/*
* Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Copyright 2009 Marek Olšák <maraeo@gmail.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
struct rc_constant * constant,
struct r300_constant_buffer * externals)
{
- static const float zero[4] = { 0.0, 0.0, 0.0, 0.0 };
+ static float vec[4] = { 0.0, 0.0, 0.0, 1.0 };
+ struct pipe_texture *tex;
+
switch(constant->Type) {
case RC_CONSTANT_EXTERNAL:
return externals->constants[constant->u.External];
case RC_CONSTANT_IMMEDIATE:
return constant->u.Immediate;
+ case RC_CONSTANT_STATE:
+ switch (constant->u.State[0]) {
+ /* Factor for converting rectangle coords to
+ * normalized coords. Should only show up on non-r500. */
+ case RC_STATE_R300_TEXRECT_FACTOR:
+ tex = &r300->textures[constant->u.State[1]]->tex;
+ vec[0] = 1.0 / tex->width0;
+ vec[1] = 1.0 / tex->height0;
+ break;
+
+ default:
+ debug_printf("r300: Implementation error: "
+ "Unknown RC_CONSTANT type %d\n", constant->u.State[0]);
+ }
+ break;
+
default:
- debug_printf("r300: Implementation error: Unhandled constant type %i\n",
- constant->Type);
- return zero;
+ debug_printf("r300: Implementation error: "
+ "Unhandled constant type %d\n", constant->Type);
}
+
+ /* This should either be (0, 0, 0, 1), which should be a relatively safe
+ * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
+ * state factors. */
+ return vec;
}
/* Convert a normal single-precision float into the 7.16 format
BEGIN_CS(13 +
((code->inst_end + 1) * 6));
- OUT_CS_REG(R500_US_CONFIG, 0);
+ OUT_CS_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
OUT_CS_REG(R500_US_PIXSIZE, code->max_temp_idx);
OUT_CS_REG(R500_US_CODE_RANGE,
R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end));
if (!query)
return;
- /* XXX This will almost certainly not return good results
- * for overlapping queries. */
BEGIN_CS(4);
if (caps->family == CHIP_FAMILY_RV530) {
OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
END_CS;
}
-void r300_emit_scissor_state(struct r300_context* r300,
- struct r300_scissor_state* scissor)
+static void r300_emit_scissor_regs(struct r300_context* r300,
+ struct r300_scissor_regs* scissor)
{
CS_LOCALS(r300);
BEGIN_CS(3);
OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
- OUT_CS(scissor->scissor_top_left);
- OUT_CS(scissor->scissor_bottom_right);
+ OUT_CS(scissor->top_left);
+ OUT_CS(scissor->bottom_right);
END_CS;
}
+void r300_emit_scissor_state(struct r300_context* r300,
+ struct r300_scissor_state* scissor)
+{
+ if (r300->rs_state->rs.scissor) {
+ r300_emit_scissor_regs(r300, &scissor->scissor);
+ } else {
+ r300_emit_scissor_regs(r300, &scissor->framebuffer);
+ }
+}
+
void r300_emit_texture(struct r300_context* r300,
struct r300_sampler_state* sampler,
struct r300_texture* tex,
unsigned offset)
{
uint32_t filter0 = sampler->filter0;
+ uint32_t format0 = tex->state.format0;
+ unsigned min_level, max_level;
CS_LOCALS(r300);
/* to emulate 1D textures through 2D ones correctly */
filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE);
}
+ /* determine min/max levels */
+ /* the MAX_MIP level is the largest (finest) one */
+ max_level = MIN2(sampler->max_lod, tex->tex.last_level);
+ min_level = MIN2(sampler->min_lod, max_level);
+ format0 |= R300_TX_NUM_LEVELS(max_level);
+ filter0 |= R300_TX_MAX_MIP_LEVEL(min_level);
+
BEGIN_CS(16);
OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), filter0 |
(offset << 28));
OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1);
OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color);
- OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), tex->state.format0);
+ OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), format0);
OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1);
OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2);
OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (offset * 4), 1);
END_CS;
}
-/* XXX I can't read this and that's not good */
-void r300_emit_aos(struct r300_context* r300, unsigned offset)
+static boolean r300_validate_aos(struct r300_context *r300)
{
struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
struct pipe_vertex_element *velem = r300->vertex_element;
- CS_LOCALS(r300);
int i;
- unsigned aos_count = r300->vertex_element_count;
+ /* Check if formats and strides are aligned to the size of DWORD. */
+ for (i = 0; i < r300->vertex_element_count; i++) {
+ if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 ||
+ pf_get_blocksize(velem[i].src_format) % 4 != 0) {
+ return FALSE;
+ }
+ }
+ return TRUE;
+}
+
+void r300_emit_aos(struct r300_context* r300, unsigned offset)
+{
+ struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer;
+ struct pipe_vertex_element *velem = r300->vertex_element;
+ int i;
