/**
* Update state for running fragment programs. Basically, load the
* program parameters with current state values.
*/
static void
_swrast_update_fragment_program( GLcontext *ctx )
{
if (ctx->FragmentProgram._Enabled) {
struct fragment_program *program = ctx->FragmentProgram.Current;
_mesa_load_state_parameters(ctx, program->Parameters);
}
}
/**
* Update state for running fragment programs. Basically, load the
* program parameters with current state values.
*/
static void
_swrast_update_fragment_program(struct gl_context *ctx, GLbitfield newState)
{
if (!_swrast_use_fragment_program(ctx))
return;
_mesa_load_state_parameters(ctx,
ctx->FragmentProgram._Current->Base.Parameters);
}
/**
* Pass the given program parameters to the graphics pipe as a
* constant buffer.
* \param shader_type either PIPE_SHADER_VERTEX or PIPE_SHADER_FRAGMENT
*/
void st_upload_constants( struct st_context *st,
struct gl_program_parameter_list *params,
unsigned shader_type)
{
assert(shader_type == PIPE_SHADER_VERTEX ||
shader_type == PIPE_SHADER_FRAGMENT ||
shader_type == PIPE_SHADER_GEOMETRY);
/* update constants */
if (params && params->NumParameters) {
struct pipe_constant_buffer cb;
const uint paramBytes = params->NumParameters * sizeof(GLfloat) * 4;
/* Update the constants which come from fixed-function state, such as
* transformation matrices, fog factors, etc. The rest of the values in
* the parameters list are explicitly set by the user with glUniform,
* glProgramParameter(), etc.
*/
_mesa_load_state_parameters(st->ctx, params);
/* We always need to get a new buffer, to keep the drivers simple and
* avoid gratuitous rendering synchronization.
* Let's use a user buffer to avoid an unnecessary copy.
*/
if (st->constbuf_uploader) {
cb.buffer = NULL;
cb.user_buffer = NULL;
u_upload_data(st->constbuf_uploader, 0, paramBytes,
params->ParameterValues, &cb.buffer_offset, &cb.buffer);
u_upload_unmap(st->constbuf_uploader);
} else {
cb.buffer = NULL;
cb.user_buffer = params->ParameterValues;
cb.buffer_offset = 0;
}
cb.buffer_size = paramBytes;
if (ST_DEBUG & DEBUG_CONSTANTS) {
debug_printf("%s(shader=%d, numParams=%d, stateFlags=0x%x)\n",
__FUNCTION__, shader_type, params->NumParameters,
params->StateFlags);
_mesa_print_parameter_list(params);
}
cso_set_constant_buffer(st->cso_context, shader_type, 0, &cb);
pipe_resource_reference(&cb.buffer, NULL);
st->state.constants[shader_type].ptr = params->ParameterValues;
st->state.constants[shader_type].size = paramBytes;
}
else if (st->state.constants[shader_type].ptr) {
/* Unbind. */
st->state.constants[shader_type].ptr = NULL;
st->state.constants[shader_type].size = 0;
cso_set_constant_buffer(st->cso_context, shader_type, 0, NULL);
}
}
void
brw_upload_vec4_pull_constants(struct brw_context *brw,
GLbitfield brw_new_constbuf,
const struct gl_program *prog,
struct brw_stage_state *stage_state,
const struct brw_vec4_prog_data *prog_data)
{
int i;
uint32_t surf_index = prog_data->base.binding_table.pull_constants_start;
/* Updates the ParamaterValues[i] pointers for all parameters of the
* basic type of PROGRAM_STATE_VAR.
*/
_mesa_load_state_parameters(&brw->ctx, prog->Parameters);
if (!prog_data->nr_pull_params) {
if (stage_state->const_bo) {
drm_intel_bo_unreference(stage_state->const_bo);
stage_state->const_bo = NULL;
stage_state->surf_offset[surf_index] = 0;
brw->state.dirty.brw |= brw_new_constbuf;
}
return;
}
/* _NEW_PROGRAM_CONSTANTS */
drm_intel_bo_unreference(stage_state->const_bo);
uint32_t size = prog_data->nr_pull_params * 4;
stage_state->const_bo = drm_intel_bo_alloc(brw->bufmgr, "vec4_const_buffer",
size, 64);
drm_intel_gem_bo_map_gtt(stage_state->const_bo);
for (i = 0; i < prog_data->nr_pull_params; i++) {
memcpy(stage_state->const_bo->virtual + i * 4,
prog_data->pull_param[i],
4);
}
if (0) {
for (i = 0; i < ALIGN(prog_data->nr_pull_params, 4) / 4; i++) {
float *row = (float *)stage_state->const_bo->virtual + i * 4;
printf("const surface %3d: %4.3f %4.3f %4.3f %4.3f\n",
i, row[0], row[1], row[2], row[3]);
}
}
drm_intel_gem_bo_unmap_gtt(stage_state->const_bo);
brw_create_constant_surface(brw, stage_state->const_bo, 0, size,
&stage_state->surf_offset[surf_index],
false);
brw->state.dirty.brw |= brw_new_constbuf;
}
static GLboolean r200VertexProgUpdateParams(GLcontext *ctx, struct r200_vertex_program *vp)
{
r200ContextPtr rmesa = R200_CONTEXT( ctx );
GLfloat *fcmd = (GLfloat *)&rmesa->hw.vpp[0].cmd[VPP_CMD_0 + 1];
int pi;
struct gl_vertex_program *mesa_vp = &vp->mesa_program;
struct gl_program_parameter_list *paramList;
drm_radeon_cmd_header_t tmp;
R200_STATECHANGE( rmesa, vpp[0] );
R200_STATECHANGE( rmesa, vpp[1] );
assert(mesa_vp->Base.Parameters);
_mesa_load_state_parameters(ctx, mesa_vp->Base.Parameters);
paramList = mesa_vp->Base.Parameters;
if(paramList->NumParameters > R200_VSF_MAX_PARAM){
fprintf(stderr, "%s:Params exhausted\n", __FUNCTION__);
return GL_FALSE;
}
for(pi = 0; pi < paramList->NumParameters; pi++) {
switch(paramList->Parameters[pi].Type) {
case PROGRAM_STATE_VAR:
case PROGRAM_NAMED_PARAM:
//fprintf(stderr, "%s", vp->Parameters->Parameters[pi].Name);
case PROGRAM_CONSTANT:
*fcmd++ = paramList->ParameterValues[pi][0];
*fcmd++ = paramList->ParameterValues[pi][1];
*fcmd++ = paramList->ParameterValues[pi][2];
*fcmd++ = paramList->ParameterValues[pi][3];
break;
default:
_mesa_problem(NULL, "Bad param type in %s", __FUNCTION__);
break;
}
if (pi == 95) {
fcmd = (GLfloat *)&rmesa->hw.vpp[1].cmd[VPP_CMD_0 + 1];
}
}
/* hack up the cmd_size so not the whole state atom is emitted always. */
rmesa->hw.vpp[0].cmd_size =
1 + 4 * ((paramList->NumParameters > 96) ? 96 : paramList->NumParameters);
tmp.i = rmesa->hw.vpp[0].cmd[VPP_CMD_0];
tmp.veclinear.count = (paramList->NumParameters > 96) ? 96 : paramList->NumParameters;
rmesa->hw.vpp[0].cmd[VPP_CMD_0] = tmp.i;
if (paramList->NumParameters > 96) {
rmesa->hw.vpp[1].cmd_size = 1 + 4 * (paramList->NumParameters - 96);
tmp.i = rmesa->hw.vpp[1].cmd[VPP_CMD_0];
tmp.veclinear.count = paramList->NumParameters - 96;
rmesa->hw.vpp[1].cmd[VPP_CMD_0] = tmp.i;
}
return GL_TRUE;
}
/**
* Creates a temporary BO containing the pull constant data for the shader
* stage, and the SURFACE_STATE struct that points at it.
*
* Pull constants are GLSL uniforms (and other constant data) beyond what we
* could fit as push constants, or that have variable-index array access
* (which is easiest to support using pull constants, and avoids filling
* register space with mostly-unused data).
*
* Compare this path to brw_curbe.c for gen4/5 push constants, and
* gen6_vs_state.c for gen6+ push constants.
*/
void
brw_upload_pull_constants(struct brw_context *brw,
GLbitfield brw_new_constbuf,
const struct gl_program *prog,
struct brw_stage_state *stage_state,
const struct brw_stage_prog_data *prog_data,
bool dword_pitch)
{
int i;
uint32_t surf_index = prog_data->binding_table.pull_constants_start;
if (!prog_data->nr_pull_params) {
if (stage_state->surf_offset[surf_index]) {
stage_state->surf_offset[surf_index] = 0;
brw->ctx.NewDriverState |= brw_new_constbuf;
}
return;
}
/* Updates the ParamaterValues[i] pointers for all parameters of the
* basic type of PROGRAM_STATE_VAR.
*/
_mesa_load_state_parameters(&brw->ctx, prog->Parameters);
/* BRW_NEW_*_PROG_DATA | _NEW_PROGRAM_CONSTANTS */
uint32_t size = prog_data->nr_pull_params * 4;
drm_intel_bo *const_bo = NULL;
uint32_t const_offset;
gl_constant_value *constants = intel_upload_space(brw, size, 64,
&const_bo, &const_offset);
STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
for (i = 0; i < prog_data->nr_pull_params; i++) {
constants[i] = *prog_data->pull_param[i];
}
if (0) {
for (i = 0; i < ALIGN(prog_data->nr_pull_params, 4) / 4; i++) {
const gl_constant_value *row = &constants[i * 4];
fprintf(stderr, "const surface %3d: %4.3f %4.3f %4.3f %4.3f\n",
i, row[0].f, row[1].f, row[2].f, row[3].f);
}
}
brw_create_constant_surface(brw, const_bo, const_offset, size,
&stage_state->surf_offset[surf_index],
dword_pitch);
drm_intel_bo_unreference(const_bo);
brw->ctx.NewDriverState |= brw_new_constbuf;
}
static void
gen6_upload_wm_push_constants(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_FRAGMENT_PROGRAM */
const struct brw_fragment_program *fp =
brw_fragment_program_const(brw->fragment_program);
/* CACHE_NEW_WM_PROG */
const struct brw_wm_prog_data *prog_data = brw->wm.prog_data;
/* Updates the ParameterValues[i] pointers for all parameters of the
* basic type of PROGRAM_STATE_VAR.
*/
/* XXX: Should this happen somewhere before to get our state flag set? */
_mesa_load_state_parameters(ctx, fp->program.Base.Parameters);
if (prog_data->base.nr_params == 0) {
brw->wm.base.push_const_size = 0;
} else {
float *constants;
unsigned int i;
constants = brw_state_batch(brw, AUB_TRACE_WM_CONSTANTS,
prog_data->base.nr_params * sizeof(float),
32, &brw->wm.base.push_const_offset);
for (i = 0; i < prog_data->base.nr_params; i++) {
constants[i] = *prog_data->base.param[i];
}
if (0) {
fprintf(stderr, "WM constants:\n");
for (i = 0; i < prog_data->base.nr_params; i++) {
if ((i & 7) == 0)
fprintf(stderr, "g%d: ", prog_data->first_curbe_grf + i / 8);
fprintf(stderr, "%8f ", constants[i]);
if ((i & 7) == 7)
fprintf(stderr, "\n");
}
if ((i & 7) != 0)
fprintf(stderr, "\n");
fprintf(stderr, "\n");
}
brw->wm.base.push_const_size = ALIGN(prog_data->base.nr_params, 8) / 8;
}
if (brw->gen >= 7) {
gen7_upload_constant_state(brw, &brw->wm.base, true,
_3DSTATE_CONSTANT_PS);
}
}
/**
* Update state for running fragment programs. Basically, load the
* program parameters with current state values.
*/
static void
_swrast_update_fragment_program(GLcontext *ctx, GLbitfield newState)
{
const struct gl_fragment_program *fp = ctx->FragmentProgram._Current;
if (fp) {
#if 0
/* XXX Need a way to trigger the initial loading of parameters
* even when there's no recent state changes.
*/
if (fp->Base.Parameters->StateFlags & newState)
#endif
_mesa_load_state_parameters(ctx, fp->Base.Parameters);
}
}
static void track_params( struct i915_fragment_program *p )
{
GLint i;
if (p->nr_params)
_mesa_load_state_parameters(p->ctx, p->FragProg.Parameters);
for (i = 0; i < p->nr_params; i++) {
GLint reg = p->param[i].reg;
COPY_4V( p->constant[reg], p->param[i].values );
}
p->params_uptodate = 1;
p->on_hardware = 0; /* overkill */
}
static void
gen6_upload_wm_push_constants(struct brw_context *brw)
{
struct intel_context *intel = &brw->intel;
struct gl_context *ctx = &intel->ctx;
/* BRW_NEW_FRAGMENT_PROGRAM */
const struct brw_fragment_program *fp =
brw_fragment_program_const(brw->fragment_program);
/* Updates the ParameterValues[i] pointers for all parameters of the
* basic type of PROGRAM_STATE_VAR.
