uint
i915_emit_const1f(struct i915_fp_compile * p, float c0)
{
struct i915_fragment_shader *ifs = p->shader;
unsigned reg, idx;
if (c0 == 0.0)
return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO);
if (c0 == 1.0)
return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE);
for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
if (ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
continue;
for (idx = 0; idx < 4; idx++) {
if (!(ifs->constant_flags[reg] & (1 << idx)) ||
ifs->constants[reg][idx] == c0) {
ifs->constants[reg][idx] = c0;
ifs->constant_flags[reg] |= 1 << idx;
if (reg + 1 > ifs->num_constants)
ifs->num_constants = reg + 1;
return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE);
}
}
}
i915_program_error(p, "i915_emit_const1f: out of constants");
return 0;
}
GLuint
i915_emit_const1f(struct i915_fragment_program * p, GLfloat c0)
{
GLint reg, idx;
if (c0 == 0.0)
return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO);
if (c0 == 1.0)
return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE);
for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
if (p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
continue;
for (idx = 0; idx < 4; idx++) {
if (!(p->constant_flags[reg] & (1 << idx)) ||
p->constant[reg][idx] == c0) {
p->constant[reg][idx] = c0;
p->constant_flags[reg] |= 1 << idx;
if (reg + 1 > p->nr_constants)
p->nr_constants = reg + 1;
return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE);
}
}
}
fprintf(stderr, "%s: out of constants\n", __FUNCTION__);
p->error = 1;
return 0;
}
void updatePolyhedron(const Vec3& current) {
const Grid& grid = m_tool->grid();
const Math::Axis::Type axis = m_plane.normal.firstComponent();
const Plane3 swizzledPlane(swizzle(m_plane.anchor(), axis), swizzle(m_plane.normal, axis));
const Vec3 theMin = swizzle(grid.snapDown(min(m_initialPoint, current)), axis);
const Vec3 theMax = swizzle(grid.snapUp (max(m_initialPoint, current)), axis);
const Vec2 topLeft2(theMin.x(), theMin.y());
const Vec2 topRight2(theMax.x(), theMin.y());
const Vec2 bottomLeft2(theMin.x(), theMax.y());
const Vec2 bottomRight2(theMax.x(), theMax.y());
const Vec3 topLeft3 = unswizzle(Vec3(topLeft2, swizzledPlane.zAt(topLeft2)), axis);
const Vec3 topRight3 = unswizzle(Vec3(topRight2, swizzledPlane.zAt(topRight2)), axis);
const Vec3 bottomLeft3 = unswizzle(Vec3(bottomLeft2, swizzledPlane.zAt(bottomLeft2)), axis);
const Vec3 bottomRight3 = unswizzle(Vec3(bottomRight2, swizzledPlane.zAt(bottomRight2)), axis);
Polyhedron3 polyhedron = m_oldPolyhedron;
polyhedron.addPoint(topLeft3);
polyhedron.addPoint(bottomLeft3);
polyhedron.addPoint(bottomRight3);
polyhedron.addPoint(topRight3);
m_tool->update(polyhedron);
}
/**
* [1, src0.y*src1.y, src0.z, src1.w]
* So basically MUL with lotsa swizzling.
