r600g: put depth and stencil into one backing buffer
[mesa.git] / src / gallium / drivers / r600 / r600_texture.c
index 86de005031b7ce52b0e18480a2948b1d86dea547..f0cf1f593d84289f43089202a3a3cfac6f9db00f 100644 (file)
 #include <util/u_math.h>
 #include <util/u_inlines.h>
 #include <util/u_memory.h>
-#include "state_tracker/drm_driver.h"
 #include "pipebuffer/pb_buffer.h"
 #include "r600_pipe.h"
 #include "r600_resource.h"
-#include "r600_state_inlines.h"
 #include "r600d.h"
 #include "r600_formats.h"
 
@@ -68,7 +66,7 @@ static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600
                                  rtransfer->staging_texture,
                                  0, &sbox);
 
-       ctx->flush(ctx, 0, NULL);
+       r600_flush(ctx, NULL, RADEON_FLUSH_ASYNC);
 }
 
 unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
@@ -79,10 +77,8 @@ unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
        switch (rtex->resource.b.b.b.target) {
        case PIPE_TEXTURE_3D:
        case PIPE_TEXTURE_CUBE:
-               return offset + layer * rtex->layer_size[level];
        default:
-               assert(layer == 0);
-               return offset;
+               return offset + layer * rtex->layer_size[level];
        }
 }
 
@@ -177,15 +173,15 @@ static unsigned r600_texture_get_nblocksx(struct pipe_screen *screen,
 {
        struct pipe_resource *ptex = &rtex->resource.b.b.b;
        unsigned nblocksx, block_align, width;
-       unsigned blocksize = util_format_get_blocksize(ptex->format);
+       unsigned blocksize = util_format_get_blocksize(rtex->real_format);
 
        if (rtex->pitch_override)
                return rtex->pitch_override / blocksize;
 
        width = mip_minify(ptex->width0, level);
-       nblocksx = util_format_get_nblocksx(ptex->format, width);
+       nblocksx = util_format_get_nblocksx(rtex->real_format, width);
 
-       block_align = r600_get_block_alignment(screen, ptex->format,
+       block_align = r600_get_block_alignment(screen, rtex->real_format,
                                              rtex->array_mode[level]);
        nblocksx = align(nblocksx, block_align);
        return nblocksx;
@@ -199,7 +195,7 @@ static unsigned r600_texture_get_nblocksy(struct pipe_screen *screen,
        unsigned height, tile_height;
 
        height = mip_minify(ptex->height0, level);
-       height = util_format_get_nblocksy(ptex->format, height);
+       height = util_format_get_nblocksy(rtex->real_format, height);
        tile_height = r600_get_height_alignment(screen,
                                                rtex->array_mode[level]);
        height = align(height, tile_height);
@@ -224,11 +220,11 @@ static void r600_texture_set_array_mode(struct pipe_screen *screen,
                unsigned w, h, tile_height, tile_width;
 
                tile_height = r600_get_height_alignment(screen, array_mode);
-               tile_width = r600_get_block_alignment(screen, ptex->format, array_mode);
+               tile_width = r600_get_block_alignment(screen, rtex->real_format, array_mode);
 
                w = mip_minify(ptex->width0, level);
                h = mip_minify(ptex->height0, level);
-               if (w < tile_width || h < tile_height)
+               if (w <= tile_width || h <= tile_height)
                        rtex->array_mode[level] = V_0280A0_ARRAY_1D_TILED_THIN1;
                else
                        rtex->array_mode[level] = array_mode;
@@ -242,13 +238,14 @@ static void r600_setup_miptree(struct pipe_screen *screen,
                               unsigned array_mode)
 {
        struct pipe_resource *ptex = &rtex->resource.b.b.b;
-       struct radeon *radeon = (struct radeon *)screen->winsys;
+       struct radeon *radeon = ((struct r600_screen*)screen)->radeon;
        enum chip_class chipc = r600_get_family_class(radeon);
        unsigned size, layer_size, i, offset;
        unsigned nblocksx, nblocksy;
 
        for (i = 0, offset = 0; i <= ptex->last_level; i++) {
-               unsigned blocksize = util_format_get_blocksize(ptex->format);
+               unsigned blocksize = util_format_get_blocksize(rtex->real_format);
+               unsigned base_align = r600_get_base_alignment(screen, rtex->real_format, array_mode);
 
                r600_texture_set_array_mode(screen, rtex, i, array_mode);
 
