/**
* Create a simple vertex shader that passes through position and the given
* attribute.
*/
static void *create_passthrough_vs(struct pipe_context *pipe, int semantic_name)
{
struct ureg_program *ureg;
struct ureg_src src[2], constants[3];
struct ureg_dst dst[2], tmp;
int i;
ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
if (!ureg)
return NULL;
/* position is in user coordinates */
src[0] = ureg_DECL_vs_input(ureg, 0);
dst[0] = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
tmp = ureg_DECL_temporary(ureg);
for (i = 0; i < Elements(constants); i++)
constants[i] = ureg_DECL_constant(ureg, i);
/* transform to clipped coordinates */
ureg_DP4(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_X), src[0], constants[0]);
ureg_DP4(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Y), src[0], constants[1]);
ureg_MOV(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_Z), src[0]);
ureg_DP4(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_W), src[0], constants[2]);
ureg_MOV(ureg, dst[0], ureg_src(tmp));
if (semantic_name >= 0) {
src[1] = ureg_DECL_vs_input(ureg, 1);
dst[1] = ureg_DECL_output(ureg, semantic_name, 0);
ureg_MOV(ureg, dst[1], src[1]);
}
ureg_END(ureg);
return ureg_create_shader_and_destroy(ureg, pipe);
}
/**
* Create a simple vertex shader that just passes through the
* vertex position and texcoord (and optionally, color).
*/
static void *
make_passthrough_vertex_shader(struct st_context *st,
GLboolean passColor)
{
if (!st->drawpix.vert_shaders[passColor]) {
struct ureg_program *ureg = ureg_create( TGSI_PROCESSOR_VERTEX );
if (ureg == NULL)
return NULL;
/* MOV result.pos, vertex.pos; */
ureg_MOV(ureg,
ureg_DECL_output( ureg, TGSI_SEMANTIC_POSITION, 0 ),
ureg_DECL_vs_input( ureg, 0 ));
/* MOV result.texcoord0, vertex.attr[1]; */
ureg_MOV(ureg,
ureg_DECL_output( ureg, TGSI_SEMANTIC_GENERIC, 0 ),
ureg_DECL_vs_input( ureg, 1 ));
if (passColor) {
/* MOV result.color0, vertex.attr[2]; */
ureg_MOV(ureg,
ureg_DECL_output( ureg, TGSI_SEMANTIC_COLOR, 0 ),
ureg_DECL_vs_input( ureg, 2 ));
}
ureg_END( ureg );
st->drawpix.vert_shaders[passColor] =
ureg_create_shader_and_destroy( ureg, st->pipe );
}
return st->drawpix.vert_shaders[passColor];
}
static struct ureg_dst
calc_position(struct vl_mc *r, struct ureg_program *shader, struct ureg_src block_scale)
{
struct ureg_src vrect, vpos;
struct ureg_dst t_vpos;
struct ureg_dst o_vpos;
vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
t_vpos = ureg_DECL_temporary(shader);
o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
/*
* block_scale = (VL_MACROBLOCK_WIDTH, VL_MACROBLOCK_HEIGHT) / (dst.width, dst.height)
*
* t_vpos = (vpos + vrect) * block_scale
* o_vpos.xy = t_vpos
* o_vpos.zw = vpos
*/
ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect);
ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), block_scale);
ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), ureg_imm1f(shader, 1.0f));
return t_vpos;
}
static void *
create_vs(struct pipe_context *pipe, unsigned vs_traits)
{
struct ureg_program *ureg;
struct ureg_src src;
struct ureg_dst dst;
struct ureg_src const0, const1;
boolean is_fill = (vs_traits & VS_FILL) != 0;
boolean is_composite = (vs_traits & VS_COMPOSITE) != 0;
boolean has_mask = (vs_traits & VS_MASK) != 0;
boolean is_yuv = (vs_traits & VS_YUV) != 0;
unsigned input_slot = 0;
ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
if (ureg == NULL)
return 0;
const0 = ureg_DECL_constant(ureg, 0);
const1 = ureg_DECL_constant(ureg, 1);
/* it has to be either a fill or a composite op */
debug_assert((is_fill ^ is_composite) ^ is_yuv);
src = ureg_DECL_vs_input(ureg, input_slot++);
dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
src = vs_normalize_coords(ureg, src, const0, const1);
ureg_MOV(ureg, dst, src);
if (is_yuv) {
src = ureg_DECL_vs_input(ureg, input_slot++);
dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 0);
ureg_MOV(ureg, dst, src);
}
if (is_composite) {
src = ureg_DECL_vs_input(ureg, input_slot++);
dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 0);
ureg_MOV(ureg, dst, src);
}
if (is_fill) {
src = ureg_DECL_vs_input(ureg, input_slot++);
dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
ureg_MOV(ureg, dst, src);
}
