |= S_02803C_ARRAY_MODE(array_mode) |
- S_02803C_PIPE_CONFIG(pipe_config) |
- S_02803C_BANK_WIDTH(bankw) |
- S_02803C_BANK_HEIGHT(bankh) |
- S_02803C_MACRO_TILE_ASPECT(macro_aspect) |
- S_02803C_NUM_BANKS(nbanks);
- z_info |= S_028040_TILE_SPLIT(tile_split);
- s_info |= S_028044_TILE_SPLIT(stile_split);
+ struct radeon_info *info = &sctx->screen->b.info;
+ unsigned index = rtex->surface.tiling_index[level];
+ unsigned stencil_index = rtex->surface.stencil_tiling_index[level];
+ unsigned macro_index = rtex->surface.macro_tile_index;
+ unsigned tile_mode = info->si_tile_mode_array[index];
+ unsigned stencil_tile_mode = info->si_tile_mode_array[stencil_index];
+ unsigned macro_mode = info->cik_macrotile_mode_array[macro_index];
+
+ db_depth_info |=
+ S_02803C_ARRAY_MODE(G_009910_ARRAY_MODE(tile_mode)) |
+ S_02803C_PIPE_CONFIG(G_009910_PIPE_CONFIG(tile_mode)) |
+ S_02803C_BANK_WIDTH(G_009990_BANK_WIDTH(macro_mode)) |
+ S_02803C_BANK_HEIGHT(G_009990_BANK_HEIGHT(macro_mode)) |
+ S_02803C_MACRO_TILE_ASPECT(G_009990_MACRO_TILE_ASPECT(macro_mode)) |
+ S_02803C_NUM_BANKS(G_009990_NUM_BANKS(macro_mode));
+ z_info |= S_028040_TILE_SPLIT(G_009910_TILE_SPLIT(tile_mode));
+ s_info |= S_028044_TILE_SPLIT(G_009910_TILE_SPLIT(stencil_tile_mode));
} else {
- tile_mode_index = si_tile_mode_index(rtex, level, false);
+ unsigned tile_mode_index = si_tile_mode_index(rtex, level, false);
z_info |= S_028040_TILE_MODE_INDEX(tile_mode_index);
tile_mode_index = si_tile_mode_index(rtex, level, true);
s_info |= S_028044_TILE_MODE_INDEX(tile_mode_index);
z_info |= S_028040_TILE_SURFACE_ENABLE(1) |
S_028040_ALLOW_EXPCLEAR(1);
- if (rtex->surface.flags & RADEON_SURF_SBUFFER)
- s_info |= S_028044_ALLOW_EXPCLEAR(1);
- else
+ if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
+ /* Workaround: For a not yet understood reason, the
+ * combination of MSAA, fast stencil clear and stencil
+ * decompress messes with subsequent stencil buffer
+ * uses. Problem was reproduced on Verde, Bonaire,
+ * Tonga, and Carrizo.
+ *
+ * Disabling EXPCLEAR works around the problem.
+ *
+ * Check piglit's arb_texture_multisample-stencil-clear
+ * test if you want to try changing this.
+ */
+ if (rtex->resource.b.b.nr_samples <= 1)
+ s_info |= S_028044_ALLOW_EXPCLEAR(1);
+ } else
/* Use all of the htile_buffer for depth if there's no stencil. */
s_info |= S_028044_TILE_STENCIL_DISABLE(1);
surf->depth_initialized = true;
}
+void si_dec_framebuffer_counters(const struct pipe_framebuffer_state *state)
+{
+ for (int i = 0; i < state->nr_cbufs; ++i) {
+ struct r600_surface *surf = NULL;
+ struct r600_texture *rtex;
+
+ if (!state->cbufs[i])
+ continue;
+ surf = (struct r600_surface*)state->cbufs[i];
+ rtex = (struct r600_texture*)surf->base.texture;
+
+ p_atomic_dec(&rtex->framebuffers_bound);
+ }
+}
+
static void si_set_framebuffer_state(struct pipe_context *ctx,
const struct pipe_framebuffer_state *state)
{
struct r600_surface *surf = NULL;
struct r600_texture *rtex;
bool old_cb0_is_integer = sctx->framebuffer.cb0_is_integer;
+ bool old_any_dst_linear = sctx->framebuffer.any_dst_linear;
unsigned old_nr_samples = sctx->framebuffer.nr_samples;
int i;
*/
sctx->b.flags |= SI_CONTEXT_INV_VMEM_L1 |
SI_CONTEXT_INV_GLOBAL_L2 |
- SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER;
+ SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER |
+ SI_CONTEXT_CS_PARTIAL_FLUSH;
/* Take the maximum of the old and new count. If the new count is lower,
* dirtying is needed to disable the unbound colorbuffers.
(1 << MAX2(sctx->framebuffer.state.nr_cbufs, state->nr_cbufs)) - 1;
sctx->framebuffer.dirty_zsbuf |= sctx->framebuffer.state.zsbuf != state->zsbuf;
+ si_dec_framebuffer_counters(&sctx->framebuffer.state);
util_copy_framebuffer_state(&sctx->framebuffer.state, state);
- sctx->framebuffer.export_16bpc = 0;
+ sctx->framebuffer.spi_shader_col_format = 0;
+ sctx->framebuffer.spi_shader_col_format_alpha = 0;
+ sctx->framebuffer.spi_shader_col_format_blend = 0;
+ sctx->framebuffer.spi_shader_col_format_blend_alpha = 0;
+ sctx->framebuffer.color_is_int8 = 0;
+
sctx->framebuffer.compressed_cb_mask = 0;
sctx->framebuffer.nr_samples = util_framebuffer_get_num_samples(state);
sctx->framebuffer.log_samples = util_logbase2(sctx->framebuffer.nr_samples);
sctx->framebuffer.cb0_is_integer = state->nr_cbufs && state->cbufs[0] &&
util_format_is_pure_integer(state->cbufs[0]->format);
+ sctx->framebuffer.any_dst_linear = false;
if (sctx->framebuffer.cb0_is_integer != old_cb0_is_integer)
si_mark_atom_dirty(sctx, &sctx->db_render_state);
si_initialize_color_surface(sctx, surf);
}
- if (surf->export_16bpc) {
- sctx->framebuffer.export_16bpc |= 1 << i;
- }
+ sctx->framebuffer.spi_shader_col_format |=
+ surf->spi_shader_col_format << (i * 4);
+ sctx->framebuffer.spi_shader_col_format_alpha |=
+ surf->spi_shader_col_format_alpha << (i * 4);
+ sctx->framebuffer.spi_shader_col_format_blend |=
+ surf->spi_shader_col_format_blend << (i * 4);
+ sctx->framebuffer.spi_shader_col_format_blend_alpha |=
+ surf->spi_shader_col_format_blend_alpha << (i * 4);
+
+ if (surf->color_is_int8)
+ sctx->framebuffer.color_is_int8 |= 1 << i;
if (rtex->fmask.size && rtex->cmask.size) {
sctx->framebuffer.compressed_cb_mask |= 1 << i;
}
+
+ if (surf->level_info->mode == RADEON_SURF_MODE_LINEAR_ALIGNED)
+ sctx->framebuffer.any_dst_linear = true;
+
r600_context_add_resource_size(ctx, surf->base.texture);
+
+ p_atomic_inc(&rtex->framebuffers_bound);
}
- /* Set the 16BPC export for possible dual-src blending. */
- if (i == 1 && surf && surf->export_16bpc) {
- sctx->framebuffer.export_16bpc |= 1 << 1;
+ /* Set the second SPI format for possible dual-src blending. */
+ if (i == 1 && surf) {
+ sctx->framebuffer.spi_shader_col_format |=
+ surf->spi_shader_col_format << (i * 4);
+ sctx->framebuffer.spi_shader_col_format_alpha |=
+ surf->spi_shader_col_format_alpha << (i * 4);
+ sctx->framebuffer.spi_shader_col_format_blend |=
+ surf->spi_shader_col_format_blend << (i * 4);
+ sctx->framebuffer.spi_shader_col_format_blend_alpha |=
+ surf->spi_shader_col_format_blend_alpha << (i * 4);
}
- assert(!(sctx->framebuffer.export_16bpc & ~0xff));
-
if (state->zsbuf) {
surf = (struct r600_surface*)state->zsbuf;
}
si_update_poly_offset_state(sctx);
- si_mark_atom_dirty(sctx, &sctx->cb_target_mask);
+ si_mark_atom_dirty(sctx, &sctx->cb_render_state);
si_mark_atom_dirty(sctx, &sctx->framebuffer.atom);
+ if (sctx->framebuffer.any_dst_linear != old_any_dst_linear)
+ si_mark_atom_dirty(sctx, &sctx->msaa_config);
+
if (sctx->framebuffer.nr_samples != old_nr_samples) {
si_mark_atom_dirty(sctx, &sctx->msaa_config);
si_mark_atom_dirty(sctx, &sctx->db_render_state);
constbuf.user_buffer = sctx->b.sample_locations_16x;
break;
default:
+ R600_ERR("Requested an invalid number of samples %i.\n",
+ sctx->framebuffer.nr_samples);
assert(0);
}
constbuf.buffer_size = sctx->framebuffer.nr_samples * 2 * 4;
- ctx->set_constant_buffer(ctx, PIPE_SHADER_FRAGMENT,
- SI_DRIVER_STATE_CONST_BUF, &constbuf);
+ si_set_rw_buffer(sctx, SI_PS_CONST_SAMPLE_POSITIONS, &constbuf);
/* Smoothing (only possible with nr_samples == 1) uses the same
* sample locations as the MSAA it simulates.
