summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/cell
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/drivers/cell')
-rw-r--r--src/gallium/drivers/cell/ppu/cell_draw_arrays.c6
-rw-r--r--src/gallium/drivers/cell/ppu/cell_gen_fragment.c2
-rw-r--r--src/gallium/drivers/cell/ppu/cell_state_emit.c2
-rw-r--r--src/gallium/drivers/cell/ppu/cell_state_shader.c2
-rw-r--r--src/gallium/drivers/cell/ppu/cell_texture.c81
-rw-r--r--src/gallium/drivers/cell/spu/Makefile5
-rw-r--r--src/gallium/drivers/cell/spu/spu_per_fragment_op.c2
-rw-r--r--src/gallium/drivers/cell/spu/spu_tri.c179
8 files changed, 138 insertions, 141 deletions
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
index 67949b73dd..644496db40 100644
--- a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
@@ -51,7 +51,7 @@ cell_map_constant_buffers(struct cell_context *sp)
struct pipe_winsys *ws = sp->pipe.winsys;
uint i;
for (i = 0; i < 2; i++) {
- if (sp->constants[i].size) {
+ if (sp->constants[i].buffer && sp->constants[i].buffer->size) {
sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i].buffer,
PIPE_BUFFER_USAGE_CPU_READ);
cell_flush_buffer_range(sp, sp->mapped_constants[i],
@@ -61,7 +61,7 @@ cell_map_constant_buffers(struct cell_context *sp)
draw_set_mapped_constant_buffer(sp->draw,
sp->mapped_constants[PIPE_SHADER_VERTEX],
- sp->constants[PIPE_SHADER_VERTEX].size);
+ sp->constants[PIPE_SHADER_VERTEX].buffer->size);
}
static void
@@ -70,7 +70,7 @@ cell_unmap_constant_buffers(struct cell_context *sp)
struct pipe_winsys *ws = sp->pipe.winsys;
uint i;
for (i = 0; i < 2; i++) {
- if (sp->constants[i].size)
+ if (sp->constants[i].buffer && sp->constants[i].buffer->size)
ws->buffer_unmap(ws, sp->constants[i].buffer);
sp->mapped_constants[i] = NULL;
}
diff --git a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
index 9bdc71b676..66d4b3b6a3 100644
--- a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
+++ b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
@@ -161,7 +161,7 @@ gen_alpha_test(const struct pipe_depth_stencil_alpha_state *dsa,
if ((dsa->alpha.func != PIPE_FUNC_NEVER) &&
(dsa->alpha.func != PIPE_FUNC_ALWAYS)) {
/* load/splat the alpha reference float value */
- spe_load_float(f, ref_reg, dsa->alpha.ref);
+ spe_load_float(f, ref_reg, dsa->alpha.ref_value);
}
/* emit code to do the alpha comparison, updating 'mask' */
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index 39b85faeb8..ff529fe22c 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -239,7 +239,7 @@ cell_emit_state(struct cell_context *cell)
if (cell->dirty & (CELL_NEW_FS_CONSTANTS)) {
const uint shader = PIPE_SHADER_FRAGMENT;
- const uint num_const = cell->constants[shader].size / sizeof(float);
+ const uint num_const = cell->constants[shader].buffer->size / sizeof(float);
uint i, j;
float *buf = cell_batch_alloc16(cell, ROUNDUP16(32 + num_const * sizeof(float)));
uint32_t *ibuf = (uint32_t *) buf;
diff --git a/src/gallium/drivers/cell/ppu/cell_state_shader.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c
index 990f23e170..bf517ea563 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c
@@ -186,7 +186,6 @@ cell_set_constant_buffer(struct pipe_context *pipe,
const struct pipe_constant_buffer *buf)
{
struct cell_context *cell = cell_context(pipe);
- struct pipe_winsys *ws = pipe->winsys;
assert(shader < PIPE_SHADER_TYPES);
assert(index == 0);
@@ -197,7 +196,6 @@ cell_set_constant_buffer(struct pipe_context *pipe,
