r600g/sb: fix debug dump code in scheduler
[mesa.git] / src / gallium / drivers / r600 / r600_texture.c
index 98cb11897635f4b63ef666308c164317d2a6aa8f..36cca17eb110ae0be0f6d4fbfcfa951f9f0a9165 100644 (file)
 #include "util/u_memory.h"
 
 
+/* Same as resource_copy_region, except that both upsampling and downsampling are allowed. */
+static void r600_copy_region_with_blit(struct pipe_context *pipe,
+                                      struct pipe_resource *dst,
+                                       unsigned dst_level,
+                                       unsigned dstx, unsigned dsty, unsigned dstz,
+                                       struct pipe_resource *src,
+                                       unsigned src_level,
+                                       const struct pipe_box *src_box)
+{
+       struct pipe_blit_info blit;
+
+       memset(&blit, 0, sizeof(blit));
+       blit.src.resource = src;
+       blit.src.format = src->format;
+       blit.src.level = src_level;
+       blit.src.box = *src_box;
+       blit.dst.resource = dst;
+       blit.dst.format = dst->format;
+       blit.dst.level = dst_level;
+       blit.dst.box.x = dstx;
+       blit.dst.box.y = dsty;
+       blit.dst.box.z = dstz;
+       blit.dst.box.width = src_box->width;
+       blit.dst.box.height = src_box->height;
+       blit.dst.box.depth = src_box->depth;
+       blit.mask = util_format_get_mask(src->format) &
+                   util_format_get_mask(dst->format);
+       blit.filter = PIPE_TEX_FILTER_NEAREST;
+
+       if (blit.mask) {
+               pipe->blit(pipe, &blit);
+       }
+}
+
 /* Copy from a full GPU texture to a transfer's staging one. */
 static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer)
 {
@@ -41,32 +75,17 @@ static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_t
        struct pipe_resource *dst = &rtransfer->staging->b.b;
        struct pipe_resource *src = transfer->resource;
 
-       if (src->nr_samples <= 1) {
-               if (!rctx->screen->dma_blit(ctx, dst, 0, 0, 0, 0,
-                                           src, transfer->level,
-                                           &transfer->box)) {
-                       /* async dma could not be use */
-                       ctx->resource_copy_region(ctx, dst, 0, 0, 0, 0,
-                                                 src, transfer->level, &transfer->box);
-               }
-       } else {
-               /* Resolve the resource. */
-               struct pipe_blit_info blit;
-
-               memset(&blit, 0, sizeof(blit));
-               blit.src.resource = src;
-               blit.src.format = src->format;
-               blit.src.level = transfer->level;
-               blit.src.box = transfer->box;
-               blit.dst.resource = dst;
-               blit.dst.format = dst->format;
-               blit.dst.box.width = transfer->box.width;
-               blit.dst.box.height = transfer->box.height;
-               blit.dst.box.depth = transfer->box.depth;
-               blit.mask = PIPE_MASK_RGBA;
-               blit.filter = PIPE_TEX_FILTER_NEAREST;
-
-               ctx->blit(ctx, &blit);
+       if (src->nr_samples > 1) {
+               r600_copy_region_with_blit(ctx, dst, 0, 0, 0, 0,
+                                          src, transfer->level, &transfer->box);
+               return;
+       }
+
+       if (!rctx->screen->dma_blit(ctx, dst, 0, 0, 0, 0,
+                                   src, transfer->level,
+                                   &transfer->box)) {
+               ctx->resource_copy_region(ctx, dst, 0, 0, 0, 0,
+                                         src, transfer->level, &transfer->box);
        }
 }
 
@@ -75,27 +94,37 @@ static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600
 {
        struct r600_context *rctx = (struct r600_context*)ctx;
        struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer;
-       struct pipe_resource *texture = transfer->resource;
+       struct pipe_resource *dst = transfer->resource;
+       struct pipe_resource *src = &rtransfer->staging->b.b;
        struct pipe_box sbox;
 
        u_box_3d(0, 0, 0, transfer->box.width, transfer->box.height, transfer->box.depth, &sbox);
 
-       if (!rctx->screen->dma_blit(ctx, texture, transfer->level,
+       if (dst->nr_samples > 1) {
+               r600_copy_region_with_blit(ctx, dst, transfer->level,
+                                          transfer->box.x, transfer->box.y, transfer->box.z,
+                                          src, 0, &sbox);
+               return;
+       }
+
+       if (!rctx->screen->dma_blit(ctx, dst, transfer->level,
                                    transfer->box.x, transfer->box.y, transfer->box.z,
-                                   &rtransfer->staging->b.b, 0, &sbox)) {
-               /* async dma could not be use */
-               ctx->resource_copy_region(ctx, texture, transfer->level,
+                                   src, 0, &sbox)) {
+               ctx->resource_copy_region(ctx, dst, transfer->level,
                                          transfer->box.x, transfer->box.y, transfer->box.z,
-                                         &rtransfer->staging->b.b,
-                                         0, &sbox);
+                                         src, 0, &sbox);
        }
 }
 