*/
static void transform_DST(struct radeon_compiler* c,
struct rc_instruction* inst)
{
emit2(c, inst->Prev, RC_OPCODE_MUL, inst->U.I.SaturateMode, inst->U.I.DstReg,
swizzle(inst->U.I.SrcReg[0], RC_SWIZZLE_ONE, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_ONE),
swizzle(inst->U.I.SrcReg[1], RC_SWIZZLE_ONE, RC_SWIZZLE_Y, RC_SWIZZLE_ONE, RC_SWIZZLE_W));
rc_remove_instruction(inst);
}
static void transform_XPD(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit2(c, inst->Prev, RC_OPCODE_MUL, 0, dst,
swizzle(inst->U.I.SrcReg[0], RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_W),
swizzle(inst->U.I.SrcReg[1], RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_W));
emit3(c, inst->Prev, RC_OPCODE_MAD, inst->U.I.SaturateMode, inst->U.I.DstReg,
swizzle(inst->U.I.SrcReg[0], RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_W),
swizzle(inst->U.I.SrcReg[1], RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_W),
negate(srcreg(RC_FILE_TEMPORARY, dst.Index)));
rc_remove_instruction(inst);
}
void
vec4_tcs_visitor::emit_output_urb_read(const dst_reg &dst,
unsigned base_offset,
unsigned first_component,
const src_reg &indirect_offset)
{
vec4_instruction *inst;
/* Set up the message header to reference the proper parts of the URB */
dst_reg header = dst_reg(this, glsl_type::uvec4_type);
inst = emit(TCS_OPCODE_SET_OUTPUT_URB_OFFSETS, header,
brw_imm_ud(dst.writemask << first_component), indirect_offset);
inst->force_writemask_all = true;
vec4_instruction *read = emit(VEC4_OPCODE_URB_READ, dst, src_reg(header));
read->offset = base_offset;
read->mlen = 1;
read->base_mrf = -1;
if (first_component) {
/* Read into a temporary and copy with a swizzle and writemask. */
read->dst = retype(dst_reg(this, glsl_type::ivec4_type), dst.type);
emit(MOV(dst, swizzle(src_reg(read->dst),
BRW_SWZ_COMP_INPUT(first_component))));
}
}
void
vec4_tcs_visitor::emit_input_urb_read(const dst_reg &dst,
const src_reg &vertex_index,
unsigned base_offset,
unsigned first_component,
const src_reg &indirect_offset)
{
vec4_instruction *inst;
dst_reg temp(this, glsl_type::ivec4_type);
temp.type = dst.type;
/* Set up the message header to reference the proper parts of the URB */
dst_reg header = dst_reg(this, glsl_type::uvec4_type);
inst = emit(TCS_OPCODE_SET_INPUT_URB_OFFSETS, header, vertex_index,
indirect_offset);
inst->force_writemask_all = true;
/* Read into a temporary, ignoring writemasking. */
inst = emit(VEC4_OPCODE_URB_READ, temp, src_reg(header));
inst->offset = base_offset;
inst->mlen = 1;
inst->base_mrf = -1;
/* Copy the temporary to the destination to deal with writemasking.
*
* Also attempt to deal with gl_PointSize being in the .w component.
*/
if (inst->offset == 0 && indirect_offset.file == BAD_FILE) {
emit(MOV(dst, swizzle(src_reg(temp), BRW_SWIZZLE_WWWW)));
} else {
src_reg src = src_reg(temp);
src.swizzle = BRW_SWZ_COMP_INPUT(first_component);
emit(MOV(dst, src));
}
}
int test(lua_State *L) {
float data[12] = {0., 1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11.};
float out_data[12];
VecN<float, 6> *v_in = (VecN<float, 6> *)data;
VecN<float, 6> *v_out = (VecN<float, 6> *)out_data;
for(int i=0; i < 2; i++) {
int oo[] = {1, 0};
int ii[] = {0, 1};
MAT(v_out) = MAT(v_in)+MAT(v_in);
swizzle(MAT(v_out), MAT(v_in), 2, oo, ii);
v_in++;
v_out++;
}
for(int j=0; j < 12; j++) {
printf("j: %f\n", out_data[j]);
}
/*
Vec3<char> v1c(64, 123, 5);
v1c = v1c*0.5;
printf("t: %d %d %d\n", v1c.x, v1c.y, v1c.z);
*/
return 0;
}
void *accessorThread(void *arg){
int *result = (int*)malloc(sizeof(int));;
*result = 0;
while(*result < MAXVAL){
swizzle(result);
usleep(10 + (rand() % 100) );
}
pthread_exit(result);
}
void *accessorThread(void *arg){
int *result = (int*)malloc(sizeof(int)); klee_make_symbolic(result, sizeof(int), "result");
*result = 0;
while(*result < MAXVAL){
swizzle(result);
usleep(10 + (rand() % 100) );
}
pthread_exit(result);
}
/* Rather than trying to intercept and jiggle depth writes during
