void
vec4_vs_visitor::emit_thread_end()
{
setup_uniform_clipplane_values();
/* Lower legacy ff and ClipVertex clipping to clip distances */
if (key->nr_userclip_plane_consts > 0) {
current_annotation = "user clip distances";
output_reg[VARYING_SLOT_CLIP_DIST0][0] =
dst_reg(this, glsl_type::vec4_type);
output_reg[VARYING_SLOT_CLIP_DIST1][0] =
dst_reg(this, glsl_type::vec4_type);
output_num_components[VARYING_SLOT_CLIP_DIST0][0] = 4;
output_num_components[VARYING_SLOT_CLIP_DIST1][0] = 4;
emit_clip_distances(output_reg[VARYING_SLOT_CLIP_DIST0][0], 0);
emit_clip_distances(output_reg[VARYING_SLOT_CLIP_DIST1][0], 4);
}
/* For VS, we always end the thread by emitting a single vertex.
* emit_urb_write_opcode() will take care of setting the eot flag on the
* SEND instruction.
*/
emit_vertex();
}
static void
emit_elt32(struct push_context *ctx, unsigned start, unsigned count)
{
uint32_t *idxbuf = ctx->idxbuf;
while (count--)
emit_vertex(ctx, idxbuf[start++]);
}
static void
emit_elt32_biased(void *priv, unsigned start, unsigned count)
{
struct push_context *ctx = priv;
uint32_t *idxbuf = ctx->idxbuf;
while (count--)
emit_vertex(ctx, idxbuf[start++] + ctx->idxbias);
}
void
vec4_vs_visitor::emit_thread_end()
{
/* For VS, we always end the thread by emitting a single vertex.
* emit_urb_write_opcode() will take care of setting the eot flag on the
* SEND instruction.
*/
emit_vertex();
}
static void
vbuf_point( struct draw_stage *stage,
struct prim_header *prim )
{
struct vbuf_stage *vbuf = vbuf_stage( stage );
check_space( vbuf, 1 );
vbuf->indices[vbuf->nr_indices++] = emit_vertex( vbuf, prim->v[0] );
}
static void
vbuf_line( struct draw_stage *stage,
struct prim_header *prim )
{
struct vbuf_stage *vbuf = vbuf_stage( stage );
unsigned i;
check_space( vbuf, 2 );
for (i = 0; i < 2; i++) {
vbuf->indices[vbuf->nr_indices++] = emit_vertex( vbuf, prim->v[i] );
}
}
/**
* Emit GLX DrawArrays protocol using a GLXRender packet.
*/
static void
emit_Render_DrawArrays(__GLXcontext * gc, const struct array_info * arrays,
GLsizei first, GLsizei count, GLsizei num_arrays, GLenum mode,
GLsizei cmdlen, GLsizei total_vertex_size)
{
GLubyte * pc = gc->pc;
GLsizei offset;
GLsizei i;
__GLX_BEGIN_VARIABLE(X_GLrop_DrawArrays, cmdlen);
emit_header(pc + 4, arrays, num_arrays, count, mode);
/* Write the actual array data.
*/
offset = __GLX_DRAWARRAYS_CMD_HDR_SIZE
+ (num_arrays * __GLX_COMPONENT_HDR_SIZE);
for ( i = 0 ; i < count ; i++ ) {
offset = emit_vertex(pc, arrays, num_arrays, i + first, offset);
}
__GLX_END(cmdlen);
}
void gen2_render_copyfunc(struct intel_batchbuffer *batch,
drm_intel_context *context,
struct igt_buf *src, unsigned src_x, unsigned src_y,
unsigned width, unsigned height,
struct igt_buf *dst, unsigned dst_x, unsigned dst_y)
{
gen2_emit_invariant(batch);
gen2_emit_copy_pipeline(batch);
gen2_emit_target(batch, dst);
gen2_emit_texture(batch, src, 0);
OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 |
I1_LOAD_S(2) | I1_LOAD_S(3) | I1_LOAD_S(8) | 2);
OUT_BATCH(1<<12);
OUT_BATCH(S3_CULLMODE_NONE | S3_VERTEXHAS_XY);
OUT_BATCH(S8_ENABLE_COLOR_BUFFER_WRITE);
OUT_BATCH(_3DSTATE_VERTEX_FORMAT_2_CMD | TEXCOORDFMT_2D << 0);
OUT_BATCH(PRIM3D_INLINE | PRIM3D_RECTLIST | (3*4 -1));
emit_vertex(batch, dst_x + width);
emit_vertex(batch, dst_y + height);
emit_vertex_normalized(batch, src_x + width, igt_buf_width(src));
emit_vertex_normalized(batch, src_y + height, igt_buf_height(src));
emit_vertex(batch, dst_x);
emit_vertex(batch, dst_y + height);
emit_vertex_normalized(batch, src_x, igt_buf_width(src));
emit_vertex_normalized(batch, src_y + height, igt_buf_height(src));
emit_vertex(batch, dst_x);
emit_vertex(batch, dst_y);
emit_vertex_normalized(batch, src_x, igt_buf_width(src));
emit_vertex_normalized(batch, src_y, igt_buf_height(src));
intel_batchbuffer_flush(batch);
}
static void
emit_seq(struct push_context *ctx, unsigned start, unsigned count)