-unsigned r600_texture_get_offset(struct r600_texture *rtex,
-                                       unsigned level, unsigned layer)
+static unsigned r600_texture_get_offset(struct r600_texture *rtex, unsigned level,
+                                       const struct pipe_box *box)
 {
+       enum pipe_format format = rtex->resource.b.b.format;
+
        return rtex->surface.level[level].offset +
-              layer * rtex->surface.level[level].slice_size;
+              box->z * rtex->surface.level[level].slice_size +
+              box->y / util_format_get_blockheight(format) * rtex->surface.level[level].pitch_bytes +
+              box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
 }
 
 static int r600_init_surface(struct r600_screen *rscreen,
@@ -284,50 +313,51 @@ void r600_texture_get_fmask_info(struct r600_screen *rscreen,
                                 unsigned nr_samples,
                                 struct r600_fmask_info *out)
 {
-       /* FMASK is allocated pretty much like an ordinary texture.
-        * Here we use bpe in the units of bits, not bytes. */
+       /* FMASK is allocated like an ordinary texture. */
        struct radeon_surface fmask = rtex->surface;
 
+       memset(out, 0, sizeof(*out));
+
+       fmask.bo_alignment = 0;
+       fmask.bo_size = 0;
+       fmask.nsamples = 1;
+       fmask.flags |= RADEON_SURF_FMASK;
+
        switch (nr_samples) {
        case 2:
-               /* This should be 8,1, but we should set nsamples > 1
-                * for the allocator to treat it as a multisample surface.
-                * Let's set 4,2 then. */
        case 4:
-               fmask.bpe = 4;
-               fmask.nsamples = 2;
+               fmask.bpe = 1;
+               fmask.bankh = 4;
                break;
        case 8:
-               fmask.bpe = 8;
-               fmask.nsamples = 4;
-               break;
-       case 16:
-               fmask.bpe = 16;
-               fmask.nsamples = 4;
+               fmask.bpe = 4;
                break;
        default:
                R600_ERR("Invalid sample count for FMASK allocation.\n");
                return;
        }
 
-       /* R600-R700 errata? Anyway, this fixes colorbuffer corruption. */
+       /* Overallocate FMASK on R600-R700 to fix colorbuffer corruption.
+        * This can be fixed by writing a separate FMASK allocator specifically
+        * for R600-R700 asics. */
        if (rscreen->chip_class <= R700) {
                fmask.bpe *= 2;
        }
 
-       if (rscreen->chip_class >= EVERGREEN) {
-               fmask.bankh = nr_samples <= 4 ? 4 : 1;
-       }
-
        if (rscreen->ws->surface_init(rscreen->ws, &fmask)) {
                R600_ERR("Got error in surface_init while allocating FMASK.\n");
                return;
        }
+
        assert(fmask.level[0].mode == RADEON_SURF_MODE_2D);
 
+       out->slice_tile_max = (fmask.level[0].nblk_x * fmask.level[0].nblk_y) / 64;
+       if (out->slice_tile_max)
+               out->slice_tile_max -= 1;
+
        out->bank_height = fmask.bankh;
        out->alignment = MAX2(256, fmask.bo_alignment);
-       out->size = (fmask.bo_size + 7) / 8;
+       out->size = fmask.bo_size;
 }
 
 static void r600_texture_allocate_fmask(struct r600_screen *rscreen,
@@ -339,6 +369,7 @@ static void r600_texture_allocate_fmask(struct r600_screen *rscreen,
                                    rtex->resource.b.b.nr_samples, &fmask);
 
        rtex->fmask_bank_height = fmask.bank_height;
+       rtex->fmask_slice_tile_max = fmask.slice_tile_max;
        rtex->fmask_offset = align(rtex->size, fmask.alignment);
        rtex->fmask_size = fmask.size;
        rtex->size = rtex->fmask_offset + rtex->fmask_size;
@@ -480,10 +511,7 @@ r600_texture_create_object(struct pipe_screen *screen,
                         */
                        R600_ERR("r600: failed to create bo for htile buffers\n");
                } else {
-                       void *ptr;
-                       ptr = rscreen->ws->buffer_map(rtex->htile->cs_buf, NULL, PIPE_TRANSFER_WRITE);
-                       memset(ptr, 0x0, htile_size);
-                       rscreen->ws->buffer_unmap(rtex->htile->cs_buf);
+                       r600_screen_clear_buffer(rscreen, &rtex->htile->b.b, 0, htile_size, 0);
                }
        }
 
