OSDN Git Service

nouveau: Swizzle textures larger than nv04 SIFM limit in parts.
authorYounes Manton <younes.m@gmail.com>
Fri, 5 Dec 2008 07:21:55 +0000 (02:21 -0500)
committerYounes Manton <younes.m@gmail.com>
Sat, 10 Jan 2009 18:52:04 +0000 (13:52 -0500)
Limit of SIFM on nv40 is 1024x1024, not sure about others.

src/gallium/drivers/nv40/nv40_miptree.c
src/gallium/winsys/drm/nouveau/nv04_surface.c

index b68967c..36e08e7 100644 (file)
@@ -80,8 +80,7 @@ nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
                /* TODO: Figure out which formats can be swizzled */
                case PIPE_FORMAT_A8R8G8B8_UNORM:
                case PIPE_FORMAT_X8R8G8B8_UNORM:
-               /* XXX: Re-enable when SIFM size limits are fixed */
-               /*case PIPE_FORMAT_R16_SNORM:*/
+               case PIPE_FORMAT_R16_SNORM:
                        break;
                default:
                        mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
index 68338eb..e9a8a2a 100644 (file)
@@ -82,6 +82,37 @@ nv04_scaled_image_format(enum pipe_format format)
        }
 }
 
+static INLINE unsigned
+nv04_swizzle_bits(unsigned x, unsigned y)
+{
+       unsigned u = (x & 0x001) << 0 |
+                    (x & 0x002) << 1 |
+                    (x & 0x004) << 2 |
+                    (x & 0x008) << 3 |
+                    (x & 0x010) << 4 |
+                    (x & 0x020) << 5 |
+                    (x & 0x040) << 6 |
+                    (x & 0x080) << 7 |
+                    (x & 0x100) << 8 |
+                    (x & 0x200) << 9 |
+                    (x & 0x400) << 10 |
+                    (x & 0x800) << 11;
+
+       unsigned v = (y & 0x001) << 1 |
+                    (y & 0x002) << 2 |
+                    (y & 0x004) << 3 |
+                    (y & 0x008) << 4 |
+                    (y & 0x010) << 5 |
+                    (y & 0x020) << 6 |
+                    (y & 0x040) << 7 |
+                    (y & 0x080) << 8 |
+                    (y & 0x100) << 9 |
+                    (y & 0x200) << 10 |
+                    (y & 0x400) << 11 |
+                    (y & 0x800) << 12;
+       return v | u;
+}
+
 static void
 nv04_surface_copy_swizzle(struct nouveau_context *nv, unsigned dx, unsigned dy,
                           unsigned sx, unsigned sy, unsigned w, unsigned h)
@@ -90,19 +121,23 @@ nv04_surface_copy_swizzle(struct nouveau_context *nv, unsigned dx, unsigned dy,
        struct pipe_surface *dst = nv->surf_dst;
        struct pipe_surface *src = nv->surf_src;
 
+       const unsigned max_w = 1024;
+       const unsigned max_h = 1024;
+       const unsigned sub_w = w > max_w ? max_w : w;
+       const unsigned sub_h = h > max_h ? max_h : h;
+       unsigned cx = 0;
+       unsigned cy = 0;
+
        /* POT or GTFO */
        assert(!(w & (w - 1)) && !(h & (h - 1)));
 
        BEGIN_RING(chan, nv->nvc->NvSwzSurf, NV04_SWIZZLED_SURFACE_DMA_IMAGE, 1);
        OUT_RELOCo(chan, nouveau_buffer(dst->buffer)->bo,
                         NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
-
-       BEGIN_RING(chan, nv->nvc->NvSwzSurf, NV04_SWIZZLED_SURFACE_FORMAT, 2);
+       BEGIN_RING(chan, nv->nvc->NvSwzSurf, NV04_SWIZZLED_SURFACE_FORMAT, 1);
        OUT_RING  (chan, nv04_surface_format(dst->format) |
                         log2i(w) << NV04_SWIZZLED_SURFACE_FORMAT_BASE_SIZE_U_SHIFT |
                         log2i(h) << NV04_SWIZZLED_SURFACE_FORMAT_BASE_SIZE_V_SHIFT);
-       OUT_RELOCl(chan, nouveau_buffer(dst->buffer)->bo, dst->offset,
-                        NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 
        BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_DMA_IMAGE, 1);
        OUT_RELOCo(chan, nouveau_buffer(src->buffer)->bo,
@@ -110,24 +145,35 @@ nv04_surface_copy_swizzle(struct nouveau_context *nv, unsigned dx, unsigned dy,
        BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_SURFACE, 1);
        OUT_RING  (chan, nv->nvc->NvSwzSurf->handle);
 
-       BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION, 9);
-       OUT_RING  (chan, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION_TRUNCATE);
-       OUT_RING  (chan, nv04_scaled_image_format(src->format));
-       OUT_RING  (chan, NV04_SCALED_IMAGE_FROM_MEMORY_OPERATION_SRCCOPY);
-       OUT_RING  (chan, 0);
-       OUT_RING  (chan, h << 16 | w);
-       OUT_RING  (chan, 0);
-       OUT_RING  (chan, h << 16 | w);
-       OUT_RING  (chan, 1 << 20);
-       OUT_RING  (chan, 1 << 20);
-       BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_SIZE, 4);
-       OUT_RING  (chan, h << 16 | w);
-       OUT_RING  (chan, src->stride |
-                        NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_ORIGIN_CENTER |
-                        NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_FILTER_POINT_SAMPLE);
-       OUT_RELOCl(chan, nouveau_buffer(src->buffer)->bo, src->offset,
-                        NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
-       OUT_RING  (chan, 0);
+       for (cy = 0; cy < h; cy += sub_h) {
+               for (cx = 0; cx < w; cx += sub_w) {
+                       BEGIN_RING(chan, nv->nvc->NvSwzSurf, NV04_SWIZZLED_SURFACE_OFFSET, 1);
+                       OUT_RELOCl(chan, nouveau_buffer(dst->buffer)->bo,
+                                        dst->offset + nv04_swizzle_bits(cx, cy) * dst->block.size,
+                                        NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+
+                       BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION, 9);
+                       OUT_RING  (chan, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION_TRUNCATE);
+                       OUT_RING  (chan, nv04_scaled_image_format(src->format));
+                       OUT_RING  (chan, NV04_SCALED_IMAGE_FROM_MEMORY_OPERATION_SRCCOPY);
+                       OUT_RING  (chan, 0);
+                       OUT_RING  (chan, sub_h << 16 | sub_w);
+                       OUT_RING  (chan, 0);
+                       OUT_RING  (chan, sub_h << 16 | sub_w);
+                       OUT_RING  (chan, 1 << 20);
+                       OUT_RING  (chan, 1 << 20);
+
+                       BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_SIZE, 4);
+                       OUT_RING  (chan, sub_h << 16 | sub_w);
+                       OUT_RING  (chan, src->stride |
+                                        NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_ORIGIN_CENTER |
+                                        NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_FILTER_POINT_SAMPLE);
+                       OUT_RELOCl(chan, nouveau_buffer(src->buffer)->bo,
+                                        src->offset + cy * src->stride + cx * src->block.size,
+                                        NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
+                       OUT_RING  (chan, 0);
+               }
+       }
 }
 
 static void