X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_texture.c;h=62c2f1fff6ceeffb3120c13cf351dae6a57244ae;hb=ce9daf6f0bda857c9ee5d021cfb444db6376bfe7;hp=57e0c6183ff84c4b1538debecce2736815ea81d2;hpb=2e6a69939c26ab8cc27bcab29ed64cec5e9428ee;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c index 57e0c6183ff..62c2f1fff6c 100644 --- a/src/gallium/drivers/r300/r300_texture.c +++ b/src/gallium/drivers/r300/r300_texture.c @@ -29,7 +29,6 @@ #include "r300_texture_desc.h" #include "r300_transfer.h" #include "r300_screen.h" -#include "r300_winsys.h" #include "util/u_format.h" #include "util/u_format_s3tc.h" @@ -39,6 +38,18 @@ #include "pipe/p_screen.h" +void util_format_combine_swizzles(unsigned char *dst, + const unsigned char *swz1, + const unsigned char *swz2) +{ + unsigned i; + + for (i = 0; i < 4; i++) { + dst[i] = swz2[i] <= UTIL_FORMAT_SWIZZLE_W ? + swz1[swz2[i]] : swz2[i]; + } +} + unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format, const unsigned char *swizzle_view, boolean dxtc_swizzle) @@ -61,10 +72,7 @@ unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format, if (swizzle_view) { /* Combine two sets of swizzles. */ - for (i = 0; i < 4; i++) { - swizzle[i] = swizzle_view[i] <= UTIL_FORMAT_SWIZZLE_W ? - swizzle_format[swizzle_view[i]] : swizzle_view[i]; - } + util_format_combine_swizzles(swizzle, swizzle_format, swizzle_view); } else { memcpy(swizzle, swizzle_format, 4); } @@ -171,17 +179,22 @@ uint32_t r300_translate_texformat(enum pipe_format format, } } - if (util_format_is_compressed(format) && - dxtc_swizzle && - format != PIPE_FORMAT_RGTC2_UNORM && - format != PIPE_FORMAT_RGTC2_SNORM && - format != PIPE_FORMAT_LATC2_UNORM && - format != PIPE_FORMAT_LATC2_SNORM) { - result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view, - TRUE); - } else { - result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view, - FALSE); + /* Add swizzling. */ + /* The RGTC1_SNORM and LATC1_SNORM swizzle is done in the shader. */ + if (format != PIPE_FORMAT_RGTC1_SNORM && + format != PIPE_FORMAT_LATC1_SNORM) { + if (util_format_is_compressed(format) && + dxtc_swizzle && + format != PIPE_FORMAT_RGTC2_UNORM && + format != PIPE_FORMAT_RGTC2_SNORM && + format != PIPE_FORMAT_LATC2_UNORM && + format != PIPE_FORMAT_LATC2_SNORM) { + result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view, + TRUE); + } else { + result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view, + FALSE); + } } /* S3TC formats. */ @@ -212,7 +225,6 @@ uint32_t r300_translate_texformat(enum pipe_format format, switch (format) { case PIPE_FORMAT_RGTC1_SNORM: case PIPE_FORMAT_LATC1_SNORM: - result |= sign_bit[2]; case PIPE_FORMAT_LATC1_UNORM: case PIPE_FORMAT_RGTC1_UNORM: return R500_TX_FORMAT_ATI1N | result; @@ -404,6 +416,19 @@ static uint32_t r300_translate_colorformat(enum pipe_format format) case PIPE_FORMAT_L8A8_SNORM: case PIPE_FORMAT_R8G8_UNORM: case PIPE_FORMAT_R8G8_SNORM: + /* These formats work fine with UV88 if US_OUT_FMT is set correctly. */ + case PIPE_FORMAT_A16_UNORM: + case PIPE_FORMAT_A16_SNORM: + case PIPE_FORMAT_A16_FLOAT: + case PIPE_FORMAT_L16_UNORM: + case PIPE_FORMAT_L16_SNORM: + case PIPE_FORMAT_L16_FLOAT: + case PIPE_FORMAT_I16_UNORM: + case PIPE_FORMAT_I16_SNORM: + case PIPE_FORMAT_I16_FLOAT: + case PIPE_FORMAT_R16_UNORM: + case PIPE_FORMAT_R16_SNORM: + case PIPE_FORMAT_R16_FLOAT: return R300_COLOR_FORMAT_UV88; case PIPE_FORMAT_B5G6R5_UNORM: @@ -422,31 +447,37 @@ static uint32_t r300_translate_colorformat(enum pipe_format format) /*case PIPE_FORMAT_B8G8R8A8_SNORM:*/ case PIPE_FORMAT_B8G8R8X8_UNORM: /*case PIPE_FORMAT_B8G8R8X8_SNORM:*/ - case PIPE_FORMAT_A8R8G8B8_UNORM: - /*case PIPE_FORMAT_A8R8G8B8_SNORM:*/ - case PIPE_FORMAT_X8R8G8B8_UNORM: - /*case PIPE_FORMAT_X8R8G8B8_SNORM:*/ - case PIPE_FORMAT_A8B8G8R8_UNORM: - /*case PIPE_FORMAT_A8B8G8R8_SNORM:*/ case PIPE_FORMAT_R8G8B8A8_UNORM: case PIPE_FORMAT_R8G8B8A8_SNORM: - case PIPE_FORMAT_X8B8G8R8_UNORM: - /*case PIPE_FORMAT_X8B8G8R8_SNORM:*/ case PIPE_FORMAT_R8G8B8X8_UNORM: /*case PIPE_FORMAT_R8G8B8X8_SNORM:*/ - case PIPE_FORMAT_R8SG8SB8UX8U_NORM: + /* These formats work fine with ARGB8888 if US_OUT_FMT is set + * correctly. */ + case PIPE_FORMAT_R16G16_UNORM: + case PIPE_FORMAT_R16G16_SNORM: + case PIPE_FORMAT_R16G16_FLOAT: + case PIPE_FORMAT_L16A16_UNORM: + case PIPE_FORMAT_L16A16_SNORM: + case PIPE_FORMAT_L16A16_FLOAT: + case PIPE_FORMAT_A32_FLOAT: + case PIPE_FORMAT_L32_FLOAT: + case PIPE_FORMAT_I32_FLOAT: + case PIPE_FORMAT_R32_FLOAT: return R300_COLOR_FORMAT_ARGB8888; case PIPE_FORMAT_R10G10B10A2_UNORM: case PIPE_FORMAT_R10G10B10X2_SNORM: case PIPE_FORMAT_B10G10R10A2_UNORM: - case PIPE_FORMAT_R10SG10SB10SA2U_NORM: return R500_COLOR_FORMAT_ARGB2101010; /* R5xx-only? */ /* 64-bit buffers. */ case PIPE_FORMAT_R16G16B16A16_UNORM: case PIPE_FORMAT_R16G16B16A16_SNORM: case PIPE_FORMAT_R16G16B16A16_FLOAT: + /* These formats work fine with ARGB16161616 if US_OUT_FMT is set + * correctly. */ + case PIPE_FORMAT_R32G32_FLOAT: + case PIPE_FORMAT_L32A32_FLOAT: return R300_COLOR_FORMAT_ARGB16161616; /* 128-bit buffers. */ @@ -489,12 +520,7 @@ static uint32_t r300_translate_out_fmt(enum pipe_format format) uint32_t modifier = 0; unsigned i; const struct util_format_description *desc; - static const uint32_t sign_bit[4] = { - R300_OUT_SIGN(0x4), - R300_OUT_SIGN(0x2), - R300_OUT_SIGN(0x1), - R300_OUT_SIGN(0x8), - }; + boolean uniform_sign; desc = util_format_description(format); @@ -509,34 +535,82 @@ static uint32_t r300_translate_out_fmt(enum pipe_format format) return ~0; /* Unsupported/unknown. */ /* Specifies how the shader output is written to the fog unit. */ - if (desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT) { - if (desc->channel[i].size == 32) { - modifier |= R300_US_OUT_FMT_C4_32_FP; - } else { - modifier |= R300_US_OUT_FMT_C4_16_FP; + switch (desc->channel[i].type) { + case UTIL_FORMAT_TYPE_FLOAT: + switch (desc->channel[i].size) { + case 32: + switch (desc->nr_channels) { + case 1: + modifier |= R300_US_OUT_FMT_C_32_FP; + break; + case 2: + modifier |= R300_US_OUT_FMT_C2_32_FP; + break; + case 4: + modifier |= R300_US_OUT_FMT_C4_32_FP; + break; + } + break; + + case 16: + switch (desc->nr_channels) { + case 1: + modifier |= R300_US_OUT_FMT_C_16_FP; + break; + case 2: + modifier |= R300_US_OUT_FMT_C2_16_FP; + break; + case 4: + modifier |= R300_US_OUT_FMT_C4_16_FP; + break; + } + break; } - } else { - if (desc->channel[i].size == 16) { - modifier |= R300_US_OUT_FMT_C4_16; - } else if (desc->channel[i].size == 10) { + break; + + default: + switch (desc->channel[i].size) { + case 16: + switch (desc->nr_channels) { + case 1: + modifier |= R300_US_OUT_FMT_C_16; + break; + case 2: + modifier |= R300_US_OUT_FMT_C2_16; + break; + case 4: + modifier |= R300_US_OUT_FMT_C4_16; + break; + } + break; + + case 10: modifier |= R300_US_OUT_FMT_C4_10; - } else { + break; + + default: /* C4_8 seems to be used for the formats whose pixel size * is <= 32 bits. */ modifier |= R300_US_OUT_FMT_C4_8; + break; } } /* Add sign. */ - for (i = 0; i < 4; i++) - if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) { - modifier |= sign_bit[i]; - } + uniform_sign = TRUE; + for (i = 0; i < desc->nr_channels; i++) + if (desc->channel[i].type != UTIL_FORMAT_TYPE_SIGNED) + uniform_sign = FALSE; + + if (uniform_sign) + modifier |= R300_OUT_SIGN(0xf); /* Add swizzles and return. */ switch (format) { - /* 8-bit outputs, one channel. - * COLORFORMAT_I8 stores the C2 component. */ + /*** Special cases (non-standard channel mapping) ***/ + + /* X8 + * COLORFORMAT_I8 stores the Z component (C2). */ case PIPE_FORMAT_A8_UNORM: case PIPE_FORMAT_A8_SNORM: return modifier | R300_C2_SEL_A; @@ -548,17 +622,22 @@ static uint32_t r300_translate_out_fmt(enum pipe_format format) case PIPE_FORMAT_R8_SNORM: return modifier | R300_C2_SEL_R; - /* 16-bit outputs, two channels. - * COLORFORMAT_UV88 stores C2 and C0. */ + /* X8Y8 + * COLORFORMAT_UV88 stores ZX (C2 and C0). */ case PIPE_FORMAT_L8A8_SNORM: - modifier |= sign_bit[2]; case PIPE_FORMAT_L8A8_UNORM: return modifier | R300_C0_SEL_A | R300_C2_SEL_R; case PIPE_FORMAT_R8G8_SNORM: - modifier |= sign_bit[2]; case PIPE_FORMAT_R8G8_UNORM: return modifier | R300_C0_SEL_G | R300_C2_SEL_R; + /* X32Y32 + * ARGB16161616 stores XZ for RG32F */ + case PIPE_FORMAT_R32G32_FLOAT: + return