i915g: implement staged texture uploadupload

For busy textures to avoid stalls and for smaller tiled textures
where the mapping/fencing overhead is too big.

Only works with the hw blitter atm.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>WIP: pipelined upload

stuff

foxip

4 files changed, 132 insertions, 12 deletions

diff --git a/src/gallium/drivers/i915/i915_context.c b/src/gallium/drivers/i915/i915_context.c
index 7a98ef73c1..8d991033a7 100644
--- a/src/gallium/drivers/i915/i915_context.c
+++ b/src/gallium/drivers/i915/i915_context.c
@@ -148,6 +148,9 @@ i915_create_context(struct pipe_screen *screen, void *priv)
    i915->base.draw_vbo = i915_draw_vbo;
 
    /* init this before draw */
+   util_slab_create(&i915->tex_transfer_pool, sizeof(struct i915_transfer),
+                    16, UTIL_SLAB_SINGLETHREADED);
+
    util_slab_create(&i915->transfer_pool, sizeof(struct pipe_transfer),
                     16, UTIL_SLAB_SINGLETHREADED);
 
diff --git a/src/gallium/drivers/i915/i915_context.h b/src/gallium/drivers/i915/i915_context.h
index dacf50e870..a304011333 100644
--- a/src/gallium/drivers/i915/i915_context.h
+++ b/src/gallium/drivers/i915/i915_context.h
@@ -254,6 +254,7 @@ struct i915_context {
    int num_validation_buffers;
 
    struct util_slab_mempool transfer_pool;
+   struct util_slab_mempool tex_transfer_pool;
 
    /** blitter/hw-clear */
    struct blitter_context* blitter;
diff --git a/src/gallium/drivers/i915/i915_resource.h b/src/gallium/drivers/i915/i915_resource.h
index 86620e6a12..f2689bf625 100644
--- a/src/gallium/drivers/i915/i915_resource.h
+++ b/src/gallium/drivers/i915/i915_resource.h
@@ -77,6 +77,12 @@ struct i915_texture {
    struct i915_winsys_buffer *buffer;
 };
 
+struct i915_transfer {
+   struct pipe_transfer b;
+   struct i915_winsys_buffer *upload_buf;
+   boolean staged_upload;
+};
+
 unsigned i915_texture_offset(struct i915_texture *tex,
                              unsigned level, unsigned layer);
 void i915_init_screen_resource_functions(struct i915_screen *is);
@@ -99,6 +105,12 @@ static INLINE struct i915_buffer *i915_buffer(struct pipe_resource *resource)
    return tex;
 }
 
+static INLINE struct i915_transfer *i915_transfer(struct pipe_transfer *transfer)
+{
+   struct i915_transfer *tex_transfer = (struct i915_transfer *)transfer;
+   return tex_transfer;
+}
+
 struct pipe_resource *
 i915_texture_create(struct pipe_screen *screen,
                     const struct pipe_resource *template);
diff --git a/src/gallium/drivers/i915/i915_resource_texture.c b/src/gallium/drivers/i915/i915_resource_texture.c
index c4e078176b..1aa6ff2f55 100644
--- a/src/gallium/drivers/i915/i915_resource_texture.c
+++ b/src/gallium/drivers/i915/i915_resource_texture.c
@@ -43,6 +43,7 @@
 #include "i915_screen.h"
 #include "i915_winsys.h"
 #include "i915_debug.h"
+#include "i915_blit.h"
 
 
 #define DEBUG_TEXTURES 0
@@ -719,27 +720,94 @@ i915_texture_get_transfer(struct pipe_context *pipe,
 {
    struct i915_context *i915 = i915_context(pipe);
    struct i915_texture *tex = i915_texture(resource);
-   struct pipe_transfer *transfer = util_slab_alloc(&i915->transfer_pool);
+   struct i915_transfer *transfer = util_slab_alloc(&i915->tex_transfer_pool);
 
    if (transfer == NULL)
       return NULL;
 
-   transfer->resource = resource;
-   transfer->level = level;
-   transfer->usage = usage;
-   transfer->box = *box;
-   transfer->stride = tex->stride;
+   transfer->b.resource = resource;
+   transfer->b.level = level;
+   transfer->b.usage = usage;
+   transfer->b.box = *box;
+   transfer->b.stride = tex->stride;
    /* FIXME: layer_stride */
 
