summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/r300/r300_state.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/drivers/r300/r300_state.c')
-rw-r--r--src/gallium/drivers/r300/r300_state.c205
1 files changed, 132 insertions, 73 deletions
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index c16cadd040..af063d4b20 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -20,16 +20,20 @@
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
+#include "draw/draw_context.h"
+
#include "util/u_math.h"
+#include "util/u_memory.h"
#include "util/u_pack_color.h"
-#include "util/u_debug.h"
+#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"
+#include "r300_screen.h"
#include "r300_state_inlines.h"
#include "r300_fs.h"
#include "r300_vs.h"
@@ -45,23 +49,49 @@ static void* r300_create_blend_state(struct pipe_context* pipe,
{
struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state);
- if (state->blend_enable) {
- /* XXX for now, always do separate alpha...
- * is it faster to do it with one reg? */
+ if (state->blend_enable)
+ {
+ unsigned eqRGB = state->rgb_func;
+ unsigned srcRGB = state->rgb_src_factor;
+ unsigned dstRGB = state->rgb_dst_factor;
+
+ unsigned eqA = state->alpha_func;
+ unsigned srcA = state->alpha_src_factor;
+ unsigned dstA = state->alpha_dst_factor;
+
+ /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
+ * this is just the crappy D3D naming */
blend->blend_control = R300_ALPHA_BLEND_ENABLE |
- R300_SEPARATE_ALPHA_ENABLE |
- R300_READ_ENABLE |
- r300_translate_blend_function(state->rgb_func) |
- (r300_translate_blend_factor(state->rgb_src_factor) <<
- R300_SRC_BLEND_SHIFT) |
- (r300_translate_blend_factor(state->rgb_dst_factor) <<
- R300_DST_BLEND_SHIFT);
- blend->alpha_blend_control =
- r300_translate_blend_function(state->alpha_func) |
- (r300_translate_blend_factor(state->alpha_src_factor) <<
- R300_SRC_BLEND_SHIFT) |
- (r300_translate_blend_factor(state->alpha_dst_factor) <<
- R300_DST_BLEND_SHIFT);
+ r300_translate_blend_function(eqRGB) |
+ ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) |
+ ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT);
+
+ /* optimization: some operations do not require the destination color */
+ 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)
+ blend->blend_control |= R300_READ_ENABLE;
+
+ /* XXX implement the optimization with DISCARD_SRC_PIXELS*/
+ /* XXX implement the optimization with SRC_ALPHA_?_NO_READ */
+
+ /* separate alpha */
+ if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
+ blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE;
+ blend->alpha_blend_control =
+ r300_translate_blend_function(eqA) |
+ (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) |
+ (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT);
+ }
}
/* PIPE_LOGICOP_* don't need to be translated, fortunately. */
@@ -95,25 +125,29 @@ static void r300_delete_blend_state(struct pipe_context* pipe,
FREE(state);
}
+/* Convert float to 10bit integer */
+static unsigned float_to_fixed10(float f)
+{
+ return CLAMP((unsigned)(f * 1023.9f), 0, 1023);
+}
+
/* Set blend color.
* Setup both R300 and R500 registers, figure out later which one to write. */
static void r300_set_blend_color(struct pipe_context* pipe,
const struct pipe_blend_color* color)
{
struct r300_context* r300 = r300_context(pipe);
- ubyte ur, ug, ub, ua;
-
- ur = float_to_ubyte(color->color[0]);
- ug = float_to_ubyte(color->color[1]);
- ub = float_to_ubyte(color->color[2]);
- ua = float_to_ubyte(color->color[3]);
util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM,
&r300->blend_color_state->blend_color);
- /* XXX this is wrong */
- r300->blend_color_state->blend_color_red_alpha = ur | (ua << 16);
- r300->blend_color_state->blend_color_green_blue = ub | (ug << 16);
+ /* XXX if FP16 blending is enabled, we should use the FP16 format */
+ r300->blend_color_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 =
+ float_to_fixed10(color->color[2]) |
+ (float_to_fixed10(color->color[1]) << 16);
r300->dirty_state |= R300_NEW_BLEND_COLOR;
}
@@ -132,31 +166,6 @@ static void r300_set_clip_state(struct pipe_context* pipe,
}
}
-static void
- r300_set_constant_buffer(struct pipe_context* pipe,
- uint shader, uint index,
- const struct pipe_constant_buffer* buffer)
-{
- struct r300_context* r300 = r300_context(pipe);
-
- /* This entire chunk of code seems ever-so-slightly baked.
- * It's as if I've got pipe_buffer* matryoshkas... */
- if (buffer && buffer->buffer && buffer->buffer->size) {
- void* map = pipe->winsys->buffer_map(pipe->winsys, buffer->buffer,
- PIPE_BUFFER_USAGE_CPU_READ);
- memcpy(r300->shader_constants[shader].constants,
- map, buffer->buffer->size);
- pipe->winsys->buffer_unmap(pipe->winsys, buffer->buffer);
-
- r300->shader_constants[shader].count =
- buffer->buffer->size / (sizeof(float) * 4);
- } else {
- r300->shader_constants[shader].count = 0;
- }
-
- r300->dirty_state |= R300_NEW_CONSTANTS;
-}
-
/* 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.
