nvir: bump max encoding size of instructions
[mesa.git] / src / gallium / drivers / r300 / r300_state.c
index fa256aa2c55eb300ca44998cc2be942375f780ce..6396577f03ce313d6d3878af087e0337529df033 100644 (file)
@@ -27,7 +27,6 @@
 #include "util/u_half.h"
 #include "util/u_helpers.h"
 #include "util/u_math.h"
-#include "util/u_mm.h"
 #include "util/u_memory.h"
 #include "util/u_pack_color.h"
 #include "util/u_transfer.h"
@@ -170,6 +169,59 @@ static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA
             dstA == PIPE_BLENDFACTOR_ONE);
 }
 
+static unsigned blend_discard_conditionally(unsigned eqRGB, unsigned eqA,
+                                            unsigned dstRGB, unsigned dstA,
+                                            unsigned srcRGB, unsigned srcA)
+{
+    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)
@@ -216,6 +268,70 @@ static unsigned arra_cmask(unsigned mask)
            (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 ||
+        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 (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. */
@@ -226,24 +342,52 @@ static void* r300_create_blend_state(struct pipe_context* pipe,
     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 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;
 
     blend->state = *state;
 
-    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;
+    /* 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;
+    }
 
-        unsigned eqA = state->rt[0].alpha_func;
-        unsigned srcA = state->rt[0].alpha_src_factor;
-        unsigned dstA = state->rt[0].alpha_dst_factor;
+    switch (dstRGBX) {
+    case PIPE_BLENDFACTOR_DST_ALPHA:
+        dstRGBX = PIPE_BLENDFACTOR_ONE;
+        break;
+    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+        dstRGBX = PIPE_BLENDFACTOR_ZERO;
+        break;
+    }
+
+    /* 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 */
@@ -251,123 +395,57 @@ static void* r300_create_blend_state(struct pipe_context* pipe,
             R300_ALPHA_BLEND_ENABLE |
             ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) |
             ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT);
-        blend_control |=
-            r300_translate_blend_function(eqRGB, TRUE);
-        blend_control_noclamp |=
-            r300_translate_blend_function(eqRGB, FALSE);
 
-        /* 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;
-            blend_control_noclamp |= 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) &&
-                        (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;
-                    }
-                }
-            }
-        }
+        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;
             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);
+            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);
         }
     }
 
@@ -397,22 +475,26 @@ static void* r300_create_blend_state(struct pipe_context* pipe,
             rrrr_cmask,
             aaaa_cmask,
             grrg_cmask,
-            arra_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(blend_control);
-            OUT_CB(alpha_blend_control);
+            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. */
+    /* 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);
@@ -422,6 +504,16 @@ static void* r300_create_blend_state(struct pipe_context* pipe,
     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_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;
+
     /* The same as above, but with no colorbuffer reads and writes. */
     BEGIN_CB(blend->cb_no_readwrite, 8);
     OUT_CB_REG(R300_RB3D_ROPCNTL, rop);
@@ -486,15 +578,17 @@ static void r300_set_blend_color(struct pipe_context* pipe,
     struct r300_blend_color_state *state =
         (struct r300_blend_color_state*)r300->blend_color_state.state;
     struct pipe_blend_color c;
-    enum pipe_format format = fb->nr_cbufs ? fb->cbufs[0]->format : 0;
+    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 (fb->nr_cbufs) {
-        switch (format) {
+    if (cb) {
+        switch (cb->format) {
         case PIPE_FORMAT_R8_UNORM:
         case PIPE_FORMAT_L8_UNORM:
         case PIPE_FORMAT_I8_UNORM:
@@ -510,9 +604,17 @@ static void r300_set_blend_color(struct pipe_context* pipe,
             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:;
         }
     }
@@ -521,8 +623,9 @@ static void r300_set_blend_color(struct pipe_context* pipe,
         BEGIN_CB(state->cb, 3);
         OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2);
 
-        switch (format) {
+        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]) |
@@ -542,7 +645,7 @@ static void r300_set_blend_color(struct pipe_context* pipe,
         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;
     }
 
@@ -730,42 +833,6 @@ static void r300_set_stencil_ref(struct pipe_context* pipe,
     r300_mark_atom_dirty(r300, &r300->dsa_state);
 }
 