@@ -505,9 +533,8 @@ r600_texture_create_object(struct pipe_screen *screen,
 
        if (rtex->cmask_size) {
                /* Initialize the cmask to 0xCC (= compressed state). */
-               char *ptr = rscreen->ws->buffer_map(resource->cs_buf, NULL, PIPE_TRANSFER_WRITE);
-               memset(ptr + rtex->cmask_offset, 0xCC, rtex->cmask_size);
-               rscreen->ws->buffer_unmap(resource->cs_buf);
+               r600_screen_clear_buffer(rscreen, &rtex->resource.b.b,
+                                        rtex->cmask_offset, rtex->cmask_size, 0xCC);
        }
 
        if (rscreen->debug_flags & DBG_VM) {
@@ -733,6 +760,44 @@ bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
        return true;
 }
 
+/**
+ * Initialize the pipe_resource descriptor to be of the same size as the box,
+ * which is supposed to hold a subregion of the texture "orig" at the given
+ * mipmap level.
+ */
+static void r600_init_temp_resource_from_box(struct pipe_resource *res,
+                                            struct pipe_resource *orig,
+                                            const struct pipe_box *box,
+                                            unsigned level, unsigned flags)
+{
+       memset(res, 0, sizeof(*res));
+       res->format = orig->format;
+       res->width0 = box->width;
+       res->height0 = box->height;
+       res->depth0 = 1;
+       res->array_size = 1;
+       res->usage = flags & R600_RESOURCE_FLAG_TRANSFER ? PIPE_USAGE_STAGING : PIPE_USAGE_STATIC;
+       res->flags = flags;
+
+       /* We must set the correct texture target and dimensions for a 3D box. */
+       if (box->depth > 1 && util_max_layer(orig, level) > 0)
+               res->target = orig->target;
+       else
+               res->target = PIPE_TEXTURE_2D;
+
+       switch (res->target) {
+       case PIPE_TEXTURE_1D_ARRAY:
+       case PIPE_TEXTURE_2D_ARRAY:
+       case PIPE_TEXTURE_CUBE_ARRAY:
+               res->array_size = box->depth;
+               break;
+       case PIPE_TEXTURE_3D:
+               res->depth0 = box->depth;
+               break;
+       default:;
+       }
+}
+
 static void *r600_texture_transfer_map(struct pipe_context *ctx,
                                       struct pipe_resource *texture,
                                       unsigned level,
@@ -744,7 +809,6 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
        struct r600_texture *rtex = (struct r600_texture*)texture;
        struct r600_transfer *trans;
        boolean use_staging_texture = FALSE;
-       enum pipe_format format = texture->format;
        struct r600_resource *buf;
        unsigned offset = 0;
        char *map;
@@ -786,66 +850,66 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
        trans->transfer.level = level;
        trans->transfer.usage = usage;
        trans->transfer.box = *box;
+
        if (rtex->is_depth) {
-               /* XXX: only readback the rectangle which is being mapped?
-               */
-               /* XXX: when discard is true, no need to read back from depth texture
-               */
                struct r600_texture *staging_depth;
 
-               assert(rtex->resource.b.b.nr_samples <= 1);
                if (rtex->resource.b.b.nr_samples > 1) {
-                       R600_ERR("mapping MSAA zbuffer unimplemented\n");
-                       FREE(trans);
-                       return NULL;
-               }
+                       /* MSAA depth buffers need to be converted to single sample buffers.
+                        *
+                        * Mapping MSAA depth buffers can occur if ReadPixels is called
+                        * with a multisample GLX visual.
+                        *
+                        * First downsample the depth buffer to a temporary texture,
+                        * then decompress the temporary one to staging.
+                        *
+                        * Only the region being mapped is transfered.
+                        */
+                       struct pipe_resource resource;
 
-               if (!r600_init_flushed_depth_texture(ctx, texture, &staging_depth)) {
-                       R600_ERR("failed to create temporary texture to hold untiled copy\n");
-                       FREE(trans);
-                       return NULL;
+                       r600_init_temp_resource_from_box(&resource, texture, box, level, 0);
+
+                       if (!r600_init_flushed_depth_texture(ctx, &resource, &staging_depth)) {
+                               R600_ERR("failed to create temporary texture to hold untiled copy\n");
+                               FREE(trans);
+                               return NULL;
+                       }
+
+                       if (usage & PIPE_TRANSFER_READ) {
+                               struct pipe_resource *temp = ctx->screen->resource_create(ctx->screen, &resource);
+
+                               r600_copy_region_with_blit(ctx, temp, 0, 0, 0, 0, texture, level, box);
+                               r600_blit_decompress_depth(ctx, (struct r600_texture*)temp, staging_depth,
+                                                          0, 0, 0, box->depth, 0, 0);
+                               pipe_resource_reference((struct pipe_resource**)&temp, NULL);
+                       }
                }
+               else {
+                       /* XXX: only readback the rectangle which is being mapped? */
+                       /* XXX: when discard is true, no need to read back from depth texture */
+                       if (!r600_init_flushed_depth_texture(ctx, texture, &staging_depth)) {
+                               R600_ERR("failed to create temporary texture to hold untiled copy\n");
+                               FREE(trans);
+                               return NULL;
+                       }
+
+                       r600_blit_decompress_depth(ctx, rtex, staging_depth,
+                                                  level, level,
+                                                  box->z, box->z + box->depth - 1,
+                                                  0, 0);
 