old_nr_samples != 1))
si_mark_atom_dirty(sctx, &sctx->msaa_sample_locs);
}
+
+ sctx->need_check_render_feedback = true;
}
static void si_emit_framebuffer_state(struct si_context *sctx, struct r600_atom *atom)
unsigned i, nr_cbufs = state->nr_cbufs;
struct r600_texture *tex = NULL;
struct r600_surface *cb = NULL;
- uint32_t sx_ps_downconvert = 0;
- uint32_t sx_blend_opt_epsilon = 0;
+ unsigned cb_color_info = 0;
/* Colorbuffers. */
for (i = 0; i < nr_cbufs; i++) {
+ unsigned pitch_tile_max, slice_tile_max, tile_mode_index;
+ unsigned cb_color_base, cb_color_fmask, cb_color_attrib;
+ unsigned cb_color_pitch, cb_color_slice, cb_color_fmask_slice;
+
if (!(sctx->framebuffer.dirty_cbufs & (1 << i)))
continue;
RADEON_PRIO_CMASK);
}
- if (tex->dcc_buffer && tex->dcc_buffer != &tex->resource) {
- radeon_add_to_buffer_list(&sctx->b, &sctx->b.gfx,
- tex->dcc_buffer, RADEON_USAGE_READWRITE,
- RADEON_PRIO_DCC);
+ /* Compute mutable surface parameters. */
+ pitch_tile_max = cb->level_info->nblk_x / 8 - 1;
+ slice_tile_max = cb->level_info->nblk_x *
+ cb->level_info->nblk_y / 64 - 1;
+ tile_mode_index = si_tile_mode_index(tex, cb->base.u.tex.level, false);
+
+ cb_color_base = (tex->resource.gpu_address + cb->level_info->offset) >> 8;
+ cb_color_pitch = S_028C64_TILE_MAX(pitch_tile_max);
+ cb_color_slice = S_028C68_TILE_MAX(slice_tile_max);
+ cb_color_attrib = cb->cb_color_attrib |
+ S_028C74_TILE_MODE_INDEX(tile_mode_index);
+
+ if (tex->fmask.size) {
+ if (sctx->b.chip_class >= CIK)
+ cb_color_pitch |= S_028C64_FMASK_TILE_MAX(tex->fmask.pitch_in_pixels / 8 - 1);
+ cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(tex->fmask.tile_mode_index);
+ cb_color_fmask = (tex->resource.gpu_address + tex->fmask.offset) >> 8;
+ cb_color_fmask_slice = S_028C88_TILE_MAX(tex->fmask.slice_tile_max);
+ } else {
+ /* This must be set for fast clear to work without FMASK. */
+ if (sctx->b.chip_class >= CIK)
+ cb_color_pitch |= S_028C64_FMASK_TILE_MAX(pitch_tile_max);
+ cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index);
+ cb_color_fmask = cb_color_base;
+ cb_color_fmask_slice = S_028C88_TILE_MAX(slice_tile_max);
+ }
+
+ cb_color_info = cb->cb_color_info | tex->cb_color_info;
+
+ if (tex->dcc_offset && cb->level_info->dcc_enabled) {
+ bool is_msaa_resolve_dst = state->cbufs[0] &&
+ state->cbufs[0]->texture->nr_samples > 1 &&
+ state->cbufs[1] == &cb->base &&
+ state->cbufs[1]->texture->nr_samples <= 1;
+
+ if (!is_msaa_resolve_dst)
+ cb_color_info |= S_028C70_DCC_ENABLE(1);
}
radeon_set_context_reg_seq(cs, R_028C60_CB_COLOR0_BASE + i * 0x3C,
sctx->b.chip_class >= VI ? 14 : 13);
- radeon_emit(cs, cb->cb_color_base); /* R_028C60_CB_COLOR0_BASE */
- radeon_emit(cs, cb->cb_color_pitch); /* R_028C64_CB_COLOR0_PITCH */
- radeon_emit(cs, cb->cb_color_slice); /* R_028C68_CB_COLOR0_SLICE */
+ radeon_emit(cs, cb_color_base); /* R_028C60_CB_COLOR0_BASE */
+ radeon_emit(cs, cb_color_pitch); /* R_028C64_CB_COLOR0_PITCH */
+ radeon_emit(cs, cb_color_slice); /* R_028C68_CB_COLOR0_SLICE */
radeon_emit(cs, cb->cb_color_view); /* R_028C6C_CB_COLOR0_VIEW */
- radeon_emit(cs, cb->cb_color_info | tex->cb_color_info); /* R_028C70_CB_COLOR0_INFO */
- radeon_emit(cs, cb->cb_color_attrib); /* R_028C74_CB_COLOR0_ATTRIB */
+ radeon_emit(cs, cb_color_info); /* R_028C70_CB_COLOR0_INFO */
+ radeon_emit(cs, cb_color_attrib); /* R_028C74_CB_COLOR0_ATTRIB */
radeon_emit(cs, cb->cb_dcc_control); /* R_028C78_CB_COLOR0_DCC_CONTROL */
radeon_emit(cs, tex->cmask.base_address_reg); /* R_028C7C_CB_COLOR0_CMASK */
radeon_emit(cs, tex->cmask.slice_tile_max); /* R_028C80_CB_COLOR0_CMASK_SLICE */
- radeon_emit(cs, cb->cb_color_fmask); /* R_028C84_CB_COLOR0_FMASK */
- radeon_emit(cs, cb->cb_color_fmask_slice); /* R_028C88_CB_COLOR0_FMASK_SLICE */
+ radeon_emit(cs, cb_color_fmask); /* R_028C84_CB_COLOR0_FMASK */
+ radeon_emit(cs, cb_color_fmask_slice); /* R_028C88_CB_COLOR0_FMASK_SLICE */
radeon_emit(cs, tex->color_clear_value[0]); /* R_028C8C_CB_COLOR0_CLEAR_WORD0 */
radeon_emit(cs, tex->color_clear_value[1]); /* R_028C90_CB_COLOR0_CLEAR_WORD1 */
- if (sctx->b.chip_class >= VI)
- radeon_emit(cs, cb->cb_dcc_base); /* R_028C94_CB_COLOR0_DCC_BASE */
-
- sx_ps_downconvert |= cb->sx_ps_downconvert << (4 * i);
- sx_blend_opt_epsilon |= cb->sx_blend_opt_epsilon << (4 * i);
+ if (sctx->b.chip_class >= VI) /* R_028C94_CB_COLOR0_DCC_BASE */
+ radeon_emit(cs, (tex->resource.gpu_address +
+ tex->dcc_offset +
+ tex->surface.level[cb->base.u.tex.level].dcc_offset) >> 8);
}
/* set CB_COLOR1_INFO for possible dual-src blending */
if (i == 1 && state->cbufs[0] &&
sctx->framebuffer.dirty_cbufs & (1 << 0)) {
radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + 1 * 0x3C,
- cb->cb_color_info | tex->cb_color_info);
- sx_ps_downconvert |= cb->sx_ps_downconvert << (4 * i);
- sx_blend_opt_epsilon |= cb->sx_blend_opt_epsilon << (4 * i);
+ cb_color_info);
i++;
}
for (; i < 8 ; i++)
if (sctx->framebuffer.dirty_cbufs & (1 << i))
radeon_set_context_reg(cs, R_028C70_CB_COLOR0_INFO + i * 0x3C, 0);
- if (sctx->b.family == CHIP_STONEY) {
- radeon_set_context_reg_seq(cs, R_028754_SX_PS_DOWNCONVERT, 2);
- radeon_emit(cs, sx_ps_downconvert); /* R_028754_SX_PS_DOWNCONVERT */
- radeon_emit(cs, sx_blend_opt_epsilon); /* R_028758_SX_BLEND_OPT_EPSILON */
- }
-
/* ZS buffer. */
if (state->zsbuf && sctx->framebuffer.dirty_zsbuf) {
struct r600_surface *zb = (struct r600_surface*)state->zsbuf;
static void si_emit_msaa_config(struct si_context *sctx, struct r600_atom *atom)
{
struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
+ unsigned num_tile_pipes = sctx->screen->b.info.num_tile_pipes;
+ /* 33% faster rendering to linear color buffers */
+ bool dst_is_linear = sctx->framebuffer.any_dst_linear;
+ unsigned sc_mode_cntl_1 =
+ S_028A4C_WALK_SIZE(dst_is_linear) |
+ S_028A4C_WALK_FENCE_ENABLE(!dst_is_linear) |
+ S_028A4C_WALK_FENCE_SIZE(num_tile_pipes == 2 ? 2 : 3) |
+ /* always 1: */
+ S_028A4C_WALK_ALIGN8_PRIM_FITS_ST(1) |
+ S_028A4C_SUPERTILE_WALK_ORDER_ENABLE(1) |
+ S_028A4C_TILE_WALK_ORDER_ENABLE(1) |
+ S_028A4C_MULTI_SHADER_ENGINE_PRIM_DISCARD_ENABLE(1) |
+ S_028A4C_FORCE_EOV_CNTDWN_ENABLE(1) |
+ S_028A4C_FORCE_EOV_REZ_ENABLE(1);
cayman_emit_msaa_config(cs, sctx->framebuffer.nr_samples,
sctx->ps_iter_samples,
- sctx->smoothing_enabled ? SI_NUM_SMOOTH_AA_SAMPLES : 0);
+ sctx->smoothing_enabled ? SI_NUM_SMOOTH_AA_SAMPLES : 0,
+ sc_mode_cntl_1);
}
-
static void si_set_min_samples(struct pipe_context *ctx, unsigned min_samples)
{
struct si_context *sctx = (struct si_context *)ctx;
*/
/**
- * Create a sampler view.
