diff options
Diffstat (limited to 'src/gallium/drivers/nv50/nv50_tex.c')
-rw-r--r-- | src/gallium/drivers/nv50/nv50_tex.c | 202 |
1 files changed, 112 insertions, 90 deletions
diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c index 4c48b12cd8..85ab947c00 100644 --- a/src/gallium/drivers/nv50/nv50_tex.c +++ b/src/gallium/drivers/nv50/nv50_tex.c @@ -29,26 +29,16 @@ #include "util/u_format.h" #define _MIXED(pf, t0, t1, t2, t3, cr, cg, cb, ca, f) \ -{ \ - PIPE_FORMAT_##pf, \ +[PIPE_FORMAT_##pf] = ( \ NV50TIC_0_0_MAPR_##cr | NV50TIC_0_0_TYPER_##t0 | \ NV50TIC_0_0_MAPG_##cg | NV50TIC_0_0_TYPEG_##t1 | \ NV50TIC_0_0_MAPB_##cb | NV50TIC_0_0_TYPEB_##t2 | \ NV50TIC_0_0_MAPA_##ca | NV50TIC_0_0_TYPEA_##t3 | \ - NV50TIC_0_0_FMT_##f \ -} + NV50TIC_0_0_FMT_##f) #define _(pf, t, cr, cg, cb, ca, f) _MIXED(pf, t, t, t, t, cr, cg, cb, ca, f) -struct nv50_texture_format { - enum pipe_format pf; - uint32_t hw; -}; - -#define NV50_TEX_FORMAT_LIST_SIZE \ - (sizeof(nv50_tex_format_list) / sizeof(struct nv50_texture_format)) - -static const struct nv50_texture_format nv50_tex_format_list[] = +static const uint32_t nv50_texture_formats[PIPE_FORMAT_COUNT] = { _(B8G8R8A8_UNORM, UNORM, C2, C1, C0, C3, 8_8_8_8), _(B8G8R8A8_SRGB, UNORM, C2, C1, C0, C3, 8_8_8_8), @@ -60,10 +50,12 @@ static const struct nv50_texture_format nv50_tex_format_list[] = _(B5G6R5_UNORM, UNORM, C2, C1, C0, ONE, 5_6_5), _(L8_UNORM, UNORM, C0, C0, C0, ONE, 8), + _(L8_SRGB, UNORM, C0, C0, C0, ONE, 8), _(A8_UNORM, UNORM, ZERO, ZERO, ZERO, C0, 8), _(I8_UNORM, UNORM, C0, C0, C0, C0, 8), _(L8A8_UNORM, UNORM, C0, C0, C0, C1, 8_8), + _(L8A8_SRGB, UNORM, C0, C0, C0, C1, 8_8), _(DXT1_RGB, UNORM, C0, C1, C2, ONE, DXT1), _(DXT1_RGBA, UNORM, C0, C1, C2, C3, DXT1), @@ -81,117 +73,143 @@ static const struct nv50_texture_format nv50_tex_format_list[] = _(R16G16_UNORM, UNORM, C0, C1, ZERO, ONE, 16_16), _MIXED(Z32_FLOAT, FLOAT, UINT, UINT, UINT, C0, C0, C0, ONE, 32_DEPTH) - }; #undef _ #undef _MIXED -static int -nv50_tex_construct(struct nv50_context *nv50, struct nouveau_stateobj *so, - struct nv50_miptree *mt, int unit, unsigned p) +static INLINE uint32_t +nv50_tic_swizzle(uint32_t tc, unsigned swz) +{ + switch (swz) { + case PIPE_SWIZZLE_RED: + return (tc & NV50TIC_0_0_MAPR_MASK) >> NV50TIC_0_0_MAPR_SHIFT; + case PIPE_SWIZZLE_GREEN: + return (tc & NV50TIC_0_0_MAPG_MASK) >> NV50TIC_0_0_MAPG_SHIFT; + case PIPE_SWIZZLE_BLUE: + return (tc & NV50TIC_0_0_MAPB_MASK) >> NV50TIC_0_0_MAPB_SHIFT; + case PIPE_SWIZZLE_ALPHA: + return (tc & NV50TIC_0_0_MAPA_MASK) >> NV50TIC_0_0_MAPA_SHIFT; + case PIPE_SWIZZLE_ONE: + return 7; + case PIPE_SWIZZLE_ZERO: + default: + return 0; + } +} + +boolean +nv50_tex_construct(struct nv50_sampler_view *view) { - unsigned i; - uint32_t mode; const struct util_format_description *desc; + struct nv50_miptree *mt = nv50_miptree(view->pipe.texture); + uint32_t swz[4], *tic = view->tic; - for (i = 0; i < NV50_TEX_FORMAT_LIST_SIZE; i++) - if (nv50_tex_format_list[i].pf == mt->base.base.format) - break; - if (i == NV50_TEX_FORMAT_LIST_SIZE) - return 1; - - if (nv50->sampler[p][unit]->normalized) - mode = 0x50001000 | (1 << 31); - else { - mode = 0x50001000 | (7 << 14); - assert(mt->base.base.target == PIPE_TEXTURE_2D); - } + tic[0] = nv50_texture_formats[view->pipe.format]; - mode |= ((mt->base.bo->tile_mode & 0x0f) << 22) | - ((mt->base.bo->tile_mode & 