/**
* Emit the TGSI instructions for inverting and adjusting WPOS.
* This code is unavoidable because it also depends on whether
* a FBO is bound (STATE_FB_WPOS_Y_TRANSFORM).
*/
static void
emit_wpos_adjustment(struct gl_context *ctx,
struct st_translate *t,
const struct gl_program *program,
boolean invert,
GLfloat adjX, GLfloat adjY[2])
{
struct ureg_program *ureg = t->ureg;
/* Fragment program uses fragment position input.
* Need to replace instances of INPUT[WPOS] with temp T
* where T = INPUT[WPOS] by y is inverted.
*/
static const gl_state_index wposTransformState[STATE_LENGTH]
= { STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM, 0, 0, 0 };
/* XXX: note we are modifying the incoming shader here! Need to
* do this before emitting the constant decls below, or this
* will be missed:
*/
unsigned wposTransConst = _mesa_add_state_reference(program->Parameters,
wposTransformState);
struct ureg_src wpostrans = ureg_DECL_constant( ureg, wposTransConst );
struct ureg_dst wpos_temp = ureg_DECL_temporary( ureg );
struct ureg_src *wpos =
ctx->Const.GLSLFragCoordIsSysVal ?
&t->systemValues[SYSTEM_VALUE_FRAG_COORD] :
&t->inputs[t->inputMapping[VARYING_SLOT_POS]];
struct ureg_src wpos_input = *wpos;
/* First, apply the coordinate shift: */
if (adjX || adjY[0] || adjY[1]) {
if (adjY[0] != adjY[1]) {
/* Adjust the y coordinate by adjY[1] or adjY[0] respectively
* depending on whether inversion is actually going to be applied
* or not, which is determined by testing against the inversion
* state variable used below, which will be either +1 or -1.
*/
struct ureg_dst adj_temp = ureg_DECL_temporary(ureg);
ureg_CMP(ureg, adj_temp,
ureg_scalar(wpostrans, invert ? 2 : 0),
ureg_imm4f(ureg, adjX, adjY[0], 0.0f, 0.0f),
ureg_imm4f(ureg, adjX, adjY[1], 0.0f, 0.0f));
ureg_ADD(ureg, wpos_temp, wpos_input, ureg_src(adj_temp));
} else {
ureg_ADD(ureg, wpos_temp, wpos_input,
ureg_imm4f(ureg, adjX, adjY[0], 0.0f, 0.0f));
}
wpos_input = ureg_src(wpos_temp);
} else {
/* MOV wpos_temp, input[wpos]
*/
ureg_MOV( ureg, wpos_temp, wpos_input );
}
/* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be
* inversion/identity, or the other way around if we're drawing to an FBO.
*/
if (invert) {
/* MAD wpos_temp.y, wpos_input, wpostrans.xxxx, wpostrans.yyyy
*/
ureg_MAD( ureg,
ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ),
wpos_input,
ureg_scalar(wpostrans, 0),
ureg_scalar(wpostrans, 1));
} else {
/* MAD wpos_temp.y, wpos_input, wpostrans.zzzz, wpostrans.wwww
*/
ureg_MAD( ureg,
ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ),
wpos_input,
ureg_scalar(wpostrans, 2),
ureg_scalar(wpostrans, 3));
}
/* Use wpos_temp as position input from here on:
*/
*wpos = ureg_src(wpos_temp);
}
/**
* 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[])
{
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);
}
/*
* 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);
}
}
}
/* 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;
for (i = 0; sysInputs; i++) {
if (sysInputs & (1 << i)) {
unsigned semName = _mesa_sysval_to_semantic[i];
t->systemValues[i] = ureg_DECL_system_value(ureg, 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);
}
}
if (procType == TGSI_PROCESSOR_FRAGMENT &&
semName == TGSI_SEMANTIC_POSITION)
emit_wpos(st_context(ctx), t, program, ureg);
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]);
}
/* 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;
}
static void *
create_deint_frag_shader(struct vl_deint_filter *filter, unsigned field,
struct vertex2f *sizes, bool spatial_filter)
{
struct ureg_program *shader;
struct ureg_src i_vtex;
struct ureg_src sampler_cur;
struct ureg_src sampler_prevprev;
struct ureg_src sampler_prev;
struct ureg_src sampler_next;
struct ureg_dst o_fragment;
struct ureg_dst t_tex;
struct ureg_dst t_comp_top, t_comp_bot;
struct ureg_dst t_diff;
struct ureg_dst t_a, t_b;
struct ureg_dst t_weave, t_linear;
shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
if (!shader) {
return NULL;
}
t_tex = ureg_DECL_temporary(shader);
t_comp_top = ureg_DECL_temporary(shader);
t_comp_bot = ureg_DECL_temporary(shader);
t_diff = ureg_DECL_temporary(shader);
t_a = ureg_DECL_temporary(shader);
t_b = ureg_DECL_temporary(shader);
t_weave = ureg_DECL_temporary(shader);
t_linear = ureg_DECL_temporary(shader);
i_vtex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX, TGSI_INTERPOLATE_LINEAR);
sampler_prevprev = ureg_DECL_sampler(shader, 0);
sampler_prev = ureg_DECL_sampler(shader, 1);
sampler_cur = ureg_DECL_sampler(shader, 2);
sampler_next = ureg_DECL_sampler(shader, 3);
o_fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
// we don't care about ZW interpolation (allows better optimization)
ureg_MOV(shader, t_tex, i_vtex);
ureg_MOV(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_ZW),
ureg_imm1f(shader, 0));
// sample between texels for cheap lowpass
ureg_ADD(shader, t_comp_top, ureg_src(t_tex),
ureg_imm4f(shader, sizes->x * 0.5f, sizes->y * -0.5f, 0, 0));
ureg_ADD(shader, t_comp_bot, ureg_src(t_tex),
ureg_imm4f(shader, sizes->x * -0.5f, sizes->y * 0.5f, 1.0f, 0));
if (field == 0) {
/* interpolating top field -> current field is a bottom field */
// cur vs prev2
ureg_TEX(shader, t_a, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_bot), sampler_cur);
ureg_TEX(shader, t_b, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_bot), sampler_prevprev);
ureg_SUB(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_X), ureg_src(t_a), ureg_src(t_b));
// prev vs next
ureg_TEX(shader, t_a, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_top), sampler_prev);
