[Mesa-dev] [PATCH 4/4] radeonsi: implement EXT_window_rectangles
Marek Olšák
maraeo at gmail.com
Thu Aug 9 01:29:45 UTC 2018
From: Marek Olšák <marek.olsak at amd.com>
---
docs/relnotes/18.3.0.html | 1 +
src/amd/common/sid.h | 16 ++++
src/gallium/drivers/radeonsi/si_blit.c | 4 +
src/gallium/drivers/radeonsi/si_get.c | 2 +-
src/gallium/drivers/radeonsi/si_gfx_cs.c | 4 +
src/gallium/drivers/radeonsi/si_pipe.h | 3 +
src/gallium/drivers/radeonsi/si_state.c | 1 -
src/gallium/drivers/radeonsi/si_state.h | 3 +
.../drivers/radeonsi/si_state_viewport.c | 80 +++++++++++++++++++
9 files changed, 112 insertions(+), 2 deletions(-)
diff --git a/docs/relnotes/18.3.0.html b/docs/relnotes/18.3.0.html
index 786145ad8da..ac2cc1e893b 100644
--- a/docs/relnotes/18.3.0.html
+++ b/docs/relnotes/18.3.0.html
@@ -45,20 +45,21 @@ TBD.
<h2>New features</h2>
<p>
Note: some of the new features are only available with certain drivers.
</p>
<ul>
<li>GL_AMD_framebuffer_multisample_advanced on radeonsi.</li>
+<li>GL_EXT_window_rectangles on radeonsi.</li>
</ul>
<h2>Bug fixes</h2>
<ul>
<li>TBD</li>
</ul>
<h2>Changes</h2>
diff --git a/src/amd/common/sid.h b/src/amd/common/sid.h
index 3c3bc541b4f..0671f7d3998 100644
--- a/src/amd/common/sid.h
+++ b/src/amd/common/sid.h
@@ -5275,20 +5275,36 @@
#define S_028208_BR_X(x) (((unsigned)(x) & 0x7FFF) << 0)
#define G_028208_BR_X(x) (((x) >> 0) & 0x7FFF)
#define C_028208_BR_X 0xFFFF8000
#define S_028208_BR_Y(x) (((unsigned)(x) & 0x7FFF) << 16)
#define G_028208_BR_Y(x) (((x) >> 16) & 0x7FFF)
#define C_028208_BR_Y 0x8000FFFF
#define R_02820C_PA_SC_CLIPRECT_RULE 0x02820C
#define S_02820C_CLIP_RULE(x) (((unsigned)(x) & 0xFFFF) << 0)
#define G_02820C_CLIP_RULE(x) (((x) >> 0) & 0xFFFF)
#define C_02820C_CLIP_RULE 0xFFFF0000
+#define V_02820C_OUT 0x0001
+#define V_02820C_IN_0 0x0002
+#define V_02820C_IN_1 0x0004
+#define V_02820C_IN_10 0x0008
+#define V_02820C_IN_2 0x0010
+#define V_02820C_IN_20 0x0020
+#define V_02820C_IN_21 0x0040
+#define V_02820C_IN_210 0x0080
+#define V_02820C_IN_3 0x0100
+#define V_02820C_IN_30 0x0200
+#define V_02820C_IN_31 0x0400
+#define V_02820C_IN_310 0x0800
+#define V_02820C_IN_32 0x1000
+#define V_02820C_IN_320 0x2000
+#define V_02820C_IN_321 0x4000
+#define V_02820C_IN_3210 0x8000
#define R_028210_PA_SC_CLIPRECT_0_TL 0x028210
#define S_028210_TL_X(x) (((unsigned)(x) & 0x7FFF) << 0)
