From 1c50ea2cd9ab8752793c99b4a7a2a6656bdde1ac Mon Sep 17 00:00:00 2001 From: Ian Romanick Date: Wed, 27 Feb 2008 13:40:23 -0800 Subject: cell: Use unified data cache for textures too --- src/gallium/drivers/cell/spu/spu_main.c | 2 + src/gallium/drivers/cell/spu/spu_main.h | 3 +- src/gallium/drivers/cell/spu/spu_texture.c | 184 +++++++++++------------------ 3 files changed, 72 insertions(+), 117 deletions(-) (limited to 'src') diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c index cc4bafdb3a..59300028d4 100644 --- a/src/gallium/drivers/cell/spu/spu_main.c +++ b/src/gallium/drivers/cell/spu/spu_main.c @@ -286,6 +286,8 @@ cmd_state_texture(const struct cell_command_texture *texture) { spu.texture.width, spu.texture.height, 0.0, 0.0}; spu.tex_size_mask = (vector unsigned int) { spu.texture.width - 1, spu.texture.height - 1, 0, 0 }; + spu.tex_size_x_mask = spu_splats(spu.texture.width - 1); + spu.tex_size_y_mask = spu_splats(spu.texture.height - 1); } diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h index d14f1abbe7..a13edd1702 100644 --- a/src/gallium/drivers/cell/spu/spu_main.h +++ b/src/gallium/drivers/cell/spu/spu_main.h @@ -107,6 +107,8 @@ struct spu_global vector float tex_size; vector unsigned int tex_size_mask; /**< == int(size - 1) */ + vector unsigned int tex_size_x_mask; /**< == int(size - 1) */ + vector unsigned int tex_size_y_mask; /**< == int(size - 1) */ vector float (*sample_texture)(vector float texcoord); @@ -130,7 +132,6 @@ extern boolean Debug; #define TAG_INDEX_BUFFER 16 #define TAG_BATCH_BUFFER 17 #define TAG_MISC 18 -#define TAG_TEXTURE_TILE 19 #define TAG_DCACHE0 20 #define TAG_DCACHE1 21 #define TAG_DCACHE2 22 diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c index 3962aaa4a9..67eb08196a 100644 --- a/src/gallium/drivers/cell/spu/spu_texture.c +++ b/src/gallium/drivers/cell/spu/spu_texture.c @@ -31,19 +31,7 @@ #include "spu_texture.h" #include "spu_tile.h" #include "spu_colorpack.h" - - -/** - * Number of texture tiles to cache. - * Note that this will probably be the largest consumer of SPU local store/ - * memory for this driver! - */ -#define CACHE_SIZE 16 - -static tile_t tex_tiles[CACHE_SIZE] ALIGN16_ATTRIB; - -static vector unsigned int tex_tile_xy[CACHE_SIZE]; - +#include "spu_dcache.h" /** @@ -52,78 +40,60 @@ static vector unsigned int tex_tile_xy[CACHE_SIZE]; void invalidate_tex_cache(void) { - /* XXX memset? */ - uint i; - for (i = 0; i < CACHE_SIZE; i++) { - tex_tile_xy[i] = ((vector unsigned int) { ~0U, ~0U, ~0U, ~0U }); - } + spu_dcache_mark_dirty((unsigned) spu.texture.start, + 4 * spu.texture.width * spu.texture.height); } -/** - * Return the cache pos/index which corresponds to tile (tx,ty) - */ -static INLINE uint -cache_pos(vector unsigned int txty) +static uint +get_texel(vec_uint4 coordinate) { - uint pos = (spu_extract(txty,0) + spu_extract(txty,1) * 4) % CACHE_SIZE; - return pos; + vec_uint4 tmp; + unsigned x = spu_extract(coordinate, 0); + unsigned y = spu_extract(coordinate, 1); + const unsigned tiles_per_row = spu.texture.width / TILE_SIZE; + unsigned tile_offset = sizeof(tile_t) * ((y / TILE_SIZE * tiles_per_row) + + (x / TILE_SIZE)); + unsigned texel_offset = 4 * (((y % TILE_SIZE) * TILE_SIZE) + + (x % TILE_SIZE)); + + spu_dcache_fetch_unaligned((qword *) & tmp, + spu.texture.start + tile_offset + texel_offset, + 4); + return spu_extract(tmp, 0); } -/** - * Make sure the tile for texel (i,j) is present, return its position/index - * in the cache. - */ -static uint -get_tex_tile(vector unsigned int ij) +static void +get_four_texels(vec_uint4 x, vec_uint4 y, vec_uint4 *texels) { - /* tile address: tx,ty */ - const vector unsigned int txty = spu_rlmask(ij, -5); /* divide by 32 */ - const uint pos = cache_pos(txty); - - if ((spu_extract(tex_tile_xy[pos], 0) != spu_extract(txty, 0)) || - (spu_extract(tex_tile_xy[pos], 1) != spu_extract(txty, 1))) { - - /* texture cache miss, fetch tile from main memory */ - const uint tiles_per_row = spu.texture.width / TILE_SIZE; - const uint bytes_per_tile = sizeof(tile_t); - const void *src = (const ubyte *) spu.texture.start - + (spu_extract(txty,1) * tiles_per_row + spu_extract(txty,0)) * bytes_per_tile; - - printf("SPU %u: tex cache miss at %d, %d pos=%u old=%d,%d\n", - spu.init.id, - spu_extract(txty,0), - spu_extract(txty,1), - pos, - spu_extract(tex_tile_xy[pos],0), - spu_extract(tex_tile_xy[pos],1)); - - ASSERT_ALIGN16(tex_tiles[pos].ui); - ASSERT_ALIGN16(src); - - mfc_get(tex_tiles[pos].ui, /* dest */ - (unsigned int) src, - bytes_per_tile, /* size */ - TAG_TEXTURE_TILE, - 0, /* tid */ - 0 /* rid */); - - wait_on_mask(1 << TAG_TEXTURE_TILE); - - tex_tile_xy[pos] = txty; - } - else { -#if 0 - printf("SPU %u: tex cache HIT at %d, %d\n", - spu.init.id, tx, ty); -#endif - } - - return pos; + const unsigned texture_ea = (uintptr_t) spu.texture.start; + vec_uint4 tile_x = spu_rlmask(x, -5); + vec_uint4 tile_y = spu_rlmask(y, -5); + const qword offset_x = si_andi((qword) x, 0x1f); + const qword offset_y = si_andi((qword) y, 0x1f); + + const qword tiles_per_row = (qword) spu_splats(spu.texture.width / TILE_SIZE); + const qword tile_size = (qword) spu_splats(sizeof(tile_t)); + + qword tile_offset = si_mpya((qword) tile_y, tiles_per_row, (qword) tile_x); + tile_offset = si_mpy((qword) tile_offset, tile_size); + + qword texel_offset = si_a(si_mpyui(offset_y, 32), offset_x); + texel_offset = si_mpyui(texel_offset, 4); + + vec_uint4 offset = (vec_uint4) si_a(tile_offset, texel_offset); + + spu_dcache_fetch_unaligned((qword *) & texels[0], + texture_ea + spu_extract(offset, 0), 4); + spu_dcache_fetch_unaligned((qword *) & texels[1], + texture_ea + spu_extract(offset, 1), 4); + spu_dcache_fetch_unaligned((qword *) & texels[2], + texture_ea + spu_extract(offset, 2), 4); + spu_dcache_fetch_unaligned((qword *) & texels[3], + texture_ea + spu_extract(offset, 3), 4); } - /** * Get texture sample at texcoord. * XXX this is extremely primitive for now. @@ -134,9 +104,7 @@ sample_texture_nearest(vector float texcoord) vector float tc = spu_mul(texcoord, spu.tex_size); vector unsigned int itc = spu_convtu(tc, 0); /* convert to int */ itc = spu_and(itc, spu.tex_size_mask); /* mask (GL_REPEAT) */ - vector unsigned int ij = spu_and(itc, TILE_SIZE-1); /* intra tile addr */ - uint pos = get_tex_tile(itc); - uint texel = tex_tiles[pos].ui[spu_extract(ij, 1)][spu_extract(ij, 0)]; + uint texel = get_texel(itc); return spu_unpack_A8R8G8B8(texel); } @@ -144,49 +112,33 @@ sample_texture_nearest(vector float texcoord) vector float sample_texture_bilinear(vector float texcoord) { - static const vector unsigned int offset10 = {1, 0, 0, 0}; - static const vector unsigned int offset01 = {0, 1, 0, 0}; + static const vec_uint4 offset_x = {0, 0, 1, 1}; + static const vec_uint4 offset_y = {0, 1, 0, 1}; vector float tc = spu_mul(texcoord, spu.tex_size); tc = spu_add(tc, spu_splats(-0.5f)); /* half texel bias */ /* integer texcoords S,T: */ - vector unsigned int itc00 = spu_convtu(tc, 0); /* convert to int */ - vector unsigned int itc01 = spu_add(itc00, offset01); - vector unsigned int itc10 = spu_add(itc00, offset10); - vector unsigned int itc11 = spu_add(itc10, offset01); - - /* mask (GL_REPEAT) */ - itc00 = spu_and(itc00, spu.tex_size_mask); - itc01 = spu_and(itc01, spu.tex_size_mask); - itc10 = spu_and(itc10, spu.tex_size_mask); - itc11 = spu_and(itc11, spu.tex_size_mask); - - /* intra tile addr */ - vector unsigned int ij00 = spu_and(itc00, TILE_SIZE-1); - vector unsigned int ij01 = spu_and(itc01, TILE_SIZE-1); - vector unsigned int ij10 = spu_and(itc10, TILE_SIZE-1); - vector unsigned int ij11 = spu_and(itc11, TILE_SIZE-1); - - /* get tile cache positions */ - uint pos00 = get_tex_tile(itc00); - uint pos01, pos10, pos11; - if ((spu_extract(ij00, 0) < TILE_SIZE-1) && - (spu_extract(ij00, 1) < TILE_SIZE-1)) { - /* all texels are in the same tile */ - pos01 = pos10 = pos11 = pos00; - } - else { - pos01 = get_tex_tile(itc01); - pos10 = get_tex_tile(itc10); - pos11 = get_tex_tile(itc11); - } - - /* get texels from tiles and convert to float[4] */ - vector float texel00 = spu_unpack_A8R8G8B8(tex_tiles[pos00].ui[spu_extract(ij00, 1)][spu_extract(ij00, 0)]); - vector float texel01 = spu_unpack_A8R8G8B8(tex_tiles[pos01].ui[spu_extract(ij01, 1)][spu_extract(ij01, 0)]); - vector float texel10 = spu_unpack_A8R8G8B8(tex_tiles[pos10].ui[spu_extract(ij10, 1)][spu_extract(ij10, 0)]); - vector float texel11 = spu_unpack_A8R8G8B8(tex_tiles[pos11].ui[spu_extract(ij11, 1)][spu_extract(ij11, 0)]); + vec_uint4 itc = spu_convtu(tc, 0); /* convert to int */ + + vec_uint4 texels[4]; + + vec_uint4 x = spu_splats(spu_extract(itc, 0)); + vec_uint4 y = spu_splats(spu_extract(itc, 1)); + + x = spu_add(x, offset_x); + y = spu_add(y, offset_y); + + x = spu_and(x, spu.tex_size_x_mask); + y = spu_and(y, spu.tex_size_y_mask); + + get_four_texels(x, y, texels); + + vector float texel00 = spu_unpack_A8R8G8B8(spu_extract(texels[0], 0)); + vector float texel01 = spu_unpack_A8R8G8B8(spu_extract(texels[1], 0)); + vector float texel10 = spu_unpack_A8R8G8B8(spu_extract(texels[2], 0)); + vector float texel11 = spu_unpack_A8R8G8B8(spu_extract(texels[3], 0)); + /* Compute weighting factors in [0,1] * Multiply texcoord by 1024, AND with 1023, convert back to float. -- cgit v1.2.3