*/
/* XXX: Should this happen somewhere before to get our state flag set? */
_mesa_load_state_parameters(ctx, fp->program.Base.Parameters);
/* CACHE_NEW_WM_PROG */
if (brw->wm.prog_data->nr_params != 0) {
float *constants;
unsigned int i;
constants = brw_state_batch(brw, AUB_TRACE_WM_CONSTANTS,
brw->wm.prog_data->nr_params *
sizeof(float),
32, &brw->wm.push_const_offset);
for (i = 0; i < brw->wm.prog_data->nr_params; i++) {
constants[i] = convert_param(brw->wm.prog_data->param_convert[i],
brw->wm.prog_data->param[i]);
}
if (0) {
printf("WM constants:\n");
for (i = 0; i < brw->wm.prog_data->nr_params; i++) {
if ((i & 7) == 0)
printf("g%d: ", brw->wm.prog_data->first_curbe_grf + i / 8);
printf("%8f ", constants[i]);
if ((i & 7) == 7)
printf("\n");
}
if ((i & 7) != 0)
printf("\n");
printf("\n");
}
}
}
/**
* Pass the given program parameters to the graphics pipe as a
* constant buffer.
* \param shader_type either PIPE_SHADER_VERTEX or PIPE_SHADER_FRAGMENT
*/
void st_upload_constants( struct st_context *st,
struct gl_program_parameter_list *params,
unsigned shader_type)
{
struct pipe_context *pipe = st->pipe;
struct pipe_resource **cbuf = &st->state.constants[shader_type];
assert(shader_type == PIPE_SHADER_VERTEX ||
shader_type == PIPE_SHADER_FRAGMENT ||
shader_type == PIPE_SHADER_GEOMETRY);
/* update constants */
if (params && params->NumParameters) {
const uint paramBytes = params->NumParameters * sizeof(GLfloat) * 4;
_mesa_load_state_parameters(st->ctx, params);
/* We always need to get a new buffer, to keep the drivers simple and
* avoid gratuitous rendering synchronization.
*/
pipe_resource_reference(cbuf, NULL );
*cbuf = pipe_buffer_create(pipe->screen,
PIPE_BIND_CONSTANT_BUFFER,
paramBytes );
if (ST_DEBUG & DEBUG_CONSTANTS) {
debug_printf("%s(shader=%d, numParams=%d, stateFlags=0x%x)\n",
__FUNCTION__, shader_type, params->NumParameters,
params->StateFlags);
_mesa_print_parameter_list(params);
}
/* load Mesa constants into the constant buffer */
pipe_buffer_write(st->pipe, *cbuf,
0, paramBytes,
params->ParameterValues);
st->pipe->set_constant_buffer(st->pipe, shader_type, 0, *cbuf);
}
else {
st->constants.tracked_state[shader_type].dirty.mesa = 0x0;
}
}
/**
* Pass the given program parameters to the graphics pipe as a
* constant buffer.
* \param id either PIPE_SHADER_VERTEX or PIPE_SHADER_FRAGMENT
*/
void st_upload_constants( struct st_context *st,
struct gl_program_parameter_list *params,
unsigned id)
{
struct pipe_context *pipe = st->pipe;
struct pipe_constant_buffer *cbuf = &st->state.constants[id];
assert(id == PIPE_SHADER_VERTEX || id == PIPE_SHADER_FRAGMENT);
/* update constants */
if (params && params->NumParameters) {
const uint paramBytes = params->NumParameters * sizeof(GLfloat) * 4;
_mesa_load_state_parameters(st->ctx, params);
/* We always need to get a new buffer, to keep the drivers simple and
* avoid gratuitous rendering synchronization.
*/
pipe_buffer_reference(&cbuf->buffer, NULL );
cbuf->buffer = pipe_buffer_create(pipe->screen, 16, PIPE_BUFFER_USAGE_CONSTANT,
paramBytes );
if (0)
{
printf("%s(shader=%d, numParams=%d, stateFlags=0x%x)\n",
__FUNCTION__, id, params->NumParameters, params->StateFlags);
_mesa_print_parameter_list(params);
}
/* load Mesa constants into the constant buffer */
if (cbuf->buffer)
st_no_flush_pipe_buffer_write(st, cbuf->buffer,
0, paramBytes,
params->ParameterValues);
st->pipe->set_constant_buffer(st->pipe, id, 0, cbuf);
}
else {
st->constants.tracked_state[id].dirty.mesa = 0;
// st->pipe->set_constant_buffer(st->pipe, id, 0, NULL);
}
}
/**
* Write parameter array for the given vertex program into dst.
* Return the total number of components written.
*/
static int r300VertexProgUpdateParams(GLcontext * ctx, struct r300_vertex_program *vp, float *dst)
{
int i;
if (vp->Base->IsNVProgram) {
_mesa_load_tracked_matrices(ctx);
} else {
if (vp->Base->Base.Parameters) {
_mesa_load_state_parameters(ctx, vp->Base->Base.Parameters);
}
}
for(i = 0; i < vp->code.constants.Count; ++i) {
const float * src = 0;
const struct rc_constant * constant = &vp->code.constants.Constants[i];
switch(constant->Type) {
case RC_CONSTANT_EXTERNAL:
if (vp->Base->IsNVProgram) {
src = ctx->VertexProgram.Parameters[constant->u.External];
} else {
src = vp->Base->Base.Parameters->ParameterValues[constant->u.External];
}
break;
case RC_CONSTANT_IMMEDIATE:
src = constant->u.Immediate;
break;
}
dst[4*i] = src[0];
dst[4*i + 1] = src[1];
dst[4*i + 2] = src[2];
dst[4*i + 3] = src[3];
}
return 4 * vp->code.constants.Count;
}
/**
* Print all of a program's parameters/fields to given file.
*/
static void
_mesa_fprint_program_parameters(FILE *f,
struct gl_context *ctx,
const struct gl_program *prog)
{
GLuint i;
fprintf(f, "InputsRead: 0x%llx (0b%s)\n",
(unsigned long long) prog->InputsRead, binary(prog->InputsRead));
fprintf(f, "OutputsWritten: 0x%llx (0b%s)\n",
(unsigned long long)prog->OutputsWritten,
binary(prog->OutputsWritten));
fprintf(f, "NumInstructions=%d\n", prog->NumInstructions);
fprintf(f, "NumTemporaries=%d\n", prog->NumTemporaries);
fprintf(f, "NumParameters=%d\n", prog->NumParameters);
fprintf(f, "NumAttributes=%d\n", prog->NumAttributes);
fprintf(f, "NumAddressRegs=%d\n", prog->NumAddressRegs);
fprintf(f, "IndirectRegisterFiles: 0x%x (0b%s)\n",
prog->IndirectRegisterFiles, binary(prog->IndirectRegisterFiles));
fprintf(f, "SamplersUsed: 0x%x (0b%s)\n",
prog->SamplersUsed, binary(prog->SamplersUsed));
fprintf(f, "Samplers=[ ");
for (i = 0; i < MAX_SAMPLERS; i++) {
fprintf(f, "%d ", prog->SamplerUnits[i]);
}
fprintf(f, "]\n");
_mesa_load_state_parameters(ctx, prog->Parameters);
#if 0
fprintf(f, "Local Params:\n");
for (i = 0; i < MAX_PROGRAM_LOCAL_PARAMS; i++){
const GLfloat *p = prog->LocalParams[i];
fprintf(f, "%2d: %f, %f, %f, %f\n", i, p[0], p[1], p[2], p[3]);
}
#endif
_mesa_print_parameter_list(prog->Parameters);
}
/**
* This function executes vertex programs
*/
static GLboolean
run_vp( struct gl_context *ctx, struct tnl_pipeline_stage *stage )
{
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vp_stage_data *store = VP_STAGE_DATA(stage);
struct vertex_buffer *VB = &tnl->vb;
struct gl_vertex_program *program = ctx->VertexProgram._Current;
struct gl_program_machine *machine = &store->machine;
GLuint outputs[VARYING_SLOT_MAX], numOutputs;
GLuint i, j;
if (!program)
return GL_TRUE;
/* ARB program or vertex shader */
_mesa_load_state_parameters(ctx, program->Base.Parameters);
/* make list of outputs to save some time below */
numOutputs = 0;
for (i = 0; i < VARYING_SLOT_MAX; i++) {
if (program->Base.OutputsWritten & BITFIELD64_BIT(i)) {
outputs[numOutputs++] = i;
}
}
/* Allocate result vectors. We delay this until now to avoid allocating
* memory that would never be used if we don't run the software tnl pipeline.
*/
if (!store->results[0].storage) {
for (i = 0; i < VARYING_SLOT_MAX; i++) {
assert(!store->results[i].storage);
_mesa_vector4f_alloc( &store->results[i], 0, VB->Size, 32 );
store->results[i].size = 4;
}
}
map_textures(ctx, program);
for (i = 0; i < VB->Count; i++) {
GLuint attr;
init_machine(ctx, machine, tnl->CurInstance);
#if 0
printf("Input %d: %f, %f, %f, %f\n", i,
VB->AttribPtr[0]->data[i][0],
VB->AttribPtr[0]->data[i][1],
VB->AttribPtr[0]->data[i][2],
VB->AttribPtr[0]->data[i][3]);
printf(" color: %f, %f, %f, %f\n",
VB->AttribPtr[3]->data[i][0],
VB->AttribPtr[3]->data[i][1],
VB->AttribPtr[3]->data[i][2],
VB->AttribPtr[3]->data[i][3]);
printf(" normal: %f, %f, %f, %f\n",
VB->AttribPtr[2]->data[i][0],
VB->AttribPtr[2]->data[i][1],
VB->AttribPtr[2]->data[i][2],
VB->AttribPtr[2]->data[i][3]);
#endif
/* the vertex array case */
for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
if (program->Base.InputsRead & BITFIELD64_BIT(attr)) {
const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
const GLuint size = VB->AttribPtr[attr]->size;
const GLuint stride = VB->AttribPtr[attr]->stride;
const GLfloat *data = (GLfloat *) (ptr + stride * i);
#ifdef NAN_CHECK
check_float(data[0]);
check_float(data[1]);
check_float(data[2]);
check_float(data[3]);
#endif
COPY_CLEAN_4V(machine->VertAttribs[attr], size, data);
}
}
/* execute the program */
_mesa_execute_program(ctx, &program->Base, machine);
/* copy the output registers into the VB->attribs arrays */
for (j = 0; j < numOutputs; j++) {
const GLuint attr = outputs[j];
#ifdef NAN_CHECK
check_float(machine->Outputs[attr][0]);
check_float(machine->Outputs[attr][1]);
check_float(machine->Outputs[attr][2]);
check_float(machine->Outputs[attr][3]);
#endif
COPY_4V(store->results[attr].data[i], machine->Outputs[attr]);
}
/* FOGC is a special case. Fragment shader expects (f,0,0,1) */
if (program->Base.OutputsWritten & BITFIELD64_BIT(VARYING_SLOT_FOGC)) {
store->results[VARYING_SLOT_FOGC].data[i][1] = 0.0;
store->results[VARYING_SLOT_FOGC].data[i][2] = 0.0;
store->results[VARYING_SLOT_FOGC].data[i][3] = 1.0;
}
#ifdef NAN_CHECK
ASSERT(machine->Outputs[0][3] != 0.0F);
#endif
#if 0
printf("HPOS: %f %f %f %f\n",
machine->Outputs[0][0],
machine->Outputs[0][1],
machine->Outputs[0][2],
machine->Outputs[0][3]);
#endif
}
unmap_textures(ctx, program);
if (program->IsPositionInvariant) {
/* We need the exact same transform as in the fixed function path here
* to guarantee invariance, depending on compiler optimization flags
* results could be different otherwise.
*/
VB->ClipPtr = TransformRaw( &store->results[0],
&ctx->_ModelProjectMatrix,
VB->AttribPtr[0] );
/* Drivers expect this to be clean to element 4...
*/
switch (VB->ClipPtr->size) {
case 1:
/* impossible */
case 2:
_mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 2 );
/* fall-through */
case 3:
_mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 3 );
/* fall-through */
case 4:
break;
}
}
else {
/* Setup the VB pointers so that the next pipeline stages get
* their data from the right place (the program output arrays).
