static void
writer_decode_clip_viewport(const struct ilo_builder *builder,
enum ilo_builder_writer_type which,
const struct ilo_builder_item *item)
{
const unsigned state_size = sizeof(uint32_t) * 4;
const unsigned count = item->size / state_size;
unsigned offset = item->offset;
unsigned i;
for (i = 0; i < count; i++) {
uint32_t dw;
dw = writer_dw(builder, which, offset, 0, "CLIP VP%d", i);
ilo_printf("xmin = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 1, "CLIP VP%d", i);
ilo_printf("xmax = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 2, "CLIP VP%d", i);
ilo_printf("ymin = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 3, "CLIP VP%d", i);
ilo_printf("ymax = %f\n", uif(dw));
offset += state_size;
}
}
static void
writer_decode_color_calc(const struct ilo_builder *builder,
enum ilo_builder_writer_type which,
const struct ilo_builder_item *item)
{
uint32_t dw;
dw = writer_dw(builder, which, item->offset, 0, "CC");
ilo_printf("alpha test format %s, round disable %d, "
"stencil ref %d, bf stencil ref %d\n",
GEN_EXTRACT(dw, GEN6_CC_DW0_ALPHATEST) ? "FLOAT32" : "UNORM8",
(bool) (dw & GEN6_CC_DW0_ROUND_DISABLE_DISABLE),
GEN_EXTRACT(dw, GEN6_CC_DW0_STENCIL0_REF),
GEN_EXTRACT(dw, GEN6_CC_DW0_STENCIL1_REF));
writer_dw(builder, which, item->offset, 1, "CC\n");
dw = writer_dw(builder, which, item->offset, 2, "CC");
ilo_printf("constant red %f\n", uif(dw));
dw = writer_dw(builder, which, item->offset, 3, "CC");
ilo_printf("constant green %f\n", uif(dw));
dw = writer_dw(builder, which, item->offset, 4, "CC");
ilo_printf("constant blue %f\n", uif(dw));
dw = writer_dw(builder, which, item->offset, 5, "CC");
ilo_printf("constant alpha %f\n", uif(dw));
}
static void
dump_VC4_PACKET_CLIPPER_XY_SCALING(void *cl, uint32_t offset, uint32_t hw_offset)
{
uint32_t *scale = cl + offset;
fprintf(stderr, "0x%08x 0x%08x: %f, %f (%f, %f, 0x%08x, 0x%08x)\n",
offset, hw_offset,
uif(scale[0]) / 16.0, uif(scale[1]) / 16.0,
uif(scale[0]), uif(scale[1]),
scale[0], scale[1]);
}
static void
dump_VC4_PACKET_CLIPPER_Z_SCALING(void *cl, uint32_t offset, uint32_t hw_offset)
{
uint32_t *translate = cl + offset;
uint32_t *scale = cl + offset + 8;
fprintf(stderr, "0x%08x 0x%08x: %f, %f (0x%08x, 0x%08x)\n",
offset, hw_offset,
uif(translate[0]), uif(translate[1]),
translate[0], translate[1]);
fprintf(stderr, "0x%08x 0x%08x: %f, %f (0x%08x, 0x%08x)\n",
offset + 8, hw_offset + 8,
uif(scale[0]), uif(scale[1]),
scale[0], scale[1]);
}
static void
writer_decode_sf_clip_viewport_gen7(const struct ilo_builder *builder,
enum ilo_builder_writer_type which,
const struct ilo_builder_item *item)
{
const unsigned state_size = sizeof(uint32_t) * 16;
const unsigned count = item->size / state_size;
unsigned offset = item->offset;
unsigned i;
for (i = 0; i < count; i++) {
uint32_t dw;
dw = writer_dw(builder, which, offset, 0, "SF_CLIP VP%d", i);
ilo_printf("m00 = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 1, "SF_CLIP VP%d", i);
ilo_printf("m11 = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 2, "SF_CLIP VP%d", i);
ilo_printf("m22 = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 3, "SF_CLIP VP%d", i);
ilo_printf("m30 = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 4, "SF_CLIP VP%d", i);
ilo_printf("m31 = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 5, "SF_CLIP VP%d", i);