-static void r300_tex_set_tiling_flags(struct r300_context *r300,
-                                      struct r300_resource *tex,
-                                      unsigned level)
-{
-    /* Check if the macrotile flag needs to be changed.
-     * Skip changing the flags otherwise. */
-    if (tex->tex.macrotile[tex->surface_level] !=
-        tex->tex.macrotile[level]) {
-        r300->rws->buffer_set_tiling(tex->buf, r300->cs,
-                tex->tex.microtile, tex->tex.macrotile[level],
-                0, 0, 0, 0, 0,
-                tex->tex.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_resource(state->cbufs[i]->texture),
-                                  state->cbufs[i]->u.tex.level);
-    }
-    if (state->zsbuf) {
-        r300_tex_set_tiling_flags(r300,
-                                  r300_resource(state->zsbuf->texture),
-                                  state->zsbuf->u.tex.level);
-    }
-}
-
 static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index,
                                     const char *binding)
 {
@@ -827,37 +894,21 @@ void r300_mark_fb_state_dirty(struct r300_context *r300,
 
     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 unsigned r300_get_num_samples(struct r300_context *r300)
-{
-    struct pipe_framebuffer_state* fb =
-            (struct pipe_framebuffer_state*)r300->fb_state.state;
-    unsigned num_samples;
-
-    if (fb->nr_cbufs)
-        num_samples = fb->cbufs[0]->texture->nr_samples;
-    else if (fb->zsbuf)
-        num_samples = fb->zsbuf->texture->nr_samples;
-    else
-        num_samples = 1;
-
-    if (!num_samples)
-        num_samples = 1;
-
-    return num_samples;
-}
-
 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_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;
@@ -876,17 +927,17 @@ r300_set_framebuffer_state(struct pipe_context* pipe,
         return;
     }
 
-    if (old_state->zsbuf && r300->zmask_in_use && !r300->locked_zbuffer) {
+    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(old_state->zsbuf, 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, old_state->zsbuf);
+            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? */
@@ -904,27 +955,27 @@ r300_set_framebuffer_state(struct pipe_context* pipe,
     }
     assert(state->zsbuf || (r300->locked_zbuffer && !unlock_zbuffer) || !r300->zmask_in_use);
 
+    /* If zsbuf is set from NULL to non-NULL or vice versa.. */
+    if (!!current_state->zsbuf != !!state->zsbuf) {
+        r300_mark_atom_dirty(r300, &r300->dsa_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->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);
 
-    /* If zsbuf is set from NULL to non-NULL or vice versa.. */
-    if (!!old_state->zsbuf != !!state->zsbuf) {
-        r300_mark_atom_dirty(r300, &r300->dsa_state);
-    }
-
-    if (r300->screen->info.drm_minor < 12) {
-       /* The tiling flags are dependent on the surface miplevel, unfortunately.
-        * This workarounds a bad design decision in old kernels which were
-        * rewriting tile fields in registers. */
-        r300_fb_set_tiling_flags(r300, state);
-    }
-
-    util_copy_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);
@@ -951,7 +1002,7 @@ r300_set_framebuffer_state(struct pipe_context* pipe,
         }
     }
 
-    r300->num_samples = r300_get_num_samples(r300);
+    r300->num_samples = util_framebuffer_get_num_samples(state);
 
     /* Set up AA config. */
     if (r300->num_samples > 1) {
@@ -976,7 +1027,8 @@ r300_set_framebuffer_state(struct pipe_context* pipe,
     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");
@@ -1026,7 +1078,7 @@ 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;
     }
@@ -1057,7 +1109,6 @@ 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.
@@ -1089,8 +1140,7 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
     float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */
     float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */
     float point_texcoord_top = 0;   /* R300_GA_POINT_T1: 0x420c */
-    boolean vclamp = state->clamp_vertex_color ||
-                     !r300_context(pipe)->screen->caps.is_r500;
+    boolean vclamp = !r300_context(pipe)->screen->caps.is_r500;
     CB_LOCALS;
 
     /* Copy rasterizer state. */
@@ -1107,7 +1157,7 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
     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;
@@ -1221,8 +1271,7 @@ 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 |
-                       (state->depth_clip ? 0 : R300_CLIP_DISABLE);
+                       R300_PS_UCP_MODE_CLIP_AS_TRIFAN;
     } else {
        vap_clip_cntl = R300_CLIP_DISABLE;
     }
@@ -1292,6 +1341,8 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* 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_set_rasterizer_state(r300->draw, &rs->rs_draw, state);
@@ -1302,18 +1353,23 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
         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 = 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) {
+        last_two_sided_color != r300->two_sided_color ||
+        last_flatshade != r300->flatshade) {
         r300_mark_atom_dirty(r300, &r300->rs_block_state);
     }
 
@@ -1327,6 +1383,10 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
             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);
+    }
 }
 
 /* Free rasterizer state. */
@@ -1408,7 +1468,8 @@ 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);
@@ -1416,9 +1477,13 @@ 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;
@@ -1426,12 +1491,6 @@ static void r300_bind_sampler_states(struct pipe_context* pipe,
     r300_mark_atom_dirty(r300, &r300->textures_state);
 }
 