pipe_buffer_reference(pipe->screen,
&cell->constants[shader].buffer,
buf->buffer);
- cell->constants[shader].size = buf->size;
if (shader == PIPE_SHADER_VERTEX)
cell->dirty |= CELL_NEW_VS_CONSTANTS;
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index 4f16e2c6af..9ba995ab7d 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -307,9 +307,8 @@ cell_twiddle_texture(struct pipe_screen *screen,
const uint texHeight = ct->base.height[level];
const uint bufWidth = align(texWidth, TILE_SIZE);
const uint bufHeight = align(texHeight, TILE_SIZE);
- const void *map = pipe_buffer_map(screen, surface->buffer,
- PIPE_BUFFER_USAGE_CPU_READ);
- const uint *src = (const uint *) ((const ubyte *) map + surface->offset);
+ const void *map = screen->surface_map(screen, surface, PIPE_BUFFER_USAGE_CPU_READ);
+ const uint *src = (const uint *) map;
switch (ct->base.format) {
case PIPE_FORMAT_A8R8G8B8_UNORM:
@@ -324,12 +323,12 @@ cell_twiddle_texture(struct pipe_screen *screen,
/* allocate buffer for tiled data now */
struct pipe_winsys *ws = screen->winsys;
uint bytes = bufWidth * bufHeight * 4 * numFaces;
- ct->tiled_buffer[level] = ws->buffer_create(ws, 16,
- PIPE_BUFFER_USAGE_PIXEL,
- bytes);
+ ct->tiled_buffer[level] =
+ ws->buffer_create(ws, 16, PIPE_BUFFER_USAGE_PIXEL, bytes);
/* and map it */
- ct->tiled_mapped[level] = ws->buffer_map(ws, ct->tiled_buffer[level],
- PIPE_BUFFER_USAGE_GPU_READ);
+ ct->tiled_mapped[level] =
+ ws->buffer_map(ws, ct->tiled_buffer[level],
+ PIPE_BUFFER_USAGE_GPU_READ);
}
dst = (uint *) ((ubyte *) ct->tiled_mapped[level] + offset);
@@ -338,11 +337,11 @@ cell_twiddle_texture(struct pipe_screen *screen,
}
break;
default:
- printf("Cell: twiddle unsupported texture format %s\n", pf_name(ct->base.format));
- ;
+ printf("Cell: twiddle unsupported texture format %s\n",
+ pf_name(ct->base.format));
}
- pipe_buffer_unmap(screen, surface->buffer);
+ screen->surface_unmap(screen, surface);
}
@@ -357,8 +356,7 @@ cell_untwiddle_texture(struct pipe_screen *screen,
const uint level = surface->level;
const uint texWidth = ct->base.width[level];
const uint texHeight = ct->base.height[level];
- const void *map = pipe_buffer_map(screen, surface->buffer,
- PIPE_BUFFER_USAGE_CPU_READ);
+ const void *map = screen->surface_map(screen, surface, PIPE_BUFFER_USAGE_CPU_READ);
const uint *src = (const uint *) ((const ubyte *) map + surface->offset);
switch (ct->base.format) {
@@ -384,11 +382,12 @@ cell_untwiddle_texture(struct pipe_screen *screen,
default:
{
ct->untiled_data[level] = NULL;
- printf("Cell: untwiddle unsupported texture format %s\n", pf_name(ct->base.format));
+ printf("Cell: untwiddle unsupported texture format %s\n",
+ pf_name(ct->base.format));
}
}
- pipe_buffer_unmap(screen, surface->buffer);
+ screen->surface_unmap(screen, surface);
}
@@ -398,15 +397,13 @@ cell_get_tex_surface(struct pipe_screen *screen,
unsigned face, unsigned level, unsigned zslice,
unsigned usage)
{
- struct pipe_winsys *ws = screen->winsys;
struct cell_texture *ct = cell_texture(pt);
struct pipe_surface *ps;
- ps = ws->surface_alloc(ws);
+ ps = CALLOC_STRUCT(pipe_surface);
if (ps) {
- assert(ps->refcount);
- assert(ps->winsys);
- pipe_buffer_reference(screen, &ps->buffer, ct->buffer);
+ ps->refcount = 1;
+ pipe_texture_reference(&ps->texture, pt);
ps->format = pt->format;
ps->block = pt->block;
ps->width = pt->width[level];
@@ -425,9 +422,9 @@ cell_get_tex_surface(struct pipe_screen *screen,
