X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_state.c;h=2d8e5b1832ad115fcf629fab966c6fce6f5caf25;hb=4087b5343d8d85f46ead6656f58c94cec0289cb0;hp=6e2a6ca0e42d87f850ecc3b409d5beeeacac975f;hpb=d779a5d16ae6a17b3fc0c097f4eb477a80e54566;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index 6e2a6ca0e42..2d8e5b1832a 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -23,10 +23,14 @@ #include "draw/draw_context.h" -#include "util/u_blitter.h" +#include "util/u_framebuffer.h" +#include "util/u_half.h" +#include "util/u_helpers.h" #include "util/u_math.h" #include "util/u_memory.h" #include "util/u_pack_color.h" +#include "util/u_transfer.h" +#include "util/u_blend.h" #include "tgsi/tgsi_parse.h" @@ -42,7 +46,6 @@ #include "r300_fs.h" #include "r300_texture.h" #include "r300_vs.h" -#include "r300_winsys.h" /* r300_state: Functions used to intialize state context by translating * Gallium state objects into semi-native r300 state objects. */ @@ -50,7 +53,7 @@ #define UPDATE_STATE(cso, atom) \ if (cso != atom.state) { \ atom.state = cso; \ - atom.dirty = TRUE; \ + r300_mark_atom_dirty(r300, &(atom)); \ } static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, @@ -167,15 +170,162 @@ static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA dstA == PIPE_BLENDFACTOR_ONE); } -static unsigned bgra_cmask(unsigned mask) +static unsigned blend_discard_conditionally(unsigned eqRGB, unsigned eqA, + unsigned dstRGB, unsigned dstA, + unsigned srcRGB, unsigned srcA) { - /* Gallium uses RGBA color ordering while R300 expects BGRA. */ + unsigned blend_control = 0; + + /* Optimization: discard pixels which don't change the colorbuffer. + * + * The code below is non-trivial and some math is involved. + * + * Discarding pixels must be disabled when FP16 AA is enabled. + * This is a hardware bug. Also, this implementation wouldn't work + * with FP blending enabled and equation clamping disabled. + * + * Equations other than ADD are rarely used and therefore won't be + * optimized. */ + if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && + (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { + /* ADD: X+Y + * REVERSE_SUBTRACT: Y-X + * + * The idea is: + * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, + * then CB will not be changed. + * + * Given the srcFactor and dstFactor variables, we can derive + * what src and dst should be equal to and discard appropriate + * pixels. + */ + if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { + blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; + } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, + dstRGB, dstA)) { + blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; + } else if (blend_discard_if_src_color_0(srcRGB, srcA, + dstRGB, dstA)) { + blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; + } else if (blend_discard_if_src_color_1(srcRGB, srcA, + dstRGB, dstA)) { + blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; + } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, + dstRGB, dstA)) { + blend_control |= + R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; + } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, + dstRGB, dstA)) { + blend_control |= + R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; + } + } + return blend_control; +} +/* The hardware colormask is clunky a must be swizzled depending on the format. + * This was figured out by trial-and-error. */ +static unsigned bgra_cmask(unsigned mask) +{ return ((mask & PIPE_MASK_R) << 2) | ((mask & PIPE_MASK_B) >> 2) | (mask & (PIPE_MASK_G | PIPE_MASK_A)); } +static unsigned rgba_cmask(unsigned mask) +{ + return mask & PIPE_MASK_RGBA; +} + +static unsigned rrrr_cmask(unsigned mask) +{ + return (mask & PIPE_MASK_R) | + ((mask & PIPE_MASK_R) << 1) | + ((mask & PIPE_MASK_R) << 2) | + ((mask & PIPE_MASK_R) << 3); +} + +static unsigned aaaa_cmask(unsigned mask) +{ + return ((mask & PIPE_MASK_A) >> 3) | + ((mask & PIPE_MASK_A) >> 2) | + ((mask & PIPE_MASK_A) >> 1) | + (mask & PIPE_MASK_A); +} + +static unsigned grrg_cmask(unsigned mask) +{ + return ((mask & PIPE_MASK_R) << 1) | + ((mask & PIPE_MASK_R) << 2) | + ((mask & PIPE_MASK_G) >> 1) | + ((mask & PIPE_MASK_G) << 2); +} + +static unsigned arra_cmask(unsigned mask) +{ + return ((mask & PIPE_MASK_R) << 1) | + ((mask & PIPE_MASK_R) << 2) | + ((mask & PIPE_MASK_A) >> 3) | + (mask & PIPE_MASK_A); +} + +static unsigned blend_read_enable(unsigned eqRGB, unsigned eqA, + unsigned dstRGB, unsigned dstA, + unsigned srcRGB, unsigned srcA, + boolean src_alpha_optz) +{ + unsigned blend_control = 0; + + /* Optimization: some operations do not require the destination color. + * + * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, + * otherwise blending gives incorrect results. It seems to be + * a hardware bug. */ + if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || + eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || + dstRGB != PIPE_BLENDFACTOR_ZERO || + dstA != PIPE_BLENDFACTOR_ZERO || + util_blend_factor_uses_dest(srcRGB, false) || + util_blend_factor_uses_dest(srcA, true)) { + /* Enable reading from the colorbuffer. */ + blend_control |= R300_READ_ENABLE; + + if (src_alpha_optz) { + /* Optimization: Depending on incoming pixels, we can + * conditionally disable the reading in hardware... */ + if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && + eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { + /* Disable reading if SRC_ALPHA == 0. */ + if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || + dstRGB == PIPE_BLENDFACTOR_ZERO) && + (dstA == PIPE_BLENDFACTOR_SRC_COLOR || + dstA == PIPE_BLENDFACTOR_SRC_ALPHA || + dstA == PIPE_BLENDFACTOR_ZERO) && + (srcRGB != PIPE_BLENDFACTOR_DST_COLOR && + srcRGB != PIPE_BLENDFACTOR_DST_ALPHA && + srcRGB != PIPE_BLENDFACTOR_INV_DST_COLOR && + srcRGB != PIPE_BLENDFACTOR_INV_DST_ALPHA)) { + blend_control |= R500_SRC_ALPHA_0_NO_READ; + } + + /* Disable reading if SRC_ALPHA == 1. */ + if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + dstRGB == PIPE_BLENDFACTOR_ZERO) && + (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || + dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + dstA == PIPE_BLENDFACTOR_ZERO) && + (srcRGB != PIPE_BLENDFACTOR_DST_COLOR && + srcRGB != PIPE_BLENDFACTOR_DST_ALPHA && + srcRGB != PIPE_BLENDFACTOR_INV_DST_COLOR && + srcRGB != PIPE_BLENDFACTOR_INV_DST_ALPHA)) { + blend_control |= R500_SRC_ALPHA_1_NO_READ; + } + } + } + } + return blend_control; +} + /* Create a new blend state based on the CSO blend state. * * This encompasses alpha blending, logic/raster ops, and blend dithering. */ @@ -185,128 +335,111 @@ static void* r300_create_blend_state(struct pipe_context* pipe, struct r300_screen* r300screen = r300_screen(pipe->screen); struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); uint32_t blend_control = 0; /* R300_RB3D_CBLEND: 0x4e04 */ + uint32_t blend_control_noclamp = 0; /* R300_RB3D_CBLEND: 0x4e04 */ + uint32_t blend_control_noalpha = 0; /* R300_RB3D_CBLEND: 0x4e04 */ + uint32_t blend_control_noalpha_noclamp = 0; /* R300_RB3D_CBLEND: 0x4e04 */ uint32_t alpha_blend_control = 0; /* R300_RB3D_ABLEND: 0x4e08 */ - uint32_t color_channel_mask = 0; /* R300_RB3D_COLOR_CHANNEL_MASK: 0x4e0c */ + uint32_t alpha_blend_control_noclamp = 0; /* R300_RB3D_ABLEND: 0x4e08 */ + uint32_t alpha_blend_control_noalpha = 0; /* R300_RB3D_ABLEND: 0x4e08 */ + uint32_t alpha_blend_control_noalpha_noclamp = 0; /* R300_RB3D_ABLEND: 0x4e08 */ uint32_t rop = 0; /* R300_RB3D_ROPCNTL: 0x4e18 */ uint32_t dither = 0; /* R300_RB3D_DITHER_CTL: 0x4e50 */ + int i; + + const unsigned eqRGB = state->rt[0].rgb_func; + const unsigned srcRGB = state->rt[0].rgb_src_factor; + const unsigned dstRGB = state->rt[0].rgb_dst_factor; + + const unsigned eqA = state->rt[0].alpha_func; + const unsigned srcA = state->rt[0].alpha_src_factor; + const unsigned dstA = state->rt[0].alpha_dst_factor; + + unsigned srcRGBX = srcRGB; + unsigned dstRGBX = dstRGB; CB_LOCALS; - if (state->rt[0].blend_enable) - { - unsigned eqRGB = state->rt[0].rgb_func; - unsigned srcRGB = state->rt[0].rgb_src_factor; - unsigned dstRGB = state->rt[0].rgb_dst_factor; + blend->state = *state; + + /* force DST_ALPHA to ONE where we can */ + switch (srcRGBX) { + case PIPE_BLENDFACTOR_DST_ALPHA: + srcRGBX = PIPE_BLENDFACTOR_ONE; + break; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + srcRGBX = PIPE_BLENDFACTOR_ZERO; + break; + } + + switch (dstRGBX) { + case PIPE_BLENDFACTOR_DST_ALPHA: + dstRGBX = PIPE_BLENDFACTOR_ONE; + break; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + dstRGBX = PIPE_BLENDFACTOR_ZERO; + break; + } - unsigned eqA = state->rt[0].alpha_func; - unsigned