* emit, just move the value into its correct position at the end of
* the program:
*/
static void
fixup_depth_write(struct i915_fragment_program *p)
{
if (p->depth_written) {
GLuint depth = UREG(REG_TYPE_OD, 0);
i915_emit_arith(p,
A0_MOV,
depth, A0_DEST_CHANNEL_W, 0,
swizzle(depth, X, Y, Z, Z), 0, 0);
}
}
static GLuint emit_combine_source( struct i915_fragment_program *p,
GLuint mask,
GLuint unit,
GLenum source,
GLenum operand )
{
GLuint arg, src;
src = get_source(p, source, unit);
switch (operand) {
case GL_ONE_MINUS_SRC_COLOR:
/* Get unused tmp,
* Emit tmp = 1.0 + arg.-x-y-z-w
*/
arg = i915_get_temp( p );
return i915_emit_arith( p, A0_ADD, arg, mask, 0,
swizzle(src, ONE, ONE, ONE, ONE ),
negate(src, 1,1,1,1), 0);
case GL_SRC_ALPHA:
if (mask == A0_DEST_CHANNEL_W)
return src;
else
return swizzle( src, W, W, W, W );
case GL_ONE_MINUS_SRC_ALPHA:
/* Get unused tmp,
* Emit tmp = 1.0 + arg.-w-w-w-w
*/
arg = i915_get_temp( p );
return i915_emit_arith( p, A0_ADD, arg, mask, 0,
swizzle(src, ONE, ONE, ONE, ONE ),
negate( swizzle(src,W,W,W,W), 1,1,1,1), 0);
case GL_SRC_COLOR:
default:
return src;
}
}
uint
i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1)
{
struct i915_fragment_shader *ifs = p->shader;
unsigned reg, idx;
if (c0 == 0.0)
return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W);
if (c0 == 1.0)
return swizzle(i915_emit_const1f(p, c1), ONE, X, Z, W);
if (c1 == 0.0)
return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W);
if (c1 == 1.0)
return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W);
// XXX emit swizzle here for 0, 1, -1 and any combination thereof
// we can use swizzle + neg for that
for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
if (ifs->constant_flags[reg] == 0xf ||
ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
continue;
for (idx = 0; idx < 3; idx++) {
if (!(ifs->constant_flags[reg] & (3 << idx))) {
ifs->constants[reg][idx + 0] = c0;
ifs->constants[reg][idx + 1] = c1;
ifs->constant_flags[reg] |= 3 << idx;
if (reg + 1 > ifs->num_constants)
ifs->num_constants = reg + 1;
return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO, ONE);
}
}
}
i915_program_error(p, "i915_emit_const2f: out of constants");
return 0;
}
GLuint
i915_emit_const2f(struct i915_fragment_program * p, GLfloat c0, GLfloat c1)
{
GLint reg, idx;
if (c0 == 0.0)
return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W);
if (c0 == 1.0)
return swizzle(i915_emit_const1f(p, c1), ONE, X, Z, W);
if (c1 == 0.0)
return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W);
if (c1 == 1.0)
return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W);
for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
if (p->constant_flags[reg] == 0xf ||
p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
continue;
for (idx = 0; idx < 3; idx++) {
if (!(p->constant_flags[reg] & (3 << idx))) {
p->constant[reg][idx] = c0;
p->constant[reg][idx + 1] = c1;
p->constant_flags[reg] |= 3 << idx;
if (reg + 1 > p->nr_constants)
p->nr_constants = reg + 1;
return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO,
ONE);
}
}
}
fprintf(stderr, "%s: out of constants\n", __func__);
p->error = 1;
return 0;
}
/**
* Rather than trying to intercept and jiggle depth writes during
* emit, just move the value into its correct position at the end of
* the program:
*/
static void
i915_fixup_depth_write(struct i915_fp_compile *p)
{
/* XXX assuming pos/depth is always in output[0] */
if (p->shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
const uint depth = UREG(REG_TYPE_OD, 0);
i915_emit_arith(p,
A0_MOV, /* opcode */
depth, /* dest reg */
A0_DEST_CHANNEL_W, /* write mask */
0, /* saturate? */
swizzle(depth, X, Y, Z, Z), /* src0 */
0, 0 /* src1, src2 */);
}
}
static void emit_program_fini( struct i915_fragment_program *p )
{
int cf = get_source( p, GL_PREVIOUS, 0 );
int out = UREG( REG_TYPE_OC, 0 );
if (p->ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) {
/* Emit specular add.