- *
- * @param ctx context
- * @param texture texture
- * @param state sampler view template
- * @param width0 width0 override (for compressed textures as int)
- * @param height0 height0 override (for compressed textures as int)
- * @param force_level set the base address to the level (for compressed textures)
+ * Build the sampler view descriptor for a buffer texture.
+ * @param state 256-bit descriptor; only the high 128 bits are filled in
*/
-struct pipe_sampler_view *
-si_create_sampler_view_custom(struct pipe_context *ctx,
- struct pipe_resource *texture,
- const struct pipe_sampler_view *state,
- unsigned width0, unsigned height0,
- unsigned force_level)
+void
+si_make_buffer_descriptor(struct si_screen *screen, struct r600_resource *buf,
+ enum pipe_format format,
+ unsigned first_element, unsigned last_element,
+ uint32_t *state)
{
- struct si_context *sctx = (struct si_context*)ctx;
- struct si_sampler_view *view = CALLOC_STRUCT(si_sampler_view);
- struct r600_texture *tmp = (struct r600_texture*)texture;
const struct util_format_description *desc;
- unsigned format, num_format, base_level, first_level, last_level;
- uint32_t pitch = 0;
- unsigned char state_swizzle[4], swizzle[4];
- unsigned height, depth, width;
- enum pipe_format pipe_format = state->format;
- struct radeon_surf_level *surflevel;
int first_non_void;
uint64_t va;
- unsigned last_layer = state->u.tex.last_layer;
-
- if (!view)
- return NULL;
+ unsigned stride;
+ unsigned num_records;
+ unsigned num_format, data_format;
- /* initialize base object */
- view->base = *state;
- view->base.texture = NULL;
- view->base.reference.count = 1;
- view->base.context = ctx;
-
- /* NULL resource, obey swizzle (only ZERO and ONE make sense). */
- if (!texture) {
- view->state[3] = S_008F1C_DST_SEL_X(si_map_swizzle(state->swizzle_r)) |
- S_008F1C_DST_SEL_Y(si_map_swizzle(state->swizzle_g)) |
- S_008F1C_DST_SEL_Z(si_map_swizzle(state->swizzle_b)) |
- S_008F1C_DST_SEL_W(si_map_swizzle(state->swizzle_a)) |
- S_008F1C_TYPE(V_008F1C_SQ_RSRC_IMG_1D);
- return &view->base;
- }
-
- 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;
-
- desc = util_format_description(state->format);
- first_non_void = util_format_get_first_non_void_channel(state->format);
- stride = desc->block.bits / 8;
- va = tmp->resource.gpu_address + state->u.buf.first_element*stride;
- format = si_translate_buffer_dataformat(ctx->screen, desc, first_non_void);
- num_format = si_translate_buffer_numformat(ctx->screen, desc, first_non_void);
-
- num_records = state->u.buf.last_element + 1 - state->u.buf.first_element;
- num_records = MIN2(num_records, texture->width0 / stride);
-
- if (sctx->b.chip_class >= VI)
- num_records *= stride;
-
- view->state[4] = va;
- view->state[5] = S_008F04_BASE_ADDRESS_HI(va >> 32) |
- S_008F04_STRIDE(stride);
- view->state[6] = num_records;
- view->state[7] = S_008F0C_DST_SEL_X(si_map_swizzle(desc->swizzle[0])) |
- S_008F0C_DST_SEL_Y(si_map_swizzle(desc->swizzle[1])) |
- S_008F0C_DST_SEL_Z(si_map_swizzle(desc->swizzle[2])) |
- S_008F0C_DST_SEL_W(si_map_swizzle(desc->swizzle[3])) |
- S_008F0C_NUM_FORMAT(num_format) |
- S_008F0C_DATA_FORMAT(format);
+ desc = util_format_description(format);
+ first_non_void = util_format_get_first_non_void_channel(format);
+ stride = desc->block.bits / 8;
+ va = buf->gpu_address + first_element * stride;
+ num_format = si_translate_buffer_numformat(&screen->b.b, desc, first_non_void);
+ data_format = si_translate_buffer_dataformat(&screen->b.b, desc, first_non_void);
- LIST_ADDTAIL(&view->list, &sctx->b.texture_buffers);
- return &view->base;
- }
+ num_records = last_element + 1 - first_element;
+ num_records = MIN2(num_records, buf->b.b.width0 / stride);
- state_swizzle[0] = state->swizzle_r;
- state_swizzle[1] = state->swizzle_g;
- state_swizzle[2] = state->swizzle_b;
- state_swizzle[3] = state->swizzle_a;
+ if (screen->b.chip_class >= VI)
+ num_records *= stride;
- surflevel = tmp->surface.level;
+ state[4] = va;
+ state[5] = S_008F04_BASE_ADDRESS_HI(va >> 32) |
+ S_008F04_STRIDE(stride);
+ state[6] = num_records;
+ state[7] = S_008F0C_DST_SEL_X(si_map_swizzle(desc->swizzle[0])) |
+ S_008F0C_DST_SEL_Y(si_map_swizzle(desc->swizzle[1])) |
+ S_008F0C_DST_SEL_Z(si_map_swizzle(desc->swizzle[2])) |
+ S_008F0C_DST_SEL_W(si_map_swizzle(desc->swizzle[3])) |
+ S_008F0C_NUM_FORMAT(num_format) |
+ S_008F0C_DATA_FORMAT(data_format);
+}
- /* Texturing with separate depth and stencil. */
- if (tmp->is_depth && !tmp->is_flushing_texture) {
- switch (pipe_format) {
- case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
- pipe_format = PIPE_FORMAT_Z32_FLOAT;
- break;
- case PIPE_FORMAT_X8Z24_UNORM:
- case PIPE_FORMAT_S8_UINT_Z24_UNORM:
- /* Z24 is always stored like this. */
- pipe_format = PIPE_FORMAT_Z24X8_UNORM;
- break;
- case PIPE_FORMAT_X24S8_UINT:
- case PIPE_FORMAT_S8X24_UINT:
- case PIPE_FORMAT_X32_S8X24_UINT:
- pipe_format = PIPE_FORMAT_S8_UINT;
- surflevel = tmp->surface.stencil_level;
- break;
- default:;
- }
- }
+/**
+ * Build the sampler view descriptor for a texture.