srcA = state->rt[0].alpha_src_factor; - unsigned dstA = state->rt[0].alpha_dst_factor; + /* Get blending register values. */ + if (state->rt[0].blend_enable) { + unsigned blend_eq, blend_eq_noclamp; /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, * this is just the crappy D3D naming */ - blend_control = R300_ALPHA_BLEND_ENABLE | - r300_translate_blend_function(eqRGB) | + blend_control = blend_control_noclamp = + R300_ALPHA_BLEND_ENABLE | ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); - /* Optimization: some operations do not require the destination color. - * - * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, - * otherwise blending gives incorrect results. It seems to be - * a hardware bug. */ - if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || - eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || - dstRGB != PIPE_BLENDFACTOR_ZERO || - dstA != PIPE_BLENDFACTOR_ZERO || - srcRGB == PIPE_BLENDFACTOR_DST_COLOR || - srcRGB == PIPE_BLENDFACTOR_DST_ALPHA || - srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR || - srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA || - srcA == PIPE_BLENDFACTOR_DST_COLOR || - srcA == PIPE_BLENDFACTOR_DST_ALPHA || - srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || - srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || - srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { - /* Enable reading from the colorbuffer. */ - blend_control |= R300_READ_ENABLE; - - if (r300screen->caps.is_r500) { - /* Optimization: Depending on incoming pixels, we can - * conditionally disable the reading in hardware... */ - if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && - eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { - /* Disable reading if SRC_ALPHA == 0. */ - if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || - dstRGB == PIPE_BLENDFACTOR_ZERO) && - (dstA == PIPE_BLENDFACTOR_SRC_COLOR || - dstA == PIPE_BLENDFACTOR_SRC_ALPHA || - dstA == PIPE_BLENDFACTOR_ZERO)) { - blend_control |= R500_SRC_ALPHA_0_NO_READ; - } - - /* Disable reading if SRC_ALPHA == 1. */ - if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || - dstRGB == PIPE_BLENDFACTOR_ZERO) && - (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || - dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || - dstA == PIPE_BLENDFACTOR_ZERO)) { - blend_control |= R500_SRC_ALPHA_1_NO_READ; - } - } - } - } + blend_control_noalpha = blend_control_noalpha_noclamp = + R300_ALPHA_BLEND_ENABLE | + ( r300_translate_blend_factor(srcRGBX) << R300_SRC_BLEND_SHIFT) | + ( r300_translate_blend_factor(dstRGBX) << R300_DST_BLEND_SHIFT); + + blend_eq = r300_translate_blend_function(eqRGB, TRUE); + blend_eq_noclamp = r300_translate_blend_function(eqRGB, FALSE); + + blend_control |= blend_eq; + blend_control_noalpha |= blend_eq; + blend_control_noclamp |= blend_eq_noclamp; + blend_control_noalpha_noclamp |= blend_eq_noclamp; + + /* Optimization: some operations do not require the destination color. */ + blend_control |= blend_read_enable(eqRGB, eqA, dstRGB, dstA, + srcRGB, srcA, r300screen->caps.is_r500); + blend_control_noclamp |= blend_read_enable(eqRGB, eqA, dstRGB, dstA, + srcRGB, srcA, FALSE); + blend_control_noalpha |= blend_read_enable(eqRGB, eqA, dstRGBX, dstA, + srcRGBX, srcA, r300screen->caps.is_r500); + blend_control_noalpha_noclamp |= blend_read_enable(eqRGB, eqA, dstRGBX, dstA, + srcRGBX, srcA, FALSE); /* Optimization: discard pixels which don't change the colorbuffer. - * - * The code below is non-trivial and some math is involved. - * - * Discarding pixels must be disabled when FP16 AA is enabled. - * This is a hardware bug. Also, this implementation wouldn't work - * with FP blending enabled and equation clamping disabled. - * - * Equations other than ADD are rarely used and therefore won't be - * optimized. */ - if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && - (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { - /* ADD: X+Y - * REVERSE_SUBTRACT: Y-X - * - * The idea is: - * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, - * then CB will not be changed. - * - * Given the srcFactor and dstFactor variables, we can derive - * what src and dst should be equal to and discard appropriate - * pixels. - */ - if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { - blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; - } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, - dstRGB, dstA)) { - blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; - } else if (blend_discard_if_src_color_0(srcRGB, srcA, - dstRGB, dstA)) { - blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; - } else if (blend_discard_if_src_color_1(srcRGB, srcA, - dstRGB, dstA)) { - blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; - } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, - dstRGB, dstA)) { - blend_control |= - R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; - } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, - dstRGB, dstA)) { - blend_control |= - R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; - } - } + * It cannot be used with FP16 AA. */ + blend_control |= blend_discard_conditionally(eqRGB, eqA, dstRGB, dstA, + srcRGB, srcA); + blend_control_noalpha |= blend_discard_conditionally(eqRGB, eqA, dstRGBX, dstA, + srcRGBX, srcA); /* separate alpha */ if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { blend_control |= R300_SEPARATE_ALPHA_ENABLE; - alpha_blend_control = - r300_translate_blend_function(eqA) | + blend_control_noclamp |= R300_SEPARATE_ALPHA_ENABLE; + + alpha_blend_control = alpha_blend_control_noclamp = (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); + alpha_blend_control |= r300_translate_blend_function(eqA, TRUE); + alpha_blend_control_noclamp |= r300_translate_blend_function(eqA, FALSE); + } + if (srcA != srcRGBX || dstA != dstRGBX || eqA != eqRGB) { + blend_control_noalpha |= R300_SEPARATE_ALPHA_ENABLE; + blend_control_noalpha_noclamp |= R300_SEPARATE_ALPHA_ENABLE; + + alpha_blend_control_noalpha = alpha_blend_control_noalpha_noclamp = + (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | + (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); + alpha_blend_control_noalpha |= r300_translate_blend_function(eqA, TRUE); + alpha_blend_control_noalpha_noclamp |= r300_translate_blend_function(eqA, FALSE); } } @@ -316,20 +449,6 @@ static void* r300_create_blend_state(struct pipe_context* pipe, (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; } - /* Color channel masks for all MRTs. */ - color_channel_mask = bgra_cmask(state->rt[0].colormask); - if (r300screen->caps.is_r500 && state->independent_blend_enable) { - if (state->rt[1].blend_enable) { - color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; - } - if (state->rt[2].blend_enable) { - color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; - } - if (state->rt[3].blend_enable) { - color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; - } - } - /* Neither fglrx nor classic r300 ever set this, regardless of dithering * state. Since it's an optional implementation detail, we can leave it * out and never dither. @@ -343,12 +462,49 @@ static void* r300_create_blend_state(struct pipe_context* pipe, */ /* Build a command buffer. */ - BEGIN_CB(blend->cb, 8); + { + unsigned (*func[COLORMASK_NUM_SWIZZLES])(unsigned) = { + bgra_cmask, + rgba_cmask, + rrrr_cmask, + aaaa_cmask, + grrg_cmask, + arra_cmask, + bgra_cmask, + rgba_cmask + }; + + for (i = 0; i < COLORMASK_NUM_SWIZZLES; i++) { + boolean has_alpha = i != COLORMASK_RGBX && i != COLORMASK_BGRX; + + BEGIN_CB(blend->cb_clamp[i], 8); + OUT_CB_REG(R300_RB3D_ROPCNTL, rop); + OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); + OUT_CB(has_alpha ? blend_control : blend_control_noalpha); + OUT_CB(has_alpha ? alpha_blend_control : alpha_blend_control_noalpha); + OUT_CB(func[i](state->rt[0].colormask)); + OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); + END_CB; + } + } + + /* Build a command buffer (for RGBA16F). */ + BEGIN_CB(blend->cb_noclamp, 8); + OUT_CB_REG(R300_RB3D_ROPCNTL, rop); + OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); + OUT_CB(blend_control_noclamp); + OUT_CB(alpha_blend_control_noclamp); + OUT_CB(rgba_cmask(state->rt[0].colormask)); + OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); + END_CB; + + /* Build a command buffer (for RGB16F). */ + BEGIN_CB(blend->cb_noclamp_noalpha, 8); OUT_CB_REG(R300_RB3D_ROPCNTL, rop); OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); - OUT_CB(blend_control); - OUT_CB(alpha_blend_control); - OUT_CB(color_channel_mask); + OUT_CB(blend_control_noalpha_noclamp); + OUT_CB(alpha_blend_control_noalpha_noclamp); + OUT_CB(rgba_cmask(state->rt[0].colormask)); OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); END_CB; @@ -370,8 +526,27 @@ static void r300_bind_blend_state(struct pipe_context* pipe, void* state) { struct r300_context* r300 = r300_context(pipe); + struct r300_blend_state *blend = (struct r300_blend_state*)state; + boolean last_alpha_to_one = r300->alpha_to_one; + boolean last_alpha_to_coverage = r300->alpha_to_coverage; UPDATE_STATE(state, r300->blend_state); + + if (!blend) + return; + + r300->alpha_to_one = blend->state.alpha_to_one; + r300->alpha_to_coverage = blend->state.alpha_to_coverage; + + if (r300->alpha_to_one != last_alpha_to_one && r300->msaa_enable && + r300->fs_status == FRAGMENT_SHADER_VALID) { + r300->fs_status = FRAGMENT_SHADER_MAYBE_DIRTY; + } + + if (r300->alpha_to_coverage != last_alpha_to_coverage && + r300->msaa_enable) { + r300_mark_atom_dirty(r300, &r300->dsa_state); + } } /* Free blend state. */ @@ -393,29 +568,82 @@ static void r300_set_blend_color(struct pipe_context* pipe, const struct pipe_blend_color* color) { struct r300_context* r300 = r300_context(pipe); - struct r300_blend_color_state* state = + struct pipe_framebuffer_state *fb = r300->fb_state.state; + struct r300_blend_color_state *state = (struct r300_blend_color_state*)r300->blend_color_state.state; + struct pipe_blend_color c; + struct pipe_surface *cb; + float tmp; CB_LOCALS; + state->state = *color; /* Save it, so that we can reuse it in set_fb_state */ + c = *color; + cb = fb->nr_cbufs ? r300_get_nonnull_cb(fb, 0) : NULL; + + /* The blend color is dependent on the colorbuffer format. */ + if (cb) { + switch (cb->format) { + case PIPE_FORMAT_R8_UNORM: + case PIPE_FORMAT_L8_UNORM: + case PIPE_FORMAT_I8_UNORM: + c.color[1] = c.color[0]; + break; + + case PIPE_FORMAT_A8_UNORM: + c.color[1] = c.color[3]; + break; + + case PIPE_FORMAT_R8G8_UNORM: + c.color[2] = c.color[1]; + break; + + case PIPE_FORMAT_L8A8_UNORM: + case PIPE_FORMAT_R8A8_UNORM: + c.color[2] = c.color[3]; + break; + + case PIPE_FORMAT_R8G8B8A8_UNORM: + case PIPE_FORMAT_R8G8B8X8_UNORM: + tmp = c.color[0]; + c.color[0] = c.color[2]; + c.color[2] = tmp; + break; + + default:; + } + } + if (r300->screen->caps.is_r500) { - /* XXX if FP16 blending is enabled, we should use the FP16 format */ BEGIN_CB(state->cb, 3); OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2); - OUT_CB(float_to_fixed10(color->color[0]) | - (float_to_fixed10(color->color[3]) << 16)); - OUT_CB(float_to_fixed10(color->color[2]) | - (float_to_fixed10(color->color[1]) << 16)); + + switch (cb ? cb->format : 0) { + case PIPE_FORMAT_R16G16B16A16_FLOAT: + case PIPE_FORMAT_R16G16B16X16_FLOAT: + OUT_CB(util_float_to_half(c.color[2]) | + (util_float_to_half(c.color[3]) << 16)); + OUT_CB(util_float_to_half(c.color[0]) | + (util_float_to_half(c.color[1]) << 16)); + break; + + default: + OUT_CB(float_to_fixed10(c.color[0]) | + (float_to_fixed10(c.color[3]) << 16)); + OUT_CB(float_to_fixed10(c.color[2]) | + (float_to_fixed10(c.color[1]) << 16)); + } + END_CB; } else { union util_color uc; - util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); + util_pack_color(c.color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); BEGIN_CB(state->cb, 2); - OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui); + OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui[0]); END_CB; } - r300->blend_color_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->blend_color_state); } static void r300_set_clip_state(struct pipe_context* pipe, @@ -426,70 +654,57 @@ static void r300_set_clip_state(struct pipe_context* pipe, (struct r300_clip_state*)r300->clip_state.state; CB_LOCALS; - clip->clip = *state; - if (r300->screen->caps.has_tcl) { - r300->clip_state.size = 2 + !!state->nr * 3 + state->nr * 4; - BEGIN_CB(clip->cb, r300->clip_state.size); - if (state->nr) { - OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG, + OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG, (r300->screen->caps.is_r500 ? R500_PVS_UCP_START : R300_PVS_UCP_START)); - OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, state->nr * 4); - OUT_CB_TABLE(state->ucp, state->nr * 4); - } - OUT_CB_REG(R300_VAP_CLIP_CNTL, ((1 << state->nr) - 1) | - R300_PS_UCP_MODE_CLIP_AS_TRIFAN | - (state->depth_clamp ? R300_CLIP_DISABLE : 0)); + OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, 6 * 4); + OUT_CB_TABLE(state->ucp, 6 * 4); END_CB; - r300->clip_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->clip_state); } else { - draw_flush(r300->draw); draw_set_clip_state(r300->draw, state); } } -static void -r300_set_sample_mask(struct pipe_context *pipe, - unsigned sample_mask) -{ -} - - /* Create a new depth, stencil, and alpha state based on the CSO dsa state. * * This contains the depth buffer, stencil buffer, alpha test, and such. * On the Radeon, depth and stencil buffer setup are intertwined, which is * the reason for some of the strange-looking assignments across registers. */ -static void* - r300_create_dsa_state(struct pipe_context* pipe, - const struct pipe_depth_stencil_alpha_state* state) +static void* r300_create_dsa_state(struct pipe_context* pipe, + const struct pipe_depth_stencil_alpha_state* state) { - struct r300_capabilities *caps = &r300_screen(pipe->screen)->caps; + boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500; struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); CB_LOCALS; + uint32_t alpha_value_fp16 = 0; + uint32_t z_buffer_control = 0; + uint32_t z_stencil_control = 0; + uint32_t stencil_ref_mask = 0; + uint32_t stencil_ref_bf = 0; dsa->dsa = *state; - /* Depth test setup. */ - if (state->depth.enabled) { - dsa->z_buffer_control |= R300_Z_ENABLE; + /* Depth test setup. - separate write mask depth for decomp flush */ + if (state->depth.writemask) { + z_buffer_control |= R300_Z_WRITE_ENABLE; + } - if (state->depth.writemask) { - dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; - } + if (state->depth.enabled) { + z_buffer_control |= R300_Z_ENABLE; - dsa->z_stencil_control |= + z_stencil_control |= (r300_translate_depth_stencil_function(state->depth.func) << R300_Z_FUNC_SHIFT); } /* Stencil buffer setup. */ if (state->stencil[0].enabled) { - dsa->z_buffer_control |= R300_STENCIL_ENABLE; - dsa->z_stencil_control |= + z_buffer_control |= R300_STENCIL_ENABLE; + z_stencil_control |= (r300_translate_depth_stencil_function(state->stencil[0].func) << R300_S_FRONT_FUNC_SHIFT) | (r300_translate_stencil_op(state->stencil[0].fail_op) << @@ -499,15 +714,15 @@ static void* (r300_translate_stencil_op(state->stencil[0].zfail_op) << R300_S_FRONT_ZFAIL_OP_SHIFT); - dsa->stencil_ref_mask = + stencil_ref_mask = (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); if (state->stencil[1].enabled) { dsa->two_sided = TRUE; - dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK; - dsa->z_stencil_control |= + z_buffer_control |= R300_STENCIL_FRONT_BACK; + z_stencil_control |= (r300_translate_depth_stencil_function(state->stencil[1].func) << R300_S_BACK_FUNC_SHIFT) | (r300_translate_stencil_op(state->stencil[1].fail_op) << @@ -517,12 +732,12 @@ static void* (r300_translate_stencil_op(state->stencil[1].zfail_op) << R300_S_BACK_ZFAIL_OP_SHIFT); - dsa->stencil_ref_bf = + stencil_ref_bf = (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) | (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT); - if (caps->is_r500) { - dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; + if (is_r500) { + z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; } else { dsa->two_sided_stencil_ref = (state->stencil[0].valuemask != state->stencil[1].valuemask || @@ -537,29 +752,26 @@ static void* r300_translate_alpha_function(state->alpha.func) | R300_FG_ALPHA_FUNC_ENABLE; - /* We could use 10bit alpha ref but who needs that? */ dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); - - if (caps->is_r500) - dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; + alpha_value_fp16 = util_float_to_half(state->alpha.ref_value); } BEGIN_CB(&dsa->cb_begin, 8); - OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); - OUT_CB(dsa->z_buffer_control); - OUT_CB(dsa->z_stencil_control); - OUT_CB(dsa->stencil_ref_mask); - OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); + OUT_CB(z_buffer_control); + OUT_CB(z_stencil_control); + OUT_CB(stencil_ref_mask); + OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, stencil_ref_bf); + OUT_CB_REG(R500_FG_ALPHA_VALUE, alpha_value_fp16); END_CB; - BEGIN_CB(dsa->cb_no_readwrite, 8); - OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); + BEGIN_CB(dsa->cb_zb_no_readwrite, 8); OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); OUT_CB(0); OUT_CB(0); OUT_CB(0); OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0); + OUT_CB_REG(R500_FG_ALPHA_VALUE, alpha_value_fp16); END_CB; return (void*)dsa; @@ -593,6 +805,7 @@ static void r300_bind_dsa_state(struct pipe_context* pipe, UPDATE_STATE(state, r300->dsa_state); + r300_mark_atom_dirty(r300, &r300->hyperz_state); /* Will be updated before the emission. */ r300_dsa_inject_stencilref(r300); } @@ -611,71 +824,30 @@ static void r300_set_stencil_ref(struct pipe_context* pipe, r300->stencil_ref = *sr; r300_dsa_inject_stencilref(r300); - r300->dsa_state.dirty = TRUE; -} - -static void r300_tex_set_tiling_flags(struct r300_context *r300, - struct r300_texture *tex, unsigned level) -{ - /* Check if the macrotile flag needs to be changed. - * Skip changing the flags otherwise. */ - if (tex->desc.macrotile[tex->surface_level] != - tex->desc.macrotile[level]) { - /* Tiling determines how DRM treats the buffer data. - * We must flush CS when changing it if the buffer is referenced. */ - if (r300->rws->cs_is_buffer_referenced(r300->cs, - tex->buffer, R300_REF_CS)) - r300->context.flush(&r300->context, 0, NULL); - - r300->rws->buffer_set_tiling(r300->rws, tex->buffer, - tex->desc.microtile, tex->desc.macrotile[level], - tex->desc.stride_in_bytes[0]); - - tex->surface_level = level; - } -} - -/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ -static void r300_fb_set_tiling_flags(struct r300_context *r300, - const struct pipe_framebuffer_state *state) -{ - unsigned i; - - /* Set tiling flags for new surfaces. */ - for (i = 0; i < state->nr_cbufs; i++) { - r300_tex_set_tiling_flags(r300, - r300_texture(state->cbufs[i]->texture), - state->cbufs[i]->level); - } - if (state->zsbuf) { - r300_tex_set_tiling_flags(r300, - r300_texture(state->zsbuf->texture), - state->zsbuf->level); - } + r300_mark_atom_dirty(r300, &r300->dsa_state); } static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index, const char *binding) { struct pipe_resource *tex = surf->texture; - struct r300_texture *rtex = r300_texture(tex); + struct r300_resource *rtex = r300_resource(tex); fprintf(stderr, - "r300: %s[%i] Dim: %ix%i, Offset: %i, ZSlice: %i, " - "Face: %i, Level: %i, Format: %s\n" + "r300: %s[%i] Dim: %ix%i, Firstlayer: %i, " + "Lastlayer: %i, Level: %i, Format: %s\n" - "r300: TEX: Macro: %s, Micro: %s, Pitch: %i, " + "r300: TEX: Macro: %s, Micro: %s, " "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n", - binding, index, surf->width, surf->height, surf->offset, - surf->zslice, surf->face, surf->level, + binding, index, surf->width, surf->height, + surf->u.tex.first_layer, surf->u.tex.last_layer, surf->u.tex.level, util_format_short_name(surf->format), - rtex->desc.macrotile[0] ? "YES" : " NO", - rtex->desc.microtile ? "YES" : " NO", - rtex->desc.stride_in_pixels[0], + rtex->tex.macrotile[0] ? "YES" : " NO", + rtex->tex.microtile ? "YES" : " NO", tex->width0, tex->height0, tex->depth0, - tex->last_level, util_format_short_name(tex->format)); + tex->last_level, util_format_short_name(surf->format)); } void r300_mark_fb_state_dirty(struct r300_context *r300, @@ -683,14 +855,24 @@ void r300_mark_fb_state_dirty(struct r300_context *r300, { struct pipe_framebuffer_state *state = r300->fb_state.state; - /* What is marked as dirty depends on the enum r300_fb_state_change. */ - r300->gpu_flush.dirty = TRUE; - r300->fb_state.dirty = TRUE; - r300->hyperz_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->gpu_flush); + r300_mark_atom_dirty(r300, &r300->fb_state); + /* What is marked as dirty depends on the enum r300_fb_state_change. */ if (change == R300_CHANGED_FB_STATE) { - r300->aa_state.dirty = TRUE; - r300->fb_state_pipelined.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->aa_state); + r300_mark_atom_dirty(r300, &r300->dsa_state); /* for AlphaRef */ + r300_set_blend_color(&r300->context, r300->blend_color_state.state); + } + + if (change == R300_CHANGED_FB_STATE || + change == R300_CHANGED_HYPERZ_FLAG) { + r300_mark_atom_dirty(r300, &r300->hyperz_state); + } + + if (change == R300_CHANGED_FB_STATE || + change == R300_CHANGED_MULTIWRITE) { + r300_mark_atom_dirty(r300, &r300->fb_state_pipelined); } /* Now compute the fb_state atom size. */ @@ -698,21 +880,32 @@ void r300_mark_fb_state_dirty(struct r300_context *r300, if (r300->cbzb_clear) r300->fb_state.size += 10; - else if (state->zsbuf) - r300->fb_state.size += r300->screen->caps.has_hiz ? 18 : 14; + else if (state->zsbuf) { + r300->fb_state.size += 10; + if (r300->hyperz_enabled) + r300->fb_state.size += 8; + } + + if (r300->cmask_in_use) { + r300->fb_state.size += 6; + if (r300->screen->caps.is_r500 && r300->screen->info.drm_minor >= 29) { + r300->fb_state.size += 3; + } + } /* The size of the rest of atoms stays the same. */ } static void - r300_set_framebuffer_state(struct pipe_context* pipe, - const struct pipe_framebuffer_state* state) +r300_set_framebuffer_state(struct pipe_context* pipe, + const struct pipe_framebuffer_state* state) { struct r300_context* r300 = r300_context(pipe); struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; - struct pipe_framebuffer_state *old_state = r300->fb_state.state; + struct pipe_framebuffer_state *current_state = r300->fb_state.state; unsigned max_width, max_height, i; uint32_t zbuffer_bpp = 0; + boolean unlock_zbuffer = FALSE; if (r300->screen->caps.is_r500) { max_width = max_height = 4096; @@ -728,71 +921,108 @@ static void return; } - if (r300->draw) { - draw_flush(r300->draw); + if (current_state->zsbuf && r300->zmask_in_use && !r300->locked_zbuffer) { + /* There is a zmask in use, what are we gonna do? */ + if (state->zsbuf) { + if (!pipe_surface_equal(current_state->zsbuf, state->zsbuf)) { + /* Decompress the currently bound zbuffer before we bind another one. */ + r300_decompress_zmask(r300); + r300->hiz_in_use = FALSE; + } + } else { + /* We don't bind another zbuffer, so lock the current one. */ + pipe_surface_reference(&r300->locked_zbuffer, current_state->zsbuf); + } + } else if (r300->locked_zbuffer) { + /* We have a locked zbuffer now, what are we gonna do? */ + if (state->zsbuf) { + if (!pipe_surface_equal(r300->locked_zbuffer, state->zsbuf)) { + /* We are binding some other zbuffer, so decompress the locked one, + * it gets unlocked automatically. */ + r300_decompress_zmask_locked_unsafe(r300); + r300->hiz_in_use = FALSE; + } else { + /* We are binding the locked zbuffer again, so unlock it. */ + unlock_zbuffer = TRUE; + } + } } + assert(state->zsbuf || (r300->locked_zbuffer && !unlock_zbuffer) || !r300->zmask_in_use); - /* If nr_cbufs is changed from zero to non-zero or vice versa... */ - if (!!old_state->nr_cbufs != !!state->nr_cbufs) { - r300->blend_state.dirty = TRUE; - } /* If zsbuf is set from NULL to non-NULL or vice versa.. */ - if (!!old_state->zsbuf != !!state->zsbuf) { - r300->dsa_state.dirty = TRUE; + if (!!current_state->zsbuf != !!state->zsbuf) { + r300_mark_atom_dirty(r300, &r300->dsa_state); } - /* The tiling flags are dependent on the surface miplevel, unfortunately. */ - r300_fb_set_tiling_flags(r300, state); + util_copy_framebuffer_state(r300->fb_state.state, state); + + /* Remove trailing NULL colorbuffers. */ + while (current_state->nr_cbufs && !current_state->cbufs[current_state->nr_cbufs-1]) + current_state->nr_cbufs--; + + /* Set whether CMASK can be used. */ + r300->cmask_in_use = + state->nr_cbufs == 1 && state->cbufs[0] && + r300->screen->cmask_resource == state->cbufs[0]->texture; + + /* Need to reset clamping or colormask. */ + r300_mark_atom_dirty(r300, &r300->blend_state); - util_assign_framebuffer_state(r300->fb_state.state, state); + /* Re-swizzle the blend color. */ + r300_set_blend_color(pipe, &((struct r300_blend_color_state*)r300->blend_color_state.state)->state); + + if (unlock_zbuffer) { + pipe_surface_reference(&r300->locked_zbuffer, NULL); + } r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE); - /* Polygon offset depends on the zbuffer bit depth. */ - if (state->zsbuf && r300->polygon_offset_enabled) { - switch (util_format_get_blocksize(state->zsbuf->texture->format)) { - case 2: - zbuffer_bpp = 16; - break; - case 4: - zbuffer_bpp = 24; - break; + if (state->zsbuf) { + switch (util_format_get_blocksize(state->zsbuf->format)) { + case 2: + zbuffer_bpp = 16; + break; + case 4: + zbuffer_bpp = 24; + break; } + /* Polygon offset depends on the zbuffer bit depth. */ if (r300->zbuffer_bpp != zbuffer_bpp) { r300->zbuffer_bpp = zbuffer_bpp; - r300->rs_state.dirty = TRUE; + + if (r300->polygon_offset_enabled) + r300_mark_atom_dirty(r300, &r300->rs_state); } } + r300->num_samples = util_framebuffer_get_num_samples(state); + /* Set up AA config. */ - if (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) { - if (state->nr_cbufs && state->cbufs[0]->texture->nr_samples > 1) { - aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE; - - switch (state->cbufs[0]->texture->nr_samples) { - case 2: - aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2; - break; - case 3: - aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3; - break; - case 4: - aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4; - break; - case 6: - aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6; - break; - } - } else { - aa->aa_config = 0; + if (r300->num_samples > 1) { + switch (r300->num_samples) { + case 2: + aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE | + R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2; + break; + case 4: + aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE | + R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4; + break; + case 6: + aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE | + R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6; + break; } + } else { + aa->aa_config = 0; } if (DBG_ON(r300, DBG_FB)) { fprintf(stderr, "r300: set_framebuffer_state:\n"); for (i = 0; i < state->nr_cbufs; i++) { - r300_print_fb_surf_info(state->cbufs[i], i, "CB"); + if (state->cbufs[i]) + r300_print_fb_surf_info(state->cbufs[i], i, "CB"); } if (state->zsbuf) { r300_print_fb_surf_info(state->zsbuf, 0, "ZB"); @@ -819,9 +1049,9 @@ void r300_mark_fs_code_dirty(struct r300_context *r300) { struct r300_fragment_shader* fs = r300_fs(r300); - r300->fs.dirty = TRUE; - r300->fs_rc_constant_state.dirty = TRUE; - r300->fs_constants.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->fs); + r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); + r300_mark_atom_dirty(r300, &r300->fs_constants); r300->fs.size = fs->shader->cb_code_size; if (r300->screen->caps.is_r500) { @@ -831,6 +1061,9 @@ void r300_mark_fs_code_dirty(struct r300_context *r300) r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5; r300->fs_constants.size = fs->shader->externals_count * 4 + 1; } + + ((struct r300_constant_buffer*)r300->fs_constants.state)->remap_table = + fs->shader->code.constants_remap_table; } /* Bind fragment shader state. */ @@ -839,16 +1072,15 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) struct r300_context* r300 = r300_context(pipe); struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; - if (fs == NULL) { + if (!fs) { r300->fs.state = NULL; return; } r300->fs.state = fs; - r300_pick_fragment_shader(r300); - r300_mark_fs_code_dirty(r300); + r300->fs_status = FRAGMENT_SHADER_DIRTY; - r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ + r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ } /* Delete fragment shader state. */ @@ -871,7 +1103,6 @@ static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) static void r300_set_polygon_stipple(struct pipe_context* pipe, const struct pipe_poly_stipple* state) { - /* XXX no idea how to set this up, but not terribly important */ } /* Create a new rasterizer state based on the CSO rasterizer state. @@ -885,9 +1116,8 @@ static void* r300_create_rs_state(struct pipe_context* pipe, const struct pipe_rasterizer_state* state) { struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); - int i; - float psiz; uint32_t vap_control_status; /* R300_VAP_CNTL_STATUS: 0x2140 */ + uint32_t vap_clip_cntl; /* R300_VAP_CLIP_CNTL: 0x221C */ uint32_t point_size; /* R300_GA_POINT_SIZE: 0x421c */ uint32_t point_minmax; /* R300_GA_POINT_MINMAX: 0x4230 */ uint32_t line_control; /* R300_GA_LINE_CNTL: 0x4234 */ @@ -897,26 +1127,31 @@ static void* r300_create_rs_state(struct pipe_context* pipe, uint32_t line_stipple_value; /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */ uint32_t polygon_mode; /* R300_GA_POLY_MODE: 0x4288 */ uint32_t clip_rule; /* R300_SC_CLIP_RULE: 0x43D0 */ - - /* Specifies top of Raster pipe specific enable controls, - * i.e. texture coordinates stuffing for points, lines, triangles */ - uint32_t stuffing_enable; /* R300_GB_ENABLE: 0x4008 */ + uint32_t round_mode; /* R300_GA_ROUND_MODE: 0x428c */ /* Point sprites texture coordinates, 0: lower left, 1: upper right */ - float point_texcoord_left; /* R300_GA_POINT_S0: 0x4200 */ + float point_texcoord_left = 0; /* R300_GA_POINT_S0: 0x4200 */ float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */ - float point_texcoord_right; /* R300_GA_POINT_S1: 0x4208 */ + float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */ float point_texcoord_top = 0; /* R300_GA_POINT_T1: 0x420c */ + boolean vclamp = !r300_context(pipe)->screen->caps.is_r500; CB_LOCALS; /* Copy rasterizer state. */ rs->rs = *state; rs->rs_draw = *state; + rs->rs.sprite_coord_enable = state->point_quad_rasterization * + state->sprite_coord_enable; + /* Override some states for Draw. */ rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */ + rs->rs_draw.offset_point = 0; + rs->rs_draw.offset_line = 0; + rs->rs_draw.offset_tri = 0; + rs->rs_draw.offset_clamp = 0; -#ifdef PIPE_ARCH_LITTLE_ENDIAN +#if UTIL_ARCH_LITTLE_ENDIAN vap_control_status = R300_VC_NO_SWAP; #else vap_control_status = R300_VC_32BIT_SWAP; @@ -936,14 +1171,16 @@ static void* r300_create_rs_state(struct pipe_context* pipe, if (state->point_size_per_vertex) { /* Per-vertex point size. * Clamp to [0, max FB size] */ - psiz = pipe->screen->get_paramf(pipe->screen, - PIPE_CAP_MAX_POINT_WIDTH); + float min_psiz = util_get_min_point_size(state); + float max_psiz = pipe->screen->get_paramf(pipe->screen, + PIPE_CAPF_MAX_POINT_WIDTH); point_minmax = - pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT; + (pack_float_16_6x(min_psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) | + (pack_float_16_6x(max_psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT); } else { /* We cannot disable the point-size vertex output, * so clamp it. */ - psiz = state->point_size; + float psiz = state->point_size; point_minmax = (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) | (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT); @@ -1012,19 +1249,8 @@ static void* r300_create_rs_state(struct pipe_context* pipe, clip_rule = state->scissor ? 0xAAAA : 0xFFFF; - /* Point sprites */ - stuffing_enable = 0; - if (state->sprite_coord_enable) { - stuffing_enable = R300_GB_POINT_STUFF_ENABLE; - for (i = 0; i < 8; i++) { - if (state->sprite_coord_enable & (1 << i)) - stuffing_enable |= - R300_GB_TEX_STR << (R300_GB_TEX0_SOURCE_SHIFT + (i*2)); - } - - point_texcoord_left = 0.0f; - point_texcoord_right = 1.0f; - + /* Point sprites coord mode */ + if (rs->rs.sprite_coord_enable) { switch (state->sprite_coord_mode) { case PIPE_SPRITE_COORD_UPPER_LEFT: point_texcoord_top = 0.0f; @@ -1037,22 +1263,36 @@ static void* r300_create_rs_state(struct pipe_context* pipe, } } + if (r300_screen(pipe->screen)->caps.has_tcl) { + vap_clip_cntl = (state->clip_plane_enable & 63) | + R300_PS_UCP_MODE_CLIP_AS_TRIFAN; + } else { + vap_clip_cntl = R300_CLIP_DISABLE; + } + + /* Vertex color clamping. FP20 means no clamping. */ + round_mode = + R300_GA_ROUND_MODE_GEOMETRY_ROUND_NEAREST | + (!vclamp ? (R300_GA_ROUND_MODE_RGB_CLAMP_FP20 | + R300_GA_ROUND_MODE_ALPHA_CLAMP_FP20) : 0); + /* Build the main command buffer. */ - BEGIN_CB(rs->cb_main, 25); + BEGIN_CB(rs->cb_main, RS_STATE_MAIN_SIZE); OUT_CB_REG(R300_VAP_CNTL_STATUS, vap_control_status); + OUT_CB_REG(R300_VAP_CLIP_CNTL, vap_clip_cntl); OUT_CB_REG(R300_GA_POINT_SIZE, point_size); OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX, 2); OUT_CB(point_minmax); OUT_CB(line_control); OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2); OUT_CB(polygon_offset_enable); - rs->cull_mode_index = 9; + rs->cull_mode_index = 11; OUT_CB(cull_mode); OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG, line_stipple_config); OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE, line_stipple_value); OUT_CB_REG(R300_GA_POLY_MODE, polygon_mode); + OUT_CB_REG(R300_GA_ROUND_MODE, round_mode); OUT_CB_REG(R300_SC_CLIP_RULE, clip_rule); - OUT_CB_REG(R300_GB_ENABLE, stuffing_enable); OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4); OUT_CB_32F(point_texcoord_left); OUT_CB_32F(point_texcoord_bottom); @@ -1094,30 +1334,52 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) struct r300_rs_state* rs = (struct r300_rs_state*)state; int last_sprite_coord_enable = r300->sprite_coord_enable; boolean last_two_sided_color = r300->two_sided_color; + boolean last_msaa_enable = r300->msaa_enable; + boolean last_flatshade = r300->flatshade; + boolean last_clip_halfz = r300->clip_halfz; if (r300->draw && rs) { - draw_flush(r300->draw); draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state); } if (rs) { - r300->polygon_offset_enabled = (rs->rs.offset_point || - rs->rs.offset_line || - rs->rs.offset_tri); + r300->polygon_offset_enabled = rs->polygon_offset_enable; r300->sprite_coord_enable = rs->rs.sprite_coord_enable; r300->two_sided_color = rs->rs.light_twoside; + r300->msaa_enable = rs->rs.multisample; + r300->flatshade = rs->rs.flatshade; + r300->clip_halfz = rs->rs.clip_halfz; } else { r300->polygon_offset_enabled = FALSE; r300->sprite_coord_enable = 0; r300->two_sided_color = FALSE; + r300->msaa_enable = FALSE; + r300->flatshade = FALSE; + r300->clip_halfz = FALSE; } UPDATE_STATE(state, r300->rs_state); - r300->rs_state.size = 25 + (r300->polygon_offset_enabled ? 