+ unsigned size1, size2, aos_count = r300->vertex_element_count;
unsigned packet_size = (aos_count * 3 + 1) / 2;
+ CS_LOCALS(r300);
+
+ /* XXX Move this checking to a more approriate place. */
+ if (!r300_validate_aos(r300)) {
+ /* XXX We should fallback using Draw. */
+ assert(0);
+ }
+
BEGIN_CS(2 + packet_size + aos_count * 2);
OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
OUT_CS(aos_count);
+
for (i = 0; i < aos_count - 1; i += 2) {
- int buf_num1 = velem[i].vertex_buffer_index;
- int buf_num2 = velem[i+1].vertex_buffer_index;
- assert(vbuf[buf_num1].stride % 4 == 0 && pf_get_size(velem[i].src_format) % 4 == 0);
- assert(vbuf[buf_num2].stride % 4 == 0 && pf_get_size(velem[i+1].src_format) % 4 == 0);
- OUT_CS((pf_get_size(velem[i].src_format) >> 2) | (vbuf[buf_num1].stride << 6) |
- (pf_get_size(velem[i+1].src_format) << 14) | (vbuf[buf_num2].stride << 22));
- OUT_CS(vbuf[buf_num1].buffer_offset + velem[i].src_offset +
- offset * vbuf[buf_num1].stride);
- OUT_CS(vbuf[buf_num2].buffer_offset + velem[i+1].src_offset +
- offset * vbuf[buf_num2].stride);
+ vb1 = &vbuf[velem[i].vertex_buffer_index];
+ vb2 = &vbuf[velem[i+1].vertex_buffer_index];
+ size1 = pf_get_blocksize(velem[i].src_format);
+ size2 = pf_get_blocksize(velem[i+1].src_format);
+
+ OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
+ R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
+ OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
+ OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
}
+
if (aos_count & 1) {
- int buf_num = velem[i].vertex_buffer_index;
- assert(vbuf[buf_num].stride % 4 == 0 && pf_get_size(velem[i].src_format) % 4 == 0);
- OUT_CS((pf_get_size(velem[i].src_format) >> 2) | (vbuf[buf_num].stride << 6));
- OUT_CS(vbuf[buf_num].buffer_offset + velem[i].src_offset +
- offset * vbuf[buf_num].stride);
+ vb1 = &vbuf[velem[i].vertex_buffer_index];
+ size1 = pf_get_blocksize(velem[i].src_format);
+
+ OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
+ OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
}
- /* XXX bare CS reloc */
for (i = 0; i < aos_count; i++) {
- cs_winsys->write_cs_reloc(cs_winsys,
- vbuf[velem[i].vertex_buffer_index].buffer,
- RADEON_GEM_DOMAIN_GTT,
- 0,
- 0);
- cs_count -= 2;
+ OUT_CS_RELOC_NO_OFFSET(vbuf[velem[i].vertex_buffer_index].buffer,
+ RADEON_GEM_DOMAIN_GTT, 0, 0);
}
END_CS;
}
+
#if 0
void r300_emit_draw_packet(struct r300_context* r300)
{
END_CS;
}
-/* XXX This should probably go to util ... */
-/* Return the number of bits set in the given number. */
-static unsigned bitcount(unsigned n)
-{
- unsigned bits;
- for (bits = 0; n > 0; n = n >> 1) {
- bits += n & 1;
- }
- return bits;
-}
-
-/* XXX ... and this one too. */
-#define MIN3(x, y, z) MIN2(MIN2(x, y), z)
void r300_emit_vertex_program_code(struct r300_context* r300,
struct r300_vertex_program_code* code)
unsigned instruction_count = code->length / 4;
int vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72;
- int input_count = MAX2(bitcount(code->InputsRead), 1);
- int output_count = MAX2(bitcount(code->OutputsWritten), 1);
+ int input_count = MAX2(util_bitcount(code->InputsRead), 1);
+ int output_count = MAX2(util_bitcount(code->OutputsWritten), 1);
int temp_count = MAX2(code->num_temporaries, 1);
int pvs_num_slots = MIN3(vtx_mem_size / input_count,
vtx_mem_size / output_count, 10);
- int pvs_num_controllers = MIN2(6, vtx_mem_size / temp_count);
+ int pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6);
CS_LOCALS(r300);
END_CS;
}
+void r300_emit_texture_count(struct r300_context* r300)
+{
+ CS_LOCALS(r300);
+
+ BEGIN_CS(2);
+ OUT_CS_REG(R300_TX_ENABLE, (1 << r300->texture_count) - 1);
+ END_CS;
+
+}
+
void r300_flush_textures(struct r300_context* r300)
{
CS_LOCALS(r300);
- BEGIN_CS(4);
+ BEGIN_CS(2);
OUT_CS_REG(R300_TX_INVALTAGS, 0);
- OUT_CS_REG(R300_TX_ENABLE, (1 << r300->texture_count) - 1);
END_CS;
}
return;
}
+ /* Check size of CS. */
+ /* Make sure we have at least 8*1024 spare dwords. */
+ /* XXX It would be nice to know the number of dwords we really need to
+ * XXX emit. */
+ if (!r300->winsys->check_cs(r300->winsys, 8*1024)) {
+ r300->context.flush(&r300->context, 0, NULL);
+ }
+
/* Clean out BOs. */
r300->winsys->reset_bos(r300->winsys);
- /* XXX check size */
validate:
/* Color buffers... */
for (i = 0; i < r300->framebuffer_state.nr_cbufs; i++) {
/* Samplers and textures are tracked separately but emitted together. */
if (r300->dirty_state &
(R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) {
+ r300_emit_texture_count(r300);
+
for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) {
if (r300->dirty_state &
((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) {
projects / mesa.git / blobdiff