+ */
+void
+si_make_texture_descriptor(struct si_screen *screen,
+ struct r600_texture *tex,
+ bool sampler,
+ enum pipe_texture_target target,
+ enum pipe_format pipe_format,
+ const unsigned char state_swizzle[4],
+ unsigned first_level, unsigned last_level,
+ unsigned first_layer, unsigned last_layer,
+ unsigned width, unsigned height, unsigned depth,
+ uint32_t *state,
+ uint32_t *fmask_state)
+{
+ struct pipe_resource *res = &tex->resource.b.b;
+ const struct util_format_description *desc;
+ unsigned char swizzle[4];
+ int first_non_void;
+ unsigned num_format, data_format, type;
+ uint64_t va;
desc = util_format_description(pipe_format);
case PIPE_FORMAT_DXT3_SRGBA:
case PIPE_FORMAT_DXT5_SRGBA:
case PIPE_FORMAT_BPTC_SRGBA:
+ case PIPE_FORMAT_ETC2_SRGB8:
+ case PIPE_FORMAT_ETC2_SRGB8A1:
+ case PIPE_FORMAT_ETC2_SRGBA8:
num_format = V_008F14_IMG_NUM_FORMAT_SRGB;
break;
case PIPE_FORMAT_RGTC1_SNORM:
case PIPE_FORMAT_LATC1_SNORM:
case PIPE_FORMAT_RGTC2_SNORM:
case PIPE_FORMAT_LATC2_SNORM:
+ case PIPE_FORMAT_ETC2_R11_SNORM:
+ case PIPE_FORMAT_ETC2_RG11_SNORM:
/* implies float, so use SNORM/UNORM to determine
whether data is signed or not */
case PIPE_FORMAT_BPTC_RGB_FLOAT:
}
}
- format = si_translate_texformat(ctx->screen, pipe_format, desc, first_non_void);
- if (format == ~0) {
- format = 0;
+ data_format = si_translate_texformat(&screen->b.b, pipe_format, desc, first_non_void);
+ if (data_format == ~0) {
+ data_format = 0;
}
+ if (!sampler &&
+ (res->target == PIPE_TEXTURE_CUBE ||
+ res->target == PIPE_TEXTURE_CUBE_ARRAY ||
+ res->target == PIPE_TEXTURE_3D)) {
+ /* For the purpose of shader images, treat cube maps and 3D
+ * textures as 2D arrays. For 3D textures, the address
+ * calculations for mipmaps are different, so we rely on the
+ * caller to effectively disable mipmaps.
+ */
+ type = V_008F1C_SQ_RSRC_IMG_2D_ARRAY;
+
+ assert(res->target != PIPE_TEXTURE_3D || (first_level == 0 && last_level == 0));
+ } else {
+ type = si_tex_dim(res->target, target, res->nr_samples);
+ }
+
+ if (type == V_008F1C_SQ_RSRC_IMG_1D_ARRAY) {
+ height = 1;
+ depth = res->array_size;
+ } else if (type == V_008F1C_SQ_RSRC_IMG_2D_ARRAY ||
+ type == V_008F1C_SQ_RSRC_IMG_2D_MSAA_ARRAY) {
+ if (sampler || res->target != PIPE_TEXTURE_3D)
+ depth = res->array_size;
+ } else if (type == V_008F1C_SQ_RSRC_IMG_CUBE)
+ depth = res->array_size / 6;
+
+ state[0] = 0;
+ state[1] = (S_008F14_DATA_FORMAT(data_format) |
+ S_008F14_NUM_FORMAT(num_format));
+ state[2] = (S_008F18_WIDTH(width - 1) |
+ S_008F18_HEIGHT(height - 1));
+ state[3] = (S_008F1C_DST_SEL_X(si_map_swizzle(swizzle[0])) |
+ S_008F1C_DST_SEL_Y(si_map_swizzle(swizzle[1])) |
+ S_008F1C_DST_SEL_Z(si_map_swizzle(swizzle[2])) |
+ S_008F1C_DST_SEL_W(si_map_swizzle(swizzle[3])) |
+ S_008F1C_BASE_LEVEL(res->nr_samples > 1 ?
+ 0 : first_level) |
+ S_008F1C_LAST_LEVEL(res->nr_samples > 1 ?
+ util_logbase2(res->nr_samples) :
+ last_level) |
+ S_008F1C_POW2_PAD(res->last_level > 0) |
+ S_008F1C_TYPE(type));
+ state[4] = S_008F20_DEPTH(depth - 1);
+ state[5] = (S_008F24_BASE_ARRAY(first_layer) |
+ S_008F24_LAST_ARRAY(last_layer));
+ state[6] = 0;
+ state[7] = 0;
+
+ if (tex->dcc_offset) {
+ unsigned swap = r600_translate_colorswap(pipe_format, FALSE);
+
+ state[6] = S_008F28_ALPHA_IS_ON_MSB(swap <= 1);
+ } else {
+ /* The last dword is unused by hw. The shader uses it to clear
+ * bits in the first dword of sampler state.
+ */
+ if (screen->b.chip_class <= CIK && res->nr_samples <= 1) {
+ if (first_level == last_level)
+ state[7] = C_008F30_MAX_ANISO_RATIO;
+ else
+ state[7] = 0xffffffff;
+ }
+ }
+
+ /* Initialize the sampler view for FMASK. */
+ if (tex->fmask.size) {
+ uint32_t fmask_format;
+
+ va = tex->resource.gpu_address + tex->fmask.offset;
+
+ switch (res->nr_samples) {
+ case 2:
+ fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK8_S2_F2;
+ break;
+ case 4:
+ fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK8_S4_F4;
+ break;
+ case 8:
+ fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK32_S8_F8;
+ break;
+ default:
+ assert(0);
+ fmask_format = V_008F14_IMG_DATA_FORMAT_INVALID;
+ }
+
+ fmask_state[0] = va >> 8;
+ fmask_state[1] = S_008F14_BASE_ADDRESS_HI(va >> 40) |
+ S_008F14_DATA_FORMAT(fmask_format) |
+ S_008F14_NUM_FORMAT(V_008F14_IMG_NUM_FORMAT_UINT);
+ fmask_state[2] = S_008F18_WIDTH(width - 1) |
+ S_008F18_HEIGHT(height - 1);
+ fmask_state[3] = S_008F1C_DST_SEL_X(V_008F1C_SQ_SEL_X) |
+ S_008F1C_DST_SEL_Y(V_008F1C_SQ_SEL_X) |
+ S_008F1C_DST_SEL_Z(V_008F1C_SQ_SEL_X) |
+ S_008F1C_DST_SEL_W(V_008F1C_SQ_SEL_X) |
+ S_008F1C_TILING_INDEX(tex->fmask.tile_mode_index) |
+ S_008F1C_TYPE(si_tex_dim(res->target, target, 0));
+ fmask_state[4] = S_008F20_DEPTH(depth - 1) |
+ S_008F20_PITCH(tex->fmask.pitch_in_pixels - 1);
+ fmask_state[5] = S_008F24_BASE_ARRAY(first_layer) |
+ S_008F24_LAST_ARRAY(last_layer);
+ fmask_state[6] = 0;
+ fmask_state[7] = 0;
+ }
+}
+
+/**
+ * Create a sampler view.
