X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Ffreedreno%2Ffreedreno_resource.c;h=0f48e5878ffa315a3e2c763e6da74a87dacabdb0;hb=f9f5d3eb554e6096b762744125580a239d3809b0;hp=0044b9804b3eee20a87eb7b6f06f9809b057150b;hpb=2e62a622e78d2268fb123e2551690ef9749ef055;p=mesa.git

diff --git a/src/gallium/drivers/freedreno/freedreno_resource.c b/src/gallium/drivers/freedreno/freedreno_resource.c
index 0044b9804b3..0f48e5878ff 100644
--- a/src/gallium/drivers/freedreno/freedreno_resource.c
+++ b/src/gallium/drivers/freedreno/freedreno_resource.c
@@ -132,7 +132,7 @@ realloc_bo(struct fd_resource *rsc, uint32_t size)
 		fd_bo_del(rsc->bo);
 
 	rsc->bo = fd_bo_new(screen->dev, size, flags, "%ux%ux%u@%u:%x",
-			prsc->width0, prsc->height0, prsc->depth0, rsc->cpp, prsc->bind);
+			prsc->width0, prsc->height0, prsc->depth0, rsc->layout.cpp, prsc->bind);
 	rsc->seqno = p_atomic_inc_return(&screen->rsc_seqno);
 	util_range_set_empty(&rsc->valid_buffer_range);
 	fd_bc_invalidate_resource(rsc, true);
@@ -223,10 +223,11 @@ fd_try_shadow_resource(struct fd_context *ctx, struct fd_resource *rsc,
 	/* TODO valid_buffer_range?? */
 	swap(rsc->bo,        shadow->bo);
 	swap(rsc->write_batch,   shadow->write_batch);
-	swap(rsc->offset, shadow->offset);
-	swap(rsc->ubwc_offset, shadow->ubwc_offset);
-	swap(rsc->ubwc_pitch, shadow->ubwc_pitch);
-	swap(rsc->ubwc_size, shadow->ubwc_size);
+	for (int level = 0; level <= prsc->last_level; level++) {
+		swap(rsc->layout.slices[level], shadow->layout.slices[level]);
+		swap(rsc->layout.ubwc_slices[level], shadow->layout.ubwc_slices[level]);
+	}
+	swap(rsc->layout.ubwc_layer_size, shadow->layout.ubwc_layer_size);
 	rsc->seqno = p_atomic_inc_return(&ctx->screen->rsc_seqno);
 
 	/* at this point, the newly created shadow buffer is not referenced
@@ -449,15 +450,14 @@ flush_resource(struct fd_context *ctx, struct fd_resource *rsc, unsigned usage)
 		mtx_unlock(&ctx->screen->lock);
 
 		foreach_batch(batch, &ctx->screen->batch_cache, batch_mask)
-			fd_batch_flush(batch, false);
+			fd_batch_flush(batch);
 
 		foreach_batch(batch, &ctx->screen->batch_cache, batch_mask) {
-			fd_batch_sync(batch);
 			fd_batch_reference(&batches[batch->idx], NULL);
 		}
 		assert(rsc->batch_mask == 0);
 	} else if (write_batch) {
-		fd_batch_flush(write_batch, true);
+		fd_batch_flush(write_batch);
 	}
 
 	fd_batch_reference(&write_batch, NULL);
@@ -530,7 +530,7 @@ fd_resource_transfer_map(struct pipe_context *pctx,
 	ptrans->level = level;
 	ptrans->usage = usage;
 	ptrans->box = *box;
-	ptrans->stride = util_format_get_nblocksx(format, slice->pitch) * rsc->cpp;
+	ptrans->stride = util_format_get_nblocksx(format, slice->pitch) * rsc->layout.cpp;
 	ptrans->layer_stride = fd_resource_layer_stride(rsc, level);
 