if (has_mask) {
src = ureg_DECL_vs_input(ureg, input_slot++);
dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 1);
ureg_MOV(ureg, dst, src);
}
ureg_END(ureg);
return ureg_create_shader_and_destroy(ureg, pipe);
}
static void *
create_ref_vert_shader(struct vl_mc *r)
{
struct ureg_program *shader;
struct ureg_src mv_scale;
struct ureg_src vmv[2];
struct ureg_dst t_vpos;
struct ureg_dst o_vmv[2];
unsigned i;
shader = ureg_create(TGSI_PROCESSOR_VERTEX);
if (!shader)
return NULL;
vmv[0] = ureg_DECL_vs_input(shader, VS_I_MV_TOP);
vmv[1] = ureg_DECL_vs_input(shader, VS_I_MV_BOTTOM);
t_vpos = calc_position(r, shader, ureg_imm2f(shader,
(float)VL_MACROBLOCK_WIDTH / r->buffer_width,
(float)VL_MACROBLOCK_HEIGHT / r->buffer_height)
);
o_vmv[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTOP);
o_vmv[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VBOTTOM);
/*
* mv_scale.xy = 0.5 / (dst.width, dst.height);
* mv_scale.z = 1.0f / 4.0f
* mv_scale.w = 1.0f / 255.0f
*
* // Apply motion vectors
* o_vmv[0..1].xy = vmv[0..1] * mv_scale + t_vpos
* o_vmv[0..1].zw = vmv[0..1] * mv_scale
*
*/
mv_scale = ureg_imm4f(shader,
0.5f / r->buffer_width,
0.5f / r->buffer_height,
1.0f / 4.0f,
1.0f / PIPE_VIDEO_MV_WEIGHT_MAX);
for (i = 0; i < 2; ++i) {
ureg_MAD(shader, ureg_writemask(o_vmv[i], TGSI_WRITEMASK_XY), mv_scale, vmv[i], ureg_src(t_vpos));
ureg_MUL(shader, ureg_writemask(o_vmv[i], TGSI_WRITEMASK_ZW), mv_scale, vmv[i]);
}
ureg_release_temporary(shader, t_vpos);
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, r->pipe);
}
void *
util_make_vertex_passthrough_shader_with_so(struct pipe_context *pipe,
uint num_attribs,
const uint *semantic_names,
const uint *semantic_indexes,
bool window_space,
const struct pipe_stream_output_info *so)
{
struct ureg_program *ureg;
uint i;
ureg = ureg_create( PIPE_SHADER_VERTEX );
if (!ureg)
return NULL;
if (window_space)
ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, TRUE);
for (i = 0; i < num_attribs; i++) {
struct ureg_src src;
struct ureg_dst dst;
src = ureg_DECL_vs_input( ureg, i );
dst = ureg_DECL_output( ureg,
semantic_names[i],
semantic_indexes[i]);
ureg_MOV( ureg, dst, src );
}
ureg_END( ureg );
return ureg_create_shader_with_so_and_destroy( ureg, pipe, so );
}
void *
util_make_vertex_passthrough_shader_with_so(struct pipe_context *pipe,
uint num_attribs,
const uint *semantic_names,
const uint *semantic_indexes,
const struct pipe_stream_output_info *so)
{
struct ureg_program *ureg;
uint i;
ureg = ureg_create( TGSI_PROCESSOR_VERTEX );
if (ureg == NULL)
return NULL;
for (i = 0; i < num_attribs; i++) {
struct ureg_src src;
struct ureg_dst dst;
src = ureg_DECL_vs_input( ureg, i );
dst = ureg_DECL_output( ureg,
semantic_names[i],
semantic_indexes[i]);
ureg_MOV( ureg, dst, src );
}
ureg_END( ureg );
return ureg_create_shader_with_so_and_destroy( ureg, pipe, so );
}
void *
st_pbo_create_vs(struct st_context *st)
{
struct pipe_screen *pscreen = st->pipe->screen;
bool use_nir = PIPE_SHADER_IR_NIR ==
pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX,
PIPE_SHADER_CAP_PREFERRED_IR);
if (use_nir) {
unsigned inputs[] = { VERT_ATTRIB_POS, SYSTEM_VALUE_INSTANCE_ID, };
unsigned outputs[] = { VARYING_SLOT_POS, VARYING_SLOT_LAYER };
return st_nir_make_passthrough_shader(st, "st/pbo VS",
MESA_SHADER_VERTEX,
st->pbo.layers ? 2 : 1,
inputs, outputs, NULL, (1 << 1));
}
struct ureg_program *ureg;
struct ureg_src in_pos;
struct ureg_src in_instanceid;
struct ureg_dst out_pos;
struct ureg_dst out_layer;
ureg = ureg_create(PIPE_SHADER_VERTEX);
if (!ureg)
return NULL;
in_pos = ureg_DECL_vs_input(ureg, TGSI_SEMANTIC_POSITION);
out_pos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
if (st->pbo.layers) {
in_instanceid = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_INSTANCEID, 0);
if (!st->pbo.use_gs)
out_layer = ureg_DECL_output(ureg, TGSI_SEMANTIC_LAYER, 0);
}
/* out_pos = in_pos */
ureg_MOV(ureg, out_pos, in_pos);
if (st->pbo.layers) {
if (st->pbo.use_gs) {
/* out_pos.z = i2f(gl_InstanceID) */
ureg_I2F(ureg, ureg_writemask(out_pos, TGSI_WRITEMASK_Z),
ureg_scalar(in_instanceid, TGSI_SWIZZLE_X));
} else {
/* out_layer = gl_InstanceID */
ureg_MOV(ureg, ureg_writemask(out_layer, TGSI_WRITEMASK_X),
