X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Ffreedreno%2Ffreedreno_gmem.c;h=648db9baee5f6b180b5dc9455ba0bd2142a14acc;hb=aab3912f21508b0681962c68fdaca1435c06b2ea;hp=7f6c8476cdb1a8614cd477686d4ff02c75449580;hpb=01b757e2b0fb97a146b0ef278b449cecab0d15e8;p=mesa.git

diff --git a/src/gallium/drivers/freedreno/freedreno_gmem.c b/src/gallium/drivers/freedreno/freedreno_gmem.c
index 7f6c8476cdb..648db9baee5 100644
--- a/src/gallium/drivers/freedreno/freedreno_gmem.c
+++ b/src/gallium/drivers/freedreno/freedreno_gmem.c
@@ -69,11 +69,39 @@
 
 static uint32_t bin_width(struct fd_context *ctx)
 {
-	if (ctx->screen->gpu_id >= 300)
+	if (is_a4xx(ctx->screen))
+		return 1024;
+	if (is_a3xx(ctx->screen))
 		return 992;
 	return 512;
 }
 
+static uint32_t
+total_size(uint8_t cbuf_cpp[], uint8_t zsbuf_cpp[2],
+		   uint32_t bin_w, uint32_t bin_h, struct fd_gmem_stateobj *gmem)
+{
+	uint32_t total = 0, i;
+
+	for (i = 0; i < MAX_RENDER_TARGETS; i++) {
+		if (cbuf_cpp[i]) {
+			gmem->cbuf_base[i] = align(total, 0x4000);
+			total = gmem->cbuf_base[i] + cbuf_cpp[i] * bin_w * bin_h;
+		}
+	}
+
+	if (zsbuf_cpp[0]) {
+		gmem->zsbuf_base[0] = align(total, 0x4000);
+		total = gmem->zsbuf_base[0] + zsbuf_cpp[0] * bin_w * bin_h;
+	}
+
+	if (zsbuf_cpp[1]) {
+		gmem->zsbuf_base[1] = align(total, 0x4000);
+		total = gmem->zsbuf_base[1] + zsbuf_cpp[1] * bin_w * bin_h;
+	}
+
+	return total;
+}
+
 static void
 calculate_tiles(struct fd_context *ctx)
 {
@@ -85,27 +113,33 @@ calculate_tiles(struct fd_context *ctx)
 	uint32_t nbins_x = 1, nbins_y = 1;
 	uint32_t bin_w, bin_h;
 	uint32_t max_width = bin_width(ctx);
-	uint32_t cpp = 4;
+	uint8_t cbuf_cpp[MAX_RENDER_TARGETS] = {0}, zsbuf_cpp[2] = {0};
 	uint32_t i, j, t, xoff, yoff;
 	uint32_t tpp_x, tpp_y;
 	bool has_zs = !!(ctx->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL));
+	int tile_n[ARRAY_SIZE(ctx->pipe)];
 
-	if (pfb->cbufs[0])
-		cpp = util_format_get_blocksize(pfb->cbufs[0]->format);
+	if (has_zs) {
+		struct fd_resource *rsc = fd_resource(pfb->zsbuf->texture);
+		zsbuf_cpp[0] = rsc->cpp;
+		if (rsc->stencil)
+			zsbuf_cpp[1] = rsc->stencil->cpp;
+	}
+	for (i = 0; i < pfb->nr_cbufs; i++) {
+		if (pfb->cbufs[i])
+			cbuf_cpp[i] = util_format_get_blocksize(pfb->cbufs[i]->format);
+		else
+			cbuf_cpp[i] = 4;
+	}
 
-	if ((gmem->cpp == cpp) && (gmem->has_zs == has_zs) &&
-			!memcmp(&gmem->scissor, scissor, sizeof(gmem->scissor))) {
+	if (!memcmp(gmem->zsbuf_cpp, zsbuf_cpp, sizeof(zsbuf_cpp)) &&
+		!memcmp(gmem->cbuf_cpp, cbuf_cpp, sizeof(cbuf_cpp)) &&
+		!memcmp(&gmem->scissor, scissor, sizeof(gmem->scissor))) {
 		/* everything is up-to-date */
 		return;
 	}
 