{
while (count--)
emit_vertex(ctx, start++);
}
static void
emit_verts(void *priv, unsigned start, unsigned count)
{
while (count--)
emit_vertex(priv, start++);
}
void
vec4_gs_visitor::visit(ir_emit_vertex *ir)
{
this->current_annotation = "emit vertex: safety check";
/* To ensure that we don't output more vertices than the shader specified
* using max_vertices, do the logic inside a conditional of the form "if
* (vertex_count < MAX)"
*/
unsigned num_output_vertices = c->gp->program.VerticesOut;
emit(CMP(dst_null_d(), this->vertex_count,
src_reg(num_output_vertices), BRW_CONDITIONAL_L));
emit(IF(BRW_PREDICATE_NORMAL));
{
/* If we're outputting 32 control data bits or less, then we can wait
* until the shader is over to output them all. Otherwise we need to
* output them as we go. Now is the time to do it, since we're about to
* output the vertex_count'th vertex, so it's guaranteed that the
* control data bits associated with the (vertex_count - 1)th vertex are
* correct.
*/
if (c->control_data_header_size_bits > 32) {
this->current_annotation = "emit vertex: emit control data bits";
/* Only emit control data bits if we've finished accumulating a batch
* of 32 bits. This is the case when:
*
* (vertex_count * bits_per_vertex) % 32 == 0
*
* (in other words, when the last 5 bits of vertex_count *
* bits_per_vertex are 0). Assuming bits_per_vertex == 2^n for some
* integer n (which is always the case, since bits_per_vertex is
* always 1 or 2), this is equivalent to requiring that the last 5-n
* bits of vertex_count are 0:
*
* vertex_count & (2^(5-n) - 1) == 0
*
* 2^(5-n) == 2^5 / 2^n == 32 / bits_per_vertex, so this is
* equivalent to:
*
* vertex_count & (32 / bits_per_vertex - 1) == 0
*/
vec4_instruction *inst =
emit(AND(dst_null_d(), this->vertex_count,
(uint32_t) (32 / c->control_data_bits_per_vertex - 1)));
inst->conditional_mod = BRW_CONDITIONAL_Z;
emit(IF(BRW_PREDICATE_NORMAL));
{
emit_control_data_bits();
/* Reset control_data_bits to 0 so we can start accumulating a new
* batch.
*
* Note: in the case where vertex_count == 0, this neutralizes the
* effect of any call to EndPrimitive() that the shader may have
* made before outputting its first vertex.
*/
inst = emit(MOV(dst_reg(this->control_data_bits), 0u));
inst->force_writemask_all = true;
}
emit(BRW_OPCODE_ENDIF);
}
this->current_annotation = "emit vertex: vertex data";
emit_vertex();
/* In stream mode we have to set control data bits for all vertices
* unless we have disabled control data bits completely (which we do
* do for GL_POINTS outputs that don't use streams).
*/
if (c->control_data_header_size_bits > 0 &&
c->prog_data.control_data_format ==
GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_SID) {
this->current_annotation = "emit vertex: Stream control data bits";
set_stream_control_data_bits(ir->stream_id());
}
this->current_annotation = "emit vertex: increment vertex count";
emit(ADD(dst_reg(this->vertex_count), this->vertex_count,
src_reg(1u)));
}
emit(BRW_OPCODE_ENDIF);
this->current_annotation = NULL;
}
/**
* Emit GLX DrawArrays protocol using a GLXRenderLarge packet.
*/
static void
emit_RenderLarge_DrawArrays(__GLXcontext * gc, const struct array_info * arrays,
GLsizei first, GLsizei count, GLsizei num_arrays, GLenum mode,
GLsizei cmdlen, GLsizei total_vertex_size)
{
GLubyte * pc = gc->pc;
GLsizei offset;
GLsizei i;
GLint maxSize;
GLint totalRequests;
GLint requestNumber;
GLsizei elements_per_request;
/* Calculate the maximum amount of data can be stuffed into a single
* packet. sz_xGLXRenderReq is added because bufSize is the maximum
* packet size minus sz_xGLXRenderReq.
*
* The important value here is elements_per_request. This is the number
* of complete array elements that will fit in a single buffer. There
* may be some wasted space at the end of the buffer, but splitting
* elements across buffer boundries would be painful.