ureg_TEX(shader, t_b, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_top), sampler_next);
ureg_SUB(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_Y), ureg_src(t_a), ureg_src(t_b));
} else {
/* interpolating bottom field -> current field is a top field */
// cur vs prev2
ureg_TEX(shader, t_a, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_top), sampler_cur);
ureg_TEX(shader, t_b, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_top), sampler_prevprev);
ureg_SUB(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_X), ureg_src(t_a), ureg_src(t_b));
// prev vs next
ureg_TEX(shader, t_a, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_bot), sampler_prev);
ureg_TEX(shader, t_b, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_bot), sampler_next);
ureg_SUB(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_Y), ureg_src(t_a), ureg_src(t_b));
}
// absolute maximum of differences
ureg_MAX(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_X), ureg_abs(ureg_src(t_diff)),
ureg_scalar(ureg_abs(ureg_src(t_diff)), TGSI_SWIZZLE_Y));
if (field == 0) {
/* weave with prev top field */
ureg_TEX(shader, t_weave, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_tex), sampler_prev);
/* get linear interpolation from current bottom field */
ureg_ADD(shader, t_comp_top, ureg_src(t_tex), ureg_imm4f(shader, 0, sizes->y * -1.0f, 1.0f, 0));
ureg_TEX(shader, t_linear, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_top), sampler_cur);
} else {
/* weave with prev bottom field */
ureg_ADD(shader, t_comp_bot, ureg_src(t_tex), ureg_imm4f(shader, 0, 0, 1.0f, 0));
ureg_TEX(shader, t_weave, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_bot), sampler_prev);
/* get linear interpolation from current top field */
ureg_ADD(shader, t_comp_bot, ureg_src(t_tex), ureg_imm4f(shader, 0, sizes->y * 1.0f, 0, 0));
ureg_TEX(shader, t_linear, TGSI_TEXTURE_2D_ARRAY, ureg_src(t_comp_bot), sampler_cur);
}
// mix between weave and linear
// fully weave if diff < 6 (0.02353), fully interpolate if diff > 14 (0.05490)
ureg_ADD(shader, ureg_writemask(t_diff, TGSI_WRITEMASK_X), ureg_src(t_diff),
ureg_imm4f(shader, -0.02353f, 0, 0, 0));
ureg_MUL(shader, ureg_saturate(ureg_writemask(t_diff, TGSI_WRITEMASK_X)),
ureg_src(t_diff), ureg_imm4f(shader, 31.8750f, 0, 0, 0));
ureg_LRP(shader, ureg_writemask(o_fragment, TGSI_WRITEMASK_X), ureg_src(t_diff),
ureg_src(t_linear), ureg_src(t_weave));
ureg_release_temporary(shader, t_tex);
ureg_release_temporary(shader, t_comp_top);
ureg_release_temporary(shader, t_comp_bot);
ureg_release_temporary(shader, t_diff);
ureg_release_temporary(shader, t_a);
ureg_release_temporary(shader, t_b);
ureg_release_temporary(shader, t_weave);
ureg_release_temporary(shader, t_linear);
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, filter->pipe);
}
/* Create a compute shader implementing clear_buffer or copy_buffer. */
void *si_create_dma_compute_shader(struct pipe_context *ctx,
unsigned num_dwords_per_thread,
bool dst_stream_cache_policy, bool is_copy)
{
assert(util_is_power_of_two_nonzero(num_dwords_per_thread));
unsigned store_qualifier = TGSI_MEMORY_COHERENT | TGSI_MEMORY_RESTRICT;
if (dst_stream_cache_policy)
store_qualifier |= TGSI_MEMORY_STREAM_CACHE_POLICY;
/* Don't cache loads, because there is no reuse. */
unsigned load_qualifier = store_qualifier | TGSI_MEMORY_STREAM_CACHE_POLICY;
unsigned num_mem_ops = MAX2(1, num_dwords_per_thread / 4);
unsigned *inst_dwords = alloca(num_mem_ops * sizeof(unsigned));
for (unsigned i = 0; i < num_mem_ops; i++) {
if (i*4 < num_dwords_per_thread)
inst_dwords[i] = MIN2(4, num_dwords_per_thread - i*4);
}
struct ureg_program *ureg = ureg_create(PIPE_SHADER_COMPUTE);
if (!ureg)
return NULL;
ureg_property(ureg, TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH, 64);
ureg_property(ureg, TGSI_PROPERTY_CS_FIXED_BLOCK_HEIGHT, 1);
ureg_property(ureg, TGSI_PROPERTY_CS_FIXED_BLOCK_DEPTH, 1);
struct ureg_src value;
if (!is_copy) {
ureg_property(ureg, TGSI_PROPERTY_CS_USER_DATA_DWORDS, inst_dwords[0]);
value = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_CS_USER_DATA, 0);
}
struct ureg_src tid = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_THREAD_ID, 0);
struct ureg_src blk = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_BLOCK_ID, 0);
struct ureg_dst store_addr = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_X);
struct ureg_dst load_addr = ureg_writemask(ureg_DECL_temporary(ureg), TGSI_WRITEMASK_X);
struct ureg_dst dstbuf = ureg_dst(ureg_DECL_buffer(ureg, 0, false));
struct ureg_src srcbuf;
struct ureg_src *values = NULL;
if (is_copy) {
srcbuf = ureg_DECL_buffer(ureg, 1, false);
values = malloc(num_mem_ops * sizeof(struct ureg_src));
}
/* If there are multiple stores, the first store writes into 0+tid,
* the 2nd store writes into 64+tid, the 3rd store writes into 128+tid, etc.
*/
ureg_UMAD(ureg, store_addr, blk, ureg_imm1u(ureg, 64 * num_mem_ops), tid);
/* Convert from a "store size unit" into bytes. */
ureg_UMUL(ureg, store_addr, ureg_src(store_addr),
ureg_imm1u(ureg, 4 * inst_dwords[0]));
ureg_MOV(ureg, load_addr, ureg_src(store_addr));
/* Distance between a load and a store for latency hiding. */
unsigned load_store_distance = is_copy ? 8 : 0;
for (unsigned i = 0; i < num_mem_ops + load_store_distance; i++) {
int d = i - load_store_distance;
if (is_copy && i < num_mem_ops) {
if (i) {
ureg_UADD(ureg, load_addr, ureg_src(load_addr),
ureg_imm1u(ureg, 4 * inst_dwords[i] * 64));
}
values[i] = ureg_src(ureg_DECL_temporary(ureg));
struct ureg_dst dst =
ureg_writemask(ureg_dst(values[i]),
u_bit_consecutive(0, inst_dwords[i]));
