1 files changed, 114 insertions, 83 deletions

diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 435e613ddf..d07e90860c 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) {
@@ -340,6 +349,7 @@ 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;
@@ -355,6 +365,7 @@ static void r300_set_blend_color(struct pipe_context* pipe,
         float_to_fixed10(color->color[2]) |
         (float_to_fixed10(color->color[1]) << 16);
 
+    r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2;
     r300->blend_color_state.dirty = TRUE;
 }
 
@@ -365,11 +376,14 @@ static void r300_set_clip_state(struct pipe_context* pipe,
 
     if (r300_screen(pipe->screen)->caps->has_tcl) {
         memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state));
-        r300->clip_state.dirty = TRUE;
+        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.
@@ -427,7 +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;
@@ -446,8 +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)
@@ -462,8 +474,10 @@ 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.state = state;
+    r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6;
     r300->dsa_state.dirty = TRUE;
 }
 
@@ -474,56 +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);
-    }
-}
-
 static void
     r300_set_framebuffer_state(struct pipe_context* pipe,
                                const struct pipe_framebuffer_state* state)
 {
     struct r300_context* r300 = r300_context(pipe);
-    struct r300_scissor_state* scissor =
-        (struct r300_scissor_state*)r300->scissor_state.state;
-    struct pipe_scissor_state pscissor;
+    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;
-
-    /* XXX Arg. This is silly. */
-    pscissor.minx = pscissor.miny = 0;
-    pscissor.maxx = state->width;
-    pscissor.maxy = state->height;
-    r300_set_scissor_regs(&pscissor, &scissor->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. */
-    r300->dirty_state |= R300_NEW_FRAMEBUFFERS;
-
+    /* 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;
+        }
+    }
 }
 
 /* Create fragment shader state. */
@@ -559,7 +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;
@@ -629,9 +639,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
     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) {
@@ -684,10 +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) {
@@ -719,7 +724,13 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
         draw_set_rasterizer_state(r300->draw, &rs->rs);
     }
 
-    r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport;
+    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;
@@ -728,7 +739,6 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
     r300->viewport_state.dirty = TRUE;
 
     /* XXX Clean these up when we move to atom emits */
-    r300->dirty_state |= R300_NEW_RS_BLOCK;
     if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
         r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
     }
@@ -866,11 +876,9 @@ static void r300_set_scissor_state(struct pipe_context* pipe,
                                    const struct pipe_scissor_state* state)
 {
     struct r300_context* r300 = r300_context(pipe);
-    struct r300_scissor_state* scissor =
-        (struct r300_scissor_state*)r300->scissor_state.state;
 
-    r300_set_scissor_regs(state, &scissor->scissor,
-                          r300_screen(r300->context.screen)->caps->is_r500);
+    memcpy(r300->scissor_state.state, state,
+        sizeof(struct pipe_scissor_state));
 
     r300->scissor_state.dirty = TRUE;
 }
@@ -931,7 +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,
@@ -949,6 +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,
@@ -989,9 +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,
@@ -1017,22 +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;