*/
maxSize = (gc->bufSize + sz_xGLXRenderReq) - sz_xGLXRenderLargeReq;
elements_per_request = maxSize / total_vertex_size;
totalRequests = ((count + (elements_per_request - 1))
/ elements_per_request) + 1;
/* Fill in the header data and send it away.
*/
__GLX_BEGIN_VARIABLE_LARGE(X_GLrop_DrawArrays, cmdlen+4);
emit_header(pc + 8, arrays, num_arrays, count, mode);
gc->pc = pc + (__GLX_DRAWARRAYS_CMD_HDR_SIZE + 4)
+ (__GLX_COMPONENT_HDR_SIZE * num_arrays);
__glXSendLargeChunk(gc, 1, totalRequests, gc->buf, gc->pc - gc->buf);
/* Write the actual array data.
*/
offset = 0;
requestNumber = 2;
for ( i = 0 ; i < count ; i++ ) {
if ( i == elements_per_request ) {
__glXSendLargeChunk(gc, requestNumber, totalRequests,
gc->buf, offset);
requestNumber++;
offset = 0;
count -= i;
first += i;
i = 0;
}
offset = emit_vertex(gc->buf, arrays, num_arrays, i + first, offset);
}
/* If the buffer isn't empty, emit the last, partial request.
*/
if ( offset != 0 ) {
assert(requestNumber == totalRequests);
__glXSendLargeChunk(gc, requestNumber, totalRequests, gc->buf, offset);
}
gc->pc = gc->buf;
}
void
vec4_gs_visitor::gs_emit_vertex(int stream_id)
{
this->current_annotation = "emit vertex: safety check";
/* Haswell and later hardware ignores the "Render Stream Select" bits
* from the 3DSTATE_STREAMOUT packet when the SOL stage is disabled,
* and instead sends all primitives down the pipeline for rasterization.
* If the SOL stage is enabled, "Render Stream Select" is honored and
* primitives bound to non-zero streams are discarded after stream output.
*
* Since the only purpose of primives sent to non-zero streams is to
* be recorded by transform feedback, we can simply discard all geometry
* bound to these streams when transform feedback is disabled.
*/
if (stream_id > 0 && !nir->info->has_transform_feedback_varyings)
return;
/* If we're outputting 32 control data bits or less, then we can wait
* until the shader is over to output them all. Otherwise we need to
* output them as we go. Now is the time to do it, since we're about to
* output the vertex_count'th vertex, so it's guaranteed that the
* control data bits associated with the (vertex_count - 1)th vertex are
* correct.
*/
if (c->control_data_header_size_bits > 32) {
this->current_annotation = "emit vertex: emit control data bits";
/* Only emit control data bits if we've finished accumulating a batch
* of 32 bits. This is the case when:
*
* (vertex_count * bits_per_vertex) % 32 == 0
*
* (in other words, when the last 5 bits of vertex_count *
* bits_per_vertex are 0). Assuming bits_per_vertex == 2^n for some
* integer n (which is always the case, since bits_per_vertex is
* always 1 or 2), this is equivalent to requiring that the last 5-n
* bits of vertex_count are 0:
*
* vertex_count & (2^(5-n) - 1) == 0
*
* 2^(5-n) == 2^5 / 2^n == 32 / bits_per_vertex, so this is
* equivalent to:
*
* vertex_count & (32 / bits_per_vertex - 1) == 0
*/
vec4_instruction *inst =
emit(AND(dst_null_ud(), this->vertex_count,
brw_imm_ud(32 / c->control_data_bits_per_vertex - 1)));
inst->conditional_mod = BRW_CONDITIONAL_Z;
emit(IF(BRW_PREDICATE_NORMAL));
{
/* If vertex_count is 0, then no control data bits have been
* accumulated yet, so we skip emitting them.
*/
emit(CMP(dst_null_ud(), this->vertex_count, brw_imm_ud(0u),
BRW_CONDITIONAL_NEQ));
emit(IF(BRW_PREDICATE_NORMAL));
emit_control_data_bits();
emit(BRW_OPCODE_ENDIF);
/* Reset control_data_bits to 0 so we can start accumulating a new
* batch.
*
* Note: in the case where vertex_count == 0, this neutralizes the
* effect of any call to EndPrimitive() that the shader may have
* made before outputting its first vertex.
*/
inst = emit(MOV(dst_reg(this->control_data_bits), brw_imm_ud(0u)));
inst->force_writemask_all = true;
}
emit(BRW_OPCODE_ENDIF);
}
this->current_annotation = "emit vertex: vertex data";
emit_vertex();
/* In stream mode we have to set control data bits for all vertices
* unless we have disabled control data bits completely (which we do
* do for GL_POINTS outputs that don't use streams).
*/
if (c->control_data_header_size_bits > 0 &&
gs_prog_data->control_data_format ==
GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_SID) {
this->current_annotation = "emit vertex: Stream control data bits";
set_stream_control_data_bits(stream_id);
}
this->current_annotation = NULL;
}