struct ureg_src srcs[] = {srcbuf, ureg_src(load_addr)};
ureg_memory_insn(ureg, TGSI_OPCODE_LOAD, &dst, 1, srcs, 2,
load_qualifier, TGSI_TEXTURE_BUFFER, 0);
}
if (d >= 0) {
if (d) {
ureg_UADD(ureg, store_addr, ureg_src(store_addr),
ureg_imm1u(ureg, 4 * inst_dwords[d] * 64));
}
struct ureg_dst dst =
ureg_writemask(dstbuf, u_bit_consecutive(0, inst_dwords[d]));
struct ureg_src srcs[] =
{ureg_src(store_addr), is_copy ? values[d] : value};
ureg_memory_insn(ureg, TGSI_OPCODE_STORE, &dst, 1, srcs, 2,
store_qualifier, TGSI_TEXTURE_BUFFER, 0);
}
}
ureg_END(ureg);
struct pipe_compute_state state = {};
state.ir_type = PIPE_SHADER_IR_TGSI;
state.prog = ureg_get_tokens(ureg, NULL);
void *cs = ctx->create_compute_state(ctx, &state);
ureg_destroy(ureg);
free(values);
return cs;
}
static void *
create_mismatch_frag_shader(struct vl_idct *idct)
{
struct ureg_program *shader;
struct ureg_src addr[2];
struct ureg_dst m[8][2];
struct ureg_dst fragment;
unsigned i;
shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
if (!shader)
return NULL;
addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
for (i = 0; i < 8; ++i) {
m[i][0] = ureg_DECL_temporary(shader);
m[i][1] = ureg_DECL_temporary(shader);
}
for (i = 0; i < 8; ++i) {
increment_addr(shader, m[i], addr, false, false, i, idct->buffer_height);
}
for (i = 0; i < 8; ++i) {
struct ureg_src s_addr[2];
s_addr[0] = ureg_src(m[i][0]);
s_addr[1] = ureg_src(m[i][1]);
fetch_four(shader, m[i], s_addr, ureg_DECL_sampler(shader, 0), false);
}
for (i = 1; i < 8; ++i) {
ureg_ADD(shader, m[0][0], ureg_src(m[0][0]), ureg_src(m[i][0]));
ureg_ADD(shader, m[0][1], ureg_src(m[0][1]), ureg_src(m[i][1]));
}
ureg_ADD(shader, m[0][0], ureg_src(m[0][0]), ureg_src(m[0][1]));
ureg_DP4(shader, m[0][0], ureg_abs(ureg_src(m[0][0])), ureg_imm1f(shader, 1 << 14));
ureg_MUL(shader, ureg_writemask(m[0][0], TGSI_WRITEMASK_W), ureg_abs(ureg_src(m[7][1])), ureg_imm1f(shader, 1 << 14));
ureg_FRC(shader, m[0][0], ureg_src(m[0][0]));
ureg_SGT(shader, m[0][0], ureg_imm1f(shader, 0.5f), ureg_abs(ureg_src(m[0][0])));
ureg_CMP(shader, ureg_writemask(m[0][0], TGSI_WRITEMASK_W), ureg_negate(ureg_src(m[0][0])),
ureg_imm1f(shader, 1.0f / (1 << 15)), ureg_imm1f(shader, -1.0f / (1 << 15)));
ureg_MUL(shader, ureg_writemask(m[0][0], TGSI_WRITEMASK_W), ureg_src(m[0][0]),
ureg_scalar(ureg_src(m[0][0]), TGSI_SWIZZLE_X));
ureg_MOV(shader, ureg_writemask(fragment, TGSI_WRITEMASK_XYZ), ureg_src(m[7][1]));
ureg_ADD(shader, ureg_writemask(fragment, TGSI_WRITEMASK_W), ureg_src(m[0][0]), ureg_src(m[7][1]));
for (i = 0; i < 8; ++i) {
ureg_release_temporary(shader, m[i][0]);
ureg_release_temporary(shader, m[i][1]);
}
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, idct->pipe);
}
static void
radial_gradient(struct ureg_program *ureg,
struct ureg_dst out,
struct ureg_src pos,
struct ureg_src sampler,
struct ureg_src coords,
struct ureg_src const0124,
struct ureg_src matrow0,
struct ureg_src matrow1, struct ureg_src matrow2)
{
struct ureg_dst temp0 = ureg_DECL_temporary(ureg);
struct ureg_dst temp1 = ureg_DECL_temporary(ureg);
struct ureg_dst temp2 = ureg_DECL_temporary(ureg);
struct ureg_dst temp3 = ureg_DECL_temporary(ureg);
struct ureg_dst temp4 = ureg_DECL_temporary(ureg);
struct ureg_dst temp5 = ureg_DECL_temporary(ureg);
ureg_MOV(ureg, ureg_writemask(temp0, TGSI_WRITEMASK_XY), pos);
ureg_MOV(ureg,
ureg_writemask(temp0, TGSI_WRITEMASK_Z),
ureg_scalar(const0124, TGSI_SWIZZLE_Y));
ureg_DP3(ureg, temp1, matrow0, ureg_src(temp0));
ureg_DP3(ureg, temp2, matrow1, ureg_src(temp0));
ureg_DP3(ureg, temp3, matrow2, ureg_src(temp0));
ureg_RCP(ureg, temp3, ureg_src(temp3));
ureg_MUL(ureg, temp1, ureg_src(temp1), ureg_src(temp3));
ureg_MUL(ureg, temp2, ureg_src(temp2), ureg_src(temp3));
ureg_MOV(ureg, ureg_writemask(temp5, TGSI_WRITEMASK_X), ureg_src(temp1));
ureg_MOV(ureg, ureg_writemask(temp5, TGSI_WRITEMASK_Y), ureg_src(temp2));
ureg_MUL(ureg, temp0, ureg_scalar(coords, TGSI_SWIZZLE_Y),
ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_Y));
ureg_MAD(ureg, temp1,
ureg_scalar(coords, TGSI_SWIZZLE_X),
ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_X), ureg_src(temp0));
ureg_ADD(ureg, temp1, ureg_src(temp1), ureg_src(temp1));
ureg_MUL(ureg, temp3,
ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_Y),
ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_Y));
ureg_MAD(ureg, temp4,
ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_X),
ureg_scalar(ureg_src(temp5), TGSI_SWIZZLE_X), ureg_src(temp3));
ureg_MOV(ureg, temp4, ureg_negate(ureg_src(temp4)));
ureg_MUL(ureg, temp2, ureg_scalar(coords, TGSI_SWIZZLE_Z), ureg_src(temp4));
ureg_MUL(ureg, temp0,
ureg_scalar(const0124, TGSI_SWIZZLE_W), ureg_src(temp2));
ureg_MUL(ureg, temp3, ureg_src(temp1), ureg_src(temp1));
ureg_SUB(ureg, temp2, ureg_src(temp3), ureg_src(temp0));
ureg_RSQ(ureg, temp2, ureg_abs(ureg_src(temp2)));
ureg_RCP(ureg, temp2, ureg_src(temp2));
ureg_SUB(ureg, temp1, ureg_src(temp2), ureg_src(temp1));
ureg_ADD(ureg, temp0,
ureg_scalar(coords, TGSI_SWIZZLE_Z),
ureg_scalar(coords, TGSI_SWIZZLE_Z));
ureg_RCP(ureg, temp0, ureg_src(temp0));
ureg_MUL(ureg, temp2, ureg_src(temp1), ureg_src(temp0));
ureg_TEX(ureg, out, TGSI_TEXTURE_1D, ureg_src(temp2), sampler);
ureg_release_temporary(ureg, temp0);
ureg_release_temporary(ureg, temp1);
ureg_release_temporary(ureg, temp2);
ureg_release_temporary(ureg, temp3);
ureg_release_temporary(ureg, temp4);
ureg_release_temporary(ureg, temp5);
}
/**
* Make simple fragment texture shader:
* IMM {0,0,0,1} // (if writemask != 0xf)
* MOV TEMP[0], IMM[0] // (if writemask != 0xf)
* TEX TEMP[0].writemask, IN[0], SAMP[0], 2D;
* .. optional SINT <-> UINT clamping ..