 	/* we always need a staging texture for tiled buffers:
@@ -539,7 +539,7 @@ fd_resource_transfer_map(struct pipe_context *pctx,
 	 * splitting a batch.. for ex, mid-frame texture uploads to a tiled
 	 * texture.
 	 */
-	if (rsc->tile_mode) {
+	if (rsc->layout.tile_mode) {
 		struct fd_resource *staging_rsc;
 
 		staging_rsc = fd_alloc_staging(ctx, rsc, level, box);
@@ -549,7 +549,7 @@ fd_resource_transfer_map(struct pipe_context *pctx,
 			// TODO for PIPE_TRANSFER_READ, need to do untiling blit..
 			trans->staging_prsc = &staging_rsc->base;
 			trans->base.stride = util_format_get_nblocksx(format,
-				staging_slice->pitch) * staging_rsc->cpp;
+				staging_slice->pitch) * staging_rsc->layout.cpp;
 			trans->base.layer_stride = fd_resource_layer_stride(staging_rsc, 0);
 			trans->staging_box = *box;
 			trans->staging_box.x = 0;
@@ -559,21 +559,6 @@ fd_resource_transfer_map(struct pipe_context *pctx,
 			if (usage & PIPE_TRANSFER_READ) {
 				fd_blit_to_staging(ctx, trans);
 
-				struct fd_batch *batch = NULL;
-
-				fd_context_lock(ctx);
-				fd_batch_reference_locked(&batch, staging_rsc->write_batch);
-				fd_context_unlock(ctx);
-
-				/* we can't fd_bo_cpu_prep() until the blit to staging
-				 * is submitted to kernel.. in that case write_batch
-				 * wouldn't be NULL yet:
-				 */
-				if (batch) {
-					fd_batch_sync(batch);
-					fd_batch_reference(&batch, NULL);
-				}
-
 				fd_bo_cpu_prep(staging_rsc->bo, ctx->pipe,
 						DRM_FREEDRENO_PREP_READ);
 			}
@@ -665,7 +650,7 @@ fd_resource_transfer_map(struct pipe_context *pctx,
 						fd_resource_slice(staging_rsc, 0);
 					trans->staging_prsc = &staging_rsc->base;
 					trans->base.stride = util_format_get_nblocksx(format,
-						staging_slice->pitch) * staging_rsc->cpp;
+						staging_slice->pitch) * staging_rsc->layout.cpp;
 					trans->base.layer_stride =
 						fd_resource_layer_stride(staging_rsc, 0);
 					trans->staging_box = *box;
@@ -707,7 +692,7 @@ fd_resource_transfer_map(struct pipe_context *pctx,
 	buf = fd_bo_map(rsc->bo);
 	offset =
 		box->y / util_format_get_blockheight(format) * ptrans->stride +
-		box->x / util_format_get_blockwidth(format) * rsc->cpp +
+		box->x / util_format_get_blockwidth(format) * rsc->layout.cpp +
 		fd_resource_offset(rsc, level, box->z);
 
 	if (usage & PIPE_TRANSFER_WRITE)
@@ -740,10 +725,10 @@ fd_resource_destroy(struct pipe_screen *pscreen,
 static uint64_t
 fd_resource_modifier(struct fd_resource *rsc)
 {
-	if (!rsc->tile_mode)
+	if (!rsc->layout.tile_mode)
 		return DRM_FORMAT_MOD_LINEAR;
 
-	if (rsc->ubwc_size)
+	if (rsc->layout.ubwc_layer_size)
 		return DRM_FORMAT_MOD_QCOM_COMPRESSED;
 
 	/* TODO invent a modifier for tiled but not UBWC buffers: */
@@ -762,7 +747,7 @@ fd_resource_get_handle(struct pipe_screen *pscreen,
 	handle->modifier = fd_resource_modifier(rsc);
 
 	return fd_screen_bo_get_handle(pscreen, rsc->bo, rsc->scanout,
-			fd_resource_slice(rsc, 0)->pitch * rsc->cpp, handle);
+			fd_resource_slice(rsc, 0)->pitch * rsc->layout.cpp, handle);
 }
 
 static uint32_t
@@ -779,7 +764,7 @@ setup_slices(struct fd_resource *rsc, uint32_t alignment, enum pipe_format forma
 	/* in layer_first layout, the level (slice) contains just one
 	 * layer (since in fact the layer contains the slices)
 	 */
-	uint32_t layers_in_level = rsc->layer_first ? 1 : prsc->array_size;
+	uint32_t layers_in_level = rsc->layout.layer_first ? 1 : prsc->array_size;
 