-               r600_blit_decompress_depth(ctx, rtex, staging_depth,
-                                          level, level,
-                                          box->z, box->z + box->depth - 1,
-                                          0, 0);
+                       offset = r600_texture_get_offset(staging_depth, level, box);
+               }
 
                trans->transfer.stride = staging_depth->surface.level[level].pitch_bytes;
                 trans->transfer.layer_stride = staging_depth->surface.level[level].slice_size;
-               trans->offset = r600_texture_get_offset(staging_depth, level, box->z);
                trans->staging = (struct r600_resource*)staging_depth;
        } else if (use_staging_texture) {
                struct pipe_resource resource;
                struct r600_texture *staging;
 
-               memset(&resource, 0, sizeof(resource));
-               resource.format = texture->format;
-               resource.width0 = box->width;
-               resource.height0 = box->height;
-               resource.depth0 = 1;
-               resource.array_size = 1;
-               resource.usage = PIPE_USAGE_STAGING;
-               resource.flags = R600_RESOURCE_FLAG_TRANSFER;
-
-               /* We must set the correct texture target and dimensions if needed for a 3D transfer. */
-               if (box->depth > 1 && util_max_layer(texture, level) > 0)
-                       resource.target = texture->target;
-               else
-                       resource.target = PIPE_TEXTURE_2D;
-
-               switch (resource.target) {
-               case PIPE_TEXTURE_1D_ARRAY:
-               case PIPE_TEXTURE_2D_ARRAY:
-               case PIPE_TEXTURE_CUBE_ARRAY:
-                       resource.array_size = box->depth;
-                       break;
-               case PIPE_TEXTURE_3D:
-                       resource.depth0 = box->depth;
-                       break;
-               default:;
-               }
-
+               r600_init_temp_resource_from_box(&resource, texture, box, level,
+                                                R600_RESOURCE_FLAG_TRANSFER);
 
                /* Create the temporary texture. */
                staging = (struct r600_texture*)ctx->screen->resource_create(ctx->screen, &resource);
@@ -859,16 +923,12 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
                trans->transfer.layer_stride = staging->surface.level[0].slice_size;
                if (usage & PIPE_TRANSFER_READ) {
                        r600_copy_to_staging_texture(ctx, trans);
-                       /* flush gfx & dma ring, order does not matter as only one can be live */
-                       if (rctx->rings.dma.cs) {
-                               rctx->rings.dma.flush(rctx, 0);
-                       }
-                       rctx->rings.gfx.flush(rctx, 0);
                }
        } else {
+               /* the resource is mapped directly */
                trans->transfer.stride = rtex->surface.level[level].pitch_bytes;
                trans->transfer.layer_stride = rtex->surface.level[level].slice_size;
-               trans->offset = r600_texture_get_offset(rtex, level, box->z);
+               offset = r600_texture_get_offset(rtex, level, box);
        }
 
        if (trans->staging) {
@@ -877,11 +937,6 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx,
                buf = &rtex->resource;
        }
 
-       if (rtex->is_depth || !trans->staging)
-               offset = trans->offset +
-                       box->y / util_format_get_blockheight(format) * trans->transfer.stride +
-                       box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
-
        if (!(map = r600_buffer_mmap_sync_with_rings(rctx, buf, usage))) {
                pipe_resource_reference((struct pipe_resource**)&trans->staging, NULL);
                FREE(trans);
@@ -913,7 +968,7 @@ static void r600_texture_transfer_unmap(struct pipe_context *ctx,
        rctx->ws->buffer_unmap(buf);
 
        if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtransfer->staging) {
-               if (rtex->is_depth) {
+               if (rtex->is_depth && rtex->resource.b.b.nr_samples <= 1) {
                        ctx->resource_copy_region(ctx, texture, transfer->level,
                                                  transfer->box.x, transfer->box.y, transfer->box.z,
                                                  &rtransfer->staging->b.b, transfer->level,