* MOV OUT[0], TEMP[0]
* END;
*
* \param tex_target one of PIPE_TEXTURE_x
* \parma interp_mode either TGSI_INTERPOLATE_LINEAR or PERSPECTIVE
* \param writemask mask of TGSI_WRITEMASK_x
*/
void *
util_make_fragment_tex_shader_writemask(struct pipe_context *pipe,
unsigned tex_target,
unsigned interp_mode,
unsigned writemask,
enum tgsi_return_type stype,
enum tgsi_return_type dtype,
bool load_level_zero,
bool use_txf)
{
struct ureg_program *ureg;
struct ureg_src sampler;
struct ureg_src tex;
struct ureg_dst temp;
struct ureg_dst out;
assert((stype == TGSI_RETURN_TYPE_FLOAT) == (dtype == TGSI_RETURN_TYPE_FLOAT));
assert(interp_mode == TGSI_INTERPOLATE_LINEAR ||
interp_mode == TGSI_INTERPOLATE_PERSPECTIVE);
ureg = ureg_create( PIPE_SHADER_FRAGMENT );
if (!ureg)
return NULL;
sampler = ureg_DECL_sampler( ureg, 0 );
ureg_DECL_sampler_view(ureg, 0, tex_target, stype, stype, stype, stype);
tex = ureg_DECL_fs_input( ureg,
TGSI_SEMANTIC_GENERIC, 0,
interp_mode );
out = ureg_DECL_output( ureg,
TGSI_SEMANTIC_COLOR,
0 );
temp = ureg_DECL_temporary(ureg);
if (writemask != TGSI_WRITEMASK_XYZW) {
struct ureg_src imm = ureg_imm4f( ureg, 0, 0, 0, 1 );
ureg_MOV( ureg, out, imm );
}
if (tex_target == TGSI_TEXTURE_BUFFER)
ureg_TXF(ureg,
ureg_writemask(temp, writemask),
tex_target, tex, sampler);
else
ureg_load_tex(ureg, ureg_writemask(temp, writemask), tex, sampler,
tex_target, load_level_zero, use_txf);
if (stype != dtype) {
if (stype == TGSI_RETURN_TYPE_SINT) {
assert(dtype == TGSI_RETURN_TYPE_UINT);
ureg_IMAX(ureg, temp, ureg_src(temp), ureg_imm1i(ureg, 0));
} else {
assert(stype == TGSI_RETURN_TYPE_UINT);
assert(dtype == TGSI_RETURN_TYPE_SINT);
ureg_UMIN(ureg, temp, ureg_src(temp), ureg_imm1u(ureg, (1u << 31) - 1));
}
}
ureg_MOV(ureg, out, ureg_src(temp));
ureg_END( ureg );
return ureg_create_shader_and_destroy( ureg, pipe );
}
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);
}
static void *
create_frag_shader(struct vl_zscan *zscan)
{
struct ureg_program *shader;
struct ureg_src *vtex;
struct ureg_src samp_src, samp_scan, samp_quant;
struct ureg_dst *tmp;
struct ureg_dst quant, fragment;
unsigned i;
shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
if (!shader)
return NULL;
vtex = MALLOC(zscan->num_channels * sizeof(struct ureg_src));
tmp = MALLOC(zscan->num_channels * sizeof(struct ureg_dst));
for (i = 0; i < zscan->num_channels; ++i)
vtex[i] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX + i, TGSI_INTERPOLATE_LINEAR);
samp_src = ureg_DECL_sampler(shader, 0);
samp_scan = ureg_DECL_sampler(shader, 1);
samp_quant = ureg_DECL_sampler(shader, 2);
for (i = 0; i < zscan->num_channels; ++i)
tmp[i] = ureg_DECL_temporary(shader);
quant = ureg_DECL_temporary(shader);
fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
/*
* tmp.x = tex(vtex, 1)
* tmp.y = vtex.z
* fragment = tex(tmp, 0) * quant
*/
for (i = 0; i < zscan->num_channels; ++i)
ureg_TEX(shader, ureg_writemask(tmp[i], TGSI_WRITEMASK_X), TGSI_TEXTURE_2D, vtex[i], samp_scan);
for (i = 0; i < zscan->num_channels; ++i)
ureg_MOV(shader, ureg_writemask(tmp[i], TGSI_WRITEMASK_Y), ureg_scalar(vtex[i], TGSI_SWIZZLE_W));
for (i = 0; i < zscan->num_channels; ++i) {
ureg_TEX(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X << i), TGSI_TEXTURE_2D, ureg_src(tmp[i]), samp_src);
ureg_TEX(shader, ureg_writemask(quant, TGSI_WRITEMASK_X << i), TGSI_TEXTURE_3D, vtex[i], samp_quant);
}
ureg_MUL(shader, quant, ureg_src(quant), ureg_imm1f(shader, 16.0f));
ureg_MUL(shader, fragment, ureg_src(tmp[0]), ureg_src(quant));
for (i = 0; i < zscan->num_channels; ++i)
ureg_release_temporary(shader, tmp[i]);
ureg_END(shader);
FREE(vtex);
FREE(tmp);
return ureg_create_shader_and_destroy(shader, zscan->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;
signed i;
shader = ureg_create(TGSI_PROCESSOR_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);
}
static void *
create_frag_shader(struct vl_median_filter *filter,
struct vertex2f *offsets,
unsigned num_offsets)
{
struct pipe_screen *screen = filter->pipe->screen;
struct ureg_program *shader;
struct ureg_src i_vtex;
struct ureg_src sampler;
struct ureg_dst *t_array = MALLOC(sizeof(struct ureg_dst) * num_offsets);
struct ureg_dst o_fragment;
const unsigned median = num_offsets >> 1;
unsigned i, j;
assert(num_offsets & 1); /* we need an odd number of offsets */
if (!(num_offsets & 1)) { /* yeah, we REALLY need an odd number of offsets!!! */
FREE(t_array);
return NULL;
}
if (num_offsets > screen->get_shader_param(
screen, PIPE_SHADER_FRAGMENT, PIPE_SHADER_CAP_MAX_TEMPS)) {
FREE(t_array);
return NULL;
}
shader = ureg_create(PIPE_SHADER_FRAGMENT);
if (!shader) {
FREE(t_array);
return NULL;
}
i_vtex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX, TGSI_INTERPOLATE_LINEAR);
sampler = ureg_DECL_sampler(shader, 0);
ureg_DECL_sampler_view(shader, 0, TGSI_TEXTURE_2D,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT);
for (i = 0; i < num_offsets; ++i)
t_array[i] = ureg_DECL_temporary(shader);
o_fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
/*
* t_array[0..*] = vtex + offset[0..*]
* t_array[0..*] = tex(t_array[0..*], sampler)
* result = partial_bubblesort(t_array)[mid]
*/
for (i = 0; i < num_offsets; ++i) {
if (!is_vec_zero(offsets[i])) {
ureg_ADD(shader, ureg_writemask(t_array[i], TGSI_WRITEMASK_XY),
i_vtex, ureg_imm2f(shader, offsets[i].x, offsets[i].y));
ureg_MOV(shader, ureg_writemask(t_array[i], TGSI_WRITEMASK_ZW),
ureg_imm1f(shader, 0.0f));
}
}
for (i = 0; i < num_offsets; ++i) {
struct ureg_src src = is_vec_zero(offsets[i]) ? i_vtex : ureg_src(t_array[i]);
ureg_TEX(shader, t_array[i], TGSI_TEXTURE_2D, src, sampler);
}
// TODO: Couldn't this be improved even more?