ilo_printf("m32 = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 8, "SF_CLIP VP%d", i);
ilo_printf("guardband xmin = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 9, "SF_CLIP VP%d", i);
ilo_printf("guardband xmax = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 10, "SF_CLIP VP%d", i);
ilo_printf("guardband ymin = %f\n", uif(dw));
dw = writer_dw(builder, which, offset, 11, "SF_CLIP VP%d", i);
ilo_printf("guardband ymax = %f\n", uif(dw));
offset += state_size;
}
}
static void
dump_float(void *cl, uint32_t offset, uint32_t hw_offset)
{
void *f = cl + offset;
fprintf(stderr, "0x%08x 0x%08x: %f (0x%08x)\n",
offset, hw_offset, uif(*(uint32_t *)f), *(uint32_t *)f);
}
static void
writer_decode_blob(const struct ilo_builder *builder,
enum ilo_builder_writer_type which,
const struct ilo_builder_item *item)
{
const unsigned state_size = sizeof(uint32_t);
const unsigned count = item->size / state_size;
unsigned offset = item->offset;
unsigned i;
for (i = 0; i < count; i += 4) {
const uint32_t *dw = writer_pointer(builder, which, offset);
writer_dw(builder, which, offset, 0, "BLOB%d", i / 4);
switch (count - i) {
case 1:
ilo_printf("(%10.4f, %10c, %10c, %10c) "
"(0x%08x, %10c, %10c, %10c)\n",
uif(dw[0]), 'X', 'X', 'X',
dw[0], 'X', 'X', 'X');
break;
case 2:
ilo_printf("(%10.4f, %10.4f, %10c, %10c) "
"(0x%08x, 0x%08x, %10c, %10c)\n",
uif(dw[0]), uif(dw[1]), 'X', 'X',
dw[0], dw[1], 'X', 'X');
break;
case 3:
ilo_printf("(%10.4f, %10.4f, %10.4f, %10c) "
"(0x%08x, 0x%08x, 0x%08x, %10c)\n",
uif(dw[0]), uif(dw[1]), uif(dw[2]), 'X',
dw[0], dw[1], dw[2], 'X');
break;
default:
ilo_printf("(%10.4f, %10.4f, %10.4f, %10.4f) "
"(0x%08x, 0x%08x, 0x%08x, 0x%08x)\n",
uif(dw[0]), uif(dw[1]), uif(dw[2]), uif(dw[3]),
dw[0], dw[1], dw[2], dw[3]);
break;
}
offset += state_size * 4;
}
}
static void
writer_decode_blob(const struct ilo_builder *builder,
enum ilo_builder_writer_type which,
const struct ilo_builder_item *item)
{
const unsigned state_size = sizeof(uint32_t) * 4;
const unsigned count = item->size / state_size;
unsigned offset = item->offset;
unsigned i;
for (i = 0; i < count; i++) {
const uint32_t *dw = writer_pointer(builder, which, offset);
writer_dw(builder, which, offset, 0, "BLOB%d", i);
/* output a single line for all four DWords */
ilo_printf("(% f, % f, % f, % f) (0x%08x, 0x%08x, 0x%08x, 0x%08x)\n",
uif(dw[0]), uif(dw[1]), uif(dw[2]), uif(dw[3]),
dw[0], dw[1], dw[2], dw[3]);
offset += state_size;
}
}
static void print_value(FILE *file, uint32_t value, int bits)
{
/* Guess if it's int or float */
if (value <= (1 << 15))
fprintf(file, "%u\n", value);
else {
float f = uif(value);
if (fabs(f) < 100000 && f*10 == floor(f*10))
fprintf(file, "%.1ff\n", f);
else
/* Don't print more leading zeros than there are bits. */
fprintf(file, "0x%0*x\n", bits / 4, value);
}
}
static void
print_load_imm(uint64_t inst)
{
uint32_t imm = inst;
uint32_t waddr_add = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
uint32_t waddr_mul = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
uint32_t cond_add = QPU_GET_FIELD(inst, QPU_COND_ADD);
uint32_t cond_mul = QPU_GET_FIELD(inst, QPU_COND_MUL);
fprintf(stderr, "load_imm ");
print_alu_dst(inst, false);
fprintf(stderr, "%s, ", (waddr_add != QPU_W_NOP ?
DESC(qpu_condflags, cond_add) : ""));
print_alu_dst(inst, true);
fprintf(stderr, "%s, ", (waddr_mul != QPU_W_NOP ?