+ *
+ * @param ctx context
+ * @param texture texture
+ * @param state sampler view template
+ * @param width0 width0 override (for compressed textures as int)
+ * @param height0 height0 override (for compressed textures as int)
+ * @param force_level set the base address to the level (for compressed textures)
+ */
+struct pipe_sampler_view *
+si_create_sampler_view_custom(struct pipe_context *ctx,
+ struct pipe_resource *texture,
+ const struct pipe_sampler_view *state,
+ unsigned width0, unsigned height0,
+ unsigned force_level)
+{
+ struct si_context *sctx = (struct si_context*)ctx;
+ struct si_sampler_view *view = CALLOC_STRUCT(si_sampler_view);
+ struct r600_texture *tmp = (struct r600_texture*)texture;
+ unsigned base_level, first_level, last_level;
+ unsigned char state_swizzle[4];
+ unsigned height, depth, width;
+ unsigned last_layer = state->u.tex.last_layer;
+ enum pipe_format pipe_format;
+ const struct radeon_surf_level *surflevel;
+
+ if (!view)
+ return NULL;
+
+ /* initialize base object */
+ view->base = *state;
+ view->base.texture = NULL;
+ view->base.reference.count = 1;
+ view->base.context = ctx;
+
+ /* NULL resource, obey swizzle (only ZERO and ONE make sense). */
+ if (!texture) {
+ view->state[3] = S_008F1C_DST_SEL_X(si_map_swizzle(state->swizzle_r)) |
+ S_008F1C_DST_SEL_Y(si_map_swizzle(state->swizzle_g)) |
+ S_008F1C_DST_SEL_Z(si_map_swizzle(state->swizzle_b)) |
+ S_008F1C_DST_SEL_W(si_map_swizzle(state->swizzle_a)) |
+ S_008F1C_TYPE(V_008F1C_SQ_RSRC_IMG_1D);
+ return &view->base;
+ }
+
+ pipe_resource_reference(&view->base.texture, texture);
+
+ 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) {
+ si_make_buffer_descriptor(sctx->screen,
+ (struct r600_resource *)texture,
+ state->format,
+ state->u.buf.first_element,
+ state->u.buf.last_element,
+ view->state);
+
+ LIST_ADDTAIL(&view->list, &sctx->b.texture_buffers);
+ return &view->base;
+ }
+
+ state_swizzle[0] = state->swizzle_r;
+ state_swizzle[1] = state->swizzle_g;
+ state_swizzle[2] = state->swizzle_b;
+ state_swizzle[3] = state->swizzle_a;
+
base_level = 0;
first_level = state->u.tex.first_level;
last_level = state->u.tex.last_level;
depth = u_minify(depth, force_level);
}
- pitch = surflevel[base_level].nblk_x * util_format_get_blockwidth(pipe_format);
-
- if (texture->target == PIPE_TEXTURE_1D_ARRAY) {
- height = 1;
- depth = texture->array_size;
- } else if (texture->target == PIPE_TEXTURE_2D_ARRAY) {
- depth = texture->array_size;
- } 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_CUBE)
last_layer = state->u.tex.first_layer;
- va = tmp->resource.gpu_address + surflevel[base_level].offset;
-
- view->state[0] = va >> 8;
- view->state[1] = (S_008F14_BASE_ADDRESS_HI(va >> 40) |
- S_008F14_DATA_FORMAT(format) |
- S_008F14_NUM_FORMAT(num_format));
- view->state[2] = (S_008F18_WIDTH(width - 1) |
- S_008F18_HEIGHT(height - 1));
- view->state[3] = (S_008F1C_DST_SEL_X(si_map_swizzle(swizzle[0])) |
- S_008F1C_DST_SEL_Y(si_map_swizzle(swizzle[1])) |
- S_008F1C_DST_SEL_Z(si_map_swizzle(swizzle[2])) |
- S_008F1C_DST_SEL_W(si_map_swizzle(swizzle[3])) |
- S_008F1C_BASE_LEVEL(texture->nr_samples > 1 ?
- 0 : first_level) |
- S_008F1C_LAST_LEVEL(texture->nr_samples > 1 ?
- util_logbase2(texture->nr_samples) :
- 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, 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(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) {
- uint64_t va = tmp->resource.gpu_address + tmp->fmask.offset;
- uint32_t fmask_format;
+ /* Texturing with separate depth and stencil. */
+ pipe_format = state->format;
+ surflevel = tmp->surface.level;
- switch (texture->nr_samples) {
- case 2:
- fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK8_S2_F2;
+ if (tmp->is_depth && !tmp->is_flushing_texture) {
+ switch (pipe_format) {
+ case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
+ pipe_format = PIPE_FORMAT_Z32_FLOAT;
break;
- case 4:
- fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK8_S4_F4;
+ case PIPE_FORMAT_X8Z24_UNORM:
+ case PIPE_FORMAT_S8_UINT_Z24_UNORM:
+ /* Z24 is always stored like this. */
+ pipe_format = PIPE_FORMAT_Z24X8_UNORM;
break;
- case 8:
- fmask_format = V_008F14_IMG_DATA_FORMAT_FMASK32_S8_F8;
+ case PIPE_FORMAT_X24S8_UINT:
+ case PIPE_FORMAT_S8X24_UINT:
+ case PIPE_FORMAT_X32_S8X24_UINT:
+ pipe_format = PIPE_FORMAT_S8_UINT;
+ surflevel = tmp->surface.stencil_level;
break;
- default:
- assert(0);
- fmask_format = V_008F14_IMG_DATA_FORMAT_INVALID;
+ default:;
}
-
- view->fmask_state[0] = va >> 8;
- view->fmask_state[1] = S_008F14_BASE_ADDRESS_HI(va >> 40) |
- S_008F14_DATA_FORMAT(fmask_format) |
- S_008F14_NUM_FORMAT(V_008F14_IMG_NUM_FORMAT_UINT);
- view->fmask_state[2] = S_008F18_WIDTH(width - 1) |
- S_008F18_HEIGHT(height - 1);
- view->fmask_state[3] = S_008F1C_DST_SEL_X(V_008F1C_SQ_SEL_X) |
- S_008F1C_DST_SEL_Y(V_008F1C_SQ_SEL_X) |
- 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,
- state->target, 0));
- view->fmask_state[4] = S_008F20_DEPTH(depth - 1) |
- S_008F20_PITCH(tmp->fmask.pitch_in_pixels - 1);
- view->fmask_state[5] = S_008F24_BASE_ARRAY(state->u.tex.first_layer) |
- S_008F24_LAST_ARRAY(last_layer);
- view->fmask_state[6] = 0;
- view->fmask_state[7] = 0;
}
+ si_make_texture_descriptor(sctx->screen, tmp, true,
+ state->target, pipe_format, state_swizzle,
+ first_level, last_level,
+ state->u.tex.first_layer, last_layer,
+ width, height, depth,
+ view->state, view->fmask_state);
+
+ view->base_level_info = &surflevel[base_level];
+ view->base_level = base_level;
+ view->block_width = util_format_get_blockwidth(pipe_format);
return &view->base;
}
{
struct si_sampler_view *view = (struct si_sampler_view *)state;
- if (view->resource && view->resource->b.b.target == PIPE_BUFFER)
+ if (state->texture && state->texture->target == PIPE_BUFFER)
LIST_DELINIT(&view->list);
pipe_resource_reference(&state->texture, NULL);
const struct pipe_sampler_state *state)
{
struct si_context *sctx = (struct si_context *)ctx;
+ struct r600_common_screen *rscreen = sctx->b.screen;
struct si_sampler_state *rstate = CALLOC_STRUCT(si_sampler_state);
- unsigned aniso_flag_offset = state->max_anisotropy > 1 ? 2 : 0;
unsigned border_color_type, border_color_index = 0;
+ unsigned max_aniso = rscreen->force_aniso >= 0 ? rscreen->force_aniso
+ : state->max_anisotropy;
+ unsigned max_aniso_ratio = r600_tex_aniso_filter(max_aniso);
if (!rstate) {
return NULL;
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)) |
- r600_tex_aniso_filter(state->max_anisotropy) << 9 |
+ S_008F30_MAX_ANISO_RATIO(max_aniso_ratio) |
S_008F30_DEPTH_COMPARE_FUNC(si_tex_compare(state->compare_func)) |
S_008F30_FORCE_UNNORMALIZED(!state->normalized_coords) |
- S_008F30_DISABLE_CUBE_WRAP(!state->seamless_cube_map));
+ S_008F30_DISABLE_CUBE_WRAP(!state->seamless_cube_map) |
+ S_008F30_COMPAT_MODE(sctx->b.chip_class >= VI));
rstate->val[1] = (S_008F34_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 8)) |
S_008F34_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 8)));
rstate->val[2] = (S_008F38_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 8)) |
- S_008F38_XY_MAG_FILTER(si_tex_filter(state->mag_img_filter) | aniso_flag_offset) |
- S_008F38_XY_MIN_FILTER(si_tex_filter(state->min_img_filter) | aniso_flag_offset) |
- S_008F38_MIP_FILTER(si_tex_mipfilter(state->min_mip_filter)));
+ S_008F38_XY_MAG_FILTER(eg_tex_filter(state->mag_img_filter, max_aniso)) |
+ S_008F38_XY_MIN_FILTER(eg_tex_filter(state->min_img_filter, max_aniso)) |
+ S_008F38_MIP_FILTER(si_tex_mipfilter(state->min_mip_filter)) |
+ S_008F38_MIP_POINT_PRECLAMP(1) |
+ S_008F38_DISABLE_LSB_CEIL(1) |
+ S_008F38_FILTER_PREC_FIX(1) |
+ S_008F38_ANISO_OVERRIDE(sctx->b.chip_class >= VI));
rstate->val[3] = S_008F3C_BORDER_COLOR_PTR(border_color_index) |
S_008F3C_BORDER_COLOR_TYPE(border_color_type);
return rstate;
struct si_vertex_element *v = CALLOC_STRUCT(si_vertex_element);
int i;
- assert(count < SI_MAX_ATTRIBS);
+ assert(count <= SI_MAX_ATTRIBS);
if (!v)
return NULL;
struct pipe_vertex_buffer *dst = sctx->vertex_buffer + start_slot;
int i;
- assert(start_slot + count <= Elements(sctx->vertex_buffer));
+ assert(start_slot + count <= ARRAY_SIZE(sctx->vertex_buffer));
if (buffers) {
for (i = 0; i < count; i++) {
/*
* Misc
*/
-static void si_set_polygon_stipple(struct pipe_context *ctx,
- const struct pipe_poly_stipple *state)
-{
- struct si_context *sctx = (struct si_context *)ctx;
- struct pipe_resource *tex;
- struct pipe_sampler_view *view;
- bool is_zero = true;
- bool is_one = true;
- int i;
-
- /* The hardware obeys 0 and 1 swizzles in the descriptor even if
- * the resource is NULL/invalid. Take advantage of this fact and skip
- * texture allocation if the stipple pattern is constant.