*/
VB->ClipPtr = &store->results[VARYING_SLOT_POS];
VB->ClipPtr->size = 4;
VB->ClipPtr->count = VB->Count;
}
VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VARYING_SLOT_COL0];
VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VARYING_SLOT_COL1];
VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VARYING_SLOT_FOGC];
VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VARYING_SLOT_PSIZ];
VB->BackfaceColorPtr = &store->results[VARYING_SLOT_BFC0];
VB->BackfaceSecondaryColorPtr = &store->results[VARYING_SLOT_BFC1];
for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
= &store->results[VARYING_SLOT_TEX0 + i];
}
for (i = 0; i < ctx->Const.MaxVarying; i++) {
if (program->Base.OutputsWritten & BITFIELD64_BIT(VARYING_SLOT_VAR0 + i)) {
/* Note: varying results get put into the generic attributes */
VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
= &store->results[VARYING_SLOT_VAR0 + i];
}
}
/* Perform NDC and cliptest operations:
*/
return do_ndc_cliptest(ctx, store);
}
/**
* Gathers together all the uniform values into a block of memory to be
* uploaded into the CURBE, then emits the state packet telling the hardware
* the new location.
*/
static void
brw_upload_constant_buffer(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_CURBE_OFFSETS */
const GLuint sz = brw->curbe.total_size;
const GLuint bufsz = sz * 16 * sizeof(GLfloat);
gl_constant_value *buf;
GLuint i;
gl_clip_plane *clip_planes;
if (sz == 0) {
goto emit;
}
buf = intel_upload_space(brw, bufsz, 64,
&brw->curbe.curbe_bo, &brw->curbe.curbe_offset);
STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
/* fragment shader constants */
if (brw->curbe.wm_size) {
_mesa_load_state_parameters(ctx, brw->fragment_program->Base.Parameters);
/* BRW_NEW_CURBE_OFFSETS */
GLuint offset = brw->curbe.wm_start * 16;
/* BRW_NEW_FS_PROG_DATA | _NEW_PROGRAM_CONSTANTS: copy uniform values */
for (i = 0; i < brw->wm.prog_data->base.nr_params; i++) {
buf[offset + i] = *brw->wm.prog_data->base.param[i];
}
}
/* clipper constants */
if (brw->curbe.clip_size) {
GLuint offset = brw->curbe.clip_start * 16;
GLuint j;
/* If any planes are going this way, send them all this way:
*/
for (i = 0; i < 6; i++) {
buf[offset + i * 4 + 0].f = fixed_plane[i][0];
buf[offset + i * 4 + 1].f = fixed_plane[i][1];
buf[offset + i * 4 + 2].f = fixed_plane[i][2];
buf[offset + i * 4 + 3].f = fixed_plane[i][3];
}
/* Clip planes: _NEW_TRANSFORM plus _NEW_PROJECTION to get to
* clip-space:
*/
clip_planes = brw_select_clip_planes(ctx);
for (j = 0; j < MAX_CLIP_PLANES; j++) {
if (ctx->Transform.ClipPlanesEnabled & (1<<j)) {
buf[offset + i * 4 + 0].f = clip_planes[j][0];
buf[offset + i * 4 + 1].f = clip_planes[j][1];
buf[offset + i * 4 + 2].f = clip_planes[j][2];
buf[offset + i * 4 + 3].f = clip_planes[j][3];
i++;
}
}
}
/* vertex shader constants */
if (brw->curbe.vs_size) {
_mesa_load_state_parameters(ctx, brw->vertex_program->Base.Parameters);
GLuint offset = brw->curbe.vs_start * 16;
/* BRW_NEW_VS_PROG_DATA | _NEW_PROGRAM_CONSTANTS: copy uniform values */
for (i = 0; i < brw->vs.prog_data->base.base.nr_params; i++) {
buf[offset + i] = *brw->vs.prog_data->base.base.param[i];
}
}
if (0) {
for (i = 0; i < sz*16; i+=4)
fprintf(stderr, "curbe %d.%d: %f %f %f %f\n", i/8, i&4,
buf[i+0].f, buf[i+1].f, buf[i+2].f, buf[i+3].f);
}
/* Because this provokes an action (ie copy the constants into the
* URB), it shouldn't be shortcircuited if identical to the
* previous time - because eg. the urb destination may have
* changed, or the urb contents different to last time.
*
* Note that the data referred to is actually copied internally,
* not just used in place according to passed pointer.
*
* It appears that the CS unit takes care of using each available
* URB entry (Const URB Entry == CURBE) in turn, and issuing
* flushes as necessary when doublebuffering of CURBEs isn't
* possible.
*/
emit:
/* Work around mysterious 965 hangs that appear to happen if you do
* two 3DPRIMITIVEs with only a CONSTANT_BUFFER inbetween. If we
* haven't already flushed for some other reason, explicitly do so.
*
* We've found no documented reason why this should be necessary.
*/
if (brw->gen == 4 && !brw->is_g4x &&
(brw->ctx.NewDriverState & (BRW_NEW_BATCH | BRW_NEW_PSP)) == 0) {
BEGIN_BATCH(1);
OUT_BATCH(MI_FLUSH);
ADVANCE_BATCH();
}
/* BRW_NEW_URB_FENCE: From the gen4 PRM, volume 1, section 3.9.8
* (CONSTANT_BUFFER (CURBE Load)):
*
* "Modifying the CS URB allocation via URB_FENCE invalidates any
* previous CURBE entries. Therefore software must subsequently
* [re]issue a CONSTANT_BUFFER command before CURBE data can be used
* in the pipeline."
*/
BEGIN_BATCH(2);
if (brw->curbe.total_size == 0) {
OUT_BATCH((CMD_CONST_BUFFER << 16) | (2 - 2));
OUT_BATCH(0);
} else {
OUT_BATCH((CMD_CONST_BUFFER << 16) | (1 << 8) | (2 - 2));
OUT_RELOC(brw->curbe.curbe_bo,
I915_GEM_DOMAIN_INSTRUCTION, 0,
(brw->curbe.total_size - 1) + brw->curbe.curbe_offset);
}
ADVANCE_BATCH();
}
/* Upload a new set of constants. Too much variability to go into the
* cache mechanism, but maybe would benefit from a comparison against
* the current uploaded set of constants.
*/
static void upload_constant_buffer(struct brw_context *brw)
{
GLcontext *ctx = &brw->intel.ctx;
struct brw_vertex_program *vp = (struct brw_vertex_program *)brw->vertex_program;
struct brw_fragment_program *fp = (struct brw_fragment_program *)brw->fragment_program;
struct brw_mem_pool *pool = &brw->pool[BRW_GS_POOL];
GLuint sz = brw->curbe.total_size;
GLuint bufsz = sz * 16 * sizeof(GLfloat);
GLfloat *buf;
GLuint i;
/* Update our own dependency flags. This works because this
* function will also be called whenever fp or vp changes.
*/
brw->curbe.tracked_state.dirty.mesa = (_NEW_TRANSFORM|_NEW_PROJECTION);
brw->curbe.tracked_state.dirty.mesa |= vp->param_state;
brw->curbe.tracked_state.dirty.mesa |= fp->param_state;
if (sz == 0) {
struct brw_constant_buffer cb;
cb.header.opcode = CMD_CONST_BUFFER;
cb.header.length = sizeof(cb)/4 - 2;
cb.header.valid = 0;
cb.bits0.buffer_length = 0;
cb.bits0.buffer_address = 0;
BRW_BATCH_STRUCT(brw, &cb);
if (brw->curbe.last_buf) {
free(brw->curbe.last_buf);
brw->curbe.last_buf = NULL;
brw->curbe.last_bufsz = 0;
}
return;
}
buf = (GLfloat *)malloc(bufsz);
memset(buf, 0, bufsz);
if (brw->curbe.wm_size) {
GLuint offset = brw->curbe.wm_start * 16;
_mesa_load_state_parameters(ctx, fp->program.Base.Parameters);
for (i = 0; i < brw->wm.prog_data->nr_params; i++)
buf[offset + i] = brw->wm.prog_data->param[i][0];
}
/* The clipplanes are actually delivered to both CLIP and VS units.
* VS uses them to calculate the outcode bitmasks.
*/
if (brw->curbe.clip_size) {
GLuint offset = brw->curbe.clip_start * 16;
GLuint j;
/* If any planes are going this way, send them all this way:
*/
for (i = 0; i < 6; i++) {
buf[offset + i * 4 + 0] = fixed_plane[i][0];
buf[offset + i * 4 + 1] = fixed_plane[i][1];
buf[offset + i * 4 + 2] = fixed_plane[i][2];
buf[offset + i * 4 + 3] = fixed_plane[i][3];
}
/* Clip planes: _NEW_TRANSFORM plus _NEW_PROJECTION to get to
* clip-space:
*/
assert(MAX_CLIP_PLANES == 6);
for (j = 0; j < MAX_CLIP_PLANES; j++) {
if (brw->attribs.Transform->ClipPlanesEnabled & (1<<j)) {
buf[offset + i * 4 + 0] = brw->attribs.Transform->_ClipUserPlane[j][0];
buf[offset + i * 4 + 1] = brw->attribs.Transform->_ClipUserPlane[j][1];
buf[offset + i * 4 + 2] = brw->attribs.Transform->_ClipUserPlane[j][2];
buf[offset + i * 4 + 3] = brw->attribs.Transform->_ClipUserPlane[j][3];
i++;
}
}
}
if (brw->curbe.vs_size) {
GLuint offset = brw->curbe.vs_start * 16;
GLuint nr = vp->program.Base.Parameters->NumParameters;
_mesa_load_state_parameters(ctx, vp->program.Base.Parameters);
for (i = 0; i < nr; i++) {
buf[offset + i * 4 + 0] = vp->program.Base.Parameters->ParameterValues[i][0];
buf[offset + i * 4 + 1] = vp->program.Base.Parameters->ParameterValues[i][1];
buf[offset + i * 4 + 2] = vp->program.Base.Parameters->ParameterValues[i][2];
buf[offset + i * 4 + 3] = vp->program.Base.Parameters->ParameterValues[i][3];
}
}
if (0) {
for (i = 0; i < sz*16; i+=4)
_mesa_printf("curbe %d.%d: %f %f %f %f\n", i/8, i&4,
buf[i+0], buf[i+1], buf[i+2], buf[i+3]);
_mesa_printf("last_buf %p buf %p sz %d/%d cmp %d\n",
brw->curbe.last_buf, buf,
bufsz, brw->curbe.last_bufsz,
brw->curbe.last_buf ? memcmp(buf, brw->curbe.last_buf, bufsz) : -1);
}
if (brw->curbe.last_buf &&
bufsz == brw->curbe.last_bufsz &&
memcmp(buf, brw->curbe.last_buf, bufsz) == 0) {
free(buf);
/* return; */
}
else {
if (brw->curbe.last_buf)
free(brw->curbe.last_buf);
brw->curbe.last_buf = buf;
brw->curbe.last_bufsz = bufsz;
if (!brw_pool_alloc(pool,
bufsz,
6,
&brw->curbe.gs_offset)) {
_mesa_printf("out of GS memory for curbe\n");
assert(0);
return;
}
/* Copy data to the buffer:
*/
bmBufferSubDataAUB(&brw->intel,
pool->buffer,
brw->curbe.gs_offset,
bufsz,
buf,
DW_CONSTANT_BUFFER,
0);
}
/* TODO: only emit the constant_buffer packet when necessary, ie:
- contents have changed
- offset has changed
- hw requirements due to other packets emitted.
*/
{
struct brw_constant_buffer cb;
memset(&cb, 0, sizeof(cb));
cb.header.opcode = CMD_CONST_BUFFER;
cb.header.length = sizeof(cb)/4 - 2;
cb.header.valid = 1;
cb.bits0.buffer_length = sz - 1;
cb.bits0.buffer_address = brw->curbe.gs_offset >> 6;
/* Because this provokes an action (ie copy the constants into the
* URB), it shouldn't be shortcircuited if identical to the
* previous time - because eg. the urb destination may have
* changed, or the urb contents different to last time.
*
* Note that the data referred to is actually copied internally,
* not just used in place according to passed pointer.
*
* It appears that the CS unit takes care of using each available
* URB entry (Const URB Entry == CURBE) in turn, and issuing
* flushes as necessary when doublebuffering of CURBEs isn't
* possible.
*/
/* intel_batchbuffer_align(brw->intel.batch, 64, sizeof(cb)); */
BRW_BATCH_STRUCT(brw, &cb);
/* intel_batchbuffer_align(brw->intel.batch, 64, 0); */
}
}
/**
* Creates a region containing the push constants for the CS on gen7+.
*
* Push constants are constant values (such as GLSL uniforms) that are
* pre-loaded into a shader stage's register space at thread spawn time.