0xf0) << 21); + swz[0] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_r); + swz[1] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_g); + swz[2] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_b); + swz[3] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_a); + view->tic[0] = (tic[0] & ~NV50TIC_0_0_SWIZZLE_MASK) | + (swz[0] << NV50TIC_0_0_MAPR_SHIFT) | + (swz[1] << NV50TIC_0_0_MAPG_SHIFT) | + (swz[2] << NV50TIC_0_0_MAPB_SHIFT) | + (swz[3] << NV50TIC_0_0_MAPA_SHIFT); - desc = util_format_description(mt->base.base.format); - assert(desc); + tic[2] = 0x50001000; + tic[2] |= ((mt->base.bo->tile_mode & 0x0f) << 22) | + ((mt->base.bo->tile_mode & 0xf0) << 21); + desc = util_format_description(mt->base.base.format); if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) - mode |= 0x0400; + tic[2] |= NV50TIC_0_2_COLORSPACE_SRGB; switch (mt->base.base.target) { case PIPE_TEXTURE_1D: + tic[2] |= NV50TIC_0_2_TARGET_1D; break; case PIPE_TEXTURE_2D: - mode |= (1 << 14); + tic[2] |= NV50TIC_0_2_TARGET_2D; break; case PIPE_TEXTURE_3D: - mode |= (2 << 14); + tic[2] |= NV50TIC_0_2_TARGET_3D; break; case PIPE_TEXTURE_CUBE: - mode |= (3 << 14); + tic[2] |= NV50TIC_0_2_TARGET_CUBE; break; default: - assert(!"unsupported texture target"); - break; + NOUVEAU_ERR("invalid texture target: %d\n", + mt->base.base.target); + return FALSE; } - so_data (so, nv50_tex_format_list[i].hw); - so_reloc(so, mt->base.bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | - NOUVEAU_BO_RD, 0, 0); - so_data (so, mode); - so_data (so, 0x00300000); - so_data (so, mt->base.base.width0 | (1 << 31)); - so_data (so, (mt->base.base.last_level << 28) | - (mt->base.base.depth0 << 16) | mt->base.base.height0); - so_data (so, 0x03000000); - so_data (so, mt->base.base.last_level << 4); - - return 0; -} + tic[3] = 0x00300000; + + tic[4] = (1 << 31) | mt->base.base.width0; + tic[5] = (mt->base.base.last_level << 28) | + (mt->base.base.depth0 << 16) | mt->base.base.height0; + + tic[6] = 0x03000000; -#ifndef NV50TCL_BIND_TIC -#define NV50TCL_BIND_TIC(n) (0x1448 + 8 * n) -#endif + tic[7] = (view->pipe.last_level << 4) | view->pipe.first_level; + + return TRUE; +} -static boolean +static int nv50_validate_textures(struct nv50_context *nv50, struct nouveau_stateobj *so, unsigned p) { - static const unsigned p_remap[PIPE_SHADER_TYPES] = { 0, 2, 1 }; - struct nouveau_grobj *eng2d = nv50->screen->eng2d; struct nouveau_grobj *tesla = nv50->screen->tesla; - unsigned unit, j, p_hw = p_remap[p]; + unsigned unit, j; + + const unsigned rll = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW; + const unsigned rlh = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH + | NOUVEAU_BO_OR; nv50_so_init_sifc(nv50, so, nv50->screen->tic, NOUVEAU_BO_VRAM, - p * (32 * 8 * 4), nv50->miptree_nr[p] * 8 * 4); + p * (32 * 8 * 4), nv50->sampler_view_nr[p] * 8 * 4); - for (unit = 0; unit < nv50->miptree_nr[p]; ++unit) { - struct nv50_miptree *mt = nv50->miptree[p][unit]; + for (unit = 0; unit < nv50->sampler_view_nr[p]; ++unit) { + struct nv50_sampler_view *view = + nv50_sampler_view(nv50->sampler_views[p][unit]); so_method(so, eng2d, NV50_2D_SIFC_DATA | (2 << 29), 8); - if (mt) { - if (nv50_tex_construct(nv50, so, mt, unit, p)) - return FALSE; + if (view) { + uint32_t tic2 = view->tic[2]; + struct nv50_miptree *mt = + nv50_miptree(view->pipe.texture); + + if (nv50->sampler[p][unit]->normalized) + tic2 |= NV50TIC_0_2_NORMALIZED_COORDS; + + so_data (so, view->tic[0]); + so_reloc (so, mt->base.bo, 0, rll, 0, 0); + so_reloc (so, mt->base.bo, 0, rlh, tic2, tic2); + so_datap (so, &view->tic[3], 5); + /* Set TEX insn $t src binding $unit in program type p * to TIC, TSC entry (32 * p + unit), mark valid (1). */ - so_method(so, tesla, NV50TCL_BIND_TIC(p_hw), 1); + so_method(so, tesla, NV50TCL_BIND_TIC(p), 1); so_data (so, ((32 * p + unit) << 9) | (unit << 1) | 1); } else { for (j = 0; j < 8; ++j) so_data(so, 0); - so_method(so, tesla, NV50TCL_BIND_TIC(p_hw), 1); + so_method(so, tesla, NV50TCL_BIND_TIC(p), 1); so_data (so, (unit << 1) | 0); } } - for (; unit < nv50->state.miptree_nr[p]; unit++) { + for (; unit < nv50->state.sampler_view_nr[p]; unit++) { /* Make other bindings invalid. */ - so_method(so, tesla, NV50TCL_BIND_TIC(p_hw), 1); + so_method(so, tesla, NV50TCL_BIND_TIC(p), 1); so_data (so, (unit << 1) | 0); } - nv50->state.miptree_nr[p] = nv50->miptree_nr[p]; + nv50->state.sampler_view_nr[p] = nv50->sampler_view_nr[p]; return TRUE; } @@ -202,23 +220,25 @@ nv50_tex_relocs(struct nv50_context *nv50) int p, unit; p = PIPE_SHADER_FRAGMENT; - for (unit = 0; unit < nv50->miptree_nr[p]; unit++) { - if (!nv50->miptree[p][unit]) + for (unit = 0; unit < nv50->sampler_view_nr[p]; unit++) { + struct pipe_sampler_view *view = nv50->sampler_views[p][unit]; + if (!view) continue; nouveau_reloc_emit(chan, nv50->screen->tic, ((p * 32) + unit) * 32, NULL, - nv50->miptree[p][unit]->base.bo, 0, 0, + nv50_miptree(view->texture)->base.bo, 0, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | NOUVEAU_BO_RD, 0, 0); } p = PIPE_SHADER_VERTEX; - for (unit = 0; unit < nv50->miptree_nr[p]; unit++) { - if (!nv50->miptree[p][unit]) + for (unit = 0; unit < nv50->sampler_view_nr[p]; unit++) { + struct pipe_sampler_view *view = nv50->sampler_views[p][unit]; + if (!view) continue; nouveau_reloc_emit(chan, nv50->screen->tic, ((p * 32) + unit) * 32, NULL, - nv50->miptree[p][unit]->base.bo, 0, 0, + nv50_miptree(view->texture)->base.bo, 0, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | NOUVEAU_BO_RD, 0, 0); } @@ -229,21 +249,23 @@ nv50_tex_validate(struct nv50_context *nv50) { struct nouveau_stateobj *so; struct nouveau_grobj *tesla = nv50->screen->tesla; - unsigned p, start, push, nrlc; - - for (nrlc = 0, start = 0, push = 0, p = 0; p < PIPE_SHADER_TYPES; ++p) { - start += MAX2(nv50->miptree_nr[p], nv50->state.miptree_nr[p]); - push += MAX2(nv50->miptree_nr[p], nv50->state.miptree_nr[p]); - nrlc += nv50->miptree_nr[p]; + unsigned p, m = 0, d = 0, r = 0; + + for (p = 0; p < 3; ++p) { + unsigned nr = MAX2(nv50->sampler_view_nr[p], + nv50->state.sampler_view_nr[p]); + m += nr; + d += nr; + r += nv50->sampler_view_nr[p]; } - start = start * 2 + 4 * PIPE_SHADER_TYPES + 2; - push = push * 9 + 19 * PIPE_SHADER_TYPES + 2; - nrlc = nrlc * 2 + 2 * PIPE_SHADER_TYPES; + m = m * 2 + 3 * 4 + 1; + d = d * 9 + 3 * 19 + 1; + r = r * 2 + 3 * 2; - so = so_new(start, push, nrlc); + so = so_new(m, d, r); - if (nv50_validate_textures(nv50, so, PIPE_SHADER_VERTEX) == FALSE || - nv50_validate_textures(nv50, so, PIPE_SHADER_FRAGMENT) == FALSE) { + if (nv50_validate_textures(nv50, so, 0) == FALSE || + nv50_validate_textures(nv50, so, 2) == FALSE) { so_ref(NULL, &so); NOUVEAU_ERR("failed tex validate\n"); |