- *
- * This is an optimization for the common case when stippling isn't
- * used but set_polygon_stipple is still called by st/mesa.
- */
- for (i = 0; i < Elements(state->stipple); i++) {
- is_zero = is_zero && state->stipple[i] == 0;
- is_one = is_one && state->stipple[i] == 0xffffffff;
- }
-
- if (is_zero || is_one) {
- struct pipe_sampler_view templ = {{0}};
-
- templ.swizzle_r = PIPE_SWIZZLE_ZERO;
- templ.swizzle_g = PIPE_SWIZZLE_ZERO;
- templ.swizzle_b = PIPE_SWIZZLE_ZERO;
- /* The pattern should be inverted in the texture. */
- templ.swizzle_a = is_zero ? PIPE_SWIZZLE_ONE : PIPE_SWIZZLE_ZERO;
-
- view = ctx->create_sampler_view(ctx, NULL, &templ);
- } else {
- /* Create a new texture. */
- tex = util_pstipple_create_stipple_texture(ctx, state->stipple);
- if (!tex)
- return;
-
- view = util_pstipple_create_sampler_view(ctx, tex);
- pipe_resource_reference(&tex, NULL);
- }
-
- ctx->set_sampler_views(ctx, PIPE_SHADER_FRAGMENT,
- SI_POLY_STIPPLE_SAMPLER, 1, &view);
- pipe_sampler_view_reference(&view, NULL);
-
- /* Bind the sampler state if needed. */
- if (!sctx->pstipple_sampler_state) {
- sctx->pstipple_sampler_state = util_pstipple_create_sampler(ctx);
- ctx->bind_sampler_states(ctx, PIPE_SHADER_FRAGMENT,
- SI_POLY_STIPPLE_SAMPLER, 1,
- &sctx->pstipple_sampler_state);
- }
-}
static void si_set_tess_state(struct pipe_context *ctx,
const float default_outer_level[4],
(void*)array, sizeof(array),
&cb.buffer_offset);
- ctx->set_constant_buffer(ctx, PIPE_SHADER_TESS_CTRL,
- SI_DRIVER_STATE_CONST_BUF, &cb);
+ si_set_rw_buffer(sctx, SI_HS_CONST_DEFAULT_TESS_LEVELS, &cb);
pipe_resource_reference(&cb.buffer, NULL);
}
sctx->b.flags |= SI_CONTEXT_INV_VMEM_L1 |
SI_CONTEXT_INV_GLOBAL_L2 |
- SI_CONTEXT_FLUSH_AND_INV_CB;
+ SI_CONTEXT_FLUSH_AND_INV_CB |
+ SI_CONTEXT_CS_PARTIAL_FLUSH;
+}
+
+static void si_memory_barrier(struct pipe_context *ctx, unsigned flags)
+{
+ struct si_context *sctx = (struct si_context *)ctx;
+
+ /* Subsequent commands must wait for all shader invocations to
+ * complete. */
+ sctx->b.flags |= SI_CONTEXT_PS_PARTIAL_FLUSH |
+ SI_CONTEXT_CS_PARTIAL_FLUSH;
+
+ if (flags & PIPE_BARRIER_CONSTANT_BUFFER)
+ sctx->b.flags |= SI_CONTEXT_INV_SMEM_L1 |
+ SI_CONTEXT_INV_VMEM_L1;
+
+ if (flags & (PIPE_BARRIER_VERTEX_BUFFER |
+ PIPE_BARRIER_SHADER_BUFFER |
+ PIPE_BARRIER_TEXTURE |
+ PIPE_BARRIER_IMAGE |
+ PIPE_BARRIER_STREAMOUT_BUFFER |
+ PIPE_BARRIER_GLOBAL_BUFFER)) {
+ /* As far as I can tell, L1 contents are written back to L2
+ * automatically at end of shader, but the contents of other
+ * L1 caches might still be stale. */
+ sctx->b.flags |= SI_CONTEXT_INV_VMEM_L1;
+ }
+
+ if (flags & PIPE_BARRIER_INDEX_BUFFER) {
+ sctx->b.flags |= SI_CONTEXT_INV_VMEM_L1;
+
+ /* Indices are read through TC L2 since VI. */
+ if (sctx->screen->b.chip_class <= CIK)
+ sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
+ }
+
+ if (flags & PIPE_BARRIER_FRAMEBUFFER)
+ sctx->b.flags |= SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER;
+
+ if (flags & (PIPE_BARRIER_MAPPED_BUFFER |
+ PIPE_BARRIER_FRAMEBUFFER |
+ PIPE_BARRIER_INDIRECT_BUFFER)) {
+ /* Not sure if INV_GLOBAL_L2 is the best thing here.