*
* For other stages, see brw_curbe.c:brw_upload_constant_buffer for the
* equivalent gen4/5 code and gen6_vs_state.c:gen6_upload_push_constants for
* gen6+.
*/
static void
brw_upload_cs_push_constants(struct brw_context *brw,
const struct gl_program *prog,
const struct brw_cs_prog_data *cs_prog_data,
struct brw_stage_state *stage_state,
enum aub_state_struct_type type)
{
struct gl_context *ctx = &brw->ctx;
const struct brw_stage_prog_data *prog_data =
(struct brw_stage_prog_data*) cs_prog_data;
/* Updates the ParamaterValues[i] pointers for all parameters of the
* basic type of PROGRAM_STATE_VAR.
*/
/* XXX: Should this happen somewhere before to get our state flag set? */
_mesa_load_state_parameters(ctx, prog->Parameters);
if (cs_prog_data->push.total.size == 0) {
stage_state->push_const_size = 0;
return;
}
gl_constant_value *param = (gl_constant_value*)
brw_state_batch(brw, type, ALIGN(cs_prog_data->push.total.size, 64),
64, &stage_state->push_const_offset);
assert(param);
STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
if (cs_prog_data->push.cross_thread.size > 0) {
gl_constant_value *param_copy = param;
assert(cs_prog_data->thread_local_id_index < 0 ||
cs_prog_data->thread_local_id_index >=
cs_prog_data->push.cross_thread.dwords);
for (unsigned i = 0;
i < cs_prog_data->push.cross_thread.dwords;
i++) {
param_copy[i] = *prog_data->param[i];
}
}
gl_constant_value thread_id;
if (cs_prog_data->push.per_thread.size > 0) {
for (unsigned t = 0; t < cs_prog_data->threads; t++) {
unsigned dst =
8 * (cs_prog_data->push.per_thread.regs * t +
cs_prog_data->push.cross_thread.regs);
unsigned src = cs_prog_data->push.cross_thread.dwords;
for ( ; src < prog_data->nr_params; src++, dst++) {
if (src != cs_prog_data->thread_local_id_index)
param[dst] = *prog_data->param[src];
else {
thread_id.u = t * cs_prog_data->simd_size;
param[dst] = thread_id;
}
}
}
}
stage_state->push_const_size =
cs_prog_data->push.cross_thread.regs +
cs_prog_data->push.per_thread.regs;
}
GLboolean r700SetupFragmentProgram(GLcontext * ctx)
{
context_t *context = R700_CONTEXT(ctx);
R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
struct r700_fragment_program *fp = (struct r700_fragment_program *)
(ctx->FragmentProgram._Current);
r700_AssemblerBase *pAsm = &(fp->r700AsmCode);
struct gl_fragment_program *mesa_fp = &(fp->mesa_program);
struct gl_program_parameter_list *paramList;
unsigned int unNumParamData;
unsigned int ui, i;
unsigned int unNumOfReg;
unsigned int unBit;
GLuint exportCount;
GLboolean point_sprite = GL_FALSE;
if(GL_FALSE == fp->loaded)
{
if(fp->r700Shader.bNeedsAssembly == GL_TRUE)
{
Assemble( &(fp->r700Shader) );
}
/* Load fp to gpu */
r600EmitShader(ctx,
&(fp->shaderbo),
(GLvoid *)(fp->r700Shader.pProgram),
fp->r700Shader.uShaderBinaryDWORDSize,
"FS");
fp->loaded = GL_TRUE;
}
DumpHwBinary(DUMP_PIXEL_SHADER, (GLvoid *)(fp->r700Shader.pProgram),
fp->r700Shader.uShaderBinaryDWORDSize);
/* TODO : enable this after MemUse fixed *=
(context->chipobj.MemUse)(context, fp->shadercode.buf->id);
*/
R600_STATECHANGE(context, ps);
r700->ps.SQ_PGM_RESOURCES_PS.u32All = 0;
SETbit(r700->ps.SQ_PGM_RESOURCES_PS.u32All, PGM_RESOURCES__PRIME_CACHE_ON_DRAW_bit);
r700->ps.SQ_PGM_START_PS.u32All = 0; /* set from buffer obj */
R600_STATECHANGE(context, spi);
unNumOfReg = fp->r700Shader.nRegs + 1;
ui = (r700->SPI_PS_IN_CONTROL_0.u32All & NUM_INTERP_mask) / (1 << NUM_INTERP_shift);
/* PS uses fragment.position */
if (mesa_fp->Base.InputsRead & (1 << FRAG_ATTRIB_WPOS))
{
ui += 1;
SETfield(r700->SPI_PS_IN_CONTROL_0.u32All, ui, NUM_INTERP_shift, NUM_INTERP_mask);
SETfield(r700->SPI_PS_IN_CONTROL_0.u32All, CENTERS_ONLY, BARYC_SAMPLE_CNTL_shift, BARYC_SAMPLE_CNTL_mask);
SETbit(r700->SPI_PS_IN_CONTROL_0.u32All, POSITION_ENA_bit);
SETbit(r700->SPI_INPUT_Z.u32All, PROVIDE_Z_TO_SPI_bit);
}
else
{
CLEARbit(r700->SPI_PS_IN_CONTROL_0.u32All, POSITION_ENA_bit);
CLEARbit(r700->SPI_INPUT_Z.u32All, PROVIDE_Z_TO_SPI_bit);
}
if (mesa_fp->Base.InputsRead & (1 << FRAG_ATTRIB_FACE))
{
ui += 1;
SETfield(r700->SPI_PS_IN_CONTROL_0.u32All, ui, NUM_INTERP_shift, NUM_INTERP_mask);
SETbit(r700->SPI_PS_IN_CONTROL_1.u32All, FRONT_FACE_ENA_bit);
SETbit(r700->SPI_PS_IN_CONTROL_1.u32All, FRONT_FACE_ALL_BITS_bit);
SETfield(r700->SPI_PS_IN_CONTROL_1.u32All, pAsm->uiFP_AttributeMap[FRAG_ATTRIB_FACE], FRONT_FACE_ADDR_shift, FRONT_FACE_ADDR_mask);
}
else
{
CLEARbit(r700->SPI_PS_IN_CONTROL_1.u32All, FRONT_FACE_ENA_bit);
}
/* see if we need any point_sprite replacements, also increase num_interp
* as there's no vp output for them */
if (ctx->Point.PointSprite)
{
for (i = FRAG_ATTRIB_TEX0; i<= FRAG_ATTRIB_TEX7; i++)
{
if (ctx->Point.CoordReplace[i - FRAG_ATTRIB_TEX0] == GL_TRUE)
{
ui++;
point_sprite = GL_TRUE;
}
}
}
if( mesa_fp->Base.InputsRead & (1 << FRAG_ATTRIB_PNTC))
ui++;
if ((mesa_fp->Base.InputsRead & (1 << FRAG_ATTRIB_PNTC)) || point_sprite)
{
SETfield(r700->SPI_PS_IN_CONTROL_0.u32All, ui, NUM_INTERP_shift, NUM_INTERP_mask);
SETbit(r700->SPI_INTERP_CONTROL_0.u32All, PNT_SPRITE_ENA_bit);
SETfield(r700->SPI_INTERP_CONTROL_0.u32All, SPI_PNT_SPRITE_SEL_S, PNT_SPRITE_OVRD_X_shift, PNT_SPRITE_OVRD_X_mask);
SETfield(r700->SPI_INTERP_CONTROL_0.u32All, SPI_PNT_SPRITE_SEL_T, PNT_SPRITE_OVRD_Y_shift, PNT_SPRITE_OVRD_Y_mask);
SETfield(r700->SPI_INTERP_CONTROL_0.u32All, SPI_PNT_SPRITE_SEL_0, PNT_SPRITE_OVRD_Z_shift, PNT_SPRITE_OVRD_Z_mask);
SETfield(r700->SPI_INTERP_CONTROL_0.u32All, SPI_PNT_SPRITE_SEL_1, PNT_SPRITE_OVRD_W_shift, PNT_SPRITE_OVRD_W_mask);
/* Like e.g. viewport and winding, point sprite coordinates are
* inverted when rendering to FBO. */
if ((ctx->Point.SpriteOrigin == GL_LOWER_LEFT) == !ctx->DrawBuffer->Name)
SETbit(r700->SPI_INTERP_CONTROL_0.u32All, PNT_SPRITE_TOP_1_bit);
else
CLEARbit(r700->SPI_INTERP_CONTROL_0.u32All, PNT_SPRITE_TOP_1_bit);
}
else
{
CLEARbit(r700->SPI_INTERP_CONTROL_0.u32All, PNT_SPRITE_ENA_bit);
}
ui = (unNumOfReg < ui) ? ui : unNumOfReg;
SETfield(r700->ps.SQ_PGM_RESOURCES_PS.u32All, ui, NUM_GPRS_shift, NUM_GPRS_mask);
CLEARbit(r700->ps.SQ_PGM_RESOURCES_PS.u32All, UNCACHED_FIRST_INST_bit);
if(fp->r700Shader.uStackSize) /* we don't use branch for now, it should be zero. */
{
SETfield(r700->ps.SQ_PGM_RESOURCES_PS.u32All, fp->r700Shader.uStackSize,
STACK_SIZE_shift, STACK_SIZE_mask);
}
SETfield(r700->ps.SQ_PGM_EXPORTS_PS.u32All, fp->r700Shader.exportMode,
EXPORT_MODE_shift, EXPORT_MODE_mask);
// emit ps input map
struct r700_vertex_program_cont *vpc =
(struct r700_vertex_program_cont *)ctx->VertexProgram._Current;
GLbitfield OutputsWritten = vpc->mesa_program.Base.OutputsWritten;
for(ui = 0; ui < R700_MAX_SHADER_EXPORTS; ui++)
r700->SPI_PS_INPUT_CNTL[ui].u32All = 0;
unBit = 1 << FRAG_ATTRIB_WPOS;
if(mesa_fp->Base.InputsRead & unBit)
{
ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_WPOS];
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit);
SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui,
SEMANTIC_shift, SEMANTIC_mask);
if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit)
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
else
CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
}
unBit = 1 << VERT_RESULT_COL0;
if(OutputsWritten & unBit)
{
ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL0];
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit);
SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui,
SEMANTIC_shift, SEMANTIC_mask);
if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit)
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
else
CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
}
unBit = 1 << VERT_RESULT_COL1;
if(OutputsWritten & unBit)
{
ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL1];
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit);
SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui,
SEMANTIC_shift, SEMANTIC_mask);
if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit)
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
else
CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
}
unBit = 1 << VERT_RESULT_FOGC;
if(OutputsWritten & unBit)
{
ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_FOGC];
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit);
SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui,
SEMANTIC_shift, SEMANTIC_mask);
if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit)
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
else
CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
}
for(i=0; i<8; i++)
{
GLboolean coord_replace = ctx->Point.PointSprite && ctx->Point.CoordReplace[i];
unBit = 1 << (VERT_RESULT_TEX0 + i);
if ((OutputsWritten & unBit) || coord_replace)
{
ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_TEX0 + i];
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit);
SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui,
SEMANTIC_shift, SEMANTIC_mask);
CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
/* ARB_point_sprite */
if (coord_replace)
{
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, PT_SPRITE_TEX_bit);
}
}
}
unBit = 1 << FRAG_ATTRIB_FACE;
if(mesa_fp->Base.InputsRead & unBit)
{
ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_FACE];
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit);
SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui,
SEMANTIC_shift, SEMANTIC_mask);
if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit)
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
else
CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
}
unBit = 1 << FRAG_ATTRIB_PNTC;
if(mesa_fp->Base.InputsRead & unBit)
{
ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_PNTC];
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit);
SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui,
SEMANTIC_shift, SEMANTIC_mask);
if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit)
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
else
CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, PT_SPRITE_TEX_bit);
}
for(i=VERT_RESULT_VAR0; i<VERT_RESULT_MAX; i++)
{
unBit = 1 << i;
if(OutputsWritten & unBit)
{
ui = pAsm->uiFP_AttributeMap[i-VERT_RESULT_VAR0+FRAG_ATTRIB_VAR0];
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit);
SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui,
SEMANTIC_shift, SEMANTIC_mask);
if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit)
SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
else
CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit);
}
}
exportCount = (r700->ps.SQ_PGM_EXPORTS_PS.u32All & EXPORT_MODE_mask) / (1 << EXPORT_MODE_shift);
if (r700->CB_SHADER_CONTROL.u32All != ((1 << exportCount) - 1))
{
R600_STATECHANGE(context, cb);
r700->CB_SHADER_CONTROL.u32All = (1 << exportCount) - 1;
}
/* sent out shader constants. */
paramList = fp->mesa_program.Base.Parameters;
if(NULL != paramList)
{
_mesa_load_state_parameters(ctx, paramList);
if (paramList->NumParameters > R700_MAX_DX9_CONSTS)
return GL_FALSE;
R600_STATECHANGE(context, ps_consts);
r700->ps.num_consts = paramList->NumParameters;
unNumParamData = paramList->NumParameters;
for(ui=0; ui<unNumParamData; ui++) {
r700->ps.consts[ui][0].f32All = paramList->ParameterValues[ui][0];
r700->ps.consts[ui][1].f32All = paramList->ParameterValues[ui][1];
r700->ps.consts[ui][2].f32All = paramList->ParameterValues[ui][2];
r700->ps.consts[ui][3].f32All = paramList->ParameterValues[ui][3];
}
/* Load fp constants to gpu */
if( (GL_TRUE == r700->bShaderUseMemConstant) && (unNumParamData > 0) )
{
r600EmitShader(ctx,
&(fp->constbo0),
(GLvoid *)&(paramList->ParameterValues[0][0]),
unNumParamData * 4,
"FS Const");
}
} else
r700->ps.num_consts = 0;
COMPILED_SUB * pCompiledSub;
GLuint uj;
GLuint unConstOffset = r700->ps.num_consts;
for(ui=0; ui<pAsm->unNumPresub; ui++)
{
pCompiledSub = pAsm->presubs[ui].pCompiledSub;
r700->ps.num_consts += pCompiledSub->NumParameters;
for(uj=0; uj<pCompiledSub->NumParameters; uj++)
{
r700->ps.consts[uj + unConstOffset][0].f32All = pCompiledSub->ParameterValues[uj][0];
r700->ps.consts[uj + unConstOffset][1].f32All = pCompiledSub->ParameterValues[uj][1];
r700->ps.consts[uj + unConstOffset][2].f32All = pCompiledSub->ParameterValues[uj][2];
r700->ps.consts[uj + unConstOffset][3].f32All = pCompiledSub->ParameterValues[uj][3];
}
unConstOffset += pCompiledSub->NumParameters;
}
return GL_TRUE;
}
/**
* Gathers together all the uniform values into a block of memory to be
* uploaded into the CURBE, then emits the state packet telling the hardware
* the new location.