 	for (level = 0; level <= prsc->last_level; level++) {
 		struct fdl_slice *slice = fd_resource_slice(rsc, level);
@@ -801,9 +786,9 @@ setup_slices(struct fd_resource *rsc, uint32_t alignment, enum pipe_format forma
 		if (prsc->target == PIPE_TEXTURE_3D && (
 					level == 1 ||
 					(level > 1 && fd_resource_slice(rsc, level - 1)->size0 > 0xf000)))
-			slice->size0 = align(blocks * rsc->cpp, alignment);
-		else if (level == 0 || rsc->layer_first || alignment == 1)
-			slice->size0 = align(blocks * rsc->cpp, alignment);
+			slice->size0 = align(blocks * rsc->layout.cpp, alignment);
+		else if (level == 0 || rsc->layout.layer_first || alignment == 1)
+			slice->size0 = align(blocks * rsc->layout.cpp, alignment);
 		else
 			slice->size0 = fd_resource_slice(rsc, level - 1)->size0;
 
@@ -851,10 +836,10 @@ fd_setup_slices(struct fd_resource *rsc)
 	if (is_a4xx(screen)) {
 		switch (rsc->base.target) {
 		case PIPE_TEXTURE_3D:
-			rsc->layer_first = false;
+			rsc->layout.layer_first = false;
 			break;
 		default:
-			rsc->layer_first = true;
+			rsc->layout.layer_first = true;
 			alignment = 1;
 			break;
 		}
@@ -877,23 +862,18 @@ fd_resource_resize(struct pipe_resource *prsc, uint32_t sz)
 	realloc_bo(rsc, fd_screen(prsc->screen)->setup_slices(rsc));
 }
 
-// TODO common helper?
-static bool
-has_depth(enum pipe_format format)
+static void
+fd_resource_layout_init(struct pipe_resource *prsc)
 {
-	switch (format) {
-	case PIPE_FORMAT_Z16_UNORM:
-	case PIPE_FORMAT_Z32_UNORM:
-	case PIPE_FORMAT_Z32_FLOAT:
-	case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
-	case PIPE_FORMAT_Z24_UNORM_S8_UINT:
-	case PIPE_FORMAT_S8_UINT_Z24_UNORM:
-	case PIPE_FORMAT_Z24X8_UNORM:
-	case PIPE_FORMAT_X8Z24_UNORM:
-		return true;
-	default:
-		return false;
-	}
+	struct fd_resource *rsc = fd_resource(prsc);
+	struct fdl_layout *layout = &rsc->layout;
+
+	layout->width0 = prsc->width0;
+	layout->height0 = prsc->height0;
+	layout->depth0 = prsc->depth0;
+
+	layout->cpp = util_format_get_blocksize(prsc->format);
+	layout->cpp *= fd_resource_nr_samples(prsc);
 }
 
 /**
@@ -956,6 +936,7 @@ fd_resource_create_with_modifiers(struct pipe_screen *pscreen,
 		return NULL;
 
 	*prsc = *tmpl;
+	fd_resource_layout_init(prsc);
 
 #define LINEAR \
 	(PIPE_BIND_SCANOUT | \
@@ -966,6 +947,9 @@ fd_resource_create_with_modifiers(struct pipe_screen *pscreen,
 	if (tmpl->bind & LINEAR)
 		linear = true;
 
+	if (fd_mesa_debug & FD_DBG_NOTILE)
+		linear = true;
+
 	/* Normally, for non-shared buffers, allow buffer compression if
 	 * not shared, otherwise only allow if QCOM_COMPRESSED modifier
 	 * is requested:
@@ -987,46 +971,22 @@ fd_resource_create_with_modifiers(struct pipe_screen *pscreen,
 	if (screen->tile_mode &&
 			(tmpl->target != PIPE_BUFFER) &&
 			!linear) {
-		rsc->tile_mode = screen->tile_mode(prsc);
+		rsc->layout.tile_mode = screen->tile_mode(prsc);
 	}
 
 	util_range_init(&rsc->valid_buffer_range);
 