+ *
+ * We need to make sure that TC L1 & L2 are written back to
+ * memory, because neither CPU accesses nor CB fetches consider
+ * TC, but there's no need to invalidate any TC cache lines. */
+ sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
+ }
}
static void *si_create_blend_custom(struct si_context *sctx, unsigned mode)
si_init_external_atom(sctx, &sctx->b.render_cond_atom, &sctx->atoms.s.render_cond);
si_init_external_atom(sctx, &sctx->b.streamout.begin_atom, &sctx->atoms.s.streamout_begin);
si_init_external_atom(sctx, &sctx->b.streamout.enable_atom, &sctx->atoms.s.streamout_enable);
+ si_init_external_atom(sctx, &sctx->b.scissors.atom, &sctx->atoms.s.scissors);
+ si_init_external_atom(sctx, &sctx->b.viewports.atom, &sctx->atoms.s.viewports);
si_init_atom(sctx, &sctx->cache_flush, &sctx->atoms.s.cache_flush, si_emit_cache_flush);
si_init_atom(sctx, &sctx->framebuffer.atom, &sctx->atoms.s.framebuffer, si_emit_framebuffer_state);
si_init_atom(sctx, &sctx->db_render_state, &sctx->atoms.s.db_render_state, si_emit_db_render_state);
si_init_atom(sctx, &sctx->msaa_config, &sctx->atoms.s.msaa_config, si_emit_msaa_config);
si_init_atom(sctx, &sctx->sample_mask.atom, &sctx->atoms.s.sample_mask, si_emit_sample_mask);
- si_init_atom(sctx, &sctx->cb_target_mask, &sctx->atoms.s.cb_target_mask, si_emit_cb_target_mask);
+ si_init_atom(sctx, &sctx->cb_render_state, &sctx->atoms.s.cb_render_state, si_emit_cb_render_state);
si_init_atom(sctx, &sctx->blend_color.atom, &sctx->atoms.s.blend_color, si_emit_blend_color);
si_init_atom(sctx, &sctx->clip_regs, &sctx->atoms.s.clip_regs, si_emit_clip_regs);
si_init_atom(sctx, &sctx->clip_state.atom, &sctx->atoms.s.clip_state, si_emit_clip_state);
- si_init_atom(sctx, &sctx->scissors.atom, &sctx->atoms.s.scissors, si_emit_scissors);
- si_init_atom(sctx, &sctx->viewports.atom, &sctx->atoms.s.viewports, si_emit_viewports);
si_init_atom(sctx, &sctx->stencil_ref.atom, &sctx->atoms.s.stencil_ref, si_emit_stencil_ref);
sctx->b.b.create_blend_state = si_create_blend_state;
sctx->custom_blend_resolve = si_create_blend_custom(sctx, V_028808_CB_RESOLVE);
sctx->custom_blend_decompress = si_create_blend_custom(sctx, V_028808_CB_FMASK_DECOMPRESS);
sctx->custom_blend_fastclear = si_create_blend_custom(sctx, V_028808_CB_ELIMINATE_FAST_CLEAR);
+ sctx->custom_blend_dcc_decompress = si_create_blend_custom(sctx, V_028808_CB_DCC_DECOMPRESS);
sctx->b.b.set_clip_state = si_set_clip_state;
- sctx->b.b.set_scissor_states = si_set_scissor_states;
- sctx->b.b.set_viewport_states = si_set_viewport_states;
sctx->b.b.set_stencil_ref = si_set_stencil_ref;
sctx->b.b.set_framebuffer_state = si_set_framebuffer_state;
sctx->b.b.set_index_buffer = si_set_index_buffer;
sctx->b.b.texture_barrier = si_texture_barrier;
- sctx->b.b.set_polygon_stipple = si_set_polygon_stipple;
+ sctx->b.b.memory_barrier = si_memory_barrier;
sctx->b.b.set_min_samples = si_set_min_samples;
sctx->b.b.set_tess_state = si_set_tess_state;
+ sctx->b.b.set_active_query_state = si_set_active_query_state;
sctx->b.set_occlusion_query_state = si_set_occlusion_query_state;
sctx->b.need_gfx_cs_space = si_need_gfx_cs_space;
sctx->b.b.draw_vbo = si_draw_vbo;
- if (sctx->b.chip_class >= CIK) {
- sctx->b.dma_copy = cik_sdma_copy;
- } else {
- sctx->b.dma_copy = si_dma_copy;
+ si_init_config(sctx);
+}
+
+static uint32_t si_get_bo_metadata_word1(struct r600_common_screen *rscreen)
+{
+ return (ATI_VENDOR_ID << 16) | rscreen->info.pci_id;
+}
+
+static void si_query_opaque_metadata(struct r600_common_screen *rscreen,
+ struct r600_texture *rtex,
+ struct radeon_bo_metadata *md)
+{
+ struct si_screen *sscreen = (struct si_screen*)rscreen;
+ struct pipe_resource *res = &rtex->resource.b.b;
+ static const unsigned char swizzle[] = {
+ PIPE_SWIZZLE_X,
+ PIPE_SWIZZLE_Y,
+ PIPE_SWIZZLE_Z,
+ PIPE_SWIZZLE_W
+ };
+ uint32_t desc[8], i;
+ bool is_array = util_resource_is_array_texture(res);
+
+ /* DRM 2.x.x doesn't support this. */
+ if (rscreen->info.drm_major != 3)
+ return;
+
+ assert(rtex->fmask.size == 0);
+
+ /* Metadata image format format version 1:
+ * [0] = 1 (metadata format identifier)
+ * [1] = (VENDOR_ID << 16) | PCI_ID
+ * [2:9] = image descriptor for the whole resource
+ * [2] is always 0, because the base address is cleared
+ * [9] is the DCC offset bits [39:8] from the beginning of
+ * the buffer
+ * [10:10+LAST_LEVEL] = mipmap level offset bits [39:8] for each level
+ */
+
+ md->metadata[0] = 1; /* metadata image format version 1 */
+
+ /* TILE_MODE_INDEX is ambiguous without a PCI ID. */
+ md->metadata[1] = si_get_bo_metadata_word1(rscreen);
+
+ si_make_texture_descriptor(sscreen, rtex, true,
+ res->target, res->format,
+ swizzle, 0, res->last_level, 0,
+ is_array ? res->array_size - 1 : 0,
+ res->width0, res->height0, res->depth0,
+ desc, NULL);
+
+ si_set_mutable_tex_desc_fields(rtex, &rtex->surface.level[0], 0, 0,
+ rtex->surface.blk_w, false, desc);
+
+ /* Clear the base address and set the relative DCC offset. */
+ desc[0] = 0;
+ desc[1] &= C_008F14_BASE_ADDRESS_HI;
+ desc[7] = rtex->dcc_offset >> 8;
+
+ /* Dwords [2:9] contain the image descriptor. */
+ memcpy(&md->metadata[2], desc, sizeof(desc));
+
+ /* Dwords [10:..] contain the mipmap level offsets. */
+ for (i = 0; i <= res->last_level; i++)
+ md->metadata[10+i] = rtex->surface.level[i].offset >> 8;
+
+ md->size_metadata = (11 + res->last_level) * 4;
+}
+
+static void si_apply_opaque_metadata(struct r600_common_screen *rscreen,
+ struct r600_texture *rtex,
+ struct radeon_bo_metadata *md)
+{
+ uint32_t *desc = &md->metadata[2];
+
+ if (rscreen->chip_class < VI)
+ return;
+
+ /* Return if DCC is enabled. The texture should be set up with it
+ * already.
+ */
+ if (md->size_metadata >= 11 * 4 &&
+ md->metadata[0] != 0 &&
+ md->metadata[1] == si_get_bo_metadata_word1(rscreen) &&
+ G_008F28_COMPRESSION_EN(desc[6])) {
+ assert(rtex->dcc_offset == ((uint64_t)desc[7] << 8));
+ return;
}
- si_init_config(sctx);
+ /* Disable DCC. These are always set by texture_from_handle and must
+ * be cleared here.
+ */
+ rtex->dcc_offset = 0;
+}
+
+void si_init_screen_state_functions(struct si_screen *sscreen)
+{
+ sscreen->b.query_opaque_metadata = si_query_opaque_metadata;
+ sscreen->b.apply_opaque_metadata = si_apply_opaque_metadata;
}
static void
{
unsigned sh_per_se = MAX2(sctx->screen->b.info.max_sh_per_se, 1);
unsigned num_se = MAX2(sctx->screen->b.info.max_se, 1);
- unsigned rb_mask = sctx->screen->b.info.si_backend_enabled_mask;
- unsigned num_rb = MIN2(sctx->screen->b.info.r600_num_backends, 16);
+ unsigned rb_mask = sctx->screen->b.info.enabled_rb_mask;
+ unsigned num_rb = MIN2(sctx->screen->b.info.num_render_backends, 16);
