X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fnv50%2Fnv50_tex.c;h=5ea0c1d726452e6ab7b20d83f50edf4dc82f9018;hb=23808f1b5e88373534c7ff546cdd89030ce1e935;hp=2813f5447704babcf196458c4a10f9930d9c8230;hpb=b605f4ff11c894500f2d0273c5d4653ff413448d;p=mesa.git diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c index 2813f544770..5ea0c1d7264 100644 --- a/src/gallium/drivers/nv50/nv50_tex.c +++ b/src/gallium/drivers/nv50/nv50_tex.c @@ -22,52 +22,49 @@ #include "nv50_context.h" #include "nv50_texture.h" +#include "nv50_resource.h" #include "nouveau/nouveau_stateobj.h" +#include "nouveau/nouveau_reloc.h" + +#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] = { - _(A8R8G8B8_UNORM, UNORM, C2, C1, C0, C3, 8_8_8_8), - _(A8R8G8B8_SRGB, UNORM, C2, C1, C0, C3, 8_8_8_8), - _(X8R8G8B8_UNORM, UNORM, C2, C1, C0, ONE, 8_8_8_8), - _(X8R8G8B8_SRGB, UNORM, C2, C1, C0, ONE, 8_8_8_8), - _(A1R5G5B5_UNORM, UNORM, C2, C1, C0, C3, 1_5_5_5), - _(A4R4G4B4_UNORM, UNORM, C2, C1, C0, C3, 4_4_4_4), + _(B8G8R8A8_UNORM, UNORM, C2, C1, C0, C3, 8_8_8_8), + _(B8G8R8A8_SRGB, UNORM, C2, C1, C0, C3, 8_8_8_8), + _(B8G8R8X8_UNORM, UNORM, C2, C1, C0, ONE, 8_8_8_8), + _(B8G8R8X8_SRGB, UNORM, C2, C1, C0, ONE, 8_8_8_8), + _(B5G5R5A1_UNORM, UNORM, C2, C1, C0, C3, 1_5_5_5), + _(B4G4R4A4_UNORM, UNORM, C2, C1, C0, C3, 4_4_4_4), - _(R5G6B5_UNORM, UNORM, C2, C1, C0, ONE, 5_6_5), + _(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), - _(A8L8_UNORM, UNORM, C0, C0, C0, C1, 8_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), _(DXT3_RGBA, UNORM, C0, C1, C2, C3, DXT3), _(DXT5_RGBA, UNORM, C0, C1, C2, C3, DXT5), - _MIXED(Z24S8_UNORM, UINT, UNORM, UINT, UINT, C1, C1, C1, ONE, 24_8), + _MIXED(S8_USCALED_Z24_UNORM, UINT, UNORM, UINT, UINT, C1, C1, C1, ONE, 24_8), + _MIXED(Z24_UNORM_S8_USCALED, UNORM, UINT, UINT, UINT, C0, C0, C0, ONE, 8_24), _(R16G16B16A16_SNORM, UNORM, C0, C1, C2, C3, 16_16_16_16), _(R16G16B16A16_UNORM, SNORM, C0, C1, C2, C3, 16_16_16_16), @@ -77,118 +74,211 @@ 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) +static INLINE uint32_t +nv50_tic_swizzle(uint32_t tc, unsigned swz) { - unsigned i; - uint32_t mode; - - 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[unit]->normalized) - mode = 0x50001000 | (1 << 31); - else { - mode = 0x50001000 | (7 << 14); - assert(mt->base.base.target == PIPE_TEXTURE_2D); + 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; } +} - mode |= ((mt->base.bo->tile_mode & 0x0f) << 22) | - ((mt->base.bo->tile_mode & 0xf0) << 21); - - if (pf_type(mt->base.base.format) == PIPE_FORMAT_TYPE_SRGB) - mode |= 0x0400; +boolean +nv50_tex_construct(struct nv50_sampler_view *view) +{ + const struct util_format_description *desc; + struct nv50_miptree *mt = nv50_miptree(view->pipe.texture); + uint32_t swz[4], *tic = view->tic; + + tic[0] = nv50_texture_formats[view->pipe.format]; + + 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); + + 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) + 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.width[0] | (1 << 31)); - so_data (so, (mt->base.base.last_level << 28) | - (mt->base.base.depth[0] << 16) | mt->base.base.height[0]); - 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; + + tic[7] = (view->pipe.last_level << 4) | view->pipe.first_level; + + return TRUE; } -void -nv50_tex_validate(struct nv50_context *nv50) +static int +nv50_validate_textures(struct nv50_context *nv50, struct nouveau_stateobj *so, + unsigned p) { struct nouveau_grobj *eng2d = nv50->screen->eng2d; struct