modifier | R300_C0_SEL_R | R300_C2_SEL_G; + + /*** Generic cases (standard channel mapping) ***/ + /* BGRA outputs. */ case PIPE_FORMAT_B5G6R5_UNORM: case PIPE_FORMAT_B5G5R5A1_UNORM: @@ -575,40 +654,52 @@ static uint32_t r300_translate_out_fmt(enum pipe_format format) R300_C2_SEL_R | R300_C3_SEL_A; /* ARGB outputs. */ - case PIPE_FORMAT_A8R8G8B8_UNORM: - /*case PIPE_FORMAT_A8R8G8B8_SNORM:*/ - case PIPE_FORMAT_X8R8G8B8_UNORM: - /*case PIPE_FORMAT_X8R8G8B8_SNORM:*/ + case PIPE_FORMAT_A16_UNORM: + case PIPE_FORMAT_A16_SNORM: + case PIPE_FORMAT_A16_FLOAT: + case PIPE_FORMAT_A32_FLOAT: return modifier | R300_C0_SEL_A | R300_C1_SEL_R | R300_C2_SEL_G | R300_C3_SEL_B; - /* ABGR outputs. */ - case PIPE_FORMAT_A8B8G8R8_UNORM: - /*case PIPE_FORMAT_A8B8G8R8_SNORM:*/ - case PIPE_FORMAT_X8B8G8R8_UNORM: - /*case PIPE_FORMAT_X8B8G8R8_SNORM:*/ - return modifier | - R300_C0_SEL_A | R300_C1_SEL_B | - R300_C2_SEL_G | R300_C3_SEL_R; - /* RGBA outputs. */ case PIPE_FORMAT_R8G8B8X8_UNORM: /*case PIPE_FORMAT_R8G8B8X8_SNORM:*/ case PIPE_FORMAT_R8G8B8A8_UNORM: case PIPE_FORMAT_R8G8B8A8_SNORM: - case PIPE_FORMAT_R8SG8SB8UX8U_NORM: case PIPE_FORMAT_R10G10B10A2_UNORM: case PIPE_FORMAT_R10G10B10X2_SNORM: - case PIPE_FORMAT_R10SG10SB10SA2U_NORM: + case PIPE_FORMAT_R16_UNORM: + case PIPE_FORMAT_R16G16_UNORM: case PIPE_FORMAT_R16G16B16A16_UNORM: + case PIPE_FORMAT_R16_SNORM: + case PIPE_FORMAT_R16G16_SNORM: case PIPE_FORMAT_R16G16B16A16_SNORM: + case PIPE_FORMAT_R16_FLOAT: + case PIPE_FORMAT_R16G16_FLOAT: case PIPE_FORMAT_R16G16B16A16_FLOAT: + case PIPE_FORMAT_R32_FLOAT: case PIPE_FORMAT_R32G32B32A32_FLOAT: + case PIPE_FORMAT_L16_UNORM: + case PIPE_FORMAT_L16_SNORM: + case PIPE_FORMAT_L16_FLOAT: + case PIPE_FORMAT_L32_FLOAT: + case PIPE_FORMAT_I16_UNORM: + case PIPE_FORMAT_I16_SNORM: + case PIPE_FORMAT_I16_FLOAT: + case PIPE_FORMAT_I32_FLOAT: return modifier | R300_C0_SEL_R | R300_C1_SEL_G | R300_C2_SEL_B | R300_C3_SEL_A; + /* LA outputs. */ + case PIPE_FORMAT_L16A16_UNORM: + case PIPE_FORMAT_L16A16_SNORM: + case PIPE_FORMAT_L16A16_FLOAT: + case PIPE_FORMAT_L32A32_FLOAT: + return modifier | + R300_C0_SEL_R | R300_C1_SEL_A; + default: return ~0; /* Unsupported. */ } @@ -638,6 +729,16 @@ void r300_texture_setup_format_state(struct r300_screen *screen, struct pipe_resource *pt = &tex->b.b.b; struct r300_texture_desc *desc = &tex->tex; boolean is_r500 = screen->caps.is_r500; + unsigned width, height, depth; + unsigned txwidth, txheight, txdepth; + + width = u_minify(desc->width0, level); + height = u_minify(desc->height0, level); + depth = u_minify(desc->depth0, level); + + txwidth = (width - 1) & 0x7ff; + txheight = (height - 1) & 0x7ff; + txdepth = util_logbase2(depth) & 0xf; /* Mask out all the fields we change. */ out->format0 = 0; @@ -647,9 +748,9 @@ void