ureg_scalar(in_instanceid, TGSI_SWIZZLE_X));
}
}
ureg_END(ureg);
return ureg_create_shader_and_destroy(ureg, st->pipe);
}
void
vl_idct_stage2_vert_shader(struct vl_idct *idct, struct ureg_program *shader,
unsigned first_output, struct ureg_dst tex)
{
struct ureg_src vrect, vpos;
struct ureg_src scale;
struct ureg_dst t_start;
struct ureg_dst o_l_addr[2], o_r_addr[2];
vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
t_start = ureg_DECL_temporary(shader);
--first_output;
o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_L_ADDR0);
o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_L_ADDR1);
o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_R_ADDR0);
o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_R_ADDR1);
scale = ureg_imm2f(shader,
(float)VL_BLOCK_WIDTH / idct->buffer_width,
(float)VL_BLOCK_HEIGHT / idct->buffer_height);
ureg_MUL(shader, ureg_writemask(tex, TGSI_WRITEMASK_Z),
ureg_scalar(vrect, TGSI_SWIZZLE_X),
ureg_imm1f(shader, VL_BLOCK_WIDTH / idct->nr_of_render_targets));
ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), vpos, scale);
calc_addr(shader, o_l_addr, vrect, ureg_imm1f(shader, 0.0f), false, false, VL_BLOCK_WIDTH / 4);
calc_addr(shader, o_r_addr, ureg_src(tex), ureg_src(t_start), true, false, idct->buffer_height / 4);
ureg_MOV(shader, ureg_writemask(o_r_addr[0], TGSI_WRITEMASK_Z), ureg_src(tex));
ureg_MOV(shader, ureg_writemask(o_r_addr[1], TGSI_WRITEMASK_Z), ureg_src(tex));
}
static void *
create_mismatch_vert_shader(struct vl_idct *idct)
{
struct ureg_program *shader;
struct ureg_src vpos;
struct ureg_src scale;
struct ureg_dst t_tex;
struct ureg_dst o_vpos, o_addr[2];
shader = ureg_create(TGSI_PROCESSOR_VERTEX);
if (!shader)
return NULL;
vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
t_tex = ureg_DECL_temporary(shader);
o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
o_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0);
o_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1);
/*
* scale = (VL_BLOCK_WIDTH, VL_BLOCK_HEIGHT) / (dst.width, dst.height)
*
* t_vpos = vpos + 7 / VL_BLOCK_WIDTH
* o_vpos.xy = t_vpos * scale
*
* o_addr = calc_addr(...)
*
*/
scale = ureg_imm2f(shader,
(float)VL_BLOCK_WIDTH / idct->buffer_width,
(float)VL_BLOCK_HEIGHT / idct->buffer_height);
ureg_MAD(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), vpos, scale, scale);
ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), ureg_imm1f(shader, 1.0f));
ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, scale);
calc_addr(shader, o_addr, ureg_src(t_tex), ureg_src(t_tex), false, false, idct->buffer_width / 4);
ureg_release_temporary(shader, t_tex);
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, idct->pipe);
}
void *
st_pbo_create_vs(struct st_context *st)
{
struct ureg_program *ureg;
struct ureg_src in_pos;
struct ureg_src in_instanceid;
struct ureg_dst out_pos;
struct ureg_dst out_layer;
ureg = ureg_create(PIPE_SHADER_VERTEX);
if (!ureg)
return NULL;
in_pos = ureg_DECL_vs_input(ureg, TGSI_SEMANTIC_POSITION);
out_pos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
if (st->pbo.layers) {
in_instanceid = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_INSTANCEID, 0);
if (!st->pbo.use_gs)
out_layer = ureg_DECL_output(ureg, TGSI_SEMANTIC_LAYER, 0);
}
/* out_pos = in_pos */
ureg_MOV(ureg, out_pos, in_pos);
if (st->pbo.layers) {
if (st->pbo.use_gs) {
/* out_pos.z = i2f(gl_InstanceID) */
ureg_I2F(ureg, ureg_writemask(out_pos, TGSI_WRITEMASK_Z),
ureg_scalar(in_instanceid, TGSI_SWIZZLE_X));
} else {
/* out_layer = gl_InstanceID */
ureg_MOV(ureg, out_layer, in_instanceid);
}
}
ureg_END(ureg);
return ureg_create_shader_and_destroy(ureg, st->pipe);
}
static void *
create_vert_shader(struct vl_deint_filter *filter)
{
struct ureg_program *shader;
struct ureg_src i_vpos;
struct ureg_dst o_vpos, o_vtex;
shader = ureg_create(PIPE_SHADER_VERTEX);
if (!shader)
return NULL;
i_vpos = ureg_DECL_vs_input(shader, 0);
o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
o_vtex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX);
ureg_MOV(shader, o_vpos, i_vpos);
ureg_MOV(shader, o_vtex, i_vpos);
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, filter->pipe);
}
/**
* Translate Mesa program to TGSI format.
