#include "util/u_framebuffer.h"
#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"
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)
(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. */
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 */
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)) {
- 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;
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);
}
}
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);
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);
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. */
(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;
+ float tmp;
CB_LOCALS;
state->state = *color; /* Save it, so that we can reuse it in set_fb_state */
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:;
}
}
switch (format) {
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]) |
}
}
-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. - separate write mask depth for decomp flush */
if (state->depth.writemask) {
- dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
+ z_buffer_control |= R300_Z_WRITE_ENABLE;
}
if (state->depth.enabled) {
- dsa->z_buffer_control |= R300_Z_ENABLE;
+ 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);
- } else {
- /* We must enable depth test, otherwise occlusion queries won't work. */
- dsa->z_buffer_control |= R300_Z_ENABLE;
- dsa->z_stencil_control |= R300_ZS_ALWAYS;
}
/* 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) <<
(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) <<
(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 ||
R300_FG_ALPHA_FUNC_ENABLE;
dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value);
- dsa->alpha_value = util_float_to_half(state->alpha.ref_value);
-
- if (caps->is_r500) {
- dsa->alpha_function_fp16 = dsa->alpha_function |
- R500_FG_ALPHA_FUNC_FP16_ENABLE;
- dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT;
- }
+ alpha_value_fp16 = util_float_to_half(state->alpha.ref_value);
}
- BEGIN_CB(&dsa->cb_begin, 10);
- OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
+ BEGIN_CB(&dsa->cb_begin, 8);
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_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value);
+ 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_begin_fp16, 10);
- OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function_fp16);
+ BEGIN_CB(dsa->cb_zb_no_readwrite, 8);
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_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value);
- END_CB;
-
- /* We must enable depth test, otherwise occlusion queries won't work.
- * We setup a dummy zbuffer to silent the CS checker, see emit_fb_state. */
- BEGIN_CB(dsa->cb_zb_no_readwrite, 10);
- OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
- OUT_CB_REG_SEQ(R300_ZB_CNTL, 3);
- OUT_CB(R300_Z_ENABLE);
- OUT_CB(R300_ZS_ALWAYS);
OUT_CB(0);
- OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0);
- OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value);
- END_CB;
-
- BEGIN_CB(dsa->cb_fp16_zb_no_readwrite, 10);
- OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function_fp16);
- OUT_CB_REG_SEQ(R300_ZB_CNTL, 3);
- OUT_CB(R300_Z_ENABLE);
- OUT_CB(R300_ZS_ALWAYS);
+ OUT_CB(0);
OUT_CB(0);
OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0);
- OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value);
+ OUT_CB_REG(R500_FG_ALPHA_VALUE, alpha_value_fp16);
END_CB;
return (void*)dsa;
/* Now compute the fb_state atom size. */
r300->fb_state.size = 2 + (8 * state->nr_cbufs);
- if (r300->cbzb_clear) {
+ if (r300->cbzb_clear)
r300->fb_state.size += 10;
- } else if (state->zsbuf) {
+ else if (state->zsbuf) {
r300->fb_state.size += 10;
if (r300->hyperz_enabled)
r300->fb_state.size += 8;
- } else if (state->nr_cbufs) {
- r300->fb_state.size += 10;
+ }
+
+ 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 i, num_samples;
+
+ if (!fb->nr_cbufs && !fb->zsbuf)
+ return 1;
+
+ num_samples = 6;
+
+ for (i = 0; i < fb->nr_cbufs; i++)
+ num_samples = MIN2(num_samples, fb->cbufs[i]->texture->nr_samples);
+
+ if (fb->zsbuf)
+ num_samples = MIN2(num_samples, fb->zsbuf->texture->nr_samples);
+
+ 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)
}
assert(state->zsbuf || (r300->locked_zbuffer && !unlock_zbuffer) || !r300->zmask_in_use);