r300_texture_setup_format_state(struct r300_screen *screen, /* Set sampler state. */ out->format0 = - R300_TX_WIDTH((u_minify(desc->width0, level) - 1) & 0x7ff) | - R300_TX_HEIGHT((u_minify(desc->height0, level) - 1) & 0x7ff) | - R300_TX_DEPTH(util_logbase2(u_minify(desc->depth0, level)) & 0xf); + R300_TX_WIDTH(txwidth) | + R300_TX_HEIGHT(txheight) | + R300_TX_DEPTH(txdepth); if (desc->uses_stride_addressing) { /* rectangles love this */ @@ -667,12 +768,32 @@ void r300_texture_setup_format_state(struct r300_screen *screen, /* large textures on r500 */ if (is_r500) { - if (desc->width0 > 2048) { + unsigned us_width = txwidth; + unsigned us_height = txheight; + unsigned us_depth = txdepth; + + if (width > 2048) { out->format2 |= R500_TXWIDTH_BIT11; } - if (desc->height0 > 2048) { + if (height > 2048) { out->format2 |= R500_TXHEIGHT_BIT11; } + + /* The US_FORMAT register fixes an R500 TX addressing bug. + * Don't ask why it must be set like this. I don't know it either. */ + if (width > 2048) { + us_width = (0x000007FF + us_width) >> 1; + us_depth |= 0x0000000D; + } + if (height > 2048) { + us_height = (0x000007FF + us_height) >> 1; + us_depth |= 0x0000000E; + } + + out->us_format0 = + R300_TX_WIDTH(us_width) | + R300_TX_HEIGHT(us_height) | + R300_TX_DEPTH(us_depth); } out->tile_config = R300_TXO_MACRO_TILE(desc->macrotile[level]) | @@ -731,7 +852,7 @@ static void r300_texture_destroy(struct pipe_screen *screen, { struct r300_resource* tex = (struct r300_resource*)texture; - r300_winsys_bo_reference(&tex->buf, NULL); + pb_reference(&tex->buf, NULL); FREE(tex); } @@ -739,7 +860,7 @@ boolean r300_resource_get_handle(struct pipe_screen* screen, struct pipe_resource *texture, struct winsys_handle *whandle) { - struct r300_winsys_screen *rws = (struct r300_winsys_screen *)screen->winsys; + struct radeon_winsys *rws = (struct radeon_winsys *)screen->winsys; struct r300_resource* tex = (struct r300_resource*)texture; if (!tex) { @@ -766,17 +887,17 @@ static const struct u_resource_vtbl r300_texture_vtbl = static struct r300_resource* r300_texture_create_object(struct r300_screen *rscreen, const struct pipe_resource *base, - enum r300_buffer_tiling microtile, - enum r300_buffer_tiling macrotile, + enum radeon_bo_layout microtile, + enum radeon_bo_layout macrotile, unsigned stride_in_bytes_override, unsigned max_buffer_size, - struct r300_winsys_bo *buffer) + struct pb_buffer *buffer) { - struct r300_winsys_screen *rws = rscreen->rws; + struct radeon_winsys *rws = rscreen->rws; struct r300_resource *tex = CALLOC_STRUCT(r300_resource); if (!tex) { if (buffer) - r300_winsys_bo_reference(&buffer, NULL); + pb_reference(&buffer, NULL); return NULL; } @@ -790,13 +911,13 @@ r300_texture_create_object(struct r300_screen *rscreen, tex->tex.macrotile[0] = macrotile; tex->tex.stride_in_bytes_override = stride_in_bytes_override; tex->domain = base->flags & R300_RESOURCE_FLAG_TRANSFER ? - R300_DOMAIN_GTT : - R300_DOMAIN_VRAM | R300_DOMAIN_GTT; + RADEON_DOMAIN_GTT : + RADEON_DOMAIN_VRAM | RADEON_DOMAIN_GTT; tex->buf_size = max_buffer_size; if (!r300_resource_set_properties(&rscreen->screen, &tex->b.b.b, 0, base)) { if (buffer) - r300_winsys_bo_reference(&buffer, NULL); + pb_reference(&buffer, NULL); FREE(tex); return NULL; } @@ -829,15 +950,16 @@ struct pipe_resource *r300_texture_create(struct pipe_screen *screen, const struct pipe_resource *base) { struct r300_screen *rscreen = r300_screen(screen); - enum r300_buffer_tiling microtile, macrotile; + enum radeon_bo_layout microtile, macrotile; if ((base->flags & R300_RESOURCE_FLAG_TRANSFER) || (base->bind & PIPE_BIND_SCANOUT)) { - microtile = R300_BUFFER_LINEAR; - macrotile = R300_BUFFER_LINEAR; + microtile = RADEON_LAYOUT_LINEAR; + macrotile = RADEON_LAYOUT_LINEAR; } else { - microtile = R300_BUFFER_SELECT_LAYOUT; - macrotile = R300_BUFFER_SELECT_LAYOUT; + /* This will make the texture_create_function select the layout. */ + microtile = RADEON_LAYOUT_UNKNOWN; + macrotile = RADEON_LAYOUT_UNKNOWN; } return (struct pipe_resource*) @@ -849,10 +971,10 @@ struct pipe_resource *r300_texture_from_handle(struct pipe_screen *screen, const struct pipe_resource *base, struct winsys_handle *whandle) { - struct r300_winsys_screen *rws = (struct r300_winsys_screen*)screen->winsys; + struct radeon_winsys *rws = (struct radeon_winsys*)screen->winsys; struct r300_screen *rscreen = r300_screen(screen); - struct r300_winsys_bo *buffer; - enum r300_buffer_tiling microtile, macrotile; + struct pb_buffer *buffer; + enum radeon_bo_layout microtile, macrotile; unsigned stride, size; /* Support only 2D textures without mipmaps */ @@ -871,14 +993,14 @@ struct pipe_resource *r300_texture_from_handle(struct pipe_screen *screen, /* Enforce a microtiled zbuffer. */ if (util_format_is_depth_or_stencil(base->format) && - microtile == R300_BUFFER_LINEAR) { + microtile == RADEON_LAYOUT_LINEAR) { switch (util_format_get_blocksize(base->format)) { case 4: - microtile = R300_BUFFER_TILED; + microtile = RADEON_LAYOUT_TILED; break; case 2: - microtile = R300_BUFFER_SQUARETILED; + microtile = RADEON_LAYOUT_SQUARETILED; break; } } @@ -919,8 +1041,8 @@ struct pipe_surface* r300_create_surface(struct pipe_context * ctx, /* Prefer VRAM if there are multiple domains to choose from. */ surface->domain = tex->domain; - if (surface->domain & R300_DOMAIN_VRAM) - surface->domain &= ~R300_DOMAIN_GTT; + if (surface->domain & RADEON_DOMAIN_VRAM) + surface->domain &= ~RADEON_DOMAIN_GTT; surface->offset = r300_texture_get_offset(tex, level, surf_tmpl->u.tex.first_layer);