*/
static void
brw_upload_constant_buffer(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_CURBE_OFFSETS */
const GLuint sz = brw->curbe.total_size;
const GLuint bufsz = sz * 16 * sizeof(GLfloat);
gl_constant_value *buf;
GLuint i;
gl_clip_plane *clip_planes;
if (sz == 0) {
goto emit;
}
buf = intel_upload_space(brw, bufsz, 64,
&brw->curbe.curbe_bo, &brw->curbe.curbe_offset);
STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
/* fragment shader constants */
if (brw->curbe.wm_size) {
_mesa_load_state_parameters(ctx, brw->fragment_program->Parameters);
/* BRW_NEW_CURBE_OFFSETS */
GLuint offset = brw->curbe.wm_start * 16;
/* BRW_NEW_FS_PROG_DATA | _NEW_PROGRAM_CONSTANTS: copy uniform values */
for (i = 0; i < brw->wm.base.prog_data->nr_params; i++) {
buf[offset + i] = *brw->wm.base.prog_data->param[i];
}
}
/* clipper constants */
if (brw->curbe.clip_size) {
GLuint offset = brw->curbe.clip_start * 16;
GLbitfield mask;
/* If any planes are going this way, send them all this way:
*/
for (i = 0; i < 6; i++) {
buf[offset + i * 4 + 0].f = fixed_plane[i][0];
buf[offset + i * 4 + 1].f = fixed_plane[i][1];
buf[offset + i * 4 + 2].f = fixed_plane[i][2];
buf[offset + i * 4 + 3].f = fixed_plane[i][3];
}
/* Clip planes: _NEW_TRANSFORM plus _NEW_PROJECTION to get to
* clip-space:
*/
clip_planes = brw_select_clip_planes(ctx);
mask = ctx->Transform.ClipPlanesEnabled;
while (mask) {
const int j = u_bit_scan(&mask);
buf[offset + i * 4 + 0].f = clip_planes[j][0];
buf[offset + i * 4 + 1].f = clip_planes[j][1];
buf[offset + i * 4 + 2].f = clip_planes[j][2];
buf[offset + i * 4 + 3].f = clip_planes[j][3];
i++;
}
}
/* vertex shader constants */
if (brw->curbe.vs_size) {
_mesa_load_state_parameters(ctx, brw->vertex_program->Parameters);
GLuint offset = brw->curbe.vs_start * 16;
/* BRW_NEW_VS_PROG_DATA | _NEW_PROGRAM_CONSTANTS: copy uniform values */
for (i = 0; i < brw->vs.base.prog_data->nr_params; i++) {
buf[offset + i] = *brw->vs.base.prog_data->param[i];
}
}
if (0) {
for (i = 0; i < sz*16; i+=4)
fprintf(stderr, "curbe %d.%d: %f %f %f %f\n", i/8, i&4,
buf[i+0].f, buf[i+1].f, buf[i+2].f, buf[i+3].f);
}
/* Because this provokes an action (ie copy the constants into the
* URB), it shouldn't be shortcircuited if identical to the
* previous time - because eg. the urb destination may have
* changed, or the urb contents different to last time.
*
* Note that the data referred to is actually copied internally,
* not just used in place according to passed pointer.
*
* It appears that the CS unit takes care of using each available
* URB entry (Const URB Entry == CURBE) in turn, and issuing
* flushes as necessary when doublebuffering of CURBEs isn't
* possible.
*/
emit:
/* BRW_NEW_URB_FENCE: From the gen4 PRM, volume 1, section 3.9.8
* (CONSTANT_BUFFER (CURBE Load)):
*
* "Modifying the CS URB allocation via URB_FENCE invalidates any
* previous CURBE entries. Therefore software must subsequently
* [re]issue a CONSTANT_BUFFER command before CURBE data can be used
* in the pipeline."
*/
BEGIN_BATCH(2);
if (brw->curbe.total_size == 0) {
OUT_BATCH((CMD_CONST_BUFFER << 16) | (2 - 2));
OUT_BATCH(0);
} else {
OUT_BATCH((CMD_CONST_BUFFER << 16) | (1 << 8) | (2 - 2));
OUT_RELOC(brw->curbe.curbe_bo,
I915_GEM_DOMAIN_INSTRUCTION, 0,
(brw->curbe.total_size - 1) + brw->curbe.curbe_offset);
}
ADVANCE_BATCH();
/* Work around a Broadwater/Crestline depth interpolator bug. The
* following sequence will cause GPU hangs:
*
* 1. Change state so that all depth related fields in CC_STATE are
* disabled, and in WM_STATE, only "PS Use Source Depth" is enabled.
* 2. Emit a CONSTANT_BUFFER packet.
* 3. Draw via 3DPRIMITIVE.
*
* The recommended workaround is to emit a non-pipelined state change after
* emitting CONSTANT_BUFFER, in order to drain the windowizer pipeline.
*
* We arbitrarily choose 3DSTATE_GLOBAL_DEPTH_CLAMP_OFFSET (as it's small),
* and always emit it when "PS Use Source Depth" is set. We could be more
* precise, but the additional complexity is probably not worth it.
*
* BRW_NEW_FRAGMENT_PROGRAM
*/
if (brw->gen == 4 && !brw->is_g4x &&
(brw->fragment_program->info.inputs_read & (1 << VARYING_SLOT_POS))) {
BEGIN_BATCH(2);
OUT_BATCH(_3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP << 16 | (2 - 2));
OUT_BATCH(0);
ADVANCE_BATCH();
}
}
GLboolean evergreenSetupVPconstants(struct gl_context * ctx)
{
context_t *context = EVERGREEN_CONTEXT(ctx);
EVERGREEN_CHIP_CONTEXT *evergreen = GET_EVERGREEN_CHIP(context);
struct evergreen_vertex_program *vp = (struct evergreen_vertex_program *) context->selected_vp;
struct gl_program_parameter_list *paramList;
unsigned int unNumParamData;
unsigned int ui;
int alloc_size;
/* sent out shader constants. */
paramList = vp->mesa_program->Base.Parameters;
if(NULL != paramList) {
/* vp->mesa_program was cloned, not updated by glsl shader api. */
/* _mesa_reference_program has already checked glsl shProg is ok and set ctx->VertexProgem._Current */
/* so, use ctx->VertexProgem._Current */
struct gl_program_parameter_list *paramListOrginal =
ctx->VertexProgram._Current->Base.Parameters;
_mesa_load_state_parameters(ctx, paramList);
if (paramList->NumParameters > EVERGREEN_MAX_DX9_CONSTS)
return GL_FALSE;
EVERGREEN_STATECHANGE(context, vs);
evergreen->vs.num_consts = paramList->NumParameters;
unNumParamData = paramList->NumParameters;
/* alloc multiple of 16 constants */
alloc_size = ((unNumParamData * 4 * 4) + 255) & ~255;
for(ui=0; ui<unNumParamData; ui++) {
if(paramList->Parameters[ui].Type == PROGRAM_UNIFORM)
{
evergreen->vs.consts[ui][0].f32All = paramListOrginal->ParameterValues[ui][0].f;
evergreen->vs.consts[ui][1].f32All = paramListOrginal->ParameterValues[ui][1].f;
evergreen->vs.consts[ui][2].f32All = paramListOrginal->ParameterValues[ui][2].f;
evergreen->vs.consts[ui][3].f32All = paramListOrginal->ParameterValues[ui][3].f;
}
else
{
evergreen->vs.consts[ui][0].f32All = paramList->ParameterValues[ui][0].f;
evergreen->vs.consts[ui][1].f32All = paramList->ParameterValues[ui][1].f;
evergreen->vs.consts[ui][2].f32All = paramList->ParameterValues[ui][2].f;
evergreen->vs.consts[ui][3].f32All = paramList->ParameterValues[ui][3].f;
}
}
radeonAllocDmaRegion(&context->radeon,
&context->vp_Constbo,
&context->vp_bo_offset,
alloc_size,
256);
r600EmitShaderConsts(ctx,
context->vp_Constbo,
context->vp_bo_offset,
(GLvoid *)&(evergreen->vs.consts[0][0]),
unNumParamData * 4 * 4);
} else
evergreen->vs.num_consts = 0;
COMPILED_SUB * pCompiledSub;
GLuint uj;
GLuint unConstOffset = evergreen->vs.num_consts;
for(ui=0; ui<vp->r700AsmCode.unNumPresub; ui++)
{
pCompiledSub = vp->r700AsmCode.presubs[ui].pCompiledSub;
evergreen->vs.num_consts += pCompiledSub->NumParameters;
for(uj=0; uj<pCompiledSub->NumParameters; uj++)
{
evergreen->vs.consts[uj + unConstOffset][0].f32All = pCompiledSub->ParameterValues[uj][0];
evergreen->vs.consts[uj + unConstOffset][1].f32All = pCompiledSub->ParameterValues[uj][1];
evergreen->vs.consts[uj + unConstOffset][2].f32All = pCompiledSub->ParameterValues[uj][2];
evergreen->vs.consts[uj + unConstOffset][3].f32All = pCompiledSub->ParameterValues[uj][3];
}
unConstOffset += pCompiledSub->NumParameters;
}
return GL_TRUE;
}
/**
* Pass the given program parameters to the graphics pipe as a
* constant buffer.
* \param shader_type either PIPE_SHADER_VERTEX or PIPE_SHADER_FRAGMENT
*/
void st_upload_constants( struct st_context *st,
struct gl_program_parameter_list *params,
gl_shader_stage stage)
{
enum pipe_shader_type shader_type = st_shader_stage_to_ptarget(stage);
assert(shader_type == PIPE_SHADER_VERTEX ||
shader_type == PIPE_SHADER_FRAGMENT ||
shader_type == PIPE_SHADER_GEOMETRY ||
shader_type == PIPE_SHADER_TESS_CTRL ||
shader_type == PIPE_SHADER_TESS_EVAL ||
shader_type == PIPE_SHADER_COMPUTE);
/* update the ATI constants before rendering */
if (shader_type == PIPE_SHADER_FRAGMENT && st->fp->ati_fs) {
struct ati_fragment_shader *ati_fs = st->fp->ati_fs;
unsigned c;
for (c = 0; c < MAX_NUM_FRAGMENT_CONSTANTS_ATI; c++) {
if (ati_fs->LocalConstDef & (1 << c))
memcpy(params->ParameterValues[c],
ati_fs->Constants[c], sizeof(GLfloat) * 4);
else
memcpy(params->ParameterValues[c],
st->ctx->ATIFragmentShader.GlobalConstants[c], sizeof(GLfloat) * 4);
}
}
/* update constants */
if (params && params->NumParameters) {
struct pipe_constant_buffer cb;
const uint paramBytes = params->NumParameters * sizeof(GLfloat) * 4;
/* Update the constants which come from fixed-function state, such as
* transformation matrices, fog factors, etc. The rest of the values in
* the parameters list are explicitly set by the user with glUniform,
* glProgramParameter(), etc.