for (i = 0; i <= median; ++i) {
for (j = 1; j < (num_offsets - i - 1); ++j) {
struct ureg_dst tmp = ureg_DECL_temporary(shader);
ureg_MOV(shader, tmp, ureg_src(t_array[j]));
ureg_MAX(shader, t_array[j], ureg_src(t_array[j]), ureg_src(t_array[j - 1]));
ureg_MIN(shader, t_array[j - 1], ureg_src(tmp), ureg_src(t_array[j - 1]));
ureg_release_temporary(shader, tmp);
}
if (i == median)
ureg_MAX(shader, t_array[j], ureg_src(t_array[j]), ureg_src(t_array[j - 1]));
else
ureg_MIN(shader, t_array[j - 1], ureg_src(t_array[j]), ureg_src(t_array[j - 1]));
}
ureg_MOV(shader, o_fragment, ureg_src(t_array[median]));
ureg_END(shader);
FREE(t_array);
return ureg_create_shader_and_destroy(shader, filter->pipe);
}
static void *
create_fs(struct st_context *st, bool download, enum pipe_texture_target target)
{
struct pipe_context *pipe = st->pipe;
struct pipe_screen *screen = pipe->screen;
struct ureg_program *ureg;
bool have_layer;
struct ureg_dst out;
struct ureg_src sampler;
struct ureg_src pos;
struct ureg_src layer;
struct ureg_src const0;
struct ureg_src const1;
struct ureg_dst temp0;
have_layer =
st->pbo.layers &&
(!download || target == PIPE_TEXTURE_1D_ARRAY
|| target == PIPE_TEXTURE_2D_ARRAY
|| target == PIPE_TEXTURE_3D
|| target == PIPE_TEXTURE_CUBE
|| target == PIPE_TEXTURE_CUBE_ARRAY);
ureg = ureg_create(PIPE_SHADER_FRAGMENT);
if (!ureg)
return NULL;
if (!download) {
out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
} else {
struct ureg_src image;
/* writeonly images do not require an explicitly given format. */
image = ureg_DECL_image(ureg, 0, TGSI_TEXTURE_BUFFER, PIPE_FORMAT_NONE,
true, false);
out = ureg_dst(image);
}
sampler = ureg_DECL_sampler(ureg, 0);
if (screen->get_param(screen, PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL)) {
pos = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_POSITION, 0);
} else {
pos = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_POSITION, 0,
TGSI_INTERPOLATE_LINEAR);
}
if (have_layer) {
layer = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_LAYER, 0,
TGSI_INTERPOLATE_CONSTANT);
}
const0 = ureg_DECL_constant(ureg, 0);
const1 = ureg_DECL_constant(ureg, 1);
temp0 = ureg_DECL_temporary(ureg);
/* Note: const0 = [ -xoffset + skip_pixels, -yoffset, stride, image_height ] */
/* temp0.xy = f2i(temp0.xy) */
ureg_F2I(ureg, ureg_writemask(temp0, TGSI_WRITEMASK_XY),
ureg_swizzle(pos,
TGSI_SWIZZLE_X, TGSI_SWIZZLE_Y,
TGSI_SWIZZLE_Y, TGSI_SWIZZLE_Y));
/* temp0.xy = temp0.xy + const0.xy */
ureg_UADD(ureg, ureg_writemask(temp0, TGSI_WRITEMASK_XY),
ureg_swizzle(ureg_src(temp0),
TGSI_SWIZZLE_X, TGSI_SWIZZLE_Y,
TGSI_SWIZZLE_Y, TGSI_SWIZZLE_Y),
ureg_swizzle(const0,
TGSI_SWIZZLE_X, TGSI_SWIZZLE_Y,
TGSI_SWIZZLE_Y, TGSI_SWIZZLE_Y));
/* temp0.x = const0.z * temp0.y + temp0.x */
ureg_UMAD(ureg, ureg_writemask(temp0, TGSI_WRITEMASK_X),
ureg_scalar(const0, TGSI_SWIZZLE_Z),
ureg_scalar(ureg_src(temp0), TGSI_SWIZZLE_Y),
ureg_scalar(ureg_src(temp0), TGSI_SWIZZLE_X));
if (have_layer) {
/* temp0.x = const0.w * layer + temp0.x */
ureg_UMAD(ureg, ureg_writemask(temp0, TGSI_WRITEMASK_X),
ureg_scalar(const0, TGSI_SWIZZLE_W),
ureg_scalar(layer, TGSI_SWIZZLE_X),
ureg_scalar(ureg_src(temp0), TGSI_SWIZZLE_X));
}
/* temp0.w = 0 */
ureg_MOV(ureg, ureg_writemask(temp0, TGSI_WRITEMASK_W), ureg_imm1u(ureg, 0));
if (download) {
struct ureg_dst temp1;
struct ureg_src op[2];
temp1 = ureg_DECL_temporary(ureg);
/* temp1.xy = pos.xy */
ureg_F2I(ureg, ureg_writemask(temp1, TGSI_WRITEMASK_XY), pos);
/* temp1.zw = 0 */
ureg_MOV(ureg, ureg_writemask(temp1, TGSI_WRITEMASK_ZW), ureg_imm1u(ureg, 0));
if (have_layer) {
struct ureg_dst temp1_layer =
ureg_writemask(temp1, target == PIPE_TEXTURE_1D_ARRAY ? TGSI_WRITEMASK_Y
: TGSI_WRITEMASK_Z);
/* temp1.y/z = layer */
ureg_MOV(ureg, temp1_layer, ureg_scalar(layer, TGSI_SWIZZLE_X));
if (target == PIPE_TEXTURE_3D) {
/* temp1.z += layer_offset */
ureg_UADD(ureg, temp1_layer,
ureg_scalar(ureg_src(temp1), TGSI_SWIZZLE_Z),
ureg_scalar(const1, TGSI_SWIZZLE_X));
}
}
/* temp1 = txf(sampler, temp1) */
ureg_TXF(ureg, temp1, util_pipe_tex_to_tgsi_tex(target, 1),
ureg_src(temp1), sampler);
/* store(out, temp0, temp1) */
op[0] = ureg_src(temp0);
op[1] = ureg_src(temp1);
ureg_memory_insn(ureg, TGSI_OPCODE_STORE, &out, 1, op, 2, 0,
TGSI_TEXTURE_BUFFER, PIPE_FORMAT_NONE);
ureg_release_temporary(ureg, temp1);
} else {
/* out = txf(sampler, temp0.x) */