#define G_028210_TL_X(x) (((x) >> 0) & 0x7FFF)
#define C_028210_TL_X 0xFFFF8000
#define S_028210_TL_Y(x) (((unsigned)(x) & 0x7FFF) << 16)
#define G_028210_TL_Y(x) (((x) >> 16) & 0x7FFF)
#define C_028210_TL_Y 0x8000FFFF
#define R_028214_PA_SC_CLIPRECT_0_BR 0x028214
#define S_028214_BR_X(x) (((unsigned)(x) & 0x7FFF) << 0)
#define G_028214_BR_X(x) (((x) >> 0) & 0x7FFF)
diff --git a/src/gallium/drivers/radeonsi/si_blit.c b/src/gallium/drivers/radeonsi/si_blit.c
index 0fd69f3a6c5..cf6495291bd 100644
--- a/src/gallium/drivers/radeonsi/si_blit.c
+++ b/src/gallium/drivers/radeonsi/si_blit.c
@@ -52,20 +52,24 @@ void si_blitter_begin(struct si_context *sctx, enum si_blitter_op op)
(struct pipe_stream_output_target**)sctx->streamout.targets);
util_blitter_save_rasterizer(sctx->blitter, sctx->queued.named.rasterizer);
if (op & SI_SAVE_FRAGMENT_STATE) {
util_blitter_save_blend(sctx->blitter, sctx->queued.named.blend);
util_blitter_save_depth_stencil_alpha(sctx->blitter, sctx->queued.named.dsa);
util_blitter_save_stencil_ref(sctx->blitter, &sctx->stencil_ref.state);
util_blitter_save_fragment_shader(sctx->blitter, sctx->ps_shader.cso);
util_blitter_save_sample_mask(sctx->blitter, sctx->sample_mask);
util_blitter_save_scissor(sctx->blitter, &sctx->scissors.states[0]);
+ util_blitter_save_window_rectangles(sctx->blitter,
+ sctx->window_rectangles_include,
+ sctx->num_window_rectangles,
+ sctx->window_rectangles);
}
if (op & SI_SAVE_FRAMEBUFFER)
util_blitter_save_framebuffer(sctx->blitter, &sctx->framebuffer.state);
if (op & SI_SAVE_TEXTURES) {
util_blitter_save_fragment_sampler_states(
sctx->blitter, 2,
(void**)sctx->samplers[PIPE_SHADER_FRAGMENT].sampler_states);
diff --git a/src/gallium/drivers/radeonsi/si_get.c b/src/gallium/drivers/radeonsi/si_get.c
index 6e804797bab..e316703bbe3 100644
--- a/src/gallium/drivers/radeonsi/si_get.c
+++ b/src/gallium/drivers/radeonsi/si_get.c
@@ -201,20 +201,21 @@ static int si_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
return SI_MAP_BUFFER_ALIGNMENT;
case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
case PIPE_CAP_MAX_VERTEX_STREAMS:
case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
+ case PIPE_CAP_MAX_WINDOW_RECTANGLES:
return 4;
case PIPE_CAP_GLSL_FEATURE_LEVEL:
case PIPE_CAP_GLSL_FEATURE_LEVEL_COMPATIBILITY:
if (sscreen->info.has_indirect_compute_dispatch)
return param == PIPE_CAP_GLSL_FEATURE_LEVEL ?
450 : 440;
return 420;
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
@@ -236,21 +237,20 @@ static int si_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
/* Unsupported features. */
case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
case PIPE_CAP_USER_VERTEX_BUFFERS:
case PIPE_CAP_FAKE_SW_MSAA:
case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
case PIPE_CAP_VERTEXID_NOBASE:
case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
- case PIPE_CAP_MAX_WINDOW_RECTANGLES:
case PIPE_CAP_TGSI_MUL_ZERO_WINS:
case PIPE_CAP_UMA:
case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
case PIPE_CAP_POST_DEPTH_COVERAGE:
case PIPE_CAP_TILE_RASTER_ORDER:
case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
case PIPE_CAP_CONTEXT_PRIORITY_MASK:
case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_TRIANGLES:
case PIPE_CAP_CONSERVATIVE_RASTER_POST_SNAP_POINTS_LINES:
case PIPE_CAP_CONSERVATIVE_RASTER_PRE_SNAP_TRIANGLES:
diff --git a/src/gallium/drivers/radeonsi/si_gfx_cs.c b/src/gallium/drivers/radeonsi/si_gfx_cs.c
index 9dbd4c64f2a..c39564ecbe5 100644
--- a/src/gallium/drivers/radeonsi/si_gfx_cs.c
+++ b/src/gallium/drivers/radeonsi/si_gfx_cs.c
@@ -271,20 +271,23 @@ void si_begin_new_gfx_cs(struct si_context *ctx)
/* CLEAR_STATE sets zeros. */
if (!has_clear_state || ctx->blend_color.any_nonzeros)
si_mark_atom_dirty(ctx, &ctx->atoms.s.blend_color);
si_mark_atom_dirty(ctx, &ctx->atoms.s.db_render_state);
if (ctx->chip_class >= GFX9)
si_mark_atom_dirty(ctx, &ctx->atoms.s.dpbb_state);
si_mark_atom_dirty(ctx, &ctx->atoms.s.stencil_ref);
si_mark_atom_dirty(ctx, &ctx->atoms.s.spi_map);
si_mark_atom_dirty(ctx, &ctx->atoms.s.streamout_enable);
si_mark_atom_dirty(ctx, &ctx->atoms.s.render_cond);
+ /* CLEAR_STATE disables all window rectangles. */
+ if (!has_clear_state || ctx->num_window_rectangles > 0)
+ si_mark_atom_dirty(ctx, &ctx->atoms.s.window_rectangles);
si_all_descriptors_begin_new_cs(ctx);
si_all_resident_buffers_begin_new_cs(ctx);
ctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
ctx->viewports.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
ctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
si_mark_atom_dirty(ctx, &ctx->atoms.s.guardband);
si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
si_mark_atom_dirty(ctx, &ctx->atoms.s.viewports);
@@ -339,20 +342,21 @@ void si_begin_new_gfx_cs(struct si_context *ctx)
ctx->tracked_regs.reg_value[SI_TRACKED_PA_SC_MODE_CNTL_1] = 0x00000000;
ctx->tracked_regs.reg_value[SI_TRACKED_PA_SU_SMALL_PRIM_FILTER_CNTL] = 0x00000000;
ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_VS_OUT_CNTL] = 0x00000000;
ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_CLIP_CNTL] = 0x00090000;
ctx->tracked_regs.reg_value[SI_TRACKED_PA_SC_BINNER_CNTL_0] = 0x00000003;
ctx->tracked_regs.reg_value[SI_TRACKED_DB_DFSM_CONTROL] = 0x00000000;
ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_GB_VERT_CLIP_ADJ] = 0x3f800000;
ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_GB_VERT_DISC_ADJ] = 0x3f800000;
ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_GB_HORZ_CLIP_ADJ] = 0x3f800000;
ctx->tracked_regs.reg_value[SI_TRACKED_PA_CL_GB_HORZ_DISC_ADJ] = 0x3f800000;
+ ctx->tracked_regs.reg_value[SI_TRACKED_PA_SC_CLIPRECT_RULE] = 0xffff;
/* Set all saved registers state to saved. */
ctx->tracked_regs.reg_saved = 0xffffffff;
} else {
/* Set all saved registers state to unknown. */
ctx->tracked_regs.reg_saved = 0;
}
/* 0xffffffff is a impossible value to register SPI_PS_INPUT_CNTL_n */
memset(ctx->tracked_regs.spi_ps_input_cntl, 0xff, sizeof(uint32_t) * 32);
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 2f77cc741bf..cfe28c2e8f0 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -797,20 +797,23 @@ struct si_context {
unsigned sample_locs_num_samples;
uint16_t sample_mask;
unsigned last_cb_target_mask;
struct si_blend_color blend_color;
struct si_clip_state clip_state;
struct si_shader_data shader_pointers;
struct si_stencil_ref stencil_ref;
struct si_scissors scissors;
struct si_streamout streamout;
struct si_viewports viewports;
+ unsigned num_window_rectangles;
+ bool window_rectangles_include;
+ struct pipe_scissor_state window_rectangles[4];
/* Precomputed states. */
struct si_pm4_state *init_config;
struct si_pm4_state *init_config_gs_rings;
bool init_config_has_vgt_flush;
struct si_pm4_state *vgt_shader_config[4];
/* shaders */
struct si_shader_ctx_state ps_shader;
struct si_shader_ctx_state gs_shader;
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index fc1ec83931f..780d9010abc 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -4865,21 +4865,20 @@ static void si_init_config(struct si_context *sctx)
si_pm4_set_reg(pm4, R_028204_PA_SC_WINDOW_SCISSOR_TL, S_028204_WINDOW_OFFSET_DISABLE(1));
si_pm4_set_reg(pm4, R_028240_PA_SC_GENERIC_SCISSOR_TL, S_028240_WINDOW_OFFSET_DISABLE(1));
si_pm4_set_reg(pm4, R_028244_PA_SC_GENERIC_SCISSOR_BR,
S_028244_BR_X(16384) | S_028244_BR_Y(16384));
si_pm4_set_reg(pm4, R_028030_PA_SC_SCREEN_SCISSOR_TL, 0);
si_pm4_set_reg(pm4, R_028034_PA_SC_SCREEN_SCISSOR_BR,
S_028034_BR_X(16384) | S_028034_BR_Y(16384));
}
if (!has_clear_state) {
- si_pm4_set_reg(pm4, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
si_pm4_set_reg(pm4, R_028230_PA_SC_EDGERULE,
S_028230_ER_TRI(0xA) |
S_028230_ER_POINT(0xA) |
S_028230_ER_RECT(0xA) |
/* Required by DX10_DIAMOND_TEST_ENA: */
S_028230_ER_LINE_LR(0x1A) |
S_028230_ER_LINE_RL(0x26) |
S_028230_ER_LINE_TB(0xA) |
S_028230_ER_LINE_BT(0xA));
/* PA_SU_HARDWARE_SCREEN_OFFSET must be 0 due to hw bug on SI */
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 1edf5d646b6..1a8f3debf3a 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -206,20 +206,21 @@ union si_state_atoms {
struct si_atom blend_color;
struct si_atom clip_regs;
struct si_atom clip_state;
struct si_atom shader_pointers;
struct si_atom guardband;
struct si_atom scissors;
struct si_atom viewports;
struct si_atom stencil_ref;
struct si_atom spi_map;
struct si_atom scratch_state;
+ struct si_atom window_rectangles;
} s;
struct si_atom array[0];
};
#define SI_ATOM_BIT(name) (1 << (offsetof(union si_state_atoms, s.name) / \
sizeof(struct si_atom)))
#define SI_NUM_ATOMS (sizeof(union si_state_atoms)/sizeof(struct si_atom*))
static inline unsigned si_atoms_that_roll_context(void)
{
@@ -274,20 +275,22 @@ enum si_tracked_reg {
SI_TRACKED_PA_CL_CLIP_CNTL,
SI_TRACKED_PA_SC_BINNER_CNTL_0,
SI_TRACKED_DB_DFSM_CONTROL,
SI_TRACKED_PA_CL_GB_VERT_CLIP_ADJ, /* 4 consecutive registers */
SI_TRACKED_PA_CL_GB_VERT_DISC_ADJ,
SI_TRACKED_PA_CL_GB_HORZ_CLIP_ADJ,
SI_TRACKED_PA_CL_GB_HORZ_DISC_ADJ,
+ SI_TRACKED_PA_SC_CLIPRECT_RULE,
+
SI_NUM_TRACKED_REGS,
};
struct si_tracked_regs {
uint32_t reg_saved;
uint32_t reg_value[SI_NUM_TRACKED_REGS];
uint32_t spi_ps_input_cntl[32];
};
/* Private read-write buffer slots. */
diff --git a/src/gallium/drivers/radeonsi/si_state_viewport.c b/src/gallium/drivers/radeonsi/si_state_viewport.c
index 4183be0c880..8dc68b126eb 100644
--- a/src/gallium/drivers/radeonsi/si_state_viewport.c
+++ b/src/gallium/drivers/radeonsi/si_state_viewport.c
@@ -428,19 +428,99 @@ void si_update_vs_viewport_state(struct si_context *ctx)
return;
if (ctx->scissors.dirty_mask)
si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
if (ctx->viewports.dirty_mask ||
ctx->viewports.depth_range_dirty_mask)
si_mark_atom_dirty(ctx, &ctx->atoms.s.viewports);
}
+static void si_emit_window_rectangles(struct si_context *sctx)
+{
+ /* There are four clipping rectangles. Their corner coordinates are inclusive.