@@ -262,11 +259,14 @@ static void r600_setup_miptree(struct pipe_screen *screen,
                        else
                                size = layer_size * 6;
                }
-               else
+               else if (ptex->target == PIPE_TEXTURE_3D)
                        size = layer_size * u_minify(ptex->depth0, i);
+               else
+                       size = layer_size * ptex->array_size;
+
                /* align base image and start of miptree */
                if ((i == 0) || (i == 1))
-                       offset = align(offset, r600_get_base_alignment(screen, ptex->format, array_mode));
+                       offset = align(offset, base_align);
                rtex->offset[i] = offset;
                rtex->layer_size[i] = layer_size;
                rtex->pitch_in_blocks[i] = nblocksx; /* CB talks in elements */
@@ -291,24 +291,31 @@ static boolean permit_hardware_blit(struct pipe_screen *screen,
                bind = PIPE_BIND_RENDER_TARGET;
 
        /* hackaround for S3TC */
-       if (util_format_is_s3tc(res->format))
+       if (util_format_is_compressed(res->format))
                return TRUE;
            
        if (!screen->is_format_supported(screen,
                                res->format,
                                res->target,
                                res->nr_samples,
-                               bind, 0))
+                                bind))
                return FALSE;
 
        if (!screen->is_format_supported(screen,
                                res->format,
                                res->target,
                                res->nr_samples,
-                               PIPE_BIND_SAMPLER_VIEW, 0))
+                                PIPE_BIND_SAMPLER_VIEW))
                return FALSE;
 
-       return TRUE;
+       switch (res->usage) {
+       case PIPE_USAGE_STREAM:
+       case PIPE_USAGE_STAGING:
+               return FALSE;
+
+       default:
+               return TRUE;
+       }
 }
 
 static boolean r600_texture_get_handle(struct pipe_screen* screen,
@@ -317,7 +324,7 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen,
 {
        struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
        struct r600_resource *resource = &rtex->resource;
-       struct radeon *radeon = (struct radeon *)screen->winsys;
+       struct radeon *radeon = ((struct r600_screen*)screen)->radeon;
 
        return r600_bo_get_winsys_handle(radeon, resource->bo,
                        rtex->pitch_in_bytes[0], whandle);
@@ -328,30 +335,20 @@ static void r600_texture_destroy(struct pipe_screen *screen,
 {
        struct r600_resource_texture *rtex = (struct r600_resource_texture*)ptex;
        struct r600_resource *resource = &rtex->resource;
-       struct radeon *radeon = (struct radeon *)screen->winsys;
 
        if (rtex->flushed_depth_texture)
                pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL);
 
        if (resource->bo) {
-               r600_bo_reference(radeon, &resource->bo, NULL);
+               r600_bo_reference(&resource->bo, NULL);
        }
        FREE(rtex);
 }
 
-static unsigned int r600_texture_is_referenced(struct pipe_context *context,
-                                               struct pipe_resource *texture,
-                                               unsigned level, int layer)
-{
-       /* FIXME */
-       return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
-}
-
 static const struct u_resource_vtbl r600_texture_vtbl =
 {
        r600_texture_get_handle,        /* get_handle */
        r600_texture_destroy,           /* resource_destroy */
-       r600_texture_is_referenced,     /* is_resource_referenced */
        r600_texture_get_transfer,      /* get_transfer */
        r600_texture_transfer_destroy,  /* transfer_destroy */
        r600_texture_transfer_map,      /* transfer_map */
@@ -366,11 +363,12 @@ r600_texture_create_object(struct pipe_screen *screen,
                           unsigned array_mode,
                           unsigned pitch_in_bytes_override,
                           unsigned max_buffer_size,
-                          struct r600_bo *bo)
+                          struct r600_bo *bo,
+                          boolean alloc_bo)
 {
        struct r600_resource_texture *rtex;
        struct r600_resource *resource;
-       struct radeon *radeon = (struct radeon *)screen->winsys;
+       struct radeon *radeon = ((struct r600_screen*)screen)->radeon;
 