5 : 0); + r300->rs_state.size = RS_STATE_MAIN_SIZE + (r300->polygon_offset_enabled ? 5 : 0); if (last_sprite_coord_enable != r300->sprite_coord_enable || - last_two_sided_color != r300->two_sided_color) { - r300->rs_block_state.dirty = TRUE; + last_two_sided_color != r300->two_sided_color || + last_flatshade != r300->flatshade) { + r300_mark_atom_dirty(r300, &r300->rs_block_state); + } + + if (last_msaa_enable != r300->msaa_enable) { + if (r300->alpha_to_coverage) { + r300_mark_atom_dirty(r300, &r300->dsa_state); + } + + if (r300->alpha_to_one && + r300->fs_status == FRAGMENT_SHADER_VALID) { + r300->fs_status = FRAGMENT_SHADER_MAYBE_DIRTY; + } + } + + if (r300->screen->caps.has_tcl && last_clip_halfz != r300->clip_halfz) { + r300_mark_atom_dirty(r300, &r300->vs_state); } } @@ -1135,7 +1397,6 @@ static void* struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); boolean is_r500 = r300->screen->caps.is_r500; int lod_bias; - union util_color uc; sampler->state = *state; @@ -1171,14 +1432,14 @@ static void* sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, state->mag_img_filter, state->min_mip_filter, - state->max_anisotropy > 0); + state->max_anisotropy > 1); sampler->filter0 |= r300_anisotropy(state->max_anisotropy); /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ /* We must pass these to the merge function to clamp them properly. */ - sampler->min_lod = MAX2((unsigned)state->min_lod, 0); - sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); + sampler->min_lod = (unsigned)MAX2(state->min_lod, 0); + sampler->max_lod = (unsigned)MAX2(ceilf(state->max_lod), 0); lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1); @@ -1192,9 +1453,6 @@ static void* sampler->filter1 |= r500_anisotropy(state->max_anisotropy); } - util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); - sampler->border_color = uc.ui; - /* R500-specific fixups and optimizations */ if (r300->screen->caps.is_r500) { sampler->filter1 |= R500_BORDER_FIX; @@ -1204,7 +1462,8 @@ static void* } static void r300_bind_sampler_states(struct pipe_context* pipe, - unsigned count, + enum pipe_shader_type shader, + unsigned start, unsigned count, void** states) { struct r300_context* r300 = r300_context(pipe); @@ -1212,20 +1471,18 @@ static void r300_bind_sampler_states(struct pipe_context* pipe, (struct r300_textures_state*)r300->textures_state.state; unsigned tex_units = r300->screen->caps.num_tex_units; - if (count > tex_units) { - return; - } + assert(start == 0); + + if (shader != PIPE_SHADER_FRAGMENT) + return; + + if (count > tex_units) + return; memcpy(state->sampler_states, states, sizeof(void*) * count); state->sampler_state_count = count; - r300->textures_state.dirty = TRUE; -} - -static void r300_lacks_vertex_textures(struct pipe_context* pipe, - unsigned count, - void** states) -{ + r300_mark_atom_dirty(r300, &r300->textures_state); } static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) @@ -1258,18 +1515,24 @@ static uint32_t r300_assign_texture_cache_region(unsigned index, unsigned num) return R300_TX_CACHE(num + index); } -static void r300_set_fragment_sampler_views(struct pipe_context* pipe, - unsigned count, - struct pipe_sampler_view** views) +static void r300_set_sampler_views(struct pipe_context* pipe, + enum pipe_shader_type shader, + unsigned start, unsigned count, + struct pipe_sampler_view** views) { struct r300_context* r300 = r300_context(pipe); struct r300_textures_state* state = (struct r300_textures_state*)r300->textures_state.state; - struct r300_texture *texture; + struct r300_resource *texture; unsigned i, real_num_views = 0, view_index = 0; unsigned tex_units = r300->screen->caps.num_tex_units; boolean dirty_tex = FALSE; + if (shader != PIPE_SHADER_FRAGMENT) + return; + + assert(start == 0); /* non-zero not handled yet */ + if (count > tex_units) { return; } @@ -1281,29 +1544,27 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe, } for (i = 0; i < count; i++) { - if (&state->sampler_views[i]->base != views[i]) { - pipe_sampler_view_reference( - (struct pipe_sampler_view**)&state->sampler_views[i], - views[i]); + pipe_sampler_view_reference( + (struct pipe_sampler_view**)&state->sampler_views[i], + views[i]); - if (!views[i]) { - continue; - } + if (!views[i]) { + continue; + } - /* A new sampler view (= texture)... */ - dirty_tex = TRUE; + /* A new sampler view (= texture)... */ + dirty_tex = TRUE; - /* Set the texrect factor in the fragment shader. + /* Set the texrect factor in the fragment shader. * Needed for RECT and NPOT fallback. */ - texture = r300_texture(views[i]->texture); - if (texture->desc.uses_stride_addressing) { - r300->fs_rc_constant_state.dirty = TRUE; - } + texture = r300_resource(views[i]->texture); + if (texture->tex.is_npot) { + r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); + } - state->sampler_views[i]->texcache_region = + state->sampler_views[i]->texcache_region = r300_assign_texture_cache_region(view_index, real_num_views); - view_index++; - } + view_index++; } for (i = count; i < tex_units; i++) { @@ -1316,37 +1577,58 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe, state->sampler_view_count = count; - r300->textures_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->textures_state); if (dirty_tex) { - r300->texture_cache_inval.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->texture_cache_inval); } } -static struct pipe_sampler_view * -r300_create_sampler_view(struct pipe_context *pipe, +struct pipe_sampler_view * +r300_create_sampler_view_custom(struct pipe_context *pipe, struct pipe_resource *texture, - const struct pipe_sampler_view *templ) + const struct pipe_sampler_view *templ, + unsigned width0_override, + unsigned height0_override) { struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view); - struct r300_texture *tex = r300_texture(texture); + struct r300_resource *tex = r300_resource(texture); + boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500; + boolean dxtc_swizzle = r300_screen(pipe->screen)->caps.dxtc_swizzle; if (view) { + unsigned hwformat; + view->base = *templ; view->base.reference.count = 1; view->base.context = pipe; view->base.texture = NULL; pipe_resource_reference(&view->base.texture, texture); + view->width0_override = width0_override; + view->height0_override = height0_override; view->swizzle[0] = templ->swizzle_r; view->swizzle[1] = templ->swizzle_g; view->swizzle[2] = templ->swizzle_b; view->swizzle[3] = templ->swizzle_a; - view->format = tex->tx_format; - view->format.format1 |= r300_translate_texformat(templ->format, - view->swizzle); - if (r300_screen(pipe->screen)->caps.is_r500) { + hwformat = r300_translate_texformat(templ->format, + view->swizzle, + is_r500, + dxtc_swizzle); + + if (hwformat == ~0) { + fprintf(stderr, "r300: Ooops. Got unsupported format %s in %s.