ureg_TXF(ureg, out, TGSI_TEXTURE_BUFFER, ureg_src(temp0), sampler);
}
ureg_release_temporary(ureg, temp0);
ureg_END(ureg);
return ureg_create_shader_and_destroy(ureg, pipe);
}
static void *
create_fs(struct pipe_context *pipe, unsigned fs_traits)
{
struct ureg_program *ureg;
struct ureg_src /*dst_sampler, */ src_sampler, mask_sampler;
struct ureg_src /*dst_pos, */ src_input, mask_pos;
struct ureg_dst src, mask;
struct ureg_dst out;
struct ureg_src imm0 = { 0 };
unsigned has_mask = (fs_traits & FS_MASK) != 0;
unsigned is_fill = (fs_traits & FS_FILL) != 0;
unsigned is_composite = (fs_traits & FS_COMPOSITE) != 0;
unsigned is_solid = (fs_traits & FS_SOLID_FILL) != 0;
unsigned is_lingrad = (fs_traits & FS_LINGRAD_FILL) != 0;
unsigned is_radgrad = (fs_traits & FS_RADGRAD_FILL) != 0;
unsigned comp_alpha_mask = fs_traits & FS_COMPONENT_ALPHA;
unsigned is_yuv = (fs_traits & FS_YUV) != 0;
unsigned src_repeat_none = (fs_traits & FS_SRC_REPEAT_NONE) != 0;
unsigned mask_repeat_none = (fs_traits & FS_MASK_REPEAT_NONE) != 0;
unsigned src_swizzle = (fs_traits & FS_SRC_SWIZZLE_RGB) != 0;
unsigned mask_swizzle = (fs_traits & FS_MASK_SWIZZLE_RGB) != 0;
unsigned src_set_alpha = (fs_traits & FS_SRC_SET_ALPHA) != 0;
unsigned mask_set_alpha = (fs_traits & FS_MASK_SET_ALPHA) != 0;
unsigned src_luminance = (fs_traits & FS_SRC_LUMINANCE) != 0;
unsigned mask_luminance = (fs_traits & FS_MASK_LUMINANCE) != 0;
unsigned dst_luminance = (fs_traits & FS_DST_LUMINANCE) != 0;
#if 0
print_fs_traits(fs_traits);
#else
(void)print_fs_traits;
#endif
ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
if (ureg == NULL)
return 0;
/* it has to be either a fill, a composite op or a yuv conversion */
debug_assert((is_fill ^ is_composite) ^ is_yuv);
(void)is_yuv;
out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
if (src_repeat_none || mask_repeat_none ||
src_set_alpha || mask_set_alpha || src_luminance) {
imm0 = ureg_imm4f(ureg, 0, 0, 0, 1);
}
if (is_composite) {
src_sampler = ureg_DECL_sampler(ureg, 0);
src_input = ureg_DECL_fs_input(ureg,
TGSI_SEMANTIC_GENERIC, 0,
TGSI_INTERPOLATE_PERSPECTIVE);
} else if (is_fill) {
if (is_solid)
src_input = ureg_DECL_fs_input(ureg,
TGSI_SEMANTIC_COLOR, 0,
TGSI_INTERPOLATE_PERSPECTIVE);
else
src_input = ureg_DECL_fs_input(ureg,
TGSI_SEMANTIC_POSITION, 0,
TGSI_INTERPOLATE_PERSPECTIVE);
} else {
debug_assert(is_yuv);
return create_yuv_shader(pipe, ureg);
}
if (has_mask) {
mask_sampler = ureg_DECL_sampler(ureg, 1);
mask_pos = ureg_DECL_fs_input(ureg,
TGSI_SEMANTIC_GENERIC, 1,
TGSI_INTERPOLATE_PERSPECTIVE);
}
#if 0 /* unused right now */
dst_sampler = ureg_DECL_sampler(ureg, 2);
dst_pos = ureg_DECL_fs_input(ureg,
TGSI_SEMANTIC_POSITION, 2,
TGSI_INTERPOLATE_PERSPECTIVE);
#endif
if (is_composite) {
if (has_mask || src_luminance || dst_luminance)
src = ureg_DECL_temporary(ureg);
else
src = out;
xrender_tex(ureg, src, src_input, src_sampler, imm0,
src_repeat_none, src_swizzle, src_set_alpha);
} else if (is_fill) {
if (is_solid) {
if (has_mask || src_luminance || dst_luminance)
src = ureg_dst(src_input);
else
ureg_MOV(ureg, out, src_input);
} else if (is_lingrad || is_radgrad) {
struct ureg_src coords, const0124, matrow0, matrow1, matrow2;
if (has_mask || src_luminance || dst_luminance)
src = ureg_DECL_temporary(ureg);
else
src = out;
coords = ureg_DECL_constant(ureg, 0);
const0124 = ureg_DECL_constant(ureg, 1);
matrow0 = ureg_DECL_constant(ureg, 2);
matrow1 = ureg_DECL_constant(ureg, 3);
matrow2 = ureg_DECL_constant(ureg, 4);
if (is_lingrad) {
linear_gradient(ureg, src,
src_input, src_sampler,
coords, const0124, matrow0, matrow1, matrow2);
} else if (is_radgrad) {
radial_gradient(ureg, src,
src_input, src_sampler,
coords, const0124, matrow0, matrow1, matrow2);
}
} else
debug_assert(!"Unknown fill type!");
}
if (src_luminance) {
ureg_MOV(ureg, src, ureg_scalar(ureg_src(src), TGSI_SWIZZLE_X));
ureg_MOV(ureg, ureg_writemask(src, TGSI_WRITEMASK_XYZ),
ureg_scalar(imm0, TGSI_SWIZZLE_X));
if (!has_mask && !dst_luminance)
ureg_MOV(ureg, out, ureg_src(src));
}
if (has_mask) {
mask = ureg_DECL_temporary(ureg);
xrender_tex(ureg, mask, mask_pos, mask_sampler, imm0,
mask_repeat_none, mask_swizzle, mask_set_alpha);
/* src IN mask */
src_in_mask(ureg, (dst_luminance) ? src : out, ureg_src(src),
ureg_src(mask),
comp_alpha_mask, mask_luminance);
ureg_release_temporary(ureg, mask);
}
if (dst_luminance) {
/*
* Make sure the alpha channel goes into the output L8 surface.