*/
GLuint s = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_ALL);
i915_emit_arith( p, A0_ADD, out, A0_DEST_CHANNEL_ALL, 0, cf,
swizzle(s, X,Y,Z,ZERO), 0 );
}
else if (cf != out) {
/* Will wind up in here if no texture enabled or a couple of
* other scenarios (GL_REPLACE for instance).
*/
i915_emit_arith( p, A0_MOV, out, A0_DEST_CHANNEL_ALL, 0, cf, 0, 0 );
}
}
/**
* Retrieve a ureg for the given source register. Will emit
* constants, apply swizzling and negation as needed.
*/
static GLuint src_vector( const struct fp_src_register *source )
{
GLuint src;
assert(source->Index < 32); /* limitiation of UREG representation */
src = UREG( src_reg_file( source->File ), source->Index );
src = swizzle(src,
_X + source->Swizzle[0],
_X + source->Swizzle[1],
_X + source->Swizzle[2],
_X + source->Swizzle[3]);
if (source->NegateBase)
src = negate( src, 1,1,1,1 );
return src;
}
/* Remove a block from a given list. Does no sanity checking. */
static
void unlinkBlock ( Arena* a, UInt* b, Int listno )
{
vg_assert(listno >= 0 && listno < VG_N_MALLOC_LISTS);
if (get_prev_p(b) == b) {
/* Only one element in the list; treat it specially. */
vg_assert(get_next_p(b) == b);
a->freelist[listno] = NULL;
} else {
UInt* b_prev = get_prev_p(b);
UInt* b_next = get_next_p(b);
a->freelist[listno] = b_prev;
set_next_p(b_prev, b_next);
set_prev_p(b_next, b_prev);
swizzle ( a, listno );
}
set_prev_p(b, NULL);
set_next_p(b, NULL);
}
static void build_sphere_texgen( struct tnl_program *p,
struct ureg dest,
GLuint writemask )
{
struct ureg normal = get_transformed_normal(p);
struct ureg eye_hat = get_eye_position_normalized(p);
struct ureg tmp = get_temp(p);
struct ureg half = register_scalar_const(p, .5);
struct ureg r = get_temp(p);
struct ureg inv_m = get_temp(p);
struct ureg id = get_identity_param(p);
/* Could share the above calculations, but it would be
* a fairly odd state for someone to set (both sphere and
* reflection active for different texture coordinate
* components. Of course - if two texture units enable
* reflect and/or sphere, things start to tilt in favour
* of seperating this out:
*/
/* n.u */
emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
/* 2n.u */
emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
/* (-2n.u)n + u */
emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat);
/* r + 0,0,1 */
emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
/* rx^2 + ry^2 + (rz+1)^2 */
emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
/* 2/m */
emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
/* 1/m */
emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
/* r/m + 1/2 */
emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
release_temp(p, tmp);
release_temp(p, r);
release_temp(p, inv_m);
}
ir_swizzle *
swizzle_w(operand a)
{
return swizzle(a, SWIZZLE_WWWW, 1);
}
/**
* Retrieve a ureg for the given source register. Will emit
* constants, apply swizzling and negation as needed.
*/
static GLuint
src_vector(struct i915_fragment_program *p,
const struct prog_src_register *source,
const struct gl_fragment_program *program)
{
GLuint src;
switch (source->File) {
/* Registers:
*/
case PROGRAM_TEMPORARY:
if (source->Index >= I915_MAX_TEMPORARY) {
i915_program_error(p, "Exceeded max temporary reg");
return 0;
}
src = UREG(REG_TYPE_R, source->Index);
break;
case PROGRAM_INPUT:
switch (source->Index) {
case FRAG_ATTRIB_WPOS:
src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
break;
case FRAG_ATTRIB_COL0:
src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
break;
case FRAG_ATTRIB_COL1:
src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
src = swizzle(src, X, Y, Z, ONE);
break;
case FