r600: implement EXT_window_rectangles

This is a backport of 0ca8294ece ("radeonsi:
implement EXT_window_rectangles")

This change was tested and passes the piglit tests (20/20)
on rv770, palm and cayman.

Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Acked-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34295>

src/gallium/drivers/r600/evergreen_state.c

@@ -4670,6 +4670,7 @@ void evergreen_init_state_functions(struct r600_context *rctx)
 	r600_add_atom(rctx, &rctx->b.render_cond_atom, id++);
 	r600_add_atom(rctx, &rctx->b.streamout.begin_atom, id++);
 	r600_add_atom(rctx, &rctx->b.streamout.enable_atom, id++);
+	r600_add_atom(rctx, &rctx->b.window_rectangles.atom, id++);
 	for (i = 0; i < EG_NUM_HW_STAGES; i++)
 		r600_init_atom(rctx, &rctx->hw_shader_stages[i].atom, id++, r600_emit_shader, 0);
 	r600_init_atom(rctx, &rctx->shader_stages.atom, id++, evergreen_emit_shader_stages, 15);

src/gallium/drivers/r600/r600_blit.c

@@ -65,6 +65,9 @@ static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op
 		util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->dsa_state.cso);
 		util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref.pipe_state);
                 util_blitter_save_sample_mask(rctx->blitter, rctx->sample_mask.sample_mask, rctx->ps_iter_samples);
+		util_blitter_save_window_rectangles(rctx->blitter, rctx->b.window_rectangles.include,
+						    rctx->b.window_rectangles.number,
+						    rctx->b.window_rectangles.states);
 	}
 
 	if (op & R600_SAVE_CONST_BUF0) {

src/gallium/drivers/r600/r600_pipe.c

@@ -475,6 +475,7 @@ static void r600_init_screen_caps(struct r600_screen *rscreen)
 	caps->two_sided_color = false;
 	caps->cull_distance = true;
 
+	caps->max_window_rectangles = R600_MAX_WINDOW_RECTANGLES;
 	caps->shader_buffer_offset_alignment = family >= CHIP_CEDAR ?  256 : 0;
 
 	caps->max_shader_patch_varyings = family >= CHIP_CEDAR ? 30 : 0;

src/gallium/drivers/r600/r600_pipe.h

@@ -20,7 +20,7 @@
 
 #include "tgsi/tgsi_scan.h"
 
-#define R600_NUM_ATOMS 56
+#define R600_NUM_ATOMS 57
 
 #define R600_MAX_IMAGES 8
 /*

src/gallium/drivers/r600/r600_pipe_common.h

@@ -446,6 +446,16 @@ struct r600_viewports {
 	struct r600_signed_scissor	as_scissor[R600_MAX_VIEWPORTS];
 };
 
+/* EXT_window_rectangles */
+#define R600_MAX_WINDOW_RECTANGLES 4
+
+struct r600_window_rectangles {
+	unsigned			number;
+	bool				include;
+	struct pipe_scissor_state	states[R600_MAX_WINDOW_RECTANGLES];
+	struct r600_atom		atom;
+};
+
 struct r600_ring {
 	struct radeon_cmdbuf		cs;
 	void (*flush)(void *ctx, unsigned flags,
@@ -493,6 +503,7 @@ struct r600_common_context {
 	struct r600_streamout		streamout;
 	struct r600_scissors		scissors;
 	struct r600_viewports		viewports;
+	struct r600_window_rectangles	window_rectangles;
 	bool				scissor_enabled;
 	bool				clip_halfz;
 	bool				vs_writes_viewport_index;

src/gallium/drivers/r600/r600_state.c

@@ -3104,6 +3104,7 @@ void r600_init_state_functions(struct r600_context *rctx)
 	r600_add_atom(rctx, &rctx->b.render_cond_atom, id++);
 	r600_add_atom(rctx, &rctx->b.streamout.begin_atom, id++);
 	r600_add_atom(rctx, &rctx->b.streamout.enable_atom, id++);
+	r600_add_atom(rctx, &rctx->b.window_rectangles.atom, id++);
 	for (i = 0; i < R600_NUM_HW_STAGES; i++)
 		r600_init_atom(rctx, &rctx->hw_shader_stages[i].atom, id++, r600_emit_shader, 0);
 	r600_init_atom(rctx, &rctx->shader_stages.atom, id++, r600_emit_shader_stages, 0);

src/gallium/drivers/r600/r600_viewport.c

@@ -6,6 +6,7 @@
 #include "r600_cs.h"
 #include "util/u_viewport.h"
 #include "tgsi/tgsi_scan.h"
+#include "r600d.h"
 
 #define R600_R_028C0C_PA_CL_GB_VERT_CLIP_ADJ         0x028C0C
 #define CM_R_028BE8_PA_CL_GB_VERT_CLIP_ADJ           0x28be8
@@ -425,14 +426,96 @@ void r600_update_vs_writes_viewport_index(struct r600_common_context *rctx,
 	    rctx->set_atom_dirty(rctx, &rctx->viewports.atom, true);
 }
 
+static void r600_emit_window_rectangles(struct r600_common_context *rctx,
+					struct r600_atom *atom)
+{
+	/* 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 = &rctx->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 = rctx->window_rectangles.number;
+	struct pipe_scissor_state *rects = rctx->window_rectangles.states;
+	unsigned rule;
+
+	assert(num_rectangles <= R600_MAX_WINDOW_RECTANGLES);
+
+	if (num_rectangles == 0)
+		rule = disabled;
+	else if (rctx->window_rectangles.include)
+		rule = ~outside[num_rectangles - 1];
+	else
+		rule = outside[num_rectangles - 1];
+
+	radeon_set_context_reg_seq(cs, R_02820C_PA_SC_CLIPRECT_RULE, 1);
+	radeon_emit(cs, 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 r600_set_window_rectangles(struct pipe_context *ctx,
+				       bool include,
+				       unsigned num_rectangles,
+				       const struct pipe_scissor_state *rects)
+{
+	struct r600_common_context *rctx = (struct r600_common_context *)ctx;
+
+	rctx->window_rectangles.number = num_rectangles;
+	rctx->window_rectangles.include = include;
+
+	if (num_rectangles)
+		memcpy(rctx->window_rectangles.states, rects,
+		       sizeof(*rects) * num_rectangles);
+
+	rctx->set_atom_dirty(rctx, &rctx->window_rectangles.atom, true);
+}
+
 void r600_init_viewport_functions(struct r600_common_context *rctx)
 {
 	rctx->scissors.atom.emit = r600_emit_scissors;
 	rctx->viewports.atom.emit = r600_emit_viewport_states;
+	rctx->window_rectangles.atom.emit = r600_emit_window_rectangles;
 
 	rctx->scissors.atom.num_dw = (2 + 16 * 2) + 6;
 	rctx->viewports.atom.num_dw = 2 + 16 * 6;
 
 	rctx->b.set_scissor_states = r600_set_scissor_states;
 	rctx->b.set_viewport_states = r600_set_viewport_states;
+	rctx->b.set_window_rectangles = r600_set_window_rectangles;
 }

src/gallium/drivers/r600/r600d.h

@@ -2849,6 +2849,22 @@
 #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) & 0x3FFF) << 0)
 #define   G_028210_TL_X(x)                             (((x) >> 0) & 0x3FFF)