*/
ureg_MOV(ureg, out, ureg_scalar(ureg_src(src), TGSI_SWIZZLE_W));
}
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_fs_msaa_resolve_bilinear(struct pipe_context *pipe,
enum tgsi_texture_type tgsi_tex,
unsigned nr_samples,
enum tgsi_return_type stype)
{
struct ureg_program *ureg;
struct ureg_src sampler, coord;
struct ureg_dst out, tmp, top, bottom;
struct ureg_dst tmp_coord[4], tmp_sum[4];
unsigned i, c;
ureg = ureg_create(PIPE_SHADER_FRAGMENT);
if (!ureg)
return NULL;
/* Declarations. */
sampler = ureg_DECL_sampler(ureg, 0);
ureg_DECL_sampler_view(ureg, 0, tgsi_tex, stype, stype, stype, stype);
coord = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_GENERIC, 0,
TGSI_INTERPOLATE_LINEAR);
out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
for (c = 0; c < 4; c++)
tmp_sum[c] = ureg_DECL_temporary(ureg);
for (c = 0; c < 4; c++)
tmp_coord[c] = ureg_DECL_temporary(ureg);
tmp = ureg_DECL_temporary(ureg);
top = ureg_DECL_temporary(ureg);
bottom = ureg_DECL_temporary(ureg);
/* Instructions. */
for (c = 0; c < 4; c++)
ureg_MOV(ureg, tmp_sum[c], ureg_imm1f(ureg, 0));
/* Get 4 texture coordinates for the bilinear filter. */
ureg_F2U(ureg, tmp_coord[0], coord); /* top-left */
ureg_UADD(ureg, tmp_coord[1], ureg_src(tmp_coord[0]),
ureg_imm4u(ureg, 1, 0, 0, 0)); /* top-right */
ureg_UADD(ureg, tmp_coord[2], ureg_src(tmp_coord[0]),
ureg_imm4u(ureg, 0, 1, 0, 0)); /* bottom-left */
ureg_UADD(ureg, tmp_coord[3], ureg_src(tmp_coord[0]),
ureg_imm4u(ureg, 1, 1, 0, 0)); /* bottom-right */
for (i = 0; i < nr_samples; i++) {
for (c = 0; c < 4; c++) {
/* Read one sample. */
ureg_MOV(ureg, ureg_writemask(tmp_coord[c], TGSI_WRITEMASK_W),
ureg_imm1u(ureg, i));
ureg_TXF(ureg, tmp, tgsi_tex, ureg_src(tmp_coord[c]), sampler);
if (stype == TGSI_RETURN_TYPE_UINT)
ureg_U2F(ureg, tmp, ureg_src(tmp));
else if (stype == TGSI_RETURN_TYPE_SINT)
ureg_I2F(ureg, tmp, ureg_src(tmp));
/* Add it to the sum.*/
ureg_ADD(ureg, tmp_sum[c], ureg_src(tmp_sum[c]), ureg_src(tmp));
}
}
/* Calculate the average. */
for (c = 0; c < 4; c++)
ureg_MUL(ureg, tmp_sum[c], ureg_src(tmp_sum[c]),
ureg_imm1f(ureg, 1.0 / nr_samples));
/* Take the 4 average values and apply a standard bilinear filter. */
ureg_FRC(ureg, tmp, coord);
ureg_LRP(ureg, top,
ureg_scalar(ureg_src(tmp), 0),
ureg_src(tmp_sum[1]),
ureg_src(tmp_sum[0]));
ureg_LRP(ureg, bottom,
ureg_scalar(ureg_src(tmp), 0),
ureg_src(tmp_sum[3]),
ureg_src(tmp_sum[2]));
ureg_LRP(ureg, tmp,
ureg_scalar(ureg_src(tmp), 1),
ureg_src(bottom),
ureg_src(top));
/* Convert to the texture format and return. */
if (stype == TGSI_RETURN_TYPE_UINT)
ureg_F2U(ureg, out, ureg_src(tmp));
else if (stype == TGSI_RETURN_TYPE_SINT)
ureg_F2I(ureg, out, ureg_src(tmp));
else
ureg_MOV(ureg, out, ureg_src(tmp));
ureg_END(ureg);
return ureg_create_shader_and_destroy(ureg, pipe);
}
static void *
create_ref_frag_shader(struct vl_mc *r)
{
const float y_scale =
r->buffer_height / 2 *
r->macroblock_size / VL_MACROBLOCK_HEIGHT;
struct ureg_program *shader;
struct ureg_src tc[2], sampler;
struct ureg_dst ref, field;
struct ureg_dst fragment;
unsigned label;
shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
if (!shader)
return NULL;
tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTOP, TGSI_INTERPOLATE_LINEAR);
tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VBOTTOM, TGSI_INTERPOLATE_LINEAR);
sampler = ureg_DECL_sampler(shader, 0);
ref = ureg_DECL_temporary(shader);
fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
field = calc_line(shader);
/*
* ref = field.z ? tc[1] : tc[0]
*
* // Adjust tc acording to top/bottom field selection
* if (|ref.z|) {
* ref.y *= y_scale
* ref.y = floor(ref.y)
* ref.y += ref.z
* ref.y /= y_scale
* }
* fragment.xyz = tex(ref, sampler[0])
*/
ureg_CMP(shader, ureg_writemask(ref, TGSI_WRITEMASK_XYZ),
ureg_negate(ureg_scalar(ureg_src(field), TGSI_SWIZZLE_Y)),
tc[1], tc[0]);
ureg_CMP(shader, ureg_writemask(fragment, TGSI_WRITEMASK_W),
ureg_negate(ureg_scalar(ureg_src(field), TGSI_SWIZZLE_Y)),
tc[1], tc[0]);
ureg_IF(shader, ureg_scalar(ureg_src(ref), TGSI_SWIZZLE_Z), &label);
ureg_MUL(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y),
ureg_src(ref), ureg_imm1f(shader, y_scale));
ureg_FLR(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y), ureg_src(ref));
ureg_ADD(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y),
ureg_src(ref), ureg_scalar(ureg_src(ref), TGSI_SWIZZLE_Z));
ureg_MUL(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y),
ureg_src(ref), ureg_imm1f(shader, 1.0f / y_scale));
ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
ureg_ENDIF(shader);
ureg_TEX(shader, ureg_writemask(fragment, TGSI_WRITEMASK_XYZ), TGSI_TEXTURE_2D, ureg_src(ref), sampler);
ureg_release_temporary(shader, ref);
ureg_release_temporary(shader, field);
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, r->pipe);
}
static void *
create_frag_shader_weave(struct vl_compositor *c)
{
struct ureg_program *shader;
struct ureg_src i_tc[2];
struct ureg_src csc[3];
struct ureg_src sampler[3];
struct ureg_dst t_tc[2];
struct ureg_dst t_texel[2];
struct ureg_dst o_fragment;
unsigned i, j;
shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
if (!shader)
return false;
i_tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTOP, TGSI_INTERPOLATE_LINEAR);
i_tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VBOTTOM, TGSI_INTERPOLATE_LINEAR);
for (i = 0; i < 3; ++i) {
csc[i] = ureg_DECL_constant(shader, i);
sampler[i] = ureg_DECL_sampler(shader, i);
}
for (i = 0; i < 2; ++i) {
t_tc[i] = ureg_DECL_temporary(shader);
t_texel[i] = ureg_DECL_temporary(shader);
}
o_fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
/* calculate the texture offsets
* t_tc.x = i_tc.x
* t_tc.y = (round(i_tc.y - 0.5) + 0.5) / height * 2
*/
for (i = 0; i < 2; ++i) {
ureg_MOV(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_X), i_tc[i]);
ureg_SUB(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_YZ),
i_tc[i], ureg_imm1f(shader, 0.5f));
ureg_ROUND(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_YZ), ureg_src(t_tc[i]));
ureg_MOV(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_W),
ureg_imm1f(shader, i ? 1.0f : 0.0f));
ureg_ADD(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_YZ),
ureg_src(t_tc[i]), ureg_imm1f(shader, 0.5f));
ureg_MUL(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_Y),
ureg_src(t_tc[i]), ureg_scalar(i_tc[0], TGSI_SWIZZLE_W));
ureg_MUL(shader, ureg_writemask(t_tc[i], TGSI_WRITEMASK_Z),
ureg_src(t_tc[i]), ureg_scalar(i_tc[1], TGSI_SWIZZLE_W));
}
/* fetch the texels