-   return transfer;
+   return &transfer->b;
 }
 
 static void
 i915_transfer_destroy(struct pipe_context *pipe,
                       struct pipe_transfer *transfer)
 {
-   struct i915_context *i915 = i915_context(pipe);
-   util_slab_free(&i915->transfer_pool, transfer);
+   struct i915_winsys *iws = i915_screen(transfer->resource->screen)->iws;
+   struct i915_transfer *tex_transfer = i915_transfer(transfer);
+
+   if (tex_transfer->upload_buf)
+      iws->buffer_destroy(iws, tex_transfer->upload_buf);
+   if (transfer->data)
+      FREE(transfer->data);
+
+   util_slab_free(&i915_context(pipe)->tex_transfer_pool, tex_transfer);
+}
+
+static unsigned
+get_transfer_box_size_2d(struct i915_texture *tex, struct pipe_box *box)
+{
+   unsigned stride;
+   stride = align(util_format_get_stride(tex->b.b.format, box->width), 2);
+   return stride * util_format_get_nblocksy(tex->b.b.format, box->height);
+}
+
+static boolean
+staged_upload(struct i915_screen *is,
+              struct i915_texture *tex,
+              struct i915_transfer *transfer)
+{
+   if (!is->debug.use_blitter)
+      return FALSE;
+
+   if (tex->b.b.target != PIPE_TEXTURE_2D ||
+         transfer->b.usage != PIPE_TRANSFER_WRITE)
+      return FALSE;
+
+   fprintf(stderr, "texcheck, size %i, tiling %i\n",  get_transfer_box_size_2d(tex, &transfer->b.box), tex->tiling);
+   /* staged upload if a fence would be way too huge */
+#define I915_MIN_FENCE_SIZE 1024*1024
+   if (tex->tiling != I915_TILE_NONE &&
+         get_transfer_box_size_2d(tex, &transfer->b.box) < I915_MIN_FENCE_SIZE) {
+      fprintf(stderr, "tilecheck\n");
+      return TRUE;
+   }
+
+   return is->iws->buffer_busy(is->iws, tex->buffer);
+}
+
+static void
+staged_upload_prepare(struct i915_screen *is,
+                      struct i915_texture *tex,
+                      struct i915_transfer *transfer)
+{
+   unsigned size = get_transfer_box_size_2d(tex, &transfer->b.box);
+   struct i915_winsys *iws = is->iws;
+   transfer->b.stride =
+      align(util_format_get_stride(tex->b.b.format, transfer->b.box.width), 2);
+   transfer->staged_upload = FALSE;
+
+   if (!transfer->upload_buf) {
+      transfer->upload_buf = iws->buffer_create(iws, size, I915_NEW_UPLOAD);
+      if (!transfer->upload_buf)
+         return;
+   }
+
+   if (!transfer->b.data) {
+      transfer->b.data = MALLOC(size);
+      if (!transfer->b.data)
+         return;
+   }
+
+   fprintf(stderr, "staged upload!\n");
+
+   transfer->staged_upload = TRUE;
 }
 
 static void *
@@ -748,7 +816,9 @@ i915_texture_transfer_map(struct pipe_context *pipe,
 {
    struct pipe_resource *resource = transfer->resource;
    struct i915_texture *tex = i915_texture(resource);
-   struct i915_winsys *iws = i915_screen(pipe->screen)->iws;
+   struct i915_screen *is = i915_screen(tex->b.b.screen);
+   struct i915_winsys *iws = is->iws;
+   struct i915_transfer *tex_transfer = i915_transfer(transfer);
    struct pipe_box *box = &transfer->box;
    enum pipe_format format = resource->format;
    unsigned offset;
@@ -762,7 +832,14 @@ i915_texture_transfer_map(struct pipe_context *pipe,
    /* TODO this is a sledgehammer */
    pipe->flush(pipe, NULL);
 
+   if (staged_upload(is, tex, tex_transfer)) {
+      staged_upload_prepare(is, tex, tex_transfer);
+      if (tex_transfer->staged_upload)
+         return transfer->data;
+   }
+
    map = iws->buffer_map(iws, tex->buffer);
+
    if (map == NULL)
       return NULL;
 
@@ -772,12 +849,39 @@ i915_texture_transfer_map(struct pipe_context *pipe,
 }
 
 static void
+staged_upload_blt(struct i915_context *i915, struct i915_texture *tex,
+                  struct i915_transfer *transfer)
+{
+   unsigned size = get_transfer_box_size_2d(tex, &transfer->b.box);
+   struct i915_screen *is = i915_screen(tex->b.b.screen);
+   struct i915_winsys *iws = is->iws;
+   int ret;
+
+   ret = iws->buffer_write(iws, transfer->upload_buf, 0, size, transfer->b.data);
+   assert(ret == 0);
+
+   i915_copy_blit(i915, util_format_get_blocksize(tex->b.b.format),
+                  transfer->b.stride, transfer->upload_buf, 0,
+                  tex->stride, tex->buffer, 0,
+                  0, 0, transfer->b.box.x, transfer->b.box.y,
+                  transfer->b.box.width, transfer->b.box.height);
+
+   assert(is->debug.use_blitter);
+}
+
+static void
 i915_texture_transfer_unmap(struct pipe_context *pipe,
 			    struct pipe_transfer *transfer)
 {
    struct i915_texture *tex = i915_texture(transfer->resource);
-   struct i915_winsys *iws = i915_screen(tex->b.b.screen)->iws;
-   iws->buffer_unmap(iws, tex->buffer);
+   struct i915_transfer *tex_transfer = i915_transfer(transfer);
+   struct i915_screen *is = i915_screen(tex->b.b.screen);
+   struct i915_winsys *iws = is->iws;
+
+   if (tex_transfer->staged_upload)
+      staged_upload_blt(i915_context(pipe), tex, tex_transfer);
+   else
+      iws->buffer_unmap(iws, tex->buffer);
 }