nouveau_grobj *tesla = nv50->screen->tesla; - struct nouveau_stateobj *so; - unsigned i, unit, push; + unsigned unit, j; - push = MAX2(nv50->miptree_nr, nv50->state.miptree_nr) * 2 + 23 + 6; - so = so_new(nv50->miptree_nr * 9 + push, nv50->miptree_nr * 2 + 2); + 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, - nv50->miptree_nr * 8 * 4); + p * (32 * 8 * 4), nv50->sampler_view_nr[p] * 8 * 4); - for (i = 0, unit = 0; unit < nv50->miptree_nr; ++unit) { - struct nv50_miptree *mt = nv50->miptree[unit]; - - if (!mt) - continue; + 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 (nv50_tex_construct(nv50, so, mt, unit)) { - NOUVEAU_ERR("failed tex validate\n"); - so_ref(NULL, &so); - return; + if (view) { + uint32_t tic2 = view->tic[2]; + struct nv50_miptree *mt = + nv50_miptree(view->pipe.texture); + + tic2 &= ~NV50TIC_0_2_NORMALIZED_COORDS; + if (nv50->sampler[p][unit]->normalized) + tic2 |= NV50TIC_0_2_NORMALIZED_COORDS; + view->tic[2] = tic2; + + 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), 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), 1); + so_data (so, (unit << 1) | 0); } + } - so_method(so, tesla, NV50TCL_SET_SAMPLER_TEX, 1); - so_data (so, (i++ << NV50TCL_SET_SAMPLER_TEX_TIC_SHIFT) | - (unit << NV50TCL_SET_SAMPLER_TEX_SAMPLER_SHIFT) | - NV50TCL_SET_SAMPLER_TEX_VALID); + for (; unit < nv50->state.sampler_view_nr[p]; unit++) { + /* Make other bindings invalid. */ + so_method(so, tesla, NV50TCL_BIND_TIC(p), 1); + so_data (so, (unit << 1) | 0); } - for (; unit < nv50->state.miptree_nr; unit++) { - so_method(so, tesla, NV50TCL_SET_SAMPLER_TEX, 1); - so_data (so, - (unit << NV50TCL_SET_SAMPLER_TEX_SAMPLER_SHIFT) | 0); + nv50->state.sampler_view_nr[p] = nv50->sampler_view_nr[p]; + return TRUE; +} + +static void +nv50_emit_texture_relocs(struct nv50_context *nv50, int prog) +{ + struct nouveau_channel *chan = nv50->screen->base.channel; + struct nouveau_bo *tic = nv50->screen->tic; + int unit; + + for (unit = 0; unit < nv50->sampler_view_nr[prog]; unit++) { + struct nv50_sampler_view *view; + struct nv50_miptree *mt; + const unsigned base = ((prog * 32) + unit) * 32; + + view = nv50_sampler_view(nv50->sampler_views[prog][unit]); + if (!view) + continue; + mt = nv50_miptree(view->pipe.texture); + + nouveau_reloc_emit(chan, tic, base + 4, NULL, mt->base.bo, 0, 0, + NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_LOW, 0, 0); + nouveau_reloc_emit(chan, tic, base + 8, NULL, mt->base.bo, 0, 0, + NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_HIGH, view->tic[2], view->tic[2]); + } +} + +void +nv50_tex_relocs(struct nv50_context *nv50) +{ + nv50_emit_texture_relocs(nv50, 2); /* FP */ + nv50_emit_texture_relocs(nv50, 0); /* VP */ +} + +struct nouveau_stateobj * +nv50_tex_validate(struct nv50_context *nv50) +{ + struct nouveau_stateobj *so; + struct nouveau_grobj *tesla = nv50->screen->tesla; + 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]; + } + m = m * 2 + 3 * 4 + 1; + d = d * 9 + 3 * 19 + 1; + r = r * 2 + 3 * 2; + + so = so_new(m, d, r); + + 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"); + return NULL; } - /* not sure if the following really do what I think: */ - so_method(so, tesla, 0x1440, 1); /* sync SIFC */ - so_data (so, 0); so_method(so, tesla, 0x1330, 1); /* flush TIC */ so_data (so, 0); - so_method(so, tesla, 0x1338, 1); /* flush texture caches */ - so_data (so, 0x20); - so_ref(so, &nv50->state.tic_upload); - so_ref(NULL, &so); - nv50->state.miptree_nr = nv50->miptree_nr; + return so; } -