+ /* Set whether CMASK can be used. */
+ r300->cmask_in_use =
+ state->nr_cbufs == 1 &&
+ r300->screen->cmask_resource == state->cbufs[0]->texture;
+
/* Need to reset clamping or colormask. */
r300_mark_atom_dirty(r300, &r300->blend_state);
+ /* Re-swizzle the blend color. */
+ r300_set_blend_color(pipe, &((struct r300_blend_color_state*)r300->blend_color_state.state)->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);
}
}
- /* Set up AA config. */
- if (state->nr_cbufs && state->cbufs[0]->texture->nr_samples > 1) {
- aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE;
+ r300->num_samples = r300_get_num_samples(r300);
- switch (state->cbufs[0]->texture->nr_samples) {
+ /* Set up AA config. */
+ if (r300->num_samples > 1) {
+ switch (r300->num_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;
+ 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_NUM_AA_SUBSAMPLES_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_NUM_AA_SUBSAMPLES_6;
+ aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE |
+ R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6;
break;
}
} else {
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. */
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;
if (r300->draw && rs) {
draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state);
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;
} else {
r300->polygon_offset_enabled = FALSE;
r300->sprite_coord_enable = 0;
r300->two_sided_color = FALSE;
+ r300->msaa_enable = FALSE;
+ r300->flatshade = 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);
}
+
+ 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;
+ }
+ }
}
/* Free rasterizer state. */
}
static void r300_bind_sampler_states(struct pipe_context* pipe,
- unsigned count,
+ unsigned shader,
+ unsigned start, unsigned count,
void** states)
{
struct r300_context* r300 = r300_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_mark_atom_dirty(r300, &r300->textures_state);
}
+
static void r300_lacks_vertex_textures(struct pipe_context* pipe,
unsigned count,
void** states)
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);
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 =
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;
}
}
static void r300_set_vertex_buffers_hwtcl(struct pipe_context* pipe,
- unsigned count,
+ unsigned start_slot, unsigned count,
const struct pipe_vertex_buffer* buffers)
{
struct r300_context* r300 = r300_context(pipe);
+ util_set_vertex_buffers_count(r300->vertex_buffer,
+ &r300->nr_vertex_buffers,
+ buffers, start_slot, count);
+
/* There must be at least one vertex buffer set, otherwise it locks up. */
- if (!count) {
- buffers = &r300->dummy_vb;
- count = 1;
+ if (!r300->nr_vertex_buffers) {
+ util_set_vertex_buffers_count(r300->vertex_buffer,
+ &r300->nr_vertex_buffers,
+ &r300->dummy_vb, 0, 1);
}
- util_copy_vertex_buffers(r300->vertex_buffer,
- &r300->nr_vertex_buffers,
- buffers, count);
-
r300->vertex_arrays_dirty = TRUE;
}
static void r300_set_vertex_buffers_swtcl(struct pipe_context* pipe,
- unsigned count,
+ unsigned start_slot, unsigned count,
const struct pipe_vertex_buffer* buffers)
{
struct r300_context* r300 = r300_context(pipe);
unsigned i;
- util_copy_vertex_buffers(r300->vertex_buffer,
- &r300->nr_vertex_buffers,
- buffers, count);
- draw_set_vertex_buffers(r300->draw, count, buffers);
+ 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);
+
+ if (!buffers)
+ return;
for (i = 0; i < count; i++) {
if (buffers[i].user_buffer) {
- draw_set_mapped_vertex_buffer(r300->draw, i,
- buffers[i].user_buffer);
+ draw_set_mapped_vertex_buffer(r300->draw, start_slot + i,
+ buffers[i].user_buffer, ~0);
} else if (buffers[i].buffer) {
- draw_set_mapped_vertex_buffer(r300->draw, i,
- r300_resource(buffers[i].buffer)->malloced_buffer);
+ draw_set_mapped_vertex_buffer(r300->draw, start_slot + i,
+ r300_resource(buffers[i].buffer)->malloced_buffer, ~0);
}
}
}
}
draw_set_indexes(r300->draw,
(const ubyte *) buf + ib->offset,
- ib->index_size);
+ ib->index_size, ~0);
}
}
struct r300_constant_buffer *cbuf;
uint32_t *mapped;
- if (!cb)
+ if (!cb || (!cb->buffer && !cb->user_buffer))
return;
switch (shader) {
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.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;
projects / mesa.git / blobdiff