+ * Every pixel is assigned a number from 0 and 15 by setting bits 0-3 depending
+ * on whether the pixel is inside cliprects 0-3, respectively. For example,
+ * if a pixel is inside cliprects 0 and 1, but outside 2 and 3, it is assigned
+ * the number 3 (binary 0011).
+ *
+ * If CLIPRECT_RULE & (1 << number), the pixel is rasterized.
+ */
+ struct radeon_cmdbuf *cs = sctx->gfx_cs;
+ static const unsigned outside[4] = {
+ /* outside rectangle 0 */
+ V_02820C_OUT |
+ V_02820C_IN_1 |
+ V_02820C_IN_2 |
+ V_02820C_IN_21 |
+ V_02820C_IN_3 |
+ V_02820C_IN_31 |
+ V_02820C_IN_32 |
+ V_02820C_IN_321,
+ /* outside rectangles 0, 1 */
+ V_02820C_OUT |
+ V_02820C_IN_2 |
+ V_02820C_IN_3 |
+ V_02820C_IN_32,
+ /* outside rectangles 0, 1, 2 */
+ V_02820C_OUT |
+ V_02820C_IN_3,
+ /* outside rectangles 0, 1, 2, 3 */
+ V_02820C_OUT,
+ };
+ const unsigned disabled = 0xffff; /* all inside and outside cases */
+ unsigned num_rectangles = sctx->num_window_rectangles;
+ struct pipe_scissor_state *rects = sctx->window_rectangles;
+ unsigned rule;
+
+ assert(num_rectangles <= 4);
+
+ if (num_rectangles == 0)
+ rule = disabled;
+ else if (sctx->window_rectangles_include)
+ rule = ~outside[num_rectangles - 1];
+ else
+ rule = outside[num_rectangles - 1];
+
+ radeon_opt_set_context_reg(sctx, R_02820C_PA_SC_CLIPRECT_RULE,
+ SI_TRACKED_PA_SC_CLIPRECT_RULE, rule);
+ if (num_rectangles == 0)
+ return;
+
+ radeon_set_context_reg_seq(cs, R_028210_PA_SC_CLIPRECT_0_TL,
+ num_rectangles * 2);
+ for (unsigned i = 0; i < num_rectangles; i++) {
+ radeon_emit(cs, S_028210_TL_X(rects[i].minx) |
+ S_028210_TL_Y(rects[i].miny));
+ radeon_emit(cs, S_028214_BR_X(rects[i].maxx) |
+ S_028214_BR_Y(rects[i].maxy));
+ }
+}
+
+static void si_set_window_rectangles(struct pipe_context *ctx,
+ boolean include,
+ unsigned num_rectangles,
+ const struct pipe_scissor_state *rects)
+{
+ struct si_context *sctx = (struct si_context *)ctx;
+
+ sctx->num_window_rectangles = num_rectangles;
+ sctx->window_rectangles_include = include;
+ if (num_rectangles) {
+ memcpy(sctx->window_rectangles, rects,
+ sizeof(*rects) * num_rectangles);
+ }
+
+ si_mark_atom_dirty(sctx, &sctx->atoms.s.window_rectangles);
+}
+
void si_init_viewport_functions(struct si_context *ctx)
{
ctx->atoms.s.guardband.emit = si_emit_guardband;
ctx->atoms.s.scissors.emit = si_emit_scissors;
ctx->atoms.s.viewports.emit = si_emit_viewport_states;
+ ctx->atoms.s.window_rectangles.emit = si_emit_window_rectangles;
ctx->b.set_scissor_states = si_set_scissor_states;
ctx->b.set_viewport_states = si_set_viewport_states;
+ ctx->b.set_window_rectangles = si_set_window_rectangles;
}
--
2.17.1
More information about the mesa-dev
mailing list