X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_state.c;h=d07e90860c28e2f2b61a093045044eb3dfd556bd;hb=653a83445f94620673f747a4ace6847a2c7fdb4d;hp=a145a7f18a5820bc79f1c80e84a92a455dfe7d0c;hpb=1a4f242be60237fb1f1acf346b1e641167bc6cee;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index a145a7f18a5..d07e90860c2 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -30,7 +30,6 @@ #include "tgsi/tgsi_parse.h" #include "pipe/p_config.h" -#include "pipe/internal/p_winsys_screen.h" #include "r300_context.h" #include "r300_reg.h" @@ -156,23 +155,33 @@ static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA dstA == PIPE_BLENDFACTOR_ONE); } +static unsigned bgra_cmask(unsigned mask) +{ + /* Gallium uses RGBA color ordering while R300 expects BGRA. */ + + return ((mask & PIPE_MASK_R) << 2) | + ((mask & PIPE_MASK_B) >> 2) | + (mask & (PIPE_MASK_G | PIPE_MASK_A)); +} + /* Create a new blend state based on the CSO blend state. * * This encompasses alpha blending, logic/raster ops, and blend dithering. */ static void* r300_create_blend_state(struct pipe_context* pipe, const struct pipe_blend_state* state) { + struct r300_screen* r300screen = r300_screen(pipe->screen); struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); - if (state->blend_enable) + if (state->rt[0].blend_enable) { - unsigned eqRGB = state->rgb_func; - unsigned srcRGB = state->rgb_src_factor; - unsigned dstRGB = state->rgb_dst_factor; + unsigned eqRGB = state->rt[0].rgb_func; + unsigned srcRGB = state->rt[0].rgb_src_factor; + unsigned dstRGB = state->rt[0].rgb_dst_factor; - unsigned eqA = state->alpha_func; - unsigned srcA = state->alpha_src_factor; - unsigned dstA = state->alpha_dst_factor; + unsigned eqA = state->rt[0].alpha_func; + unsigned srcA = state->rt[0].alpha_src_factor; + unsigned dstA = state->rt[0].alpha_dst_factor; /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, * this is just the crappy D3D naming */ @@ -289,18 +298,18 @@ static void* r300_create_blend_state(struct pipe_context* pipe, (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; } - /* Color Channel Mask */ - if (state->colormask & PIPE_MASK_R) { - blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_RED_MASK0; - } - if (state->colormask & PIPE_MASK_G) { - blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0; - } - if (state->colormask & PIPE_MASK_B) { - blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0; - } - if (state->colormask & PIPE_MASK_A) { - blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0; + /* Color channel masks for all MRTs. */ + blend->color_channel_mask = bgra_cmask(state->rt[0].colormask); + if (r300screen->caps->is_r500 && state->independent_blend_enable) { + if (state->rt[1].blend_enable) { + blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; + } + if (state->rt[2].blend_enable) { + blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; + } + if (state->rt[3].blend_enable) { + blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; + } } if (state->dither) { @@ -317,8 +326,8 @@ static void r300_bind_blend_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); - r300->blend_state = (struct r300_blend_state*)state; - r300->dirty_state |= R300_NEW_BLEND; + r300->blend_state.state = state; + r300->blend_state.dirty = TRUE; } /* Free blend state. */ @@ -340,20 +349,24 @@ 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_screen* r300screen = r300_screen(pipe->screen); + struct r300_blend_color_state* state = + (struct r300_blend_color_state*)r300->blend_color_state.state; union util_color uc; util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); - r300->blend_color_state->blend_color = uc.ui; + state->blend_color = uc.ui; /* XXX if FP16 blending is enabled, we should use the FP16 format */ - r300->blend_color_state->blend_color_red_alpha = + state->blend_color_red_alpha = float_to_fixed10(color->color[0]) | (float_to_fixed10(color->color[3]) << 16); - r300->blend_color_state->blend_color_green_blue = + state->blend_color_green_blue = float_to_fixed10(color->color[2]) | (float_to_fixed10(color->color[1]) << 16); - r300->dirty_state |= R300_NEW_BLEND_COLOR; + r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2; + r300->blend_color_state.dirty = TRUE; } static void r300_set_clip_state(struct pipe_context* pipe, @@ -362,12 +375,15 @@ static void r300_set_clip_state(struct pipe_context* pipe, struct r300_context* r300 = r300_context(pipe); if (r300_screen(pipe->screen)->caps->has_tcl) { - r300->clip_state = *state; - r300->dirty_state |= R300_NEW_CLIP; + memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); + r300->clip_state.size = 29; } else { draw_flush(r300->draw); draw_set_clip_state(r300->draw, state); + r300->clip_state.size = 2; } + + r300->clip_state.dirty = TRUE; } /* Create a new depth, stencil, and alpha state based on the CSO dsa state. @@ -425,7 +441,6 @@ static void* (r300_translate_stencil_op(state->stencil[1].zfail_op) << R300_S_BACK_ZFAIL_OP_SHIFT); - /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */ if (caps->is_r500) { dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; @@ -444,8 +459,7 @@ static void* r300_translate_alpha_function(state->alpha.func) | R300_FG_ALPHA_FUNC_ENABLE; - /* XXX figure out why emitting 10bit alpha ref causes CS to dump */ - /* always use 8bit alpha ref */ + /* We could use 10bit alpha ref but who needs that? */ dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); if (caps->is_r500) @@ -460,9 +474,11 @@ static void r300_bind_dsa_state(struct pipe_context* pipe, void* state) { struct r300_context* r300 = r300_context(pipe); + struct r300_screen* r300screen = r300_screen(pipe->screen); - r300->dsa_state = (struct r300_dsa_state*)state; - r300->dirty_state |= R300_NEW_DSA; + r300->dsa_state.state = state; + r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6; + r300->dsa_state.dirty = TRUE; } /* Free DSA state. */ @@ -472,57 +488,52 @@ static void r300_delete_dsa_state(struct pipe_context* pipe, FREE(state); } -static void r300_set_scissor_regs(const struct pipe_scissor_state* state, - struct r300_scissor_regs *scissor, - boolean is_r500) -{ - if (is_r500) { - scissor->top_left = - (state->minx << R300_SCISSORS_X_SHIFT) | - (state->miny << R300_SCISSORS_Y_SHIFT); - scissor->bottom_right = - ((state->maxx - 1) << R300_SCISSORS_X_SHIFT) | - ((state->maxy - 1) << R300_SCISSORS_Y_SHIFT); - } else { - /* Offset of 1440 in non-R500 chipsets. */ - scissor->top_left = - ((state->minx + 1440) << R300_SCISSORS_X_SHIFT) | - ((state->miny + 1440) << R300_SCISSORS_Y_SHIFT); - scissor->bottom_right = - (((state->maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) | - (((state->maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT); - } - - scissor->empty_area = state->minx >= state->maxx || - state->miny >= state->maxy; -} - static void r300_set_framebuffer_state(struct pipe_context* pipe, const struct pipe_framebuffer_state* state) { struct r300_context* r300 = r300_context(pipe); - struct pipe_scissor_state scissor; + uint32_t zbuffer_bpp = 0; + + r300->fb_state.size = (10 * state->nr_cbufs) + + (2 * (4 - state->nr_cbufs)) + + (state->zsbuf ? 10 : 0) + 6; + + if (state->nr_cbufs > 4) { + debug_printf("r300: Implementation error: Too many MRTs in %s, " + "refusing to bind framebuffer state!\n", __FUNCTION__); + return; + } if (r300->draw) { draw_flush(r300->draw); } - r300->framebuffer_state = *state; - - scissor.minx = scissor.miny = 0; - scissor.maxx = state->width; - scissor.maxy = state->height; - r300_set_scissor_regs(&scissor, &r300->scissor_state->framebuffer, - r300_screen(r300->context.screen)->caps->is_r500); + memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); /* Don't rely on the order of states being set for the first time. */ - if (!r300->rs_state || !r300->rs_state->rs.scissor) { - r300->dirty_state |= R300_NEW_SCISSOR; + /* XXX wait what */ + r300->blend_state.dirty = TRUE; + r300->dsa_state.dirty = TRUE; + r300->fb_state.dirty = TRUE; + r300->scissor_state.dirty = TRUE; + + /* 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 (r300->zbuffer_bpp != zbuffer_bpp) { + r300->zbuffer_bpp = zbuffer_bpp; + r300->rs_state.dirty = TRUE; + } } - r300->dirty_state |= R300_NEW_FRAMEBUFFERS; - r300->dirty_state |= R300_NEW_BLEND; - r300->dirty_state |= R300_NEW_DSA; } /* Create fragment shader state. */ @@ -558,7 +569,7 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) r300_pick_fragment_shader(r300); if (r300->vs && r300_vertex_shader_setup_wpos(r300)) { - r300->dirty_state |= R300_NEW_VERTEX_FORMAT; + r300->vertex_format_state.dirty = TRUE; } r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; @@ -601,8 +612,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe, /* Copy rasterizer state for Draw. */ rs->rs = *state; - rs->enable_vte = !state->bypass_vs_clip_and_viewport; - #ifdef PIPE_ARCH_LITTLE_ENDIAN rs->vap_control_status = R300_VC_NO_SWAP; #else @@ -619,18 +628,17 @@ static void* r300_create_rs_state(struct pipe_context* pipe, rs->point_size = pack_float_16_6x(state->point_size) | (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); - rs->point_minmax = - ((int)(state->point_size_min * 6.0) << + /* set hw limits - clamping done by state tracker in vs or point_size + XXX always need to emit this? */ + rs->point_minmax = + ((int)(0.0 * 6.0) << R300_GA_POINT_MINMAX_MIN_SHIFT) | - ((int)(state->point_size_max * 6.0) << + ((int)(4096.0 * 6.0) << R300_GA_POINT_MINMAX_MAX_SHIFT); rs->line_control = pack_float_16_6x(state->line_width) | R300_GA_LINE_CNTL_END_TYPE_COMP; - /* XXX I think there is something wrong with the polygon mode, - * XXX re-test when r300g is in a better shape */ - /* Enable polygon mode */ if (state->fill_cw != PIPE_POLYGON_MODE_FILL || state->fill_ccw != PIPE_POLYGON_MODE_FILL) { @@ -683,10 +691,8 @@ static void* r300_create_rs_state(struct pipe_context* pipe, } if (rs->polygon_offset_enable) { - rs->depth_offset_front = rs->depth_offset_back = - fui(state->offset_units); - rs->depth_scale_front = rs->depth_scale_back = - fui(state->offset_scale); + rs->depth_offset = state->offset_units; + rs->depth_scale = state->offset_scale; } if (state->line_stipple_enable) { @@ -718,12 +724,21 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) draw_set_rasterizer_state(r300->draw, &rs->rs); } - r300->rs_state = rs; + if (rs) { + r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport; + r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; + } else { + r300->tcl_bypass = FALSE; + r300->polygon_offset_enabled = FALSE; + } + + r300->rs_state.state = rs; + r300->rs_state.dirty = TRUE; + /* XXX Why is this still needed, dammit!? */ + r300->scissor_state.dirty = TRUE; + r300->viewport_state.dirty = TRUE; + /* XXX Clean these up when we move to atom emits */ - r300->dirty_state |= R300_NEW_RASTERIZER; - r300->dirty_state |= R300_NEW_RS_BLOCK; - r300->dirty_state |= R300_NEW_SCISSOR; - r300->dirty_state |= R300_NEW_VIEWPORT; if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; } @@ -862,49 +877,48 @@ static void r300_set_scissor_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); - r300_set_scissor_regs(state, &r300->scissor_state->scissor, - r300_screen(r300->context.screen)->caps->is_r500); + memcpy(r300->scissor_state.state, state, + sizeof(struct pipe_scissor_state)); - /* Don't rely on the order of states being set for the first time. */ - if (!r300->rs_state || r300->rs_state->rs.scissor) { - r300->dirty_state |= R300_NEW_SCISSOR; - } + r300->scissor_state.dirty = TRUE; } static void r300_set_viewport_state(struct pipe_context* pipe, const struct pipe_viewport_state* state) { struct r300_context* r300 = r300_context(pipe); + struct r300_viewport_state* viewport = + (struct r300_viewport_state*)r300->viewport_state.state; /* Do the transform in HW. */ - r300->viewport_state->vte_control = R300_VTX_W0_FMT; + viewport->vte_control = R300_VTX_W0_FMT; if (state->scale[0] != 1.0f) { - r300->viewport_state->xscale = state->scale[0]; - r300->viewport_state->vte_control |= R300_VPORT_X_SCALE_ENA; + viewport->xscale = state->scale[0]; + viewport->vte_control |= R300_VPORT_X_SCALE_ENA; } if (state->scale[1] != 1.0f) { - r300->viewport_state->yscale = state->scale[1]; - r300->viewport_state->vte_control |= R300_VPORT_Y_SCALE_ENA; + viewport->yscale = state->scale[1]; + viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; } if (state->scale[2] != 1.0f) { - r300->viewport_state->zscale = state->scale[2]; - r300->viewport_state->vte_control |= R300_VPORT_Z_SCALE_ENA; + viewport->zscale = state->scale[2]; + viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; } if (state->translate[0] != 0.0f) { - r300->viewport_state->xoffset = state->translate[0]; - r300->viewport_state->vte_control |= R300_VPORT_X_OFFSET_ENA; + viewport->xoffset = state->translate[0]; + viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; } if (state->translate[1] != 0.0f) { - r300->viewport_state->yoffset = state->translate[1]; - r300->viewport_state->vte_control |= R300_VPORT_Y_OFFSET_ENA; + viewport->yoffset = state->translate[1]; + viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; } if (state->translate[2] != 0.0f) { - r300->viewport_state->zoffset = state->translate[2]; - r300->viewport_state->vte_control |= R300_VPORT_Z_OFFSET_ENA; + viewport->zoffset = state->translate[2]; + viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; } - r300->dirty_state |= R300_NEW_VIEWPORT; + r300->viewport_state.dirty = TRUE; if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; } @@ -925,7 +939,23 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe, draw_set_vertex_buffers(r300->draw, count, buffers); } - r300->dirty_state |= R300_NEW_VERTEX_FORMAT; + r300->vertex_format_state.dirty = TRUE; +} + +static boolean r300_validate_aos(struct r300_context *r300) +{ + struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; + struct pipe_vertex_element *velem = r300->vertex_element; + int i; + + /* Check if formats and strides are aligned to the size of DWORD. */ + for (i = 0; i < r300->vertex_element_count; i++) { + if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 || + util_format_get_blocksize(velem[i].src_format) % 4 != 0) { + return FALSE; + } + } + return TRUE; } static void r300_set_vertex_elements(struct pipe_context* pipe, @@ -943,6 +973,12 @@ static void r300_set_vertex_elements(struct pipe_context* pipe, draw_flush(r300->draw); draw_set_vertex_elements(r300->draw, count, elements); } + + if (!r300_validate_aos(r300)) { + /* XXX We should fallback using draw. */ + assert(0); + abort(); + } } static void* r300_create_vs_state(struct pipe_context* pipe, @@ -983,9 +1019,10 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) r300_vertex_shader_setup_wpos(r300); } + r300->vertex_format_state.dirty = TRUE; + r300->dirty_state |= - R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS | - R300_NEW_VERTEX_FORMAT; + R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS; } else { draw_flush(r300->draw); draw_bind_vertex_shader(r300->draw, @@ -1011,22 +1048,22 @@ 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, - const struct pipe_constant_buffer *buf) + struct pipe_buffer *buf) { struct r300_context* r300 = r300_context(pipe); void *mapped; - if (buf == NULL || buf->buffer->size == 0 || - (mapped = pipe_buffer_map(pipe->screen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) + if (buf == NULL || buf->size == 0 || + (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) { r300->shader_constants[shader].count = 0; return; } - assert((buf->buffer->size % 4 * sizeof(float)) == 0); - memcpy(r300->shader_constants[shader].constants, mapped, buf->buffer->size); - r300->shader_constants[shader].count = buf->buffer->size / (4 * sizeof(float)); - pipe_buffer_unmap(pipe->screen, buf->buffer); + assert((buf->size % 4 * sizeof(float)) == 0); + memcpy(r300->shader_constants[shader].constants, mapped, buf->size); + r300->shader_constants[shader].count = buf->size / (4 * sizeof(float)); + pipe_buffer_unmap(pipe->screen, buf); if (shader == PIPE_SHADER_VERTEX) r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;