@@ -166,6 +175,8 @@ static void*
r300_create_dsa_state(struct pipe_context* pipe,
const struct pipe_depth_stencil_alpha_state* state)
{
+ struct r300_capabilities *caps =
+ r300_screen(r300_context(pipe)->context.screen)->caps;
struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
/* Depth test setup. */
@@ -210,9 +221,16 @@ static void*
(r300_translate_stencil_op(state->stencil[1].zfail_op) <<
R300_S_BACK_ZFAIL_OP_SHIFT);
- dsa->stencil_ref_bf = (state->stencil[1].ref_value) |
- (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) |
- (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT);
+ /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */
+ if (caps->is_r500)
+ {
+ dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
+ dsa->stencil_ref_bf = (state->stencil[1].ref_value) |
+ (state->stencil[1].valuemask <<
+ R300_STENCILMASK_SHIFT) |
+ (state->stencil[1].writemask <<
+ R300_STENCILWRITEMASK_SHIFT);
+ }
}
}
@@ -221,11 +239,13 @@ static void*
dsa->alpha_function =
r300_translate_alpha_function(state->alpha.func) |
R300_FG_ALPHA_FUNC_ENABLE;
- dsa->alpha_reference = CLAMP(state->alpha.ref_value * 1023.0f,
- 0, 1023);
- } else {
- /* XXX need to fix this to be dynamically set
- dsa->z_buffer_top = R300_ZTOP_ENABLE; */
+
+ /* XXX figure out why emitting 10bit alpha ref causes CS to dump */
+ /* always use 8bit alpha ref */
+ dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value);
+
+ if (caps->is_r500)
+ dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT;
}
return (void*)dsa;
@@ -261,7 +281,9 @@ static void
{
struct r300_context* r300 = r300_context(pipe);
- draw_flush(r300->draw);
+ if (r300->draw) {
+ draw_flush(r300->draw);
+ }
r300->framebuffer_state = *state;
@@ -300,7 +322,7 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
r300->fs = fs;
- r300->dirty_state |= R300_NEW_FRAGMENT_SHADER;
+ r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS;
}
/* Delete fragment shader state. */
@@ -308,7 +330,7 @@ static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
{
struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
rc_constants_destroy(&fs->code.constants);
- FREE(fs->state.tokens);
+ FREE((void*)fs->state.tokens);
FREE(shader);
}
@@ -424,11 +446,17 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
struct r300_context* r300 = r300_context(pipe);
struct r300_rs_state* rs = (struct r300_rs_state*)state;
- draw_flush(r300->draw);
- draw_set_rasterizer_state(r300->draw, &rs->rs);
+ if (r300->draw) {
+ draw_flush(r300->draw);
+ draw_set_rasterizer_state(r300->draw, &rs->rs);
+ }
r300->rs_state = rs;
+ /* 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;
}
/* Free rasterizer state. */
@@ -562,17 +590,14 @@ static void r300_set_viewport_state(struct pipe_context* pipe,
r300->viewport_state->vte_control = R300_VTX_W0_FMT;
if (state->scale[0] != 1.0f) {
- assert(state->scale[0] != 0.0f);
r300->viewport_state->xscale = state->scale[0];
r300->viewport_state->vte_control |= R300_VPORT_X_SCALE_ENA;
}
if (state->scale[1] != 1.0f) {
- assert(state->scale[1] != 0.0f);
r300->viewport_state->yscale = state->scale[1];
r300->viewport_state->vte_control |= R300_VPORT_Y_SCALE_ENA;
}
if (state->scale[2] != 1.0f) {
- assert(state->scale[2] != 0.0f);
r300->viewport_state->zscale = state->scale[2];
r300->viewport_state->vte_control |= R300_VPORT_Z_SCALE_ENA;
}
@@ -603,8 +628,10 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
r300->vertex_buffer_count = count;
- draw_flush(r300->draw);
- draw_set_vertex_buffers(r300->draw, count, buffers);
+ if (r300->draw) {
+ draw_flush(r300->draw);
+ draw_set_vertex_buffers(r300->draw, count, buffers);
+ }
}
static void r300_set_vertex_elements(struct pipe_context* pipe,
@@ -613,8 +640,15 @@ static void r300_set_vertex_elements(struct pipe_context* pipe,
{
struct r300_context* r300 = r300_context(pipe);
- draw_flush(r300->draw);
- draw_set_vertex_elements(r300->draw, count, elements);
+ memcpy(r300->vertex_elements, elements,
+ sizeof(struct pipe_vertex_element) * count);
+
+ r300->vertex_element_count = count;
+
+ if (r300->draw) {
+ draw_flush(r300->draw);
+ draw_set_vertex_elements(r300->draw, count, elements);
+ }
}
static void* r300_create_vs_state(struct pipe_context* pipe,
@@ -657,7 +691,7 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
draw_bind_vertex_shader(r300->draw, vs->draw);
r300->vs = vs;
- r300->dirty_state |= R300_NEW_VERTEX_SHADER;
+ r300->dirty_state |= R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS;
} else {
draw_bind_vertex_shader(r300->draw,
(struct draw_vertex_shader*)shader);
@@ -673,7 +707,7 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
rc_constants_destroy(&vs->code.constants);
draw_delete_vertex_shader(r300->draw, vs->draw);
- FREE(vs->state.tokens);
+ FREE((void*)vs->state.tokens);
FREE(shader);
} else {
draw_delete_vertex_shader(r300->draw,
@@ -681,6 +715,31 @@ 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 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)
+ {
+ 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);
+
+ if (shader == PIPE_SHADER_VERTEX)
+ r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
+ else if (shader == PIPE_SHADER_FRAGMENT)
+ r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
+}
+
void r300_init_state_functions(struct r300_context* r300)
{
r300->context.create_blend_state = r300_create_blend_state;