* \param program the program to translate
* \param numInputs number of input registers used
* \param inputMapping maps Mesa fragment program inputs to TGSI generic
* input indexes
* \param inputSemanticName the TGSI_SEMANTIC flag for each input
* \param inputSemanticIndex the semantic index (ex: which texcoord) for
* each input
* \param interpMode the TGSI_INTERPOLATE_LINEAR/PERSP mode for each input
* \param numOutputs number of output registers used
* \param outputMapping maps Mesa fragment program outputs to TGSI
* generic outputs
* \param outputSemanticName the TGSI_SEMANTIC flag for each output
* \param outputSemanticIndex the semantic index (ex: which texcoord) for
* each output
*
* \return PIPE_OK or PIPE_ERROR_OUT_OF_MEMORY
*/
enum pipe_error
st_translate_mesa_program(
struct gl_context *ctx,
uint procType,
struct ureg_program *ureg,
const struct gl_program *program,
GLuint numInputs,
const GLuint inputMapping[],
const ubyte inputSemanticName[],
const ubyte inputSemanticIndex[],
const GLuint interpMode[],
GLuint numOutputs,
const GLuint outputMapping[],
const ubyte outputSemanticName[],
const ubyte outputSemanticIndex[],
boolean passthrough_edgeflags,
boolean clamp_color)
{
struct st_translate translate, *t;
unsigned i;
enum pipe_error ret = PIPE_OK;
assert(numInputs <= ARRAY_SIZE(t->inputs));
assert(numOutputs <= ARRAY_SIZE(t->outputs));
t = &translate;
memset(t, 0, sizeof *t);
t->procType = procType;
t->inputMapping = inputMapping;
t->outputMapping = outputMapping;
t->ureg = ureg;
/*_mesa_print_program(program);*/
/*
* Declare input attributes.
*/
if (procType == TGSI_PROCESSOR_FRAGMENT) {
for (i = 0; i < numInputs; i++) {
t->inputs[i] = ureg_DECL_fs_input(ureg,
inputSemanticName[i],
inputSemanticIndex[i],
interpMode[i]);
}
if (program->InputsRead & VARYING_BIT_POS) {
/* Must do this after setting up t->inputs, and before
* emitting constant references, below:
*/
emit_wpos(st_context(ctx), t, program, ureg);
}
if (program->InputsRead & VARYING_BIT_FACE) {
emit_face_var( t, program );
}
/*
* Declare output attributes.
*/
for (i = 0; i < numOutputs; i++) {
switch (outputSemanticName[i]) {
case TGSI_SEMANTIC_POSITION:
t->outputs[i] = ureg_DECL_output( ureg,
TGSI_SEMANTIC_POSITION, /* Z / Depth */
outputSemanticIndex[i] );
t->outputs[i] = ureg_writemask( t->outputs[i],
TGSI_WRITEMASK_Z );
break;
case TGSI_SEMANTIC_STENCIL:
t->outputs[i] = ureg_DECL_output( ureg,
TGSI_SEMANTIC_STENCIL, /* Stencil */
outputSemanticIndex[i] );
t->outputs[i] = ureg_writemask( t->outputs[i],
TGSI_WRITEMASK_Y );
break;
case TGSI_SEMANTIC_COLOR:
t->outputs[i] = ureg_DECL_output( ureg,
TGSI_SEMANTIC_COLOR,
outputSemanticIndex[i] );
break;
default:
debug_assert(0);
return 0;
}
}
}
else if (procType == TGSI_PROCESSOR_GEOMETRY) {
for (i = 0; i < numInputs; i++) {
t->inputs[i] = ureg_DECL_input(ureg,
inputSemanticName[i],
inputSemanticIndex[i], 0, 1);
}
for (i = 0; i < numOutputs; i++) {
t->outputs[i] = ureg_DECL_output( ureg,
outputSemanticName[i],
outputSemanticIndex[i] );
}
}
else {
assert(procType == TGSI_PROCESSOR_VERTEX);
for (i = 0; i < numInputs; i++) {
t->inputs[i] = ureg_DECL_vs_input(ureg, i);
}
for (i = 0; i < numOutputs; i++) {
t->outputs[i] = ureg_DECL_output( ureg,
outputSemanticName[i],
outputSemanticIndex[i] );
if (outputSemanticName[i] == TGSI_SEMANTIC_FOG) {
/* force register to contain a fog coordinate in the form (F, 0, 0, 1). */
ureg_MOV(ureg,
ureg_writemask(t->outputs[i], TGSI_WRITEMASK_YZW),
ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f));
t->outputs[i] = ureg_writemask(t->outputs[i], TGSI_WRITEMASK_X);
}
}
if (passthrough_edgeflags)
emit_edgeflags( t, program );
}
/* Declare address register.