        rtex = CALLOC_STRUCT(r600_resource_texture);
        if (rtex == NULL)
@@ -383,24 +381,86 @@ r600_texture_create_object(struct pipe_screen *screen,
        resource->b.b.b.screen = screen;
        resource->bo = bo;
        rtex->pitch_override = pitch_in_bytes_override;
+       rtex->real_format = base->format;
+
+       /* We must split depth and stencil into two separate buffers on Evergreen. */
+       if (r600_get_family_class(((struct r600_screen*)screen)->radeon) >= EVERGREEN &&
+           util_format_is_depth_and_stencil(base->format)) {
+               struct pipe_resource stencil;
+               unsigned stencil_pitch_override = 0;
+
+               switch (base->format) {
+               case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
+                       rtex->real_format = PIPE_FORMAT_Z24X8_UNORM;
+                       break;
+               case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
+                       rtex->real_format = PIPE_FORMAT_X8Z24_UNORM;
+                       break;
+               case PIPE_FORMAT_Z32_FLOAT_S8X24_USCALED:
+                       rtex->real_format = PIPE_FORMAT_Z32_FLOAT;
+                       break;
+               default:
+                       assert(0);
+                       FREE(rtex);
+                       return NULL;
+               }
+
+               /* Divide the pitch in bytes by 4 for stencil, because it has a smaller pixel size. */
+               if (pitch_in_bytes_override) {
+                       assert(base->format == PIPE_FORMAT_Z24_UNORM_S8_USCALED ||
+                              base->format == PIPE_FORMAT_S8_USCALED_Z24_UNORM);
+                       stencil_pitch_override = pitch_in_bytes_override / 4;
+               }
+
+               /* Allocate the stencil buffer. */
+               stencil = *base;
+               stencil.format = PIPE_FORMAT_S8_USCALED;
+               rtex->stencil = r600_texture_create_object(screen, &stencil, array_mode,
+                                                          stencil_pitch_override,
+                                                          max_buffer_size, NULL, FALSE);
+               if (!rtex->stencil) {
+                       FREE(rtex);
+                       return NULL;
+               }
+               /* Proceed in creating the depth buffer. */
+       }
+
        /* only mark depth textures the HW can hit as depth textures */
-       if (util_format_is_depth_or_stencil(base->format) && permit_hardware_blit(screen, base))
+       if (util_format_is_depth_or_stencil(rtex->real_format) && permit_hardware_blit(screen, base))
                rtex->depth = 1;
 
        r600_setup_miptree(screen, rtex, array_mode);
 
+       /* If we initialized separate stencil for Evergreen. place it after depth. */
+       if (rtex->stencil) {
+               unsigned stencil_align, stencil_offset;
+
+               stencil_align = r600_get_base_alignment(screen, rtex->stencil->real_format, array_mode);
+               stencil_offset = align(rtex->size, stencil_align);
+
+               for (unsigned i = 0; i <= rtex->stencil->resource.b.b.b.last_level; i++)
+                       rtex->stencil->offset[i] += stencil_offset;
+
+               rtex->size = stencil_offset + rtex->stencil->size;
+       }
+
        resource->size = rtex->size;
 
-       if (!resource->bo) {
+       /* Now create the backing buffer. */
+       if (!resource->bo && alloc_bo) {
                struct pipe_resource *ptex = &rtex->resource.b.b.b;
-               int base_align = r600_get_base_alignment(screen, ptex->format, array_mode);
+               unsigned base_align = r600_get_base_alignment(screen, ptex->format, array_mode);
 
                resource->bo = r600_bo(radeon, rtex->size, base_align, base->bind, base->usage);
                if (!resource->bo) {
+                       pipe_resource_reference((struct pipe_resource**)&rtex->stencil, NULL);
                        FREE(rtex);
                        return NULL;
                }
        }
+
+       if (rtex->stencil)
+               rtex->stencil->resource.bo = rtex->resource.bo;
        return rtex;
 }
 
@@ -412,8 +472,15 @@ struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
 
        /* Would like some magic "get_bool_option_once" routine.
         */
-       if (force_tiling == -1)
-               force_tiling = debug_get_bool_option("R600_FORCE_TILING", FALSE);
+       if (force_tiling == -1) {
+#if 0
+               /* reenable when 2D tiling is fixed better */
+               struct r600_screen *rscreen = (struct r600_screen *)screen;
+               if (r600_get_minor_version(rscreen->radeon) >= 9)
+                       force_tiling = debug_get_bool_option("R600_TILING", TRUE);
+#endif
+               force_tiling = debug_get_bool_option("R600_TILING", FALSE);
+       }
 
        if (force_tiling && permit_hardware_blit(screen, templ)) {
                if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
@@ -423,12 +490,11 @@ struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
        }
 
        if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
-           util_format_is_s3tc(templ->format))
+           util_format_is_compressed(templ->format))
                array_mode = V_038000_ARRAY_1D_TILED_THIN1;
 
        return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
-                                                                 0, 0, NULL);
-
+                                                                 0, 0, NULL, TRUE);
 }
 
 static struct pipe_surface *r600_create_surface(struct pipe_context *pipe,
@@ -473,8 +539,9 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
                                               const struct pipe_resource *templ,
                                               struct winsys_handle *whandle)
 {
-       struct radeon *rw = (struct radeon*)screen->winsys;
+       struct radeon *rw = ((struct r600_screen*)screen)->radeon;
        struct r600_bo *bo = NULL;
+       unsigned stride = 0;
        unsigned array_mode = 0;
 