*/
if (params->StateFlags)
_mesa_load_state_parameters(st->ctx, params);
_mesa_shader_write_subroutine_indices(st->ctx, stage);
/* We always need to get a new buffer, to keep the drivers simple and
* avoid gratuitous rendering synchronization.
* Let's use a user buffer to avoid an unnecessary copy.
*/
if (st->constbuf_uploader) {
cb.buffer = NULL;
cb.user_buffer = NULL;
u_upload_data(st->constbuf_uploader, 0, paramBytes,
st->ctx->Const.UniformBufferOffsetAlignment,
params->ParameterValues, &cb.buffer_offset, &cb.buffer);
u_upload_unmap(st->constbuf_uploader);
} else {
cb.buffer = NULL;
cb.user_buffer = params->ParameterValues;
cb.buffer_offset = 0;
}
cb.buffer_size = paramBytes;
if (ST_DEBUG & DEBUG_CONSTANTS) {
debug_printf("%s(shader=%d, numParams=%d, stateFlags=0x%x)\n",
__func__, shader_type, params->NumParameters,
params->StateFlags);
_mesa_print_parameter_list(params);
}
cso_set_constant_buffer(st->cso_context, shader_type, 0, &cb);
pipe_resource_reference(&cb.buffer, NULL);
st->state.constants[shader_type].ptr = params->ParameterValues;
st->state.constants[shader_type].size = paramBytes;
}
else if (st->state.constants[shader_type].ptr) {
/* Unbind. */
st->state.constants[shader_type].ptr = NULL;
st->state.constants[shader_type].size = 0;
cso_set_constant_buffer(st->cso_context, shader_type, 0, NULL);
}
}
static void
upload_vs_state(struct brw_context *brw)
{
struct intel_context *intel = &brw->intel;
GLcontext *ctx = &intel->ctx;
const struct brw_vertex_program *vp =
brw_vertex_program_const(brw->vertex_program);
unsigned int nr_params = vp->program.Base.Parameters->NumParameters;
drm_intel_bo *constant_bo;
int i;
if (vp->use_const_buffer || nr_params == 0) {
/* Disable the push constant buffers. */
BEGIN_BATCH(5);
OUT_BATCH(CMD_3D_CONSTANT_VS_STATE << 16 | (5 - 2));
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
} else {
if (brw->vertex_program->IsNVProgram)
_mesa_load_tracked_matrices(ctx);
/* Updates the ParamaterValues[i] pointers for all parameters of the
* basic type of PROGRAM_STATE_VAR.
*/
_mesa_load_state_parameters(ctx, vp->program.Base.Parameters);
constant_bo = drm_intel_bo_alloc(intel->bufmgr, "VS constant_bo",
nr_params * 4 * sizeof(float),
4096);
drm_intel_gem_bo_map_gtt(constant_bo);
for (i = 0; i < nr_params; i++) {
memcpy((char *)constant_bo->virtual + i * 4 * sizeof(float),
vp->program.Base.Parameters->ParameterValues[i],
4 * sizeof(float));
}
drm_intel_gem_bo_unmap_gtt(constant_bo);
BEGIN_BATCH(5);
OUT_BATCH(CMD_3D_CONSTANT_VS_STATE << 16 |
GEN6_CONSTANT_BUFFER_0_ENABLE |
(5 - 2));
OUT_RELOC(constant_bo,
I915_GEM_DOMAIN_RENDER, 0, /* XXX: bad domain */
ALIGN(nr_params, 2) / 2 - 1);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
drm_intel_bo_unreference(constant_bo);
}
intel_batchbuffer_emit_mi_flush(intel->batch);
BEGIN_BATCH(6);
OUT_BATCH(CMD_3D_VS_STATE << 16 | (6 - 2));
OUT_RELOC(brw->vs.prog_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
OUT_BATCH((0 << GEN6_VS_SAMPLER_COUNT_SHIFT) |
(brw->vs.nr_surfaces << GEN6_VS_BINDING_TABLE_ENTRY_COUNT_SHIFT));
OUT_BATCH(0); /* scratch space base offset */
OUT_BATCH((1 << GEN6_VS_DISPATCH_START_GRF_SHIFT) |
(brw->vs.prog_data->urb_read_length << GEN6_VS_URB_READ_LENGTH_SHIFT) |
(0 << GEN6_VS_URB_ENTRY_READ_OFFSET_SHIFT));
OUT_BATCH((0 << GEN6_VS_MAX_THREADS_SHIFT) |
GEN6_VS_STATISTICS_ENABLE);
ADVANCE_BATCH();
intel_batchbuffer_emit_mi_flush(intel->batch);
}
/**
* This function executes vertex programs
*/
static GLboolean
run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
{
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vp_stage_data *store = VP_STAGE_DATA(stage);
struct vertex_buffer *VB = &tnl->vb;
struct gl_vertex_program *program = ctx->VertexProgram._Current;
struct gl_program_machine machine;
GLuint outputs[VERT_RESULT_MAX], numOutputs;
GLuint i, j;
if (!program)
return GL_TRUE;
if (program->IsNVProgram) {
_mesa_load_tracked_matrices(ctx);
}
else {
/* ARB program or vertex shader */
_mesa_load_state_parameters(ctx, program->Base.Parameters);
}
/* make list of outputs to save some time below */
numOutputs = 0;
for (i = 0; i < VERT_RESULT_MAX; i++) {
if (program->Base.OutputsWritten & (1 << i)) {
outputs[numOutputs++] = i;
}
}
map_textures(ctx, program);
for (i = 0; i < VB->Count; i++) {
GLuint attr;
init_machine(ctx, &machine);
#if 0
printf("Input %d: %f, %f, %f, %f\n", i,
VB->AttribPtr[0]->data[i][0],
VB->AttribPtr[0]->data[i][1],
VB->AttribPtr[0]->data[i][2],
VB->AttribPtr[0]->data[i][3]);
printf(" color: %f, %f, %f, %f\n",
VB->AttribPtr[3]->data[i][0],
VB->AttribPtr[3]->data[i][1],
VB->AttribPtr[3]->data[i][2],
VB->AttribPtr[3]->data[i][3]);
printf(" normal: %f, %f, %f, %f\n",
VB->AttribPtr[2]->data[i][0],
VB->AttribPtr[2]->data[i][1],
VB->AttribPtr[2]->data[i][2],
VB->AttribPtr[2]->data[i][3]);
#endif
/* the vertex array case */
for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
if (program->Base.InputsRead & (1 << attr)) {
const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
const GLuint size = VB->AttribPtr[attr]->size;
const GLuint stride = VB->AttribPtr[attr]->stride;
const GLfloat *data = (GLfloat *) (ptr + stride * i);
COPY_CLEAN_4V(machine.VertAttribs[attr], size, data);
}
}
/* execute the program */
_mesa_execute_program(ctx, &program->Base, &machine);
/* copy the output registers into the VB->attribs arrays */
for (j = 0; j < numOutputs; j++) {
const GLuint attr = outputs[j];
COPY_4V(store->results[attr].data[i], machine.Outputs[attr]);
}
#if 0
printf("HPOS: %f %f %f %f\n",
machine.Outputs[0][0],
machine.Outputs[0][1],
machine.Outputs[0][2],
machine.Outputs[0][3]);
#endif
}
unmap_textures(ctx, program);
/* Fixup fog and point size results if needed */
if (program->IsNVProgram) {
if (ctx->Fog.Enabled &&
(program->Base.OutputsWritten & (1 << VERT_RESULT_FOGC)) == 0) {
for (i = 0; i < VB->Count; i++) {
store->results[VERT_RESULT_FOGC].data[i][0] = 1.0;
}
}
if (ctx->VertexProgram.PointSizeEnabled &&
(program->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) == 0) {
for (i = 0; i < VB->Count; i++) {
store->results[VERT_RESULT_PSIZ].data[i][0] = ctx->Point.Size;
}
}
}
if (program->IsPositionInvariant) {
/* We need the exact same transform as in the fixed function path here
* to guarantee invariance, depending on compiler optimization flags
* results could be different otherwise.
*/
VB->ClipPtr = TransformRaw( &store->results[0],
&ctx->_ModelProjectMatrix,
VB->AttribPtr[0] );
/* Drivers expect this to be clean to element 4...
*/
switch (VB->ClipPtr->size) {
case 1:
/* impossible */
case 2:
_mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 2 );
/* fall-through */
case 3:
_mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 3 );
/* fall-through */
case 4:
break;
}
}
else {
/* Setup the VB pointers so that the next pipeline stages get
* their data from the right place (the program output arrays).
*/
VB->ClipPtr = &store->results[VERT_RESULT_HPOS];
VB->ClipPtr->size = 4;
VB->ClipPtr->count = VB->Count;
}
VB->ColorPtr[0] = &store->results[VERT_RESULT_COL0];
VB->ColorPtr[1] = &store->results[VERT_RESULT_BFC0];
VB->SecondaryColorPtr[0] = &store->results[VERT_RESULT_COL1];
VB->SecondaryColorPtr[1] = &store->results[VERT_RESULT_BFC1];
VB->FogCoordPtr = &store->results[VERT_RESULT_FOGC];
VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VERT_RESULT_COL0];
VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VERT_RESULT_COL1];
VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VERT_RESULT_FOGC];
VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VERT_RESULT_PSIZ];
for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
VB->TexCoordPtr[i] =
VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
= &store->results[VERT_RESULT_TEX0 + i];
}
for (i = 0; i < ctx->Const.MaxVarying; i++) {
if (program->Base.OutputsWritten & (1 << (VERT_RESULT_VAR0 + i))) {
/* Note: varying results get put into the generic attributes */
VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
= &store->results[VERT_RESULT_VAR0 + i];
}
}
/* Perform NDC and cliptest operations:
*/
return do_ndc_cliptest(ctx, store);
}
/**
* This function executes vertex programs
*/
static GLboolean
run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
{
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vp_stage_data *store = VP_STAGE_DATA(stage);
struct vertex_buffer *VB = &tnl->vb;
struct gl_vertex_program *program = ctx->VertexProgram._Current;
struct gl_program_machine machine;
GLuint outputs[VERT_RESULT_MAX], numOutputs;
GLuint i, j;
if (!program)
return GL_TRUE;
if (program->IsNVProgram) {
_mesa_load_tracked_matrices(ctx);
}
else {
/* ARB program or vertex shader */
_mesa_load_state_parameters(ctx, program->Base.Parameters);
}
numOutputs = 0;
for (i = 0; i < VERT_RESULT_MAX; i++) {
if (program->Base.OutputsWritten & (1 << i)) {
outputs[numOutputs++] = i;
}
}
for (i = 0; i < VB->Count; i++) {
GLuint attr;
init_machine(ctx, &machine);
#if 0
printf("Input %d: %f, %f, %f, %f\n", i,
VB->AttribPtr[0]->data[i][0],
VB->AttribPtr[0]->data[i][1],
VB->AttribPtr[0]->data[i][2],
VB->AttribPtr[0]->data[i][3]);
printf(" color: %f, %f, %f, %f\n",
VB->AttribPtr[3]->data[i][0],
VB->AttribPtr[3]->data[i][1],
VB->AttribPtr[3]->data[i][2],
VB->AttribPtr[3]->data[i][3]);
printf(" normal: %f, %f, %f, %f\n",
VB->AttribPtr[2]->data[i][0],
VB->AttribPtr[2]->data[i][1],
VB->AttribPtr[2]->data[i][2],
VB->AttribPtr[2]->data[i][3]);
#endif
/* the vertex array case */
for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
if (program->Base.InputsRead & (1 << attr)) {
const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
const GLuint size = VB->AttribPtr[attr]->size;
const GLuint stride = VB->AttribPtr[attr]->stride;
const GLfloat *data = (GLfloat *) (ptr + stride * i);
COPY_CLEAN_4V(machine.VertAttribs[attr], size, data);
}
}
/* execute the program */
_mesa_execute_program(ctx, &program->Base, &machine);
/* copy the output registers into the VB->attribs arrays */
for (j = 0; j < numOutputs; j++) {
const GLuint attr = outputs[j];
COPY_4V(store->results[attr].data[i], machine.Outputs[attr]);
}
#if 0
printf("HPOS: %f %f %f %f\n",
machine.Outputs[0][0],
machine.Outputs[0][1],
machine.Outputs[0][2],
machine.Outputs[0][3]);
#endif
}
/* Fixup fog and point size results if needed */
if (program->IsNVProgram) {
if (ctx->Fog.Enabled &&
(program->Base.OutputsWritten & (1 << VERT_RESULT_FOGC)) == 0) {
for (i = 0; i < VB->Count; i++) {
store->results[VERT_RESULT_FOGC].data[i][0] = 1.0;
}
}
if (ctx->VertexProgram.PointSizeEnabled &&
(program->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) == 0) {
for (i = 0; i < VB->Count; i++) {
store->results[VERT_RESULT_PSIZ].data[i][0] = ctx->Point.Size;
}
}
}
/* Setup the VB pointers so that the next pipeline stages get
* their data from the right place (the program output arrays).