ps->zslice = zslice;
if (pt->target == PIPE_TEXTURE_CUBE || pt->target == PIPE_TEXTURE_3D) {
- ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) *
- ps->nblocksy *
- ps->stride;
+ ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) *
+ ps->nblocksy *
+ ps->stride;
}
else {
assert(face == 0);
@@ -449,18 +446,27 @@ cell_tex_surface_release(struct pipe_screen *screen,
{
struct cell_texture *ct = cell_texture((*s)->texture);
const uint level = (*s)->level;
+ struct pipe_surface *surf = *s;
- if (((*s)->usage & PIPE_BUFFER_USAGE_CPU_READ) && (ct->untiled_data[level]))
+ if ((surf->usage & PIPE_BUFFER_USAGE_CPU_READ) && (ct->untiled_data[level]))
{
align_free(ct->untiled_data[level]);
ct->untiled_data[level] = NULL;
}
- /* XXX if done rendering to teximage, re-tile */
+ if ((ct->base.tex_usage & PIPE_TEXTURE_USAGE_SAMPLER) &&
+ (surf->usage & PIPE_BUFFER_USAGE_CPU_WRITE)) {
+ /* convert from linear to tiled layout */
+ cell_twiddle_texture(screen, surf);
+ }
- pipe_texture_reference(&(*s)->texture, NULL);
+ /* XXX if done rendering to teximage, re-tile */
- screen->winsys->surface_release(screen->winsys, s);
+ if (--surf->refcount == 0) {
+ pipe_texture_reference(&surf->texture, NULL);
+ FREE(surf);
+ }
+ *s = NULL;
}
@@ -475,17 +481,20 @@ cell_surface_map(struct pipe_screen *screen,
assert(ct);
+#if 0
if (flags & ~surface->usage) {
assert(0);
return NULL;
}
+#endif
- map = pipe_buffer_map( screen, surface->buffer, flags );
- if (map == NULL)
+ map = pipe_buffer_map( screen, ct->buffer, flags );
+ if (map == NULL) {
return NULL;
- else
- {
- if ((surface->usage & PIPE_BUFFER_USAGE_CPU_READ) && (ct->untiled_data[level])) {
+ }
+ else {
+ if ((surface->usage & PIPE_BUFFER_USAGE_CPU_READ) &&
+ (ct->untiled_data[level])) {
return (void *) ((ubyte *) ct->untiled_data[level] + surface->offset);
}
else {
@@ -503,13 +512,7 @@ cell_surface_unmap(struct pipe_screen *screen,
assert(ct);
- if ((ct->base.tex_usage & PIPE_TEXTURE_USAGE_SAMPLER) &&
- (surface->usage & PIPE_BUFFER_USAGE_CPU_WRITE)) {
- /* convert from linear to tiled layout */
- cell_twiddle_texture(screen, surface);
- }
-
- pipe_buffer_unmap( screen, surface->buffer );
+ pipe_buffer_unmap( screen, ct->buffer );
}
diff --git a/src/gallium/drivers/cell/spu/Makefile b/src/gallium/drivers/cell/spu/Makefile
index 116453b79c..3cc52301da 100644
--- a/src/gallium/drivers/cell/spu/Makefile
+++ b/src/gallium/drivers/cell/spu/Makefile
@@ -33,9 +33,10 @@ OLD_SOURCES = \
spu_vertex_shader.c
-SPU_OBJECTS = $(SOURCES:.c=.o) \
+SPU_OBJECTS = $(SOURCES:.c=.o)
+
+SPU_ASM_OUT = $(SOURCES:.c=.s)
-SPU_ASM_OUT = $(SOURCES:.c=.s) \
INCLUDE_DIRS = \
-I$(TOP)/src/mesa \
diff --git a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
index 683664e8a4..eba9f95cf1 100644
--- a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
+++ b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
@@ -85,7 +85,7 @@ spu_fallback_fragment_ops(uint x, uint y,
* Do alpha test
*/
if (spu.depth_stencil_alpha.alpha.enabled) {
- vector float ref = spu_splats(spu.depth_stencil_alpha.alpha.ref);
+ vector float ref = spu_splats(spu.depth_stencil_alpha.alpha.ref_value);
vector unsigned int amask;
switch (spu.depth_stencil_alpha.alpha.func) {
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index 0d9fcb9997..d727268475 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -29,7 +29,6 @@
* Triangle rendering within a tile.