        /* Support only 2D textures without mipmaps */
@@ -482,15 +549,13 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen,
              templ->depth0 != 1 || templ->last_level != 0)
                return NULL;
 
-       bo = r600_bo_handle(rw, whandle->handle, &array_mode);
+       bo = r600_bo_handle(rw, whandle, &stride, &array_mode);
        if (bo == NULL) {
                return NULL;
        }
 
        return (struct pipe_resource *)r600_texture_create_object(screen, templ, array_mode,
-                                                                 whandle->stride,
-                                                                 0,
-                                                                 bo);
+                                                                 stride, 0, bo, FALSE);
 }
 
 int r600_texture_depth_flush(struct pipe_context *ctx,
@@ -507,6 +572,7 @@ int r600_texture_depth_flush(struct pipe_context *ctx,
        resource.width0 = texture->width0;
        resource.height0 = texture->height0;
        resource.depth0 = 1;
+       resource.array_size = 1;
        resource.last_level = texture->last_level;
        resource.nr_samples = 0;
        resource.usage = PIPE_USAGE_DYNAMIC;
@@ -637,11 +703,12 @@ struct pipe_transfer* r600_texture_get_transfer(struct pipe_context *ctx,
                if (usage & PIPE_TRANSFER_READ) {
                        r600_copy_to_staging_texture(ctx, trans);
                        /* Always referenced in the blit. */
-                       ctx->flush(ctx, 0, NULL);
+                       r600_flush(ctx, NULL, 0);
                }
                return &trans->transfer;
        }
        trans->transfer.stride = rtex->pitch_in_bytes[level];
+       trans->transfer.layer_stride = rtex->layer_size[level];
        trans->offset = r600_texture_get_offset(rtex, level, box->z);
        return &trans->transfer;
 }
@@ -672,12 +739,12 @@ void r600_texture_transfer_destroy(struct pipe_context *ctx,
 void* r600_texture_transfer_map(struct pipe_context *ctx,
                                struct pipe_transfer* transfer)
 {
+       struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
        struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
        struct r600_bo *bo;
        enum pipe_format format = transfer->resource->format;
-       struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
+       struct radeon *radeon = rctx->screen->radeon;
        unsigned offset = 0;
-       unsigned usage = 0;
        char *map;
 
        if (rtransfer->staging_texture) {
@@ -695,30 +762,7 @@ void* r600_texture_transfer_map(struct pipe_context *ctx,
                        transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
        }
 
-       if (transfer->usage & PIPE_TRANSFER_WRITE) {
-               usage |= PB_USAGE_CPU_WRITE;
-
-               if (transfer->usage & PIPE_TRANSFER_DISCARD) {
-               }
-
-               if (transfer->usage & PIPE_TRANSFER_FLUSH_EXPLICIT) {
-               }
-       }
-
-       if (transfer->usage & PIPE_TRANSFER_READ) {
-               usage |= PB_USAGE_CPU_READ;
-       }
-
-       if (transfer->usage & PIPE_TRANSFER_DONTBLOCK) {
-               usage |= PB_USAGE_DONTBLOCK;
-       }
-
-       if (transfer->usage & PIPE_TRANSFER_UNSYNCHRONIZED) {
-               usage |= PB_USAGE_UNSYNCHRONIZED;
-       }
-
-       map = r600_bo_map(radeon, bo, usage, ctx);
-       if (!map) {
+       if (!(map = r600_bo_map(radeon, bo, rctx->ctx.cs, transfer->usage))) {
                return NULL;
        }
 
@@ -729,7 +773,7 @@ void r600_texture_transfer_unmap(struct pipe_context *ctx,
                                 struct pipe_transfer* transfer)
 {
        struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
-       struct radeon *radeon = (struct radeon *)ctx->screen->winsys;
+       struct radeon *radeon = ((struct r600_screen*)ctx->screen)->radeon;
        struct r600_bo *bo;
 
        if (rtransfer->staging_texture) {
@@ -766,11 +810,7 @@ static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
        };
 
        if (swizzle_view) {
-               /* Combine two sets of swizzles. */
-               for (i = 0; i < 4; i++) {
-                       swizzle[i] = swizzle_view[i] <= UTIL_FORMAT_SWIZZLE_W ?
-                               swizzle_format[swizzle_view[i]] : swizzle_view[i];
-               }
+               util_format_compose_swizzles(swizzle_format, swizzle_view, swizzle);
        } else {
                memcpy(swizzle, swizzle_format, 4);
        }
@@ -801,13 +841,16 @@ static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
 }
 