unsigned rb_per_pkr = MIN2(num_rb / num_se / sh_per_se, 2);
unsigned rb_per_se = num_rb / num_se;
unsigned se_mask[4];
unsigned se;
- se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
- se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
- se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
- se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
+ se_mask[0] = ((1 << rb_per_se) - 1);
+ se_mask[1] = (se_mask[0] << rb_per_se);
+ se_mask[2] = (se_mask[1] << rb_per_se);
+ se_mask[3] = (se_mask[2] << rb_per_se);
+
+ se_mask[0] &= rb_mask;
+ se_mask[1] &= rb_mask;
+ se_mask[2] &= rb_mask;
+ se_mask[3] &= rb_mask;
assert(num_se == 1 || num_se == 2 || num_se == 4);
assert(sh_per_se == 1 || sh_per_se == 2);
* fields are for, so I'm leaving them as their default
* values. */
- if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
- (!se_mask[2] && !se_mask[3]))) {
- raster_config_1 &= C_028354_SE_PAIR_MAP;
-
- if (!se_mask[0] && !se_mask[1]) {
- raster_config_1 |=
- S_028354_SE_PAIR_MAP(V_028354_RASTER_CONFIG_SE_PAIR_MAP_3);
- } else {
- raster_config_1 |=
- S_028354_SE_PAIR_MAP(V_028354_RASTER_CONFIG_SE_PAIR_MAP_0);
- }
- }
-
for (se = 0; se < num_se; se++) {
unsigned raster_config_se = raster_config;
unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
S_030800_SE_INDEX(se) | S_030800_SH_BROADCAST_WRITES(1) |
S_030800_INSTANCE_BROADCAST_WRITES(1));
si_pm4_set_reg(pm4, R_028350_PA_SC_RASTER_CONFIG, raster_config_se);
- if (sctx->b.chip_class >= CIK)
- si_pm4_set_reg(pm4, R_028354_PA_SC_RASTER_CONFIG_1, raster_config_1);
}
/* GRBM_GFX_INDEX has a different offset on SI and CI+ */
si_pm4_set_reg(pm4, GRBM_GFX_INDEX,
SE_BROADCAST_WRITES | SH_BROADCAST_WRITES |
INSTANCE_BROADCAST_WRITES);
- else
+ else {
si_pm4_set_reg(pm4, R_030800_GRBM_GFX_INDEX,
S_030800_SE_BROADCAST_WRITES(1) | S_030800_SH_BROADCAST_WRITES(1) |
S_030800_INSTANCE_BROADCAST_WRITES(1));
+
+ if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
+ (!se_mask[2] && !se_mask[3]))) {
+ raster_config_1 &= C_028354_SE_PAIR_MAP;
+
+ if (!se_mask[0] && !se_mask[1]) {
+ raster_config_1 |=
+ S_028354_SE_PAIR_MAP(V_028354_RASTER_CONFIG_SE_PAIR_MAP_3);
+ } else {
+ raster_config_1 |=
+ S_028354_SE_PAIR_MAP(V_028354_RASTER_CONFIG_SE_PAIR_MAP_0);
+ }
+ }
+
+ si_pm4_set_reg(pm4, R_028354_PA_SC_RASTER_CONFIG_1, raster_config_1);
+ }
}
static void si_init_config(struct si_context *sctx)
{
struct si_screen *sscreen = sctx->screen;
- unsigned num_rb = MIN2(sctx->screen->b.info.r600_num_backends, 16);
- unsigned rb_mask = sctx->screen->b.info.si_backend_enabled_mask;
+ unsigned num_rb = MIN2(sctx->screen->b.info.num_render_backends, 16);
+ unsigned rb_mask = sctx->screen->b.info.enabled_rb_mask;
unsigned raster_config, raster_config_1;
uint64_t border_color_va = sctx->border_color_buffer->gpu_address;
struct si_pm4_state *pm4 = CALLOC_STRUCT(si_pm4_state);
return;
si_pm4_cmd_begin(pm4, PKT3_CONTEXT_CONTROL);
- si_pm4_cmd_add(pm4, 0x80000000);
- si_pm4_cmd_add(pm4, 0x80000000);
+ si_pm4_cmd_add(pm4, CONTEXT_CONTROL_LOAD_ENABLE(1));
+ si_pm4_cmd_add(pm4, CONTEXT_CONTROL_SHADOW_ENABLE(1));
si_pm4_cmd_end(pm4, false);
si_pm4_set_reg(pm4, R_028A18_VGT_HOS_MAX_TESS_LEVEL, fui(64));
raster_config_1 = 0x0000002e;
}
break;
+ case CHIP_POLARIS10:
+ raster_config = 0x16000012;
+ raster_config_1 = 0x0000002a;
+ break;
+ case CHIP_POLARIS11:
+ raster_config = 0x16000012;
+ raster_config_1 = 0x00000000;
+ break;
case CHIP_TONGA:
raster_config = 0x16000012;
raster_config_1 = 0x0000002a;
break;
case CHIP_ICELAND:
- raster_config = 0x00000002;
+ if (num_rb == 1)
+ raster_config = 0x00000000;
+ else
+ raster_config = 0x00000002;
raster_config_1 = 0x00000000;
break;
case CHIP_CARRIZO:
/* PA_SU_HARDWARE_SCREEN_OFFSET must be 0 due to hw bug on SI */
si_pm4_set_reg(pm4, R_028234_PA_SU_HARDWARE_SCREEN_OFFSET, 0);
si_pm4_set_reg(pm4, R_028820_PA_CL_NANINF_CNTL, 0);
- si_pm4_set_reg(pm4, R_028BE8_PA_CL_GB_VERT_CLIP_ADJ, fui(1.0));
- si_pm4_set_reg(pm4, R_028BEC_PA_CL_GB_VERT_DISC_ADJ, fui(1.0));
- si_pm4_set_reg(pm4, R_028BF0_PA_CL_GB_HORZ_CLIP_ADJ, fui(1.0));
- si_pm4_set_reg(pm4, R_028BF4_PA_CL_GB_HORZ_DISC_ADJ, fui(1.0));
si_pm4_set_reg(pm4, R_028AC0_DB_SRESULTS_COMPARE_STATE0, 0x0);
si_pm4_set_reg(pm4, R_028AC4_DB_SRESULTS_COMPARE_STATE1, 0x0);
si_pm4_set_reg(pm4, R_028AC8_DB_PRELOAD_CONTROL, 0x0);
si_pm4_set_reg(pm4, R_028408_VGT_INDX_OFFSET, 0);
if (sctx->b.chip_class >= CIK) {
- si_pm4_set_reg(pm4, R_00B51C_SPI_SHADER_PGM_RSRC3_LS, S_00B51C_CU_EN(0xfffc));
si_pm4_set_reg(pm4, R_00B41C_SPI_SHADER_PGM_RSRC3_HS, 0);
- si_pm4_set_reg(pm4, R_00B31C_SPI_SHADER_PGM_RSRC3_ES, S_00B31C_CU_EN(0xfffe));
+ si_pm4_set_reg(pm4, R_00B31C_SPI_SHADER_PGM_RSRC3_ES, S_00B31C_CU_EN(0xffff));
si_pm4_set_reg(pm4, R_00B21C_SPI_SHADER_PGM_RSRC3_GS, S_00B21C_CU_EN(0xffff));
- si_pm4_set_reg(pm4, R_00B118_SPI_SHADER_PGM_RSRC3_VS, S_00B118_CU_EN(0xffff));
- si_pm4_set_reg(pm4, R_00B11C_SPI_SHADER_LATE_ALLOC_VS, S_00B11C_LIMIT(0));
+
+ if (sscreen->b.info.num_good_compute_units /
+ (sscreen->b.info.max_se * sscreen->b.info.max_sh_per_se) <= 4) {
+ /* Too few available compute units per SH. Disallowing
+ * VS to run on CU0 could hurt us more than late VS
+ * allocation would help.
+ *
+ * LATE_ALLOC_VS = 2 is the highest safe number.
+ */
+ si_pm4_set_reg(pm4, R_00B51C_SPI_SHADER_PGM_RSRC3_LS, S_00B51C_CU_EN(0xffff));
+ si_pm4_set_reg(pm4, R_00B118_SPI_SHADER_PGM_RSRC3_VS, S_00B118_CU_EN(0xffff));
+ si_pm4_set_reg(pm4, R_00B11C_SPI_SHADER_LATE_ALLOC_VS, S_00B11C_LIMIT(2));
+ } else {
+ /* Set LATE_ALLOC_VS == 31. It should be less than
+ * the number of scratch waves. Limitations:
+ * - VS can't execute on CU0.
+ * - If HS writes outputs to LDS, LS can't execute on CU0.
+ */
+ si_pm4_set_reg(pm4, R_00B51C_SPI_SHADER_PGM_RSRC3_LS, S_00B51C_CU_EN(0xfffe));
+ si_pm4_set_reg(pm4, R_00B118_SPI_SHADER_PGM_RSRC3_VS, S_00B118_CU_EN(0xfffe));
+ si_pm4_set_reg(pm4, R_00B11C_SPI_SHADER_LATE_ALLOC_VS, S_00B11C_LIMIT(31));
+ }
+
si_pm4_set_reg(pm4, R_00B01C_SPI_SHADER_PGM_RSRC3_PS, S_00B01C_CU_EN(0xffff));
}
if (sctx->b.chip_class >= VI) {
+ unsigned vgt_tess_distribution;
+
si_pm4_set_reg(pm4, R_028424_CB_DCC_CONTROL,
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);
+
+ vgt_tess_distribution =
+ S_028B50_ACCUM_ISOLINE(32) |
+ S_028B50_ACCUM_TRI(11) |
+ S_028B50_ACCUM_QUAD(11) |
+ S_028B50_DONUT_SPLIT(16);
+
+ /* Testing with Unigine Heaven extreme tesselation yielded best results
+ * with TRAP_SPLIT = 3.
+ */
+ if (sctx->b.family == CHIP_FIJI ||
+ sctx->b.family >= CHIP_POLARIS10)
+ vgt_tess_distribution |= S_028B50_TRAP_SPLIT(3);
+
+ si_pm4_set_reg(pm4, R_028B50_VGT_TESS_DISTRIBUTION, vgt_tess_distribution);
}
if (sctx->b.family == CHIP_STONEY)
projects / mesa.git / blobdiff