*/
VB->ClipPtr = &store->results[VERT_RESULT_HPOS];
VB->ClipPtr->size = 4;
VB->ClipPtr->count = VB->Count;
VB->ColorPtr[0] = &store->results[VERT_RESULT_COL0];
VB->ColorPtr[1] = &store->results[VERT_RESULT_BFC0];
VB->SecondaryColorPtr[0] = &store->results[VERT_RESULT_COL1];
VB->SecondaryColorPtr[1] = &store->results[VERT_RESULT_BFC1];
VB->FogCoordPtr = &store->results[VERT_RESULT_FOGC];
VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VERT_RESULT_COL0];
VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VERT_RESULT_COL1];
VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VERT_RESULT_FOGC];
VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VERT_RESULT_PSIZ];
for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
VB->TexCoordPtr[i] =
VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
= &store->results[VERT_RESULT_TEX0 + i];
}
for (i = 0; i < ctx->Const.MaxVarying; i++) {
if (program->Base.OutputsWritten & (1 << (VERT_RESULT_VAR0 + i))) {
/* Note: varying results get put into the generic attributes */
VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
= &store->results[VERT_RESULT_VAR0 + i];
}
}
/* Cliptest and perspective divide. Clip functions must clear
* the clipmask.
*/
store->ormask = 0;
store->andmask = CLIP_FRUSTUM_BITS;
if (tnl->NeedNdcCoords) {
VB->NdcPtr =
_mesa_clip_tab[VB->ClipPtr->size]( VB->ClipPtr,
&store->ndcCoords,
store->clipmask,
&store->ormask,
&store->andmask );
}
else {
VB->NdcPtr = NULL;
_mesa_clip_np_tab[VB->ClipPtr->size]( VB->ClipPtr,
NULL,
store->clipmask,
&store->ormask,
&store->andmask );
}
if (store->andmask) /* All vertices are outside the frustum */
return GL_FALSE;
/* This is where we'd do clip testing against the user-defined
* clipping planes, but they're not supported by vertex programs.
*/
VB->ClipOrMask = store->ormask;
VB->ClipMask = store->clipmask;
return GL_TRUE;
}
GLboolean r700SetupVertexProgram(GLcontext * ctx)
{
context_t *context = R700_CONTEXT(ctx);
R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
struct r700_vertex_program *vp = context->selected_vp;
struct gl_program_parameter_list *paramList;
unsigned int unNumParamData;
unsigned int ui;
if(GL_FALSE == vp->loaded)
{
if(vp->r700Shader.bNeedsAssembly == GL_TRUE)
{
Assemble( &(vp->r700Shader) );
}
/* Load vp to gpu */
r600EmitShader(ctx,
&(vp->shaderbo),
(GLvoid *)(vp->r700Shader.pProgram),
vp->r700Shader.uShaderBinaryDWORDSize,
"VS");
vp->loaded = GL_TRUE;
}
DumpHwBinary(DUMP_VERTEX_SHADER, (GLvoid *)(vp->r700Shader.pProgram),
vp->r700Shader.uShaderBinaryDWORDSize);
/* TODO : enable this after MemUse fixed *=
(context->chipobj.MemUse)(context, vp->shadercode.buf->id);
*/
R600_STATECHANGE(context, vs);
R600_STATECHANGE(context, fs); /* hack */
r700->vs.SQ_PGM_RESOURCES_VS.u32All = 0;
SETbit(r700->vs.SQ_PGM_RESOURCES_VS.u32All, PGM_RESOURCES__PRIME_CACHE_ON_DRAW_bit);
r700->vs.SQ_PGM_START_VS.u32All = 0; /* set from buffer object. */
SETfield(r700->vs.SQ_PGM_RESOURCES_VS.u32All, vp->r700Shader.nRegs + 1,
NUM_GPRS_shift, NUM_GPRS_mask);
if(vp->r700Shader.uStackSize) /* we don't use branch for now, it should be zero. */
{
SETfield(r700->vs.SQ_PGM_RESOURCES_VS.u32All, vp->r700Shader.uStackSize,
STACK_SIZE_shift, STACK_SIZE_mask);
}
R600_STATECHANGE(context, spi);
SETfield(r700->SPI_VS_OUT_CONFIG.u32All,
vp->r700Shader.nParamExports ? (vp->r700Shader.nParamExports - 1) : 0,
VS_EXPORT_COUNT_shift, VS_EXPORT_COUNT_mask);
SETfield(r700->SPI_PS_IN_CONTROL_0.u32All, vp->r700Shader.nParamExports,
NUM_INTERP_shift, NUM_INTERP_mask);
/*
SETbit(r700->SPI_PS_IN_CONTROL_0.u32All, PERSP_GRADIENT_ENA_bit);
CLEARbit(r700->SPI_PS_IN_CONTROL_0.u32All, LINEAR_GRADIENT_ENA_bit);
*/
/* sent out shader constants. */
paramList = vp->mesa_program->Base.Parameters;
if(NULL != paramList) {
_mesa_load_state_parameters(ctx, paramList);
if (paramList->NumParameters > R700_MAX_DX9_CONSTS)
return GL_FALSE;
R600_STATECHANGE(context, vs_consts);
r700->vs.num_consts = paramList->NumParameters;
unNumParamData = paramList->NumParameters;
for(ui=0; ui<unNumParamData; ui++) {
r700->vs.consts[ui][0].f32All = paramList->ParameterValues[ui][0];
r700->vs.consts[ui][1].f32All = paramList->ParameterValues[ui][1];
r700->vs.consts[ui][2].f32All = paramList->ParameterValues[ui][2];
r700->vs.consts[ui][3].f32All = paramList->ParameterValues[ui][3];
}
} else
r700->vs.num_consts = 0;
return GL_TRUE;
}
GLboolean r700SetupVertexProgram(struct gl_context * ctx)
{
context_t *context = R700_CONTEXT(ctx);
R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
struct r700_vertex_program *vp = context->selected_vp;
struct gl_program_parameter_list *paramList;
unsigned int unNumParamData;
unsigned int ui;
if(GL_FALSE == vp->loaded)
{
if(vp->r700Shader.bNeedsAssembly == GL_TRUE)
{
Assemble( &(vp->r700Shader) );
}
/* Load vp to gpu */
r600EmitShader(ctx,
&(vp->shaderbo),
(GLvoid *)(vp->r700Shader.pProgram),
vp->r700Shader.uShaderBinaryDWORDSize,
"VS");
if(GL_TRUE == r700->bShaderUseMemConstant)
{
paramList = vp->mesa_program->Base.Parameters;
if(NULL != paramList)
{
unNumParamData = paramList->NumParameters;
r600AllocShaderConsts(ctx,
&(vp->constbo0),
unNumParamData *4*4,
"VSCON");
}
}
vp->loaded = GL_TRUE;
}
DumpHwBinary(DUMP_VERTEX_SHADER, (GLvoid *)(vp->r700Shader.pProgram),
vp->r700Shader.uShaderBinaryDWORDSize);
/* TODO : enable this after MemUse fixed *=
(context->chipobj.MemUse)(context, vp->shadercode.buf->id);
*/
R600_STATECHANGE(context, vs);
R600_STATECHANGE(context, fs); /* hack */
r700->vs.SQ_PGM_RESOURCES_VS.u32All = 0;
SETbit(r700->vs.SQ_PGM_RESOURCES_VS.u32All, PGM_RESOURCES__PRIME_CACHE_ON_DRAW_bit);
r700->vs.SQ_ALU_CONST_CACHE_VS_0.u32All = 0; /* set from buffer object. */
r700->vs.SQ_PGM_START_VS.u32All = 0;
SETfield(r700->vs.SQ_PGM_RESOURCES_VS.u32All, vp->r700Shader.nRegs + 1,
NUM_GPRS_shift, NUM_GPRS_mask);
if(vp->r700Shader.uStackSize) /* we don't use branch for now, it should be zero. */
{
SETfield(r700->vs.SQ_PGM_RESOURCES_VS.u32All, vp->r700Shader.uStackSize,
STACK_SIZE_shift, STACK_SIZE_mask);
}
R600_STATECHANGE(context, spi);
if(vp->mesa_program->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) {
R600_STATECHANGE(context, cl);
SETbit(r700->PA_CL_VS_OUT_CNTL.u32All, USE_VTX_POINT_SIZE_bit);
SETbit(r700->PA_CL_VS_OUT_CNTL.u32All, VS_OUT_MISC_VEC_ENA_bit);
} else if (r700->PA_CL_VS_OUT_CNTL.u32All != 0) {
R600_STATECHANGE(context, cl);
CLEARbit(r700->PA_CL_VS_OUT_CNTL.u32All, USE_VTX_POINT_SIZE_bit);
CLEARbit(r700->PA_CL_VS_OUT_CNTL.u32All, VS_OUT_MISC_VEC_ENA_bit);
}
SETfield(r700->SPI_VS_OUT_CONFIG.u32All,
vp->r700Shader.nParamExports ? (vp->r700Shader.nParamExports - 1) : 0,
VS_EXPORT_COUNT_shift, VS_EXPORT_COUNT_mask);
SETfield(r700->SPI_PS_IN_CONTROL_0.u32All, vp->r700Shader.nParamExports,
NUM_INTERP_shift, NUM_INTERP_mask);
/*
SETbit(r700->SPI_PS_IN_CONTROL_0.u32All, PERSP_GRADIENT_ENA_bit);
CLEARbit(r700->SPI_PS_IN_CONTROL_0.u32All, LINEAR_GRADIENT_ENA_bit);
*/
/* sent out shader constants. */
paramList = vp->mesa_program->Base.Parameters;
if(NULL != paramList) {
/* vp->mesa_program was cloned, not updated by glsl shader api. */
/* _mesa_reference_program has already checked glsl shProg is ok and set ctx->VertexProgem._Current */
/* so, use ctx->VertexProgem._Current */
struct gl_program_parameter_list *paramListOrginal =
ctx->VertexProgram._Current->Base.Parameters;
_mesa_load_state_parameters(ctx, paramList);
if (paramList->NumParameters > R700_MAX_DX9_CONSTS)
return GL_FALSE;
R600_STATECHANGE(context, vs_consts);
r700->vs.num_consts = paramList->NumParameters;
unNumParamData = paramList->NumParameters;
for(ui=0; ui<unNumParamData; ui++) {
if(paramList->Parameters[ui].Type == PROGRAM_UNIFORM)
{
r700->vs.consts[ui][0].f32All = paramListOrginal->ParameterValues[ui][0];
r700->vs.consts[ui][1].f32All = paramListOrginal->ParameterValues[ui][1];
r700->vs.consts[ui][2].f32All = paramListOrginal->ParameterValues[ui][2];
r700->vs.consts[ui][3].f32All = paramListOrginal->ParameterValues[ui][3];
}
else
{
r700->vs.consts[ui][0].f32All = paramList->ParameterValues[ui][0];
r700->vs.consts[ui][1].f32All = paramList->ParameterValues[ui][1];
r700->vs.consts[ui][2].f32All = paramList->ParameterValues[ui][2];
r700->vs.consts[ui][3].f32All = paramList->ParameterValues[ui][3];
}
}
/* Load vp constants to gpu */
if(GL_TRUE == r700->bShaderUseMemConstant)
{
r600EmitShaderConsts(ctx,
vp->constbo0,
0,
(GLvoid *)&(r700->vs.consts[0][0]),
unNumParamData * 4 * 4);
}
} else
r700->vs.num_consts = 0;
COMPILED_SUB * pCompiledSub;
GLuint uj;
GLuint unConstOffset = r700->vs.num_consts;
for(ui=0; ui<vp->r700AsmCode.unNumPresub; ui++)
{
pCompiledSub = vp->r700AsmCode.presubs[ui].pCompiledSub;
r700->vs.num_consts += pCompiledSub->NumParameters;
for(uj=0; uj<pCompiledSub->NumParameters; uj++)
{
r700->vs.consts[uj + unConstOffset][0].f32All = pCompiledSub->ParameterValues[uj][0];
r700->vs.consts[uj + unConstOffset][1].f32All = pCompiledSub->ParameterValues[uj][1];
r700->vs.consts[uj + unConstOffset][2].f32All = pCompiledSub->ParameterValues[uj][2];
r700->vs.consts[uj + unConstOffset][3].f32All = pCompiledSub->ParameterValues[uj][3];
}
unConstOffset += pCompiledSub->NumParameters;
}
return GL_TRUE;
}
/**
* Creates a streamed BO containing the push constants for the VS or GS on
* gen6+.