-	/* if have depth/stencil, we need to leave room: */
-	if (has_zs) {
-		gmem_size /= 2;
-		max_width /= 2;
-	}
-
-	if (fd_mesa_debug & FD_DBG_DSCIS) {
+	if (fd_mesa_debug & FD_DBG_NOSCIS) {
 		minx = 0;
 		miny = 0;
 		width = pfb->width;
@@ -128,10 +162,18 @@ calculate_tiles(struct fd_context *ctx)
 		bin_w = align(width / nbins_x, 32);
 	}
 
+	if (fd_mesa_debug & FD_DBG_MSGS) {
+		debug_printf("binning input: cbuf cpp:");
+		for (i = 0; i < pfb->nr_cbufs; i++)
+			debug_printf(" %d", cbuf_cpp[i]);
+		debug_printf(", zsbuf cpp: %d; %dx%d\n",
+				zsbuf_cpp[0], width, height);
+	}
+
 	/* then find a bin width/height that satisfies the memory
 	 * constraints:
 	 */
-	while ((bin_w * bin_h * cpp) > gmem_size) {
+	while (total_size(cbuf_cpp, zsbuf_cpp, bin_w, bin_h, gmem) > gmem_size) {
 		if (bin_w > bin_h) {
 			nbins_x++;
 			bin_w = align(width / nbins_x, 32);
@@ -144,8 +186,8 @@ calculate_tiles(struct fd_context *ctx)
 	DBG("using %d bins of size %dx%d", nbins_x*nbins_y, bin_w, bin_h);
 
 	gmem->scissor = *scissor;
-	gmem->cpp = cpp;
-	gmem->has_zs = has_zs;
+	memcpy(gmem->cbuf_cpp, cbuf_cpp, sizeof(cbuf_cpp));
+	memcpy(gmem->zsbuf_cpp, zsbuf_cpp, sizeof(zsbuf_cpp));
 	gmem->bin_h = bin_h;
 	gmem->bin_w = bin_w;
 	gmem->nbins_x = nbins_x;
@@ -211,6 +253,7 @@ calculate_tiles(struct fd_context *ctx)
 	/* configure tiles: */
 	t = 0;
 	yoff = miny;
+	memset(tile_n, 0, sizeof(tile_n));
 	for (i = 0; i < nbins_y; i++) {
 		uint32_t bw, bh;
 
@@ -221,20 +264,17 @@ calculate_tiles(struct fd_context *ctx)
 
 		for (j = 0; j < nbins_x; j++) {
 			struct fd_tile *tile = &ctx->tile[t];
-			uint32_t n, p;
+			uint32_t p;
 
 			assert(t < ARRAY_SIZE(ctx->tile));
 
 			/* pipe number: */
 			p = ((i / tpp_y) * div_round_up(nbins_x, tpp_x)) + (j / tpp_x);
 
-			/* slot number: */
-			n = ((i % tpp_y) * tpp_x) + (j % tpp_x);
-
 			/* clip bin width: */
 			bw = MIN2(bin_w, minx + width - xoff);
 
-			tile->n = n;
+			tile->n = tile_n[p]++;
 			tile->p = p;
 			tile->bin_w = bw;
 			tile->bin_h = bh;
@@ -317,14 +357,13 @@ void
 fd_gmem_render_tiles(struct fd_context *ctx)
 {
 	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
-	uint32_t timestamp = 0;
 	bool sysmem = false;
 
 	if (ctx->emit_sysmem_prep) {
 		if (ctx->cleared || ctx->gmem_reason || (ctx->num_draws > 5)) {
 			DBG("GMEM: cleared=%x, gmem_reason=%x, num_draws=%u",
 				ctx->cleared, ctx->gmem_reason, ctx->num_draws);
-		} else if (!(fd_mesa_debug & FD_DBG_DBYPASS)) {
+		} else if (!(fd_mesa_debug & FD_DBG_NOBYPASS)) {
 			sysmem = true;
 		}
 	}
@@ -369,13 +408,6 @@ fd_gmem_render_tiles(struct fd_context *ctx)
 
 	fd_reset_wfi(ctx);
 
-	/* update timestamps on render targets: */
-	timestamp = fd_ringbuffer_timestamp(ctx->ring);
-	if (pfb->cbufs[0])
-		fd_resource(pfb->cbufs[0]->texture)->timestamp = timestamp;
-	if (pfb->zsbuf)
-		fd_resource(pfb->zsbuf->texture)->timestamp = timestamp;
-
 	/* reset maximal bounds: */
 	ctx->max_scissor.minx = ctx->max_scissor.miny = ~0;
 	ctx->max_scissor.maxx = ctx->max_scissor.maxy = 0;