*/
if (program->NumAddressRegs > 0) {
debug_assert( program->NumAddressRegs == 1 );
t->address[0] = ureg_DECL_address( ureg );
}
/* Declare misc input registers
*/
{
GLbitfield sysInputs = program->SystemValuesRead;
unsigned numSys = 0;
for (i = 0; sysInputs; i++) {
if (sysInputs & (1 << i)) {
unsigned semName = _mesa_sysval_to_semantic[i];
t->systemValues[i] = ureg_DECL_system_value(ureg, numSys, semName, 0);
if (semName == TGSI_SEMANTIC_INSTANCEID ||
semName == TGSI_SEMANTIC_VERTEXID) {
/* From Gallium perspective, these system values are always
* integer, and require native integer support. However, if
* native integer is supported on the vertex stage but not the
* pixel stage (e.g, i915g + draw), Mesa will generate IR that
* assumes these system values are floats. To resolve the
* inconsistency, we insert a U2F.
*/
struct st_context *st = st_context(ctx);
struct pipe_screen *pscreen = st->pipe->screen;
assert(procType == TGSI_PROCESSOR_VERTEX);
assert(pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX, PIPE_SHADER_CAP_INTEGERS));
(void) pscreen; /* silence non-debug build warnings */
if (!ctx->Const.NativeIntegers) {
struct ureg_dst temp = ureg_DECL_local_temporary(t->ureg);
ureg_U2F( t->ureg, ureg_writemask(temp, TGSI_WRITEMASK_X), t->systemValues[i]);
t->systemValues[i] = ureg_scalar(ureg_src(temp), 0);
}
}
numSys++;
sysInputs &= ~(1 << i);
}
}
}
if (program->IndirectRegisterFiles & (1 << PROGRAM_TEMPORARY)) {
/* If temps are accessed with indirect addressing, declare temporaries
* in sequential order. Else, we declare them on demand elsewhere.
*/
for (i = 0; i < program->NumTemporaries; i++) {
/* XXX use TGSI_FILE_TEMPORARY_ARRAY when it's supported by ureg */
t->temps[i] = ureg_DECL_temporary( t->ureg );
}
}
/* Emit constants and immediates. Mesa uses a single index space
* for these, so we put all the translated regs in t->constants.
*/
if (program->Parameters) {
t->constants = calloc( program->Parameters->NumParameters,
sizeof t->constants[0] );
if (t->constants == NULL) {
ret = PIPE_ERROR_OUT_OF_MEMORY;
goto out;
}
for (i = 0; i < program->Parameters->NumParameters; i++) {
switch (program->Parameters->Parameters[i].Type) {
case PROGRAM_STATE_VAR:
case PROGRAM_UNIFORM:
t->constants[i] = ureg_DECL_constant( ureg, i );
break;
/* Emit immediates only when there's no indirect addressing of
* the const buffer.
* FIXME: Be smarter and recognize param arrays:
* indirect addressing is only valid within the referenced
* array.