*
* Push constants are constant values (such as GLSL uniforms) that are
* pre-loaded into a shader stage's register space at thread spawn time.
*
* Not all GLSL uniforms will be uploaded as push constants: The hardware has
* a limitation of 32 or 64 EU registers (256 or 512 floats) per stage to be
* uploaded as push constants, while GL 4.4 requires at least 1024 components
* to be usable for the VS. Plus, currently we always use pull constants
* instead of push constants when doing variable-index array access.
*
* See brw_curbe.c for the equivalent gen4/5 code.
*/
void
gen6_upload_push_constants(struct brw_context *brw,
const struct gl_program *prog,
const struct brw_stage_prog_data *prog_data,
struct brw_stage_state *stage_state,
enum aub_state_struct_type type)
{
struct gl_context *ctx = &brw->ctx;
/* Updates the ParamaterValues[i] pointers for all parameters of the
* basic type of PROGRAM_STATE_VAR.
*/
/* XXX: Should this happen somewhere before to get our state flag set? */
_mesa_load_state_parameters(ctx, prog->Parameters);
if (prog_data->nr_params == 0) {
stage_state->push_const_size = 0;
} else {
gl_constant_value *param;
int i;
param = brw_state_batch(brw, type,
prog_data->nr_params * sizeof(gl_constant_value),
32, &stage_state->push_const_offset);
STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
/* _NEW_PROGRAM_CONSTANTS
*
* Also _NEW_TRANSFORM -- we may reference clip planes other than as a
* side effect of dereferencing uniforms, so _NEW_PROGRAM_CONSTANTS
* wouldn't be set for them.
*/
for (i = 0; i < prog_data->nr_params; i++) {
param[i] = *prog_data->param[i];
}
if (0) {
fprintf(stderr, "%s constants:\n",
_mesa_shader_stage_to_string(stage_state->stage));
for (i = 0; i < prog_data->nr_params; i++) {
if ((i & 7) == 0)
fprintf(stderr, "g%d: ",
prog_data->dispatch_grf_start_reg + i / 8);
fprintf(stderr, "%8f ", param[i].f);
if ((i & 7) == 7)
fprintf(stderr, "\n");
}
if ((i & 7) != 0)
fprintf(stderr, "\n");
fprintf(stderr, "\n");
}
stage_state->push_const_size = ALIGN(prog_data->nr_params, 8) / 8;
/* We can only push 32 registers of constants at a time. */
/* From the SNB PRM (vol2, part 1, section 3.2.1.4: 3DSTATE_CONSTANT_VS:
*
* "The sum of all four read length fields (each incremented to
* represent the actual read length) must be less than or equal to
* 32"
*
* From the IVB PRM (vol2, part 1, section 3.2.1.3: 3DSTATE_CONSTANT_VS:
*
* "The sum of all four read length fields must be less than or
* equal to the size of 64"
*
* The other shader stages all match the VS's limits.
*/
assert(stage_state->push_const_size <= 32);
}
}
/**
* Pass the given program parameters to the graphics pipe as a
* constant buffer.
*/
void st_upload_constants(struct st_context *st, struct gl_program *prog)
{
gl_shader_stage stage = prog->info.stage;
struct gl_program_parameter_list *params = prog->Parameters;
enum pipe_shader_type shader_type = pipe_shader_type_from_mesa(stage);
assert(shader_type == PIPE_SHADER_VERTEX ||
shader_type == PIPE_SHADER_FRAGMENT ||
shader_type == PIPE_SHADER_GEOMETRY ||
shader_type == PIPE_SHADER_TESS_CTRL ||
shader_type == PIPE_SHADER_TESS_EVAL ||
shader_type == PIPE_SHADER_COMPUTE);
/* update the ATI constants before rendering */
if (shader_type == PIPE_SHADER_FRAGMENT && st->fp->ati_fs) {
struct ati_fragment_shader *ati_fs = st->fp->ati_fs;
unsigned c;
for (c = 0; c < MAX_NUM_FRAGMENT_CONSTANTS_ATI; c++) {
if (ati_fs->LocalConstDef & (1 << c))
memcpy(params->ParameterValues[c],
ati_fs->Constants[c], sizeof(GLfloat) * 4);
else
memcpy(params->ParameterValues[c],
st->ctx->ATIFragmentShader.GlobalConstants[c], sizeof(GLfloat) * 4);
}
}
/* Make all bindless samplers/images bound texture/image units resident in
* the context.
*/
st_make_bound_samplers_resident(st, prog);
st_make_bound_images_resident(st, prog);
/* update constants */
if (params && params->NumParameters) {
struct pipe_constant_buffer cb;
const uint paramBytes = params->NumParameters * sizeof(GLfloat) * 4;
/* Update the constants which come from fixed-function state, such as
* transformation matrices, fog factors, etc. The rest of the values in
* the parameters list are explicitly set by the user with glUniform,
* glProgramParameter(), etc.
*/
if (params->StateFlags)
_mesa_load_state_parameters(st->ctx, params);
_mesa_shader_write_subroutine_indices(st->ctx, stage);
cb.buffer = NULL;
cb.user_buffer = params->ParameterValues;
cb.buffer_offset = 0;
cb.buffer_size = paramBytes;
if (ST_DEBUG & DEBUG_CONSTANTS) {
debug_printf("%s(shader=%d, numParams=%d, stateFlags=0x%x)\n",
__func__, shader_type, params->NumParameters,
params->StateFlags);
_mesa_print_parameter_list(params);
}
cso_set_constant_buffer(st->cso_context, shader_type, 0, &cb);
pipe_resource_reference(&cb.buffer, NULL);
st->state.constants[shader_type].ptr = params->ParameterValues;
st->state.constants[shader_type].size = paramBytes;
}
else if (st->state.constants[shader_type].ptr) {
/* Unbind. */
st->state.constants[shader_type].ptr = NULL;
st->state.constants[shader_type].size = 0;
cso_set_constant_buffer(st->cso_context, shader_type, 0, NULL);
}
}
/* Upload a new set of constants. Too much variability to go into the
* cache mechanism, but maybe would benefit from a comparison against
* the current uploaded set of constants.
*/
static void prepare_constant_buffer(struct brw_context *brw)
{
GLcontext *ctx = &brw->intel.ctx;
const struct brw_vertex_program *vp =
brw_vertex_program_const(brw->vertex_program);
const struct brw_fragment_program *fp =
brw_fragment_program_const(brw->fragment_program);
const GLuint sz = brw->curbe.total_size;
const GLuint bufsz = sz * 16 * sizeof(GLfloat);
GLfloat *buf;
GLuint i;
if (sz == 0) {
if (brw->curbe.last_buf) {
free(brw->curbe.last_buf);
brw->curbe.last_buf = NULL;
brw->curbe.last_bufsz = 0;
}
return;
}
buf = (GLfloat *) calloc(1, bufsz);
/* fragment shader constants */
if (brw->curbe.wm_size) {
GLuint offset = brw->curbe.wm_start * 16;
_mesa_load_state_parameters(ctx, fp->program.Base.Parameters);
/* copy float constants */
for (i = 0; i < brw->wm.prog_data->nr_params; i++)
buf[offset + i] = *brw->wm.prog_data->param[i];
}
/* The clipplanes are actually delivered to both CLIP and VS units.
* VS uses them to calculate the outcode bitmasks.
*/
if (brw->curbe.clip_size) {
GLuint offset = brw->curbe.clip_start * 16;
GLuint j;
/* If any planes are going this way, send them all this way:
*/
for (i = 0; i < 6; i++) {
buf[offset + i * 4 + 0] = fixed_plane[i][0];
buf[offset + i * 4 + 1] = fixed_plane[i][1];
buf[offset + i * 4 + 2] = fixed_plane[i][2];
buf[offset + i * 4 + 3] = fixed_plane[i][3];
}
/* Clip planes: _NEW_TRANSFORM plus _NEW_PROJECTION to get to
* clip-space:
*/
assert(MAX_CLIP_PLANES == 6);
for (j = 0; j < MAX_CLIP_PLANES; j++) {
if (ctx->Transform.ClipPlanesEnabled & (1<<j)) {
buf[offset + i * 4 + 0] = ctx->Transform._ClipUserPlane[j][0];
buf[offset + i * 4 + 1] = ctx->Transform._ClipUserPlane[j][1];
buf[offset + i * 4 + 2] = ctx->Transform._ClipUserPlane[j][2];
buf[offset + i * 4 + 3] = ctx->Transform._ClipUserPlane[j][3];
i++;
}
}
}
/* vertex shader constants */
if (brw->curbe.vs_size) {
GLuint offset = brw->curbe.vs_start * 16;
GLuint nr = brw->vs.prog_data->nr_params / 4;
if (brw->vertex_program->IsNVProgram)
_mesa_load_tracked_matrices(ctx);
/* Updates the ParamaterValues[i] pointers for all parameters of the
* basic type of PROGRAM_STATE_VAR.
*/
_mesa_load_state_parameters(ctx, vp->program.Base.Parameters);
if (vp->use_const_buffer) {
/* Load the subset of push constants that will get used when
* we also have a pull constant buffer.
*/
for (i = 0; i < vp->program.Base.Parameters->NumParameters; i++) {
if (brw->vs.constant_map[i] != -1) {
assert(brw->vs.constant_map[i] <= nr);
memcpy(buf + offset + brw->vs.constant_map[i] * 4,
vp->program.Base.Parameters->ParameterValues[i],
4 * sizeof(float));
}
}
} else {
for (i = 0; i < nr; i++) {
memcpy(buf + offset + i * 4,
vp->program.Base.Parameters->ParameterValues[i],
4 * sizeof(float));
}
}
}
if (0) {
for (i = 0; i < sz*16; i+=4)
printf("curbe %d.%d: %f %f %f %f\n", i/8, i&4,
buf[i+0], buf[i+1], buf[i+2], buf[i+3]);
printf("last_buf %p buf %p sz %d/%d cmp %d\n",
brw->curbe.last_buf, buf,
bufsz, brw->curbe.last_bufsz,
brw->curbe.last_buf ? memcmp(buf, brw->curbe.last_buf, bufsz) : -1);
}
if (brw->curbe.curbe_bo != NULL &&
brw->curbe.last_buf &&
bufsz == brw->curbe.last_bufsz &&
memcmp(buf, brw->curbe.last_buf, bufsz) == 0) {
/* constants have not changed */
free(buf);
}
else {
/* constants have changed */
if (brw->curbe.last_buf)
free(brw->curbe.last_buf);
brw->curbe.last_buf = buf;
brw->curbe.last_bufsz = bufsz;
if (brw->curbe.curbe_bo != NULL &&
brw->curbe.curbe_next_offset + bufsz > brw->curbe.curbe_bo->size)
{
drm_intel_gem_bo_unmap_gtt(brw->curbe.curbe_bo);
dri_bo_unreference(brw->curbe.curbe_bo);
brw->curbe.curbe_bo = NULL;
}
if (brw->curbe.curbe_bo == NULL) {
/* Allocate a single page for CURBE entries for this batchbuffer.
* They're generally around 64b.
*/
brw->curbe.curbe_bo = dri_bo_alloc(brw->intel.bufmgr, "CURBE",
4096, 1 << 6);
brw->curbe.curbe_next_offset = 0;
drm_intel_gem_bo_map_gtt(brw->curbe.curbe_bo);
}
brw->curbe.curbe_offset = brw->curbe.curbe_next_offset;
brw->curbe.curbe_next_offset += bufsz;
brw->curbe.curbe_next_offset = ALIGN(brw->curbe.curbe_next_offset, 64);
/* Copy data to the buffer:
*/
memcpy(brw->curbe.curbe_bo->virtual + brw->curbe.curbe_offset,
buf,
bufsz);
}
brw_add_validated_bo(brw, brw->curbe.curbe_bo);
/* Because this provokes an action (ie copy the constants into the
* URB), it shouldn't be shortcircuited if identical to the
* previous time - because eg. the urb destination may have
* changed, or the urb contents different to last time.
*
* Note that the data referred to is actually copied internally,
* not just used in place according to passed pointer.
*
* It appears that the CS unit takes care of using each available
* URB entry (Const URB Entry == CURBE) in turn, and issuing
* flushes as necessary when doublebuffering of CURBEs isn't
* possible.
*/
}