DESC(qpu_condflags, cond_mul) : ""));
fprintf(stderr, "0x%08x (%f)", imm, uif(imm));
}
static void
print_load_imm(uint64_t inst)
{
uint32_t imm = inst;
uint32_t waddr_add = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
uint32_t waddr_mul = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
uint32_t cond_add = QPU_GET_FIELD(inst, QPU_COND_ADD);
uint32_t cond_mul = QPU_GET_FIELD(inst, QPU_COND_MUL);
fprintf(stderr, "load_imm ");
print_alu_dst(inst, false);
if (waddr_add != QPU_W_NOP)
vc4_qpu_disasm_cond(stderr, cond_add);
fprintf(stderr, ", ");
print_alu_dst(inst, true);
if (waddr_mul != QPU_W_NOP)
vc4_qpu_disasm_cond(stderr, cond_mul);
fprintf(stderr, ", ");
fprintf(stderr, "0x%08x (%f)", imm, uif(imm));
}
static void
write_texture_p1(struct vc5_job *job,
struct vc5_cl_out **uniforms,
struct vc5_texture_stateobj *texstate,
uint32_t unit)
{
struct pipe_sampler_view *psview = texstate->textures[unit];
struct vc5_sampler_view *sview = vc5_sampler_view(psview);
struct V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1 unpacked = {
.texture_state_record_base_address = texstate->texture_state[unit],
};
uint32_t packed;
V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1_pack(&job->indirect,
(uint8_t *)&packed,
&unpacked);
cl_aligned_u32(uniforms, packed | sview->p1);
}
struct vc5_cl_reloc
vc5_write_uniforms(struct vc5_context *vc5, struct vc5_compiled_shader *shader,
struct vc5_constbuf_stateobj *cb,
struct vc5_texture_stateobj *texstate)
{
struct v3d_uniform_list *uinfo = &shader->prog_data.base->uniforms;
struct vc5_job *job = vc5->job;
const uint32_t *gallium_uniforms = cb->cb[0].user_buffer;
struct vc5_bo *ubo = vc5_upload_ubo(vc5, shader, gallium_uniforms);
/* We always need to return some space for uniforms, because the HW
* will be prefetching, even if we don't read any in the program.
*/
vc5_cl_ensure_space(&job->indirect, MAX2(uinfo->count, 1) * 4, 4);
struct vc5_cl_reloc uniform_stream = cl_get_address(&job->indirect);
vc5_bo_reference(uniform_stream.bo);
struct vc5_cl_out *uniforms =
cl_start(&job->indirect);
for (int i = 0; i < uinfo->count; i++) {
switch (uinfo->contents[i]) {
case QUNIFORM_CONSTANT:
cl_aligned_u32(&uniforms, uinfo->data[i]);
break;
case QUNIFORM_UNIFORM:
cl_aligned_u32(&uniforms,
gallium_uniforms[uinfo->data[i]]);
break;
case QUNIFORM_VIEWPORT_X_SCALE:
cl_aligned_f(&uniforms, vc5->viewport.scale[0] * 256.0f);
break;
case QUNIFORM_VIEWPORT_Y_SCALE:
cl_aligned_f(&uniforms, vc5->viewport.scale[1] * 256.0f);
break;
case QUNIFORM_VIEWPORT_Z_OFFSET:
cl_aligned_f(&uniforms, vc5->viewport.translate[2]);
break;
case QUNIFORM_VIEWPORT_Z_SCALE:
cl_aligned_f(&uniforms, vc5->viewport.scale[2]);
break;
case QUNIFORM_USER_CLIP_PLANE:
cl_aligned_f(&uniforms,
vc5->clip.ucp[uinfo->data[i] / 4][uinfo->data[i] % 4]);
break;
case QUNIFORM_TEXTURE_CONFIG_P1:
write_texture_p1(job, &uniforms, texstate,
uinfo->data[i]);
break;
#if 0
case QUNIFORM_TEXTURE_FIRST_LEVEL:
write_texture_first_level(job, &uniforms, texstate,
uinfo->data[i]);
break;
#endif
case QUNIFORM_TEXRECT_SCALE_X:
case QUNIFORM_TEXRECT_SCALE_Y:
cl_aligned_u32(&uniforms,
get_texrect_scale(texstate,
uinfo->contents[i],
uinfo->data[i]));
break;
case QUNIFORM_TEXTURE_WIDTH:
case QUNIFORM_TEXTURE_HEIGHT:
case QUNIFORM_TEXTURE_DEPTH:
case QUNIFORM_TEXTURE_ARRAY_SIZE:
case QUNIFORM_TEXTURE_LEVELS:
cl_aligned_u32(&uniforms,
get_texture_size(texstate,
uinfo->contents[i],
uinfo->data[i]));
break;
case QUNIFORM_STENCIL:
cl_aligned_u32(&uniforms,
vc5->zsa->stencil_uniforms[uinfo->data[i]] |
(uinfo->data[i] <= 1 ?