*/
-#include <transpose_matrix4x4.h>
#include "pipe/p_compiler.h"
#include "pipe/p_format.h"
#include "util/u_math.h"
@@ -71,6 +70,12 @@ struct vertex_header {
#define MASK_ALL 0xf
+#define CHAN0 0
+#define CHAN1 1
+#define CHAN2 2
+#define CHAN3 3
+
+
#define DEBUG_VERTS 0
/**
@@ -144,99 +149,97 @@ struct setup_stage {
static struct setup_stage setup;
-/**
- * Evaluate attribute coefficients (plane equations) to compute
- * attribute values for the four fragments in a quad.
- * Eg: four colors will be computed (in AoS format).
- */
-static INLINE void
-eval_coeff(uint slot, float x, float y, vector float w, vector float result[4])
+static INLINE vector float
+splatx(vector float v)
{
- switch (spu.vertex_info.attrib[slot].interp_mode) {
- case INTERP_CONSTANT:
- result[QUAD_TOP_LEFT] =
- result[QUAD_TOP_RIGHT] =
- result[QUAD_BOTTOM_LEFT] =
- result[QUAD_BOTTOM_RIGHT] = setup.coef[slot].a0;
- break;
- case INTERP_LINEAR:
- {
- vector float dadx = setup.coef[slot].dadx;
- vector float dady = setup.coef[slot].dady;
- vector float topLeft =
- spu_add(setup.coef[slot].a0,
- spu_add(spu_mul(spu_splats(x), dadx),
- spu_mul(spu_splats(y), dady)));
-
- result[QUAD_TOP_LEFT] = topLeft;
- result[QUAD_TOP_RIGHT] = spu_add(topLeft, dadx);
- result[QUAD_BOTTOM_LEFT] = spu_add(topLeft, dady);
- result[QUAD_BOTTOM_RIGHT] = spu_add(spu_add(topLeft, dadx), dady);
- }
- break;
- case INTERP_PERSPECTIVE:
- {
- vector float dadx = setup.coef[slot].dadx;
- vector float dady = setup.coef[slot].dady;
- vector float topLeft =
- spu_add(setup.coef[slot].a0,
- spu_add(spu_mul(spu_splats(x), dadx),
- spu_mul(spu_splats(y), dady)));
-
- vector float wInv = spu_re(w); /* 1.0 / w */
-
- result[QUAD_TOP_LEFT] = spu_mul(topLeft, wInv);
- result[QUAD_TOP_RIGHT] = spu_mul(spu_add(topLeft, dadx), wInv);
- result[QUAD_BOTTOM_LEFT] = spu_mul(spu_add(topLeft, dady), wInv);
- result[QUAD_BOTTOM_RIGHT] = spu_mul(spu_add(spu_add(topLeft, dadx), dady), wInv);
- }
- break;
- case INTERP_POS:
- case INTERP_NONE:
- break;
- default:
- ASSERT(0);
- }
+ return spu_splats(spu_extract(v, CHAN0));
}
-
-/**
- * As above, but return 4 vectors in SOA format.
- * XXX this will all be re-written someday.
- */
-static INLINE void
-eval_coeff_soa(uint slot, float x, float y, vector float w, vector float result[4])
+static INLINE vector float
+splaty(vector float v)
{
- eval_coeff(slot, x, y, w, result);
- _transpose_matrix4x4(result, result);
+ return spu_splats(spu_extract(v, CHAN1));
}
+static INLINE vector float
+splatz(vector float v)
+{
+ return spu_splats(spu_extract(v, CHAN2));
+}
-/** Evalute coefficients to get Z for four pixels in a quad */
static INLINE vector float
-eval_z(float x, float y)
+splatw(vector float v)
{
- const uint slot = 0;
- const float dzdx = spu_extract(setup.coef[slot].dadx, 2);
- const float dzdy = spu_extract(setup.coef[slot].dady, 2);
- const float topLeft = spu_extract(setup.coef[slot].a0, 2) + x * dzdx + y * dzdy;
- const vector float topLeftv = spu_splats(topLeft);
- const vector float derivs = (vector float) { 0.0, dzdx, dzdy, dzdx + dzdy };
- return spu_add(topLeftv, derivs);
+ return spu_splats(spu_extract(v, CHAN3));
}
-/** Evalute coefficients to get W for four pixels in a quad */
-static INLINE vector float
-eval_w(float x, float y)
+/**
+ * Setup fragment shader inputs by evaluating triangle's vertex
+ * attribute coefficient info.
+ * \param x quad x pos
+ * \param y quad y pos
+ * \param fragZ returns quad Z values
+ * \param fragInputs returns fragment program inputs
+ * Note: this code could be incorporated into the fragment program
+ * itself to avoid the loop and switch.