-static void r300_lacks_vertex_textures(struct pipe_context* pipe,
-                                       unsigned count,
-                                       void** states)
-{
-}
-
 static void r300_delete_sampler_state(struct pipe_context* pipe, void* state)
 {
     FREE(state);
@@ -1462,9 +1521,10 @@ 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 =
@@ -1474,6 +1534,11 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
     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;
     }
@@ -1606,8 +1671,10 @@ static void r300_set_sample_mask(struct pipe_context *pipe,
     r300_mark_atom_dirty(r300, &r300->sample_mask);
 }
 
-static void r300_set_scissor_state(struct pipe_context* pipe,
-                                   const struct pipe_scissor_state* state)
+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);
 
@@ -1617,8 +1684,10 @@ static void r300_set_scissor_state(struct pipe_context* pipe,
     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 =
@@ -1627,7 +1696,7 @@ static void r300_set_viewport_state(struct pipe_context* pipe,
     r300->viewport = *state;
 
     if (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;
     }
@@ -1703,47 +1772,16 @@ static void r300_set_vertex_buffers_swtcl(struct pipe_context* pipe,
         return;
 
     for (i = 0; i < count; i++) {
-        if (buffers[i].user_buffer) {
+        if (buffers[i].is_user_buffer) {
             draw_set_mapped_vertex_buffer(r300->draw, start_slot + i,
-                                          buffers[i].user_buffer);
-        } else if (buffers[i].buffer) {
+                                          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)->malloced_buffer);
+                                          r300_resource(buffers[i].buffer.resource)->malloced_buffer, ~0);
         }
     }
 }
 
-static void r300_set_index_buffer_hwtcl(struct pipe_context* pipe,
-                                        const struct pipe_index_buffer *ib)
-{
-    struct r300_context* r300 = r300_context(pipe);
-
-    if (ib) {
-        pipe_resource_reference(&r300->index_buffer.buffer, ib->buffer);
-        memcpy(&r300->index_buffer, ib, sizeof(*ib));
-    } else {
-        pipe_resource_reference(&r300->index_buffer.buffer, NULL);
-    }
-}
-
-static void r300_set_index_buffer_swtcl(struct pipe_context* pipe,
-                                        const struct pipe_index_buffer *ib)
-{
-    struct r300_context* r300 = r300_context(pipe);
-
-    if (ib) {
-        const void *buf = NULL;
-        if (ib->user_buffer) {
-            buf = ib->user_buffer;
-        } else if (ib->buffer) {
-            buf = r300_resource(ib->buffer)->malloced_buffer;
-        }
-        draw_set_indexes(r300->draw,
-                         (const ubyte *) buf + ib->offset,
-                         ib->index_size);
-    }
-}
-
 /* Initialize the PSC tables. */
 static void r300_vertex_psc(struct r300_vertex_element_state *velems)
 {
@@ -1835,7 +1873,7 @@ 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;
     }
 
@@ -1881,7 +1919,7 @@ 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;
     }
@@ -1933,14 +1971,14 @@ 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_constant_buffer *cb)
+                                     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;
 
-    if (!cb)
+    if (!cb || (!cb->buffer && !cb->user_buffer))
         return;
 
     switch (shader) {
@@ -1998,7 +2036,7 @@ static void r300_set_constant_buffer(struct pipe_context *pipe,
     }
 }
 
-static void r300_texture_barrier(struct pipe_context *pipe)
+static void r300_texture_barrier(struct pipe_context *pipe, unsigned flags)
 {
     struct r300_context *r300 = r300_context(pipe);
 
@@ -2006,6 +2044,10 @@ static void r300_texture_barrier(struct pipe_context *pipe)
     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;
@@ -2038,24 +2080,21 @@ 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;
 
     if (r300->screen->caps.has_tcl) {
         r300->context.set_vertex_buffers = r300_set_vertex_buffers_hwtcl;
-        r300->context.set_index_buffer = r300_set_index_buffer_hwtcl;
     } else {
         r300->context.set_vertex_buffers = r300_set_vertex_buffers_swtcl;
-        r300->context.set_index_buffer = r300_set_index_buffer_swtcl;
     }
 
     r300->context.create_vertex_elements_state = r300_create_vertex_elements_state;
@@ -2067,4 +2106,5 @@ void r300_init_state_functions(struct r300_context* r300)
     r300->context.delete_vs_state = r300_delete_vs_state;
 
     r300->context.texture_barrier = r300_texture_barrier;
+    r300->context.memory_barrier = r300_memory_barrier;
 }