(vc5->stencil_ref.ref_value[uinfo->data[i]] << 8) :
0));
break;
case QUNIFORM_ALPHA_REF:
cl_aligned_f(&uniforms,
vc5->zsa->base.alpha.ref_value);
break;
case QUNIFORM_SAMPLE_MASK:
cl_aligned_u32(&uniforms, vc5->sample_mask);
break;
case QUNIFORM_UBO_ADDR:
if (uinfo->data[i] == 0) {
cl_aligned_reloc(&job->indirect, &uniforms,
ubo, 0);
} else {
int ubo_index = uinfo->data[i];
struct vc5_resource *rsc =
vc5_resource(cb->cb[ubo_index].buffer);
cl_aligned_reloc(&job->indirect, &uniforms,
rsc->bo,
cb->cb[ubo_index].buffer_offset);
}
break;
case QUNIFORM_TEXTURE_FIRST_LEVEL:
case QUNIFORM_TEXTURE_MSAA_ADDR:
case QUNIFORM_TEXTURE_BORDER_COLOR:
/* XXX */
break;
default:
assert(quniform_contents_is_texture_p0(uinfo->contents[i]));
write_texture_p0(job, &uniforms, texstate,
uinfo->contents[i] -
QUNIFORM_TEXTURE_CONFIG_P0_0,
uinfo->data[i]);
break;
}
#if 0
uint32_t written_val = *((uint32_t *)uniforms - 1);
fprintf(stderr, "shader %p[%d]: 0x%08x / 0x%08x (%f)\n",
shader, i, __gen_address_offset(&uniform_stream) + i * 4,
written_val, uif(written_val));
#endif
}
cl_end(&job->indirect, uniforms);
vc5_bo_unreference(&ubo);
return uniform_stream;
}
void
vc5_set_shader_uniform_dirty_flags(struct vc5_compiled_shader *shader)
{
uint32_t dirty = 0;
for (int i = 0; i < shader->prog_data.base->uniforms.count; i++) {
switch (shader->prog_data.base->uniforms.contents[i]) {
case QUNIFORM_CONSTANT:
break;
case QUNIFORM_UNIFORM:
case QUNIFORM_UBO_ADDR:
dirty |= VC5_DIRTY_CONSTBUF;
break;
case QUNIFORM_VIEWPORT_X_SCALE:
case QUNIFORM_VIEWPORT_Y_SCALE:
case QUNIFORM_VIEWPORT_Z_OFFSET:
case QUNIFORM_VIEWPORT_Z_SCALE:
dirty |= VC5_DIRTY_VIEWPORT;
break;
case QUNIFORM_USER_CLIP_PLANE:
dirty |= VC5_DIRTY_CLIP;
break;
case QUNIFORM_TEXTURE_CONFIG_P1:
case QUNIFORM_TEXTURE_BORDER_COLOR:
case QUNIFORM_TEXTURE_FIRST_LEVEL:
case QUNIFORM_TEXTURE_MSAA_ADDR:
case QUNIFORM_TEXRECT_SCALE_X:
case QUNIFORM_TEXRECT_SCALE_Y:
case QUNIFORM_TEXTURE_WIDTH:
case QUNIFORM_TEXTURE_HEIGHT:
case QUNIFORM_TEXTURE_DEPTH:
case QUNIFORM_TEXTURE_ARRAY_SIZE:
case QUNIFORM_TEXTURE_LEVELS:
/* We could flag this on just the stage we're
* compiling for, but it's not passed in.
*/
dirty |= VC5_DIRTY_FRAGTEX | VC5_DIRTY_VERTTEX;
break;
case QUNIFORM_STENCIL:
case QUNIFORM_ALPHA_REF:
dirty |= VC5_DIRTY_ZSA;
break;
case QUNIFORM_SAMPLE_MASK:
dirty |= VC5_DIRTY_SAMPLE_MASK;
break;
default:
assert(quniform_contents_is_texture_p0(shader->prog_data.base->uniforms.contents[i]));
dirty |= VC5_DIRTY_FRAGTEX | VC5_DIRTY_VERTTEX;
break;
}
}
shader->uniform_dirty_bits = dirty;
}