+ */
+static void
+eval_inputs(float x, float y, vector float *fragZ, vector float fragInputs[])
{
- const uint slot = 0;
- const float dwdx = spu_extract(setup.coef[slot].dadx, 3);
- const float dwdy = spu_extract(setup.coef[slot].dady, 3);
- const float topLeft = spu_extract(setup.coef[slot].a0, 3) + x * dwdx + y * dwdy;
- const vector float topLeftv = spu_splats(topLeft);
- const vector float derivs = (vector float) { 0.0, dwdx, dwdy, dwdx + dwdy };
- return spu_add(topLeftv, derivs);
+ static const vector float deltaX = (const vector float) {0, 1, 0, 1};
+ static const vector float deltaY = (const vector float) {0, 0, 1, 1};
+
+ const uint posSlot = 0;
+ const vector float pos = setup.coef[posSlot].a0;
+ const vector float dposdx = setup.coef[posSlot].dadx;
+ const vector float dposdy = setup.coef[posSlot].dady;
+ const vector float fragX = spu_splats(x) + deltaX;
+ const vector float fragY = spu_splats(y) + deltaY;
+ vector float fragW, wInv;
+ uint i;
+
+ *fragZ = splatz(pos) + fragX * splatz(dposdx) + fragY * splatz(dposdy);
+ fragW = splatw(pos) + fragX * splatw(dposdx) + fragY * splatw(dposdy);
+ wInv = spu_re(fragW); /* 1 / w */
+
+ /* loop over fragment program inputs */
+ for (i = 0; i < spu.vertex_info.num_attribs; i++) {
+ uint attr = i + 1;
+ enum interp_mode interp = spu.vertex_info.attrib[attr].interp_mode;
+
+ /* constant term */
+ vector float a0 = setup.coef[attr].a0;
+ vector float r0 = splatx(a0);
+ vector float r1 = splaty(a0);
+ vector float r2 = splatz(a0);
+ vector float r3 = splatw(a0);
+
+ if (interp == INTERP_LINEAR || interp == INTERP_PERSPECTIVE) {
+ /* linear term */
+ vector float dadx = setup.coef[attr].dadx;
+ vector float dady = setup.coef[attr].dady;
+ /* Use SPU intrinsics here to get slightly better code.
+ * originally: r0 += fragX * splatx(dadx) + fragY * splatx(dady);
+ */
+ r0 = spu_madd(fragX, splatx(dadx), spu_madd(fragY, splatx(dady), r0));
+ r1 = spu_madd(fragX, splaty(dadx), spu_madd(fragY, splaty(dady), r1));
+ r2 = spu_madd(fragX, splatz(dadx), spu_madd(fragY, splatz(dady), r2));
+ r3 = spu_madd(fragX, splatw(dadx), spu_madd(fragY, splatw(dady), r3));
+ if (interp == INTERP_PERSPECTIVE) {
+ /* perspective term */
+ r0 *= wInv;
+ r1 *= wInv;
+ r2 *= wInv;
+ r3 *= wInv;
+ }
+ }
+ fragInputs[CHAN0] = r0;
+ fragInputs[CHAN1] = r1;
+ fragInputs[CHAN2] = r2;
+ fragInputs[CHAN3] = r3;
+ fragInputs += 4;
+ }
}
@@ -262,19 +265,11 @@ emit_quad( int x, int y, mask_t mask)
* Run fragment shader, execute per-fragment ops, update fb/tile.
*/
vector float inputs[4*4], outputs[2*4];
- vector float fragZ = eval_z((float) x, (float) y);
- vector float fragW = eval_w((float) x, (float) y);
vector unsigned int kill_mask;
+ vector float fragZ;
+
+ eval_inputs((float) x, (float) y, &fragZ, inputs);
- /* setup inputs */
-#if 0
- eval_coeff_soa(1, (float) x, (float) y, fragW, inputs);
-#else
- uint i;
- for (i = 0; i < spu.vertex_info.num_attribs; i++) {
- eval_coeff_soa(i+1, (float) x, (float) y, fragW, inputs + i * 4);
- }
-#endif
ASSERT(spu.fragment_program);
ASSERT(spu.fragment_ops);
generated by cgit v1.2.3 (git 2.39.1) at 2024-10-17 20:55:03 +0000