\n", + util_format_short_name(templ->format), __func__); + } + assert(hwformat != ~0); + + r300_texture_setup_format_state(r300_screen(pipe->screen), tex, + templ->format, 0, + width0_override, height0_override, + &view->format); + view->format.format1 |= hwformat; + if (is_r500) { view->format.format2 |= r500_tx_format_msb_bit(templ->format); } } @@ -1354,6 +1636,17 @@ r300_create_sampler_view(struct pipe_context *pipe, return (struct pipe_sampler_view*)view; } +static struct pipe_sampler_view * +r300_create_sampler_view(struct pipe_context *pipe, + struct pipe_resource *texture, + const struct pipe_sampler_view *templ) +{ + return r300_create_sampler_view_custom(pipe, texture, templ, + r300_resource(texture)->tex.width0, + r300_resource(texture)->tex.height0); +} + + static void r300_sampler_view_destroy(struct pipe_context *pipe, struct pipe_sampler_view *view) @@ -1362,19 +1655,33 @@ r300_sampler_view_destroy(struct pipe_context *pipe, FREE(view); } -static void r300_set_scissor_state(struct pipe_context* pipe, - const struct pipe_scissor_state* state) +static void r300_set_sample_mask(struct pipe_context *pipe, + unsigned mask) +{ + struct r300_context* r300 = r300_context(pipe); + + *((unsigned*)r300->sample_mask.state) = mask; + + r300_mark_atom_dirty(r300, &r300->sample_mask); +} + +static void r300_set_scissor_states(struct pipe_context* pipe, + unsigned start_slot, + unsigned num_scissors, + const struct pipe_scissor_state* state) { struct r300_context* r300 = r300_context(pipe); memcpy(r300->scissor_state.state, state, sizeof(struct pipe_scissor_state)); - r300->scissor_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->scissor_state); } -static void r300_set_viewport_state(struct pipe_context* pipe, - const struct pipe_viewport_state* state) +static void r300_set_viewport_states(struct pipe_context* pipe, + unsigned start_slot, + unsigned num_viewports, + const struct pipe_viewport_state* state) { struct r300_context* r300 = r300_context(pipe); struct r300_viewport_state* viewport = @@ -1383,8 +1690,7 @@ static void r300_set_viewport_state(struct pipe_context* pipe, r300->viewport = *state; if (r300->draw) { - draw_flush(r300->draw); - draw_set_viewport_state(r300->draw, state); + draw_set_viewport_states(r300->draw, start_slot, num_viewports, state); viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT; return; } @@ -1417,90 +1723,57 @@ static void r300_set_viewport_state(struct pipe_context* pipe, viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; } - r300->viewport_state.dirty = TRUE; - if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { - r300->fs_rc_constant_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->viewport_state); + if (r300->fs.state && r300_fs(r300)->shader && + r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { + r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); } } -static void r300_set_vertex_buffers(struct pipe_context* pipe, - unsigned count, +static void r300_set_vertex_buffers_hwtcl(struct pipe_context* pipe, + unsigned start_slot, unsigned count, const struct pipe_vertex_buffer* buffers) { struct r300_context* r300 = r300_context(pipe); - struct pipe_vertex_buffer *vbo; - unsigned i, max_index = (1 << 24) - 1; - boolean any_user_buffer = FALSE; - - if (count == r300->vertex_buffer_count && - memcmp(r300->vertex_buffer, buffers, - sizeof(struct pipe_vertex_buffer) * count) == 0) { - return; - } - - if (r300->screen->caps.has_tcl) { - /* HW TCL. */ - r300->incompatible_vb_layout = FALSE; - - /* Check if the strides and offsets are aligned to the size of DWORD. */ - for (i = 0; i < count; i++) { - if (buffers[i].buffer) { - if (buffers[i].stride % 4 != 0 || - buffers[i].buffer_offset % 4 != 0) { - r300->incompatible_vb_layout = TRUE; - break; - } - } - } - for (i = 0; i < count; i++) { - /* Why, yes, I AM casting away constness. How did you know? */ - vbo = (struct pipe_vertex_buffer*)&buffers[i]; + util_set_vertex_buffers_count(r300->vertex_buffer, + &r300->nr_vertex_buffers, + buffers, start_slot, count); - /* Skip NULL buffers */ - if (!buffers[i].buffer) { - continue; - } - - if (r300_buffer_is_user_buffer(vbo->buffer)) { - any_user_buffer = TRUE; - } + /* There must be at least one vertex buffer set, otherwise it locks up. */ + if (!r300->nr_vertex_buffers) { + util_set_vertex_buffers_count(r300->vertex_buffer, + &r300->nr_vertex_buffers, + &r300->dummy_vb, 0, 1); + } - if (vbo->max_index == ~0) { - /* if no VBO stride then only one vertex value so max index is 1 */ - /* should think about converting to VS constants like svga does */ - if (!vbo->stride) - vbo->max_index = 1; - else - vbo->max_index = - (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride; - } + r300->vertex_arrays_dirty = TRUE; +} - max_index = MIN2(vbo->max_index, max_index); - } +static void r300_set_vertex_buffers_swtcl(struct pipe_context* pipe, + unsigned start_slot, unsigned count, + const struct pipe_vertex_buffer* buffers) +{ + struct r300_context* r300 = r300_context(pipe); + unsigned i; - r300->any_user_vbs = any_user_buffer; - r300->vertex_buffer_max_index = max_index; + util_set_vertex_buffers_count(r300->vertex_buffer, + &r300->nr_vertex_buffers, + buffers, start_slot, count); + draw_set_vertex_buffers(r300->draw, start_slot, count, buffers); - } else { - /* SW TCL. */ - draw_flush(r300->draw); - draw_set_vertex_buffers(r300->draw, count, buffers); - } + if (!buffers) + return; - /* Common code. */ for (i = 0; i < count; i++) { - /* Reference our buffer. */ - pipe_resource_reference(&r300->vertex_buffer[i].buffer, buffers[i].buffer); - } - for (; i < r300->vertex_buffer_count; i++) { - /* Dereference any old buffers. */ - pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL); + if (buffers[i].is_user_buffer) { + draw_set_mapped_vertex_buffer(r300->draw, start_slot + i, + buffers[i].buffer.user, ~0); + } else if (buffers[i].buffer.resource) { + draw_set_mapped_vertex_buffer(r300->draw, start_slot + i, + r300_resource(buffers[i].buffer.resource)->malloced_buffer, ~0); + } } - - memcpy(r300->vertex_buffer, buffers, - sizeof(struct pipe_vertex_buffer) * count); - r300->vertex_buffer_count = count; } /* Initialize the PSC tables. */ @@ -1511,17 +1784,11 @@ static void r300_vertex_psc(struct r300_vertex_element_state *velems) enum pipe_format format; unsigned i; - if (velems->count > 16) { - fprintf(stderr, "r300: More than 16 vertex elements are not supported," - " requested %i, using 16.\n", velems->count); - velems->count = 16; - } - /* Vertex shaders have no semantics on their inputs, * so PSC should just route stuff based on the vertex elements, * and not on attrib information. */ for (i = 0; i < velems->count; i++) { - format = velems->hw_format[i]; + format = velems->velem[i].src_format; type = r300_translate_vertex_data_type(format); if (type == R300_INVALID_FORMAT) { @@ -1553,89 +1820,44 @@ static void r300_vertex_psc(struct r300_vertex_element_state *velems) vstream->count = (i >> 1) + 1; } -#define FORMAT_REPLACE(what, withwhat) \ - case PIPE_FORMAT_##what: *format = PIPE_FORMAT_##withwhat; break - static void* r300_create_vertex_elements_state(struct pipe_context* pipe, unsigned count, const struct pipe_vertex_element* attribs) { struct r300_vertex_element_state *velems; unsigned i; - enum pipe_format *format; + struct pipe_vertex_element dummy_attrib = {0}; + + /* R300 Programmable Stream Control (PSC) doesn't support 0 vertex elements. */ + if (!count) { + dummy_attrib.src_format = PIPE_FORMAT_R8G8B8A8_UNORM; + attribs = &dummy_attrib; + count = 1; + } else if (count > 16) { + fprintf(stderr, "r300: More than 16 vertex elements are not supported," + " requested %i, using 16.\n", count); + count = 16; + } - assert(count <= PIPE_MAX_ATTRIBS); velems = CALLOC_STRUCT(r300_vertex_element_state); - if (velems != NULL) { - velems->count = count; - memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); - - if (r300_screen(pipe->screen)->caps.has_tcl) { - /* Set the best hw format in case the original format is not - * supported by hw. */ - for (i = 0; i < count; i++) { - velems->hw_format[i] = velems->velem[i].src_format; - format = &velems->hw_format[i]; - - /* This is basically the list of unsupported formats. - * For now we don't care about the alignment, that's going to - * be sorted out after the PSC setup. */ - switch (*format) { - FORMAT_REPLACE(R64_FLOAT, R32_FLOAT); - FORMAT_REPLACE(R64G64_FLOAT, R32G32_FLOAT); - FORMAT_REPLACE(R64G64B64_FLOAT, R32G32B32_FLOAT); - FORMAT_REPLACE(R64G64B64A64_FLOAT, R32G32B32A32_FLOAT); - - FORMAT_REPLACE(R32_UNORM, R32_FLOAT); - FORMAT_REPLACE(R32G32_UNORM, R32G32_FLOAT); - FORMAT_REPLACE(R32G32B32_UNORM, R32G32B32_FLOAT); - FORMAT_REPLACE(R32G32B32A32_UNORM, R32G32B32A32_FLOAT); - - FORMAT_REPLACE(R32_USCALED, R32_FLOAT); - FORMAT_REPLACE(R32G32_USCALED, R32G32_FLOAT); - FORMAT_REPLACE(R32G32B32_USCALED, R32G32B32_FLOAT); - FORMAT_REPLACE(R32G32B32A32_USCALED,R32G32B32A32_FLOAT); - - FORMAT_REPLACE(R32_SNORM, R32_FLOAT); - FORMAT_REPLACE(R32G32_SNORM, R32G32_FLOAT); - FORMAT_REPLACE(R32G32B32_SNORM, R32G32B32_FLOAT); - FORMAT_REPLACE(R32G32B32A32_SNORM, R32G32B32A32_FLOAT); - - FORMAT_REPLACE(R32_SSCALED, R32_FLOAT); - FORMAT_REPLACE(R32G32_SSCALED, R32G32_FLOAT); - FORMAT_REPLACE(R32G32B32_SSCALED, R32G32B32_FLOAT); - FORMAT_REPLACE(R32G32B32A32_SSCALED,R32G32B32A32_FLOAT); - - FORMAT_REPLACE(R32_FIXED, R32_FLOAT); - FORMAT_REPLACE(R32G32_FIXED, R32G32_FLOAT); - FORMAT_REPLACE(R32G32B32_FIXED, R32G32B32_FLOAT); - FORMAT_REPLACE(R32G32B32A32_FIXED, R32G32B32A32_FLOAT); - - default:; - } + if (!velems) + return NULL; - velems->incompatible_layout = - velems->incompatible_layout || - velems->velem[i].src_format != velems->hw_format[i] || - velems->velem[i].src_offset % 4 != 0; - } + velems->count = count; + memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); - /* Now setup PSC. - * The unused components will be replaced by (..., 0, 1). */ - r300_vertex_psc(velems); - - /* Align the formats to the size of DWORD. - * We only care about the blocksizes of the formats since - * swizzles are already set up. - * Also compute the vertex size. */ - for (i = 0; i < count; i++) { - /* This is OK because we check for aligned strides too. */ - velems->hw_format_size[i] = - align(util_format_get_blocksize(velems->hw_format[i]), 4); - velems->vertex_size_dwords += velems->hw_format_size[i] / 4; - } + if (r300_screen(pipe->screen)->caps.has_tcl) { + /* Setup PSC. + * The unused components will be replaced by (..., 0, 1). */ + r300_vertex_psc(velems); + + for (i = 0; i < count; i++) { + velems->format_size[i] = + align(util_format_get_blocksize(velems->velem[i].src_format), 4); + velems->vertex_size_dwords += velems->format_size[i] / 4; } } + return velems; } @@ -1645,25 +1867,25 @@ static void r300_bind_vertex_elements_state(struct pipe_context *pipe, struct r300_context *r300 = r300_context(pipe); struct r300_vertex_element_state *velems = state; - if (velems == NULL) { + if (!velems) { return; } r300->velems = velems; if (r300->draw) { - draw_flush(r300->draw); draw_set_vertex_elements(r300->draw, velems->count, velems->velem); return; } UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; + r300->vertex_arrays_dirty = TRUE; } static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) { - FREE(state); + FREE(state); } static void* r300_create_vs_state(struct pipe_context* pipe, @@ -1677,10 +1899,10 @@ static void* r300_create_vs_state(struct pipe_context* pipe, vs->state.tokens = tgsi_dup_tokens(shader->tokens); if (r300->screen->caps.has_tcl) { - r300_init_vs_outputs(vs); + r300_init_vs_outputs(r300, vs); r300_translate_vertex_shader(r300, vs); } else { - r300_draw_init_vertex_shader(r300->draw, vs); + r300_draw_init_vertex_shader(r300, vs); } return vs; @@ -1691,7 +1913,7 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) struct r300_context* r300 = r300_context(pipe); struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; - if (vs == NULL) { + if (!vs) { r300->vs_state.state = NULL; return; } @@ -1701,24 +1923,25 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) r300->vs_state.state = vs; /* The majority of the RS block bits is dependent on the vertex shader. */ - r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ + r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ if (r300->screen->caps.has_tcl) { - r300->vs_state.dirty = TRUE; - r300->vs_state.size = - vs->code.length + 9 + + unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2; + r300_mark_atom_dirty(r300, &r300->vs_state); + r300->vs_state.size = vs->code.length + 9 + + (R300_VS_MAX_FC_OPS * fc_op_dwords + 4); + + r300_mark_atom_dirty(r300, &r300->vs_constants); + r300->vs_constants.size = + 2 + + (vs->externals_count ? vs->externals_count * 4 + 3 : 0) + (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0); - if (vs->externals_count) { - r300->vs_constants.dirty = TRUE; - r300->vs_constants.size = vs->externals_count * 4 + 3; - } else { - r300->vs_constants.size = 0; - } + ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table = + vs->code.constants_remap_table; - r300->pvs_flush.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->pvs_flush); } else { - draw_flush(r300->draw); draw_bind_vertex_shader(r300->draw, (struct draw_vertex_shader*)vs->draw_vs); } @@ -1731,6 +1954,7 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) if (r300->screen->caps.has_tcl) { rc_constants_destroy(&vs->code.constants); + FREE(vs->code.constants_remap_table); } else { draw_delete_vertex_shader(r300->draw, (struct draw_vertex_shader*)vs->draw_vs); @@ -1741,70 +1965,83 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) } static void r300_set_constant_buffer(struct pipe_context *pipe, - uint shader, uint index, - struct pipe_resource *buf) + enum pipe_shader_type shader, uint index, + const struct pipe_constant_buffer *cb) { struct r300_context* r300 = r300_context(pipe); struct r300_constant_buffer *cbuf; - uint32_t *mapped = r300_buffer(buf)->user_buffer; - int max_size = 0, max_size_bytes = 0, clamped_size = 0; + uint32_t *mapped; + + if (!cb || (!cb->buffer && !cb->user_buffer)) + return; switch (shader) { case PIPE_SHADER_VERTEX: cbuf = (struct r300_constant_buffer*)r300->vs_constants.state; - max_size = 256; break; case PIPE_SHADER_FRAGMENT: cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; - if (r300->screen->caps.is_r500) { - max_size = 256; - } else { - max_size = 32; - } break; default: - assert(0); return; } - max_size_bytes = max_size * 4 * sizeof(float); - if (buf == NULL || buf->width0 == 0 || - (mapped = r300_buffer(buf)->constant_buffer) == NULL) { - cbuf->count = 0; - return; + + if (cb->user_buffer) + mapped = (uint32_t*)cb->user_buffer; + else { + struct r300_resource *rbuf = r300_resource(cb->buffer); + + if (rbuf && rbuf->malloced_buffer) + mapped = (uint32_t*)rbuf->malloced_buffer; + else + return; } if (shader == PIPE_SHADER_FRAGMENT || (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) { - assert((buf->width0 % (4 * sizeof(float))) == 0); - - /* Check the size of the constant buffer. */ - /* XXX Subtract immediates and RC_STATE_* variables. */ - if (buf->width0 > max_size_bytes) { - fprintf(stderr, "r300: Max size of the constant buffer is " - "%i*4 floats.\n", max_size); - } - - clamped_size = MIN2(buf->width0, max_size_bytes); - cbuf->count = clamped_size / (4 * sizeof(float)); cbuf->ptr = mapped; } if (shader == PIPE_SHADER_VERTEX) { if (r300->screen->caps.has_tcl) { - if (r300->vs_constants.size) { - r300->vs_constants.dirty = TRUE; + struct r300_vertex_shader *vs = + (struct r300_vertex_shader*)r300->vs_state.state; + + if (!vs) { + cbuf->buffer_base = 0; + return; } - r300->pvs_flush.dirty = TRUE; + + cbuf->buffer_base = r300->vs_const_base; + r300->vs_const_base += vs->code.constants.Count; + if (r300->vs_const_base > R500_MAX_PVS_CONST_VECS) { + r300->vs_const_base = vs->code.constants.Count; + cbuf->buffer_base = 0; + r300_mark_atom_dirty(r300, &r300->pvs_flush); + } + r300_mark_atom_dirty(r300, &r300->vs_constants); } else if (r300->draw) { draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, - 0, mapped, buf->width0); + 0, mapped, cb->buffer_size); } } else if (shader == PIPE_SHADER_FRAGMENT) { - r300->fs_constants.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->fs_constants); } } +static void r300_texture_barrier(struct pipe_context *pipe, unsigned flags) +{ + struct r300_context *r300 = r300_context(pipe); + + r300_mark_atom_dirty(r300, &r300->gpu_flush); + r300_mark_atom_dirty(r300, &r300->texture_cache_inval); +} + +static void r300_memory_barrier(struct pipe_context *pipe, unsigned flags) +{ +} + void r300_init_state_functions(struct r300_context* r300) { r300->context.create_blend_state = r300_create_blend_state; @@ -1837,19 +2074,22 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.delete_rasterizer_state = r300_delete_rs_state; r300->context.create_sampler_state = r300_create_sampler_state; - r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; - r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; + r300->context.bind_sampler_states = r300_bind_sampler_states; r300->context.delete_sampler_state = r300_delete_sampler_state; - r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; + r300->context.set_sampler_views = r300_set_sampler_views; r300->context.create_sampler_view = r300_create_sampler_view; r300->context.sampler_view_destroy = r300_sampler_view_destroy; - r300->context.set_scissor_state = r300_set_scissor_state; + r300->context.set_scissor_states = r300_set_scissor_states; - r300->context.set_viewport_state = r300_set_viewport_state; + r300->context.set_viewport_states = r300_set_viewport_states; - r300->context.set_vertex_buffers = r300_set_vertex_buffers; + if (r300->screen->caps.has_tcl) { + r300->context.set_vertex_buffers = r300_set_vertex_buffers_hwtcl; + } else { + r300->context.set_vertex_buffers = r300_set_vertex_buffers_swtcl; + } r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; @@ -1858,4 +2098,7 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.create_vs_state = r300_create_vs_state; r300->context.bind_vs_state = r300_bind_vs_state; r300->context.delete_vs_state = r300_delete_vs_state; + + r300->context.texture_barrier = r300_texture_barrier; + r300->context.memory_barrier = r300_memory_barrier; }