* texel[0..1].x = tex(t_tc[0..1][0])
* texel[0..1].y = tex(t_tc[0..1][1])
* texel[0..1].z = tex(t_tc[0..1][2])
*/
for (i = 0; i < 2; ++i)
for (j = 0; j < 3; ++j) {
struct ureg_src src = ureg_swizzle(ureg_src(t_tc[i]),
TGSI_SWIZZLE_X, j ? TGSI_SWIZZLE_Z : TGSI_SWIZZLE_Y, TGSI_SWIZZLE_W, TGSI_SWIZZLE_W);
ureg_TEX(shader, ureg_writemask(t_texel[i], TGSI_WRITEMASK_X << j),
TGSI_TEXTURE_2D_ARRAY, src, sampler[j]);
}
/* calculate linear interpolation factor
* factor = |round(i_tc.y) - i_tc.y| * 2
*/
ureg_ROUND(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_YZ), i_tc[0]);
ureg_ADD(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_YZ),
ureg_src(t_tc[0]), ureg_negate(i_tc[0]));
ureg_MUL(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_YZ),
ureg_abs(ureg_src(t_tc[0])), ureg_imm1f(shader, 2.0f));
ureg_LRP(shader, t_texel[0], ureg_swizzle(ureg_src(t_tc[0]),
TGSI_SWIZZLE_Y, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_Z),
ureg_src(t_texel[0]), ureg_src(t_texel[1]));
/* and finally do colour space transformation
* fragment = csc * texel
*/
ureg_MOV(shader, ureg_writemask(t_texel[0], TGSI_WRITEMASK_W), ureg_imm1f(shader, 1.0f));
for (i = 0; i < 3; ++i)
ureg_DP4(shader, ureg_writemask(o_fragment, TGSI_WRITEMASK_X << i), csc[i], ureg_src(t_texel[0]));
ureg_MOV(shader, ureg_writemask(o_fragment, TGSI_WRITEMASK_W), ureg_imm1f(shader, 1.0f));
for (i = 0; i < 2; ++i) {
ureg_release_temporary(shader, t_texel[i]);
ureg_release_temporary(shader, t_tc[i]);
}
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, c->pipe);
}
static void *
create_ycbcr_frag_shader(struct vl_mc *r, float scale, bool invert,
vl_mc_ycbcr_frag_shader fs_callback, void *callback_priv)
{
struct ureg_program *shader;
struct ureg_src flags;
struct ureg_dst tmp;
struct ureg_dst fragment;
unsigned label;
shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
if (!shader)
return NULL;
flags = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_FLAGS, TGSI_INTERPOLATE_LINEAR);
fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
tmp = calc_line(shader);
/*
* if (field == tc.w)
* kill();
* else {
* fragment.xyz = tex(tc, sampler) * scale + tc.z
* fragment.w = 1.0f
* }
*/
ureg_SEQ(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y),
ureg_scalar(flags, TGSI_SWIZZLE_W), ureg_src(tmp));
ureg_IF(shader, ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y), &label);
ureg_KILL(shader);
ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
ureg_ELSE(shader, &label);
fs_callback(callback_priv, r, shader, VS_O_VTEX, tmp);
if (scale != 1.0f)
ureg_MAD(shader, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ),
ureg_src(tmp), ureg_imm1f(shader, scale),
ureg_scalar(flags, TGSI_SWIZZLE_Z));
else
ureg_ADD(shader, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ),
ureg_src(tmp), ureg_scalar(flags, TGSI_SWIZZLE_Z));
ureg_MUL(shader, ureg_writemask(fragment, TGSI_WRITEMASK_XYZ), ureg_src(tmp), ureg_imm1f(shader, invert ? -1.0f : 1.0f));
ureg_MOV(shader, ureg_writemask(fragment, TGSI_WRITEMASK_W), ureg_imm1f(shader, 1.0f));
ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
ureg_ENDIF(shader);
ureg_release_temporary(shader, tmp);
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, r->pipe);
}
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_stage1_vert_shader(struct vl_idct *idct)
{
struct ureg_program *shader;
struct ureg_src vrect, vpos;
struct ureg_src scale;
struct ureg_dst t_tex, t_start;
struct ureg_dst o_vpos, o_l_addr[2], o_r_addr[2];
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_tex = ureg_DECL_temporary(shader);
t_start = ureg_DECL_temporary(shader);
o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0);
o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1);
o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0);
o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1);
/*
* scale = (VL_BLOCK_WIDTH, VL_BLOCK_HEIGHT) / (dst.width, dst.height)
*
* t_vpos = vpos + vrect
* o_vpos.xy = t_vpos * scale
* o_vpos.zw = vpos
*
* o_l_addr = calc_addr(...)
* o_r_addr = calc_addr(...)
*
*/
scale = ureg_imm2f(shader,
(float)VL_BLOCK_WIDTH / idct->buffer_width,
(float)VL_BLOCK_HEIGHT / idct->buffer_height);
ureg_ADD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, vrect);
ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale);
ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_tex));
ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), ureg_imm1f(shader, 1.0f));
ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), vpos, scale);
calc_addr(shader, o_l_addr, ureg_src(t_tex), ureg_src(t_start), false, false, idct->buffer_width / 4);
calc_addr(shader, o_r_addr, vrect, ureg_imm1f(shader, 0.0f), true, true, VL_BLOCK_WIDTH / 4);
ureg_release_temporary(shader, t_tex);
ureg_release_temporary(shader, t_start);
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, idct->pipe);
}
static void *
create_stage1_frag_shader(struct vl_idct *idct)
{
struct ureg_program *shader;
struct ureg_src l_addr[2], r_addr[2];
struct ureg_dst l[4][2], r[2];
struct ureg_dst *fragment;
int i, j;
shader = ureg_create(PIPE_SHADER_FRAGMENT);
if (!shader)
return NULL;
fragment = MALLOC(idct->nr_of_render_targets * sizeof(struct ureg_dst));
l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
for (i = 0; i < idct->nr_of_render_targets; ++i)
fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i);
for (i = 0; i < 4; ++i) {
l[i][0] = ureg_DECL_temporary(shader);
l[i][1] = ureg_DECL_temporary(shader);
}
r[0] = ureg_DECL_temporary(shader);
r[1] = ureg_DECL_temporary(shader);
for (i = 0; i < 4; ++i) {
increment_addr(shader, l[i], l_addr, false, false, i - 2, idct->buffer_height);
}
for (i = 0; i < 4; ++i) {
struct ureg_src s_addr[2];
s_addr[0] = ureg_src(l[i][0]);
s_addr[1] = ureg_src(l[i][1]);
fetch_four(shader, l[i], s_addr, ureg_DECL_sampler(shader, 0), false);
}
for (i = 0; i < idct->nr_of_render_targets; ++i) {
struct ureg_src s_addr[2];
increment_addr(shader, r, r_addr, true, true, i - (signed)idct->nr_of_render_targets / 2, VL_BLOCK_HEIGHT);
s_addr[0] = ureg_src(r[0]);
s_addr[1] = ureg_src(r[1]);
fetch_four(shader, r, s_addr, ureg_DECL_sampler(shader, 1), false);
for (j = 0; j < 4; ++j) {
matrix_mul(shader, ureg_writemask(fragment[i], TGSI_WRITEMASK_X << j), l[j], r);
}
}
for (i = 0; i < 4; ++i) {
ureg_release_temporary(shader, l[i][0]);
ureg_release_temporary(shader, l[i][1]);
}
ureg_release_temporary(shader, r[0]);
ureg_release_temporary(shader, r[1]);
ureg_END(shader);
FREE(fragment);
return ureg_create_shader_and_destroy(shader, idct->pipe);
}