*/
case PROGRAM_CONSTANT:
if (program->IndirectRegisterFiles & PROGRAM_ANY_CONST)
t->constants[i] = ureg_DECL_constant( ureg, i );
else
t->constants[i] =
ureg_DECL_immediate( ureg,
(const float*) program->Parameters->ParameterValues[i],
4 );
break;
default:
break;
}
}
}
/* texture samplers */
for (i = 0; i < ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits; i++) {
if (program->SamplersUsed & (1 << i)) {
t->samplers[i] = ureg_DECL_sampler( ureg, i );
}
}
/* Emit each instruction in turn:
*/
for (i = 0; i < program->NumInstructions; i++) {
set_insn_start( t, ureg_get_instruction_number( ureg ));
compile_instruction( ctx, t, &program->Instructions[i], clamp_color );
}
/* Fix up all emitted labels:
*/
for (i = 0; i < t->labels_count; i++) {
ureg_fixup_label( ureg,
t->labels[i].token,
t->insn[t->labels[i].branch_target] );
}
out:
free(t->insn);
free(t->labels);
free(t->constants);
if (t->error) {
debug_printf("%s: translate error flag set\n", __func__);
}
return ret;
}
void *si_get_blitter_vs(struct si_context *sctx, enum blitter_attrib_type type,
unsigned num_layers)
{
unsigned vs_blit_property;
void **vs;
switch (type) {
case UTIL_BLITTER_ATTRIB_NONE:
vs = num_layers > 1 ? &sctx->vs_blit_pos_layered :
&sctx->vs_blit_pos;
vs_blit_property = SI_VS_BLIT_SGPRS_POS;
break;
case UTIL_BLITTER_ATTRIB_COLOR:
vs = num_layers > 1 ? &sctx->vs_blit_color_layered :
&sctx->vs_blit_color;
vs_blit_property = SI_VS_BLIT_SGPRS_POS_COLOR;
break;
case UTIL_BLITTER_ATTRIB_TEXCOORD_XY:
case UTIL_BLITTER_ATTRIB_TEXCOORD_XYZW:
assert(num_layers == 1);
vs = &sctx->vs_blit_texcoord;
vs_blit_property = SI_VS_BLIT_SGPRS_POS_TEXCOORD;
break;
default:
assert(0);
return NULL;
}
if (*vs)
return *vs;
struct ureg_program *ureg = ureg_create(PIPE_SHADER_VERTEX);
if (!ureg)
return NULL;
/* Tell the shader to load VS inputs from SGPRs: */
ureg_property(ureg, TGSI_PROPERTY_VS_BLIT_SGPRS, vs_blit_property);
ureg_property(ureg, TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION, true);
/* This is just a pass-through shader with 1-3 MOV instructions. */
ureg_MOV(ureg,
ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0),
ureg_DECL_vs_input(ureg, 0));
if (type != UTIL_BLITTER_ATTRIB_NONE) {
ureg_MOV(ureg,
ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 0),
ureg_DECL_vs_input(ureg, 1));
}
if (num_layers > 1) {
struct ureg_src instance_id =
ureg_DECL_system_value(ureg, TGSI_SEMANTIC_INSTANCEID, 0);
struct ureg_dst layer =
ureg_DECL_output(ureg, TGSI_SEMANTIC_LAYER, 0);
ureg_MOV(ureg, ureg_writemask(layer, TGSI_WRITEMASK_X),
ureg_scalar(instance_id, TGSI_SWIZZLE_X));
}
ureg_END(ureg);
*vs = ureg_create_shader_and_destroy(ureg, &sctx->b);
return *vs;
}
static void *
create_vert_shader(struct vl_compositor *c)
{
struct ureg_program *shader;
struct ureg_src vpos, vtex, color;
struct ureg_dst tmp;
struct ureg_dst o_vpos, o_vtex, o_color;
struct ureg_dst o_vtop, o_vbottom;
shader = ureg_create(TGSI_PROCESSOR_VERTEX);
if (!shader)
return false;
vpos = ureg_DECL_vs_input(shader, 0);
vtex = ureg_DECL_vs_input(shader, 1);
color = ureg_DECL_vs_input(shader, 2);
tmp = ureg_DECL_temporary(shader);
o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
o_color = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, VS_O_COLOR);
o_vtex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX);
o_vtop = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTOP);
o_vbottom = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VBOTTOM);
/*
* o_vpos = vpos
* o_vtex = vtex
* o_color = color
*/
ureg_MOV(shader, o_vpos, vpos);
ureg_MOV(shader, o_vtex, vtex);
ureg_MOV(shader, o_color, color);
/*
* tmp.x = vtex.w / 2
* tmp.y = vtex.w / 4
*
* o_vtop.x = vtex.x
* o_vtop.y = vtex.y * tmp.x + 0.25f
* o_vtop.z = vtex.y * tmp.y + 0.25f
* o_vtop.w = 1 / tmp.x
*
* o_vbottom.x = vtex.x
* o_vbottom.y = vtex.y * tmp.x - 0.25f
* o_vbottom.z = vtex.y * tmp.y - 0.25f
* o_vbottom.w = 1 / tmp.y
*/
ureg_MUL(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X),
ureg_scalar(vtex, TGSI_SWIZZLE_W), ureg_imm1f(shader, 0.5f));
ureg_MUL(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y),
ureg_scalar(vtex, TGSI_SWIZZLE_W), ureg_imm1f(shader, 0.25f));
ureg_MOV(shader, ureg_writemask(o_vtop, TGSI_WRITEMASK_X), vtex);
ureg_MAD(shader, ureg_writemask(o_vtop, TGSI_WRITEMASK_Y), ureg_scalar(vtex, TGSI_SWIZZLE_Y),
ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X), ureg_imm1f(shader, 0.25f));
ureg_MAD(shader, ureg_writemask(o_vtop, TGSI_WRITEMASK_Z), ureg_scalar(vtex, TGSI_SWIZZLE_Y),
ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y), ureg_imm1f(shader, 0.25f));
ureg_RCP(shader, ureg_writemask(o_vtop, TGSI_WRITEMASK_W),
ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X));
ureg_MOV(shader, ureg_writemask(o_vbottom, TGSI_WRITEMASK_X), vtex);
ureg_MAD(shader, ureg_writemask(o_vbottom, TGSI_WRITEMASK_Y), ureg_scalar(vtex, TGSI_SWIZZLE_Y),
ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X), ureg_imm1f(shader, -0.25f));
ureg_MAD(shader, ureg_writemask(o_vbottom, TGSI_WRITEMASK_Z), ureg_scalar(vtex, TGSI_SWIZZLE_Y),
ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y), ureg_imm1f(shader, -0.25f));
ureg_RCP(shader, ureg_writemask(o_vbottom, TGSI_WRITEMASK_W),
ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y));
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, c->pipe);
}
static void *
create_vert_shader(struct vl_zscan *zscan)
{
struct ureg_program *shader;
struct ureg_src scale;
struct ureg_src vrect, vpos, block_num;
struct ureg_dst tmp;
struct ureg_dst o_vpos;
struct ureg_dst *o_vtex;
unsigned i;
shader = ureg_create(PIPE_SHADER_VERTEX);
if (!shader)
return NULL;
o_vtex = MALLOC(zscan->num_channels * sizeof(struct ureg_dst));
scale = ureg_imm2f(shader,
(float)VL_BLOCK_WIDTH / zscan->buffer_width,
(float)VL_BLOCK_HEIGHT / zscan->buffer_height);
vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
block_num = ureg_DECL_vs_input(shader, VS_I_BLOCK_NUM);
tmp = ureg_DECL_temporary(shader);
o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
for (i = 0; i < zscan->num_channels; ++i)
o_vtex[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX + i);
/*
* o_vpos.xy = (vpos + vrect) * scale
* o_vpos.zw = 1.0f
*
* tmp.xy = InstanceID / blocks_per_line
* tmp.x = frac(tmp.x)
* tmp.y = floor(tmp.y)
*
* o_vtex.x = vrect.x / blocks_per_line + tmp.x
* o_vtex.y = vrect.y
* o_vtex.z = tmp.z * blocks_per_line / blocks_total
*/
ureg_ADD(shader, ureg_writemask(tmp, TGSI_WRITEMASK_XY), vpos, vrect);
ureg_MUL(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(tmp), scale);
ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), ureg_imm1f(shader, 1.0f));
ureg_MUL(shader, ureg_writemask(tmp, TGSI_WRITEMASK_XW), ureg_scalar(block_num, TGSI_SWIZZLE_X),
ureg_imm1f(shader, 1.0f / zscan->blocks_per_line));
ureg_FRC(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X));
ureg_FLR(shader, ureg_writemask(tmp, TGSI_WRITEMASK_W), ureg_src(tmp));
for (i = 0; i < zscan->num_channels; ++i) {
ureg_ADD(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y),
ureg_imm1f(shader, 1.0f / (zscan->blocks_per_line * VL_BLOCK_WIDTH)
* (i - (signed)zscan->num_channels / 2)));
ureg_MAD(shader, ureg_writemask(o_vtex[i], TGSI_WRITEMASK_X), vrect,
ureg_imm1f(shader, 1.0f / zscan->blocks_per_line), ureg_src(tmp));
ureg_MOV(shader, ureg_writemask(o_vtex[i], TGSI_WRITEMASK_Y), vrect);
ureg_MOV(shader, ureg_writemask(o_vtex[i], TGSI_WRITEMASK_Z), vpos);
ureg_MUL(shader, ureg_writemask(o_vtex[i], TGSI_WRITEMASK_W), ureg_src(tmp),
ureg_imm1f(shader, (float)zscan->blocks_per_line / zscan->blocks_total));
}
ureg_release_temporary(shader, tmp);
ureg_END(shader);
FREE(o_vtex);
return ureg_create_shader_and_destroy(shader, zscan->pipe);
}
/**
* The current vertex shader is already executed by the 'draw'
* module, so we just need to generate a simple vertex shader
* to pass through all those VS outputs that will
* be consumed by the fragment shader.
* Used when we employ the 'draw' module.
*/
static enum pipe_error
compile_passthrough_vs(struct svga_context *svga,
struct svga_vertex_shader *vs,
struct svga_fragment_shader *fs,
struct svga_shader_variant **out_variant)
{
struct svga_shader_variant *variant = NULL;
unsigned num_inputs;
unsigned i;
unsigned num_elements;
struct svga_vertex_shader new_vs;
struct ureg_src src[PIPE_MAX_SHADER_INPUTS];
struct ureg_dst dst[PIPE_MAX_SHADER_OUTPUTS];
struct ureg_program *ureg;
unsigned num_tokens;
struct svga_compile_key key;
enum pipe_error ret;
assert(svga_have_vgpu10(svga));
assert(fs);
num_inputs = fs->base.info.num_inputs;
ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
if (!ureg)
return PIPE_ERROR_OUT_OF_MEMORY;
/* draw will always add position */
dst[0] = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
src[0] = ureg_DECL_vs_input(ureg, 0);
num_elements = 1;
/**
* swtnl backend redefines the input layout based on the
* fragment shader's inputs. So we only need to passthrough
* those inputs that will be consumed by the fragment shader.