 	rsc->internal_format = format;
-	rsc->cpp = util_format_get_blocksize(format);
-	rsc->cpp *= fd_resource_nr_samples(prsc);
-
-	assert(rsc->cpp);
-
-	// XXX probably need some extra work if we hit rsc shadowing path w/ lrz..
-	if ((is_a5xx(screen) || is_a6xx(screen)) &&
-		 (fd_mesa_debug & FD_DBG_LRZ) && has_depth(format)) {
-		const uint32_t flags = DRM_FREEDRENO_GEM_CACHE_WCOMBINE |
-				DRM_FREEDRENO_GEM_TYPE_KMEM; /* TODO */
-		unsigned lrz_pitch  = align(DIV_ROUND_UP(tmpl->width0, 8), 64);
-		unsigned lrz_height = DIV_ROUND_UP(tmpl->height0, 8);
-
-		/* LRZ buffer is super-sampled: */
-		switch (prsc->nr_samples) {
-		case 4:
-			lrz_pitch *= 2;
-		case 2:
-			lrz_height *= 2;
-		}
 
-		unsigned size = lrz_pitch * lrz_height * 2;
-
-		size += 0x1000; /* for GRAS_LRZ_FAST_CLEAR_BUFFER */
-
-		rsc->lrz_height = lrz_height;
-		rsc->lrz_width = lrz_pitch;
-		rsc->lrz_pitch = lrz_pitch;
-		rsc->lrz = fd_bo_new(screen->dev, size, flags, "lrz");
+	if (prsc->target == PIPE_BUFFER) {
+		assert(prsc->format == PIPE_FORMAT_R8_UNORM);
+		size = prsc->width0;
+		fdl_layout_buffer(&rsc->layout, size);
+	} else {
+		size = screen->setup_slices(rsc);
 	}
 
-	size = screen->setup_slices(rsc);
-
-	if (allow_ubwc && screen->fill_ubwc_buffer_sizes && rsc->tile_mode)
+	if (allow_ubwc && screen->fill_ubwc_buffer_sizes && rsc->layout.tile_mode)
 		size += screen->fill_ubwc_buffer_sizes(rsc);
 
 	/* special case for hw-query buffer, which we need to allocate before we
@@ -1038,11 +998,15 @@ fd_resource_create_with_modifiers(struct pipe_screen *pscreen,
 		return prsc;
 	}
 
-	if (rsc->layer_first) {
-		rsc->layer_size = align(size, 4096);
-		size = rsc->layer_size * prsc->array_size;
+	/* Set the layer size if the (non-a6xx) backend hasn't done so. */
+	if (rsc->layout.layer_first && !rsc->layout.layer_size) {
+		rsc->layout.layer_size = align(size, 4096);
+		size = rsc->layout.layer_size * prsc->array_size;
 	}
 
+	if (fd_mesa_debug & FD_DBG_LAYOUT)
+		fdl_dump_layout(&rsc->layout);
+
 	realloc_bo(rsc, size);
 	if (!rsc->bo)
 		goto fail;
@@ -1108,6 +1072,7 @@ fd_resource_from_handle(struct pipe_screen *pscreen,
 		return NULL;
 
 	*prsc = *tmpl;
+	fd_resource_layout_init(prsc);
 
 	pipe_reference_init(&prsc->reference, 1);
 
@@ -1120,9 +1085,7 @@ fd_resource_from_handle(struct pipe_screen *pscreen,
 		goto fail;
 
 	rsc->internal_format = tmpl->format;
-	rsc->cpp = util_format_get_blocksize(tmpl->format);
-	rsc->cpp *= fd_resource_nr_samples(prsc);
-	slice->pitch = handle->stride / rsc->cpp;
+	slice->pitch = handle->stride / rsc->layout.cpp;
 	slice->offset = handle->offset;
 	slice->size0 = handle->stride * prsc->height0;
 
@@ -1143,7 +1106,7 @@ fd_resource_from_handle(struct pipe_screen *pscreen,
 		goto fail;
 	}
 
-	assert(rsc->cpp);
+	assert(rsc->layout.cpp);
 
 	if (screen->ro) {
 		rsc->scanout =