 /* texture format translate */
-uint32_t r600_translate_texformat(enum pipe_format format,
+uint32_t r600_translate_texformat(struct pipe_screen *screen,
+                                 enum pipe_format format,
                                  const unsigned char *swizzle_view,
                                  uint32_t *word4_p, uint32_t *yuv_format_p)
 {
        uint32_t result = 0, word4 = 0, yuv_format = 0;
        const struct util_format_description *desc;
        boolean uniform = TRUE;
+       static int r600_enable_s3tc = -1;
+
        int i;
        const uint32_t sign_bit[4] = {
                S_038010_FORMAT_COMP_X(V_038010_SQ_FORMAT_COMP_SIGNED),
@@ -843,6 +886,12 @@ uint32_t r600_translate_texformat(enum pipe_format format,
                        result = FMT_8;
                        word4 |= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT);
                        goto out_word4;
+               case PIPE_FORMAT_Z32_FLOAT:
+                       result = FMT_32_FLOAT;
+                       goto out_word4;
+               case PIPE_FORMAT_Z32_FLOAT_S8X24_USCALED:
+                       result = FMT_X24_8_32_FLOAT;
+                       goto out_word4;
                default:
                        goto out_unknown;
                }
@@ -859,21 +908,45 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 
        case UTIL_FORMAT_COLORSPACE_SRGB:
                word4 |= S_038010_FORCE_DEGAMMA(1);
-               if (format == PIPE_FORMAT_L8A8_SRGB || format == PIPE_FORMAT_L8_SRGB)
-                       goto out_unknown; /* fails for some reason - TODO */
                break;
 
        default:
                break;
        }
 
-       /* S3TC formats. TODO */
-       if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
-               static int r600_enable_s3tc = -1;
+       if (r600_enable_s3tc == -1) {
+               struct r600_screen *rscreen = (struct r600_screen *)screen;
+               if (r600_get_minor_version(rscreen->radeon) >= 9)
+                       r600_enable_s3tc = 1;
+               else
+                       r600_enable_s3tc = debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
+       }
 
-               if (r600_enable_s3tc == -1)
-                       r600_enable_s3tc =
-                               debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
+       if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
+               if (!r600_enable_s3tc)
+                       goto out_unknown;
+
+               switch (format) {
+               case PIPE_FORMAT_RGTC1_SNORM:
+               case PIPE_FORMAT_LATC1_SNORM:
+                       word4 |= sign_bit[0];
+               case PIPE_FORMAT_RGTC1_UNORM:
+               case PIPE_FORMAT_LATC1_UNORM:
+                       result = FMT_BC4;
+                       goto out_word4;
+               case PIPE_FORMAT_RGTC2_SNORM:
+               case PIPE_FORMAT_LATC2_SNORM:
+                       word4 |= sign_bit[0] | sign_bit[1];
+               case PIPE_FORMAT_RGTC2_UNORM:
+               case PIPE_FORMAT_LATC2_UNORM:
+                       result = FMT_BC5;
+                       goto out_word4;
+               default:
+                       goto out_unknown;
+               }
+       }
+
+       if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
 
                if (!r600_enable_s3tc)
                        goto out_unknown;
@@ -902,6 +975,14 @@ uint32_t r600_translate_texformat(enum pipe_format format,
                }
        }
 
+       if (format == PIPE_FORMAT_R9G9B9E5_FLOAT) {
+               result = FMT_5_9_9_9_SHAREDEXP;
+               goto out_word4;
+       } else if (format == PIPE_FORMAT_R11G11B10_FLOAT) {
+               result = FMT_10_11_11_FLOAT;
+               goto out_word4;
+       }
+
 
        for (i = 0; i < desc->nr_channels; i++) {
                if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
@@ -911,8 +992,6 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 
        /* R8G8Bx_SNORM - TODO CxV8U8 */
 
-       /* RGTC - TODO */
-
        /* See whether the components are of the same size. */
        for (i = 1; i < desc->nr_channels; i++) {
                uniform = uniform && desc->channel[0].size == desc->channel[i].size;
@@ -1048,8 +1127,9 @@ uint32_t r600_translate_texformat(enum pipe_format format,
                                goto out_word4;
                        }
                }
-
+               goto out_unknown;
        }
+
 out_word4:
        if (word4_p)
                *word4_p = word4;
@@ -1057,6 +1137,6 @@ out_word4:
                *yuv_format_p = yuv_format;
        return result;
 out_unknown:
-//     R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format));
+       /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */
        return ~0;
 }