/**
* Declare a VS input register.
* We still make up the input semantics the same as in 2.0
*/
static boolean
vs30_input(struct svga_shader_emitter *emit,
struct tgsi_declaration_semantic semantic,
unsigned idx)
{
SVGA3DOpDclArgs dcl;
SVGA3dShaderInstToken opcode;
unsigned usage, index;
opcode = inst_token( SVGA3DOP_DCL );
dcl.values[0] = 0;
dcl.values[1] = 0;
emit->input_map[idx] = src_register( SVGA3DREG_INPUT, idx );
dcl.dst = dst_register( SVGA3DREG_INPUT, idx );
assert(dcl.dst.reserved0);
svga_generate_vdecl_semantics( idx, &usage, &index );
dcl.usage = usage;
dcl.index = index;
dcl.values[0] |= 1<<31;
return (emit_instruction(emit, opcode) &&
svga_shader_emit_dwords( emit, dcl.values, Elements(dcl.values)));
}
/**
* Declare a VS input register.
* We still make up the input semantics the same as in 2.0
*/
static boolean
vs30_input(struct svga_shader_emitter *emit,
struct tgsi_declaration_semantic semantic,
unsigned idx)
{
SVGA3DOpDclArgs dcl;
SVGA3dShaderInstToken opcode;
unsigned usage, index;
opcode = inst_token( SVGA3DOP_DCL );
dcl.values[0] = 0;
dcl.values[1] = 0;
if (emit->key.vkey.zero_stride_vertex_elements & (1 << idx)) {
unsigned i;
unsigned offset = 0;
unsigned start_idx = emit->info.file_max[TGSI_FILE_CONSTANT] + 1;
/* adjust for prescale constants */
start_idx += emit->key.vkey.need_prescale ? 2 : 0;
/* compute the offset from the start of zero stride constants */
for (i = 0; i < PIPE_MAX_ATTRIBS && i < idx; ++i) {
if (emit->key.vkey.zero_stride_vertex_elements & (1<<i))
++offset;
}
emit->input_map[idx] = src_register( SVGA3DREG_CONST,
start_idx + offset );
} else {
emit->input_map[idx] = src_register( SVGA3DREG_INPUT, idx );
dcl.dst = dst_register( SVGA3DREG_INPUT, idx );
assert(dcl.dst.reserved0);
svga_generate_vdecl_semantics( idx, &usage, &index );
dcl.usage = usage;
dcl.index = index;
dcl.values[0] |= 1<<31;
return (emit_instruction(emit, opcode) &&
svga_shader_emit_dwords( emit, dcl.values, Elements(dcl.values)));
}
return TRUE;
}
static int emit_hw_vs_vdecl( struct svga_context *svga,
unsigned dirty )
{
const struct pipe_vertex_element *ve = svga->curr.ve;
SVGA3dVertexDecl decl;
unsigned i;
assert(svga->curr.num_vertex_elements >=
svga->curr.vs->base.info.file_count[TGSI_FILE_INPUT]);
svga_hwtnl_reset_vdecl( svga->hwtnl,
svga->curr.num_vertex_elements );
for (i = 0; i < svga->curr.num_vertex_elements; i++) {
const struct pipe_vertex_buffer *vb = &svga->curr.vb[ve[i].vertex_buffer_index];
unsigned usage, index;
svga_generate_vdecl_semantics( i, &usage, &index );
/* SVGA_NEW_VELEMENT
*/
decl.identity.type = svga->state.sw.ve_format[i];
decl.identity.method = SVGA3D_DECLMETHOD_DEFAULT;
decl.identity.usage = usage;
decl.identity.usageIndex = index;
decl.array.stride = vb->stride;
decl.array.offset = (vb->buffer_offset +
ve[i].src_offset);
svga_hwtnl_vdecl( svga->hwtnl,
i,
&decl,
vb->buffer );
}
return 0;
}
static enum pipe_error
emit_hw_vs_vdecl(struct svga_context *svga, unsigned dirty)
{
const struct pipe_vertex_element *ve = svga->curr.velems->velem;
SVGA3dVertexDecl decls[SVGA3D_INPUTREG_MAX];
unsigned buffer_indexes[SVGA3D_INPUTREG_MAX];
unsigned i;
unsigned neg_bias = 0;
assert(svga->curr.velems->count >=
svga->curr.vs->base.info.file_count[TGSI_FILE_INPUT]);
/**
* We can't set the VDECL offset to something negative, so we
* must calculate a common negative additional index bias, and modify
* the VDECL offsets accordingly so they *all* end up positive.
*
* Note that the exact value of the negative index bias is not that
* important, since we compensate for it when we calculate the vertex
* buffer offset below. The important thing is that all vertex buffer
* offsets remain positive.
*
* Note that we use a negative bias variable in order to make the
* rounding maths more easy to follow, and to avoid int / unsigned
* confusion.
*/
for (i = 0; i < svga->curr.velems->count; i++) {
const struct pipe_vertex_buffer *vb =
&svga->curr.vb[ve[i].vertex_buffer_index];
struct svga_buffer *buffer;
unsigned int offset = vb->buffer_offset + ve[i].src_offset;
unsigned tmp_neg_bias = 0;
if (!vb->buffer)
continue;
buffer = svga_buffer(vb->buffer);
if (buffer->uploaded.start > offset) {
tmp_neg_bias = buffer->uploaded.start - offset;
if (vb->stride)
tmp_neg_bias = (tmp_neg_bias + vb->stride - 1) / vb->stride;
neg_bias = MAX2(neg_bias, tmp_neg_bias);
}
}
for (i = 0; i < svga->curr.velems->count; i++) {
const struct pipe_vertex_buffer *vb =
&svga->curr.vb[ve[i].vertex_buffer_index];
unsigned usage, index;
struct svga_buffer *buffer;
if (!vb->buffer)
continue;
buffer = svga_buffer(vb->buffer);
svga_generate_vdecl_semantics( i, &usage, &index );
/* SVGA_NEW_VELEMENT
*/
decls[i].identity.type = svga->curr.velems->decl_type[i];
decls[i].identity.method = SVGA3D_DECLMETHOD_DEFAULT;
decls[i].identity.usage = usage;
decls[i].identity.usageIndex = index;
decls[i].array.stride = vb->stride;
/* Compensate for partially uploaded vbo, and
* for the negative index bias.
*/
decls[i].array.offset = (vb->buffer_offset
+ ve[i].src_offset
+ neg_bias * vb->stride
- buffer->uploaded.start);
assert(decls[i].array.offset >= 0);
buffer_indexes[i] = ve[i].vertex_buffer_index;
assert(!buffer->uploaded.buffer);
}
svga_hwtnl_vertex_decls(svga->hwtnl,
svga->curr.velems->count,
decls,
buffer_indexes,
svga->curr.velems->id);
svga_hwtnl_vertex_buffers(svga->hwtnl,
svga->curr.num_vertex_buffers,
svga->curr.vb);
svga_hwtnl_set_index_bias( svga->hwtnl, -(int) neg_bias );
return PIPE_OK;
}