* Note: DX10 requires the number of vertex elements
* specified in the input layout to be no less than the
* number of inputs to the vertex shader.
*/
for (i = 0; i < num_inputs; i++) {
switch (fs->base.info.input_semantic_name[i]) {
case TGSI_SEMANTIC_COLOR:
case TGSI_SEMANTIC_GENERIC:
case TGSI_SEMANTIC_FOG:
dst[num_elements] = ureg_DECL_output(ureg,
fs->base.info.input_semantic_name[i],
fs->base.info.input_semantic_index[i]);
src[num_elements] = ureg_DECL_vs_input(ureg, num_elements);
num_elements++;
break;
default:
break;
}
}
for (i = 0; i < num_elements; i++) {
ureg_MOV(ureg, dst[i], src[i]);
}
ureg_END(ureg);
memset(&new_vs, 0, sizeof(new_vs));
new_vs.base.tokens = ureg_get_tokens(ureg, &num_tokens);
tgsi_scan_shader(new_vs.base.tokens, &new_vs.base.info);
memset(&key, 0, sizeof(key));
key.vs.undo_viewport = 1;
ret = compile_vs(svga, &new_vs, &key, &variant);
if (ret != PIPE_OK)
return ret;
ureg_free_tokens(new_vs.base.tokens);
ureg_destroy(ureg);
/* Overwrite the variant key to indicate it's a pass-through VS */
memset(&variant->key, 0, sizeof(variant->key));
variant->key.vs.passthrough = 1;
variant->key.vs.undo_viewport = 1;
*out_variant = variant;
return PIPE_OK;
}
static void *
create_ycbcr_vert_shader(struct vl_mc *r, vl_mc_ycbcr_vert_shader vs_callback, void *callback_priv)
{
struct ureg_program *shader;
struct ureg_src vrect, vpos;
struct ureg_dst t_vpos, t_vtex;
struct ureg_dst o_vpos, o_flags;
struct vertex2f scale = {
(float)VL_BLOCK_WIDTH / r->buffer_width * VL_MACROBLOCK_WIDTH / r->macroblock_size,
(float)VL_BLOCK_HEIGHT / r->buffer_height * VL_MACROBLOCK_HEIGHT / r->macroblock_size
};
unsigned label;
shader = ureg_create(TGSI_PROCESSOR_VERTEX);
if (!shader)
return NULL;
vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
t_vpos = calc_position(r, shader, ureg_imm2f(shader, scale.x, scale.y));
t_vtex = ureg_DECL_temporary(shader);
o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
o_flags = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_FLAGS);
/*
* o_vtex.xy = t_vpos
* o_flags.z = intra * 0.5
*
* if(interlaced) {
* t_vtex.xy = vrect.y ? { 0, scale.y } : { -scale.y : 0 }
* t_vtex.z = vpos.y % 2
* t_vtex.y = t_vtex.z ? t_vtex.x : t_vtex.y
* o_vpos.y = t_vtex.y + t_vpos.y
*
* o_flags.w = t_vtex.z ? 0 : 1
* }
*
*/
vs_callback(callback_priv, r, shader, VS_O_VTEX, t_vpos);
ureg_MUL(shader, ureg_writemask(o_flags, TGSI_WRITEMASK_Z),
ureg_scalar(vpos, TGSI_SWIZZLE_Z), ureg_imm1f(shader, 0.5f));
ureg_MOV(shader, ureg_writemask(o_flags, TGSI_WRITEMASK_W), ureg_imm1f(shader, -1.0f));
if (r->macroblock_size == VL_MACROBLOCK_HEIGHT) { //TODO
ureg_IF(shader, ureg_scalar(vpos, TGSI_SWIZZLE_W), &label);
ureg_CMP(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_XY),
ureg_negate(ureg_scalar(vrect, TGSI_SWIZZLE_Y)),
ureg_imm2f(shader, 0.0f, scale.y),
ureg_imm2f(shader, -scale.y, 0.0f));
ureg_MUL(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_Z),
ureg_scalar(vpos, TGSI_SWIZZLE_Y), ureg_imm1f(shader, 0.5f));
ureg_FRC(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_Z), ureg_src(t_vtex));
ureg_CMP(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_Y),
ureg_negate(ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_Z)),
ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_X),
ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_Y));
ureg_ADD(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_Y),
ureg_src(t_vpos), ureg_src(t_vtex));
ureg_CMP(shader, ureg_writemask(o_flags, TGSI_WRITEMASK_W),
ureg_negate(ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_Z)),
ureg_imm1f(shader, 0.0f), ureg_imm1f(shader, 1.0f));
ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
ureg_ENDIF(shader);
}
ureg_release_temporary(shader, t_vtex);
ureg_release_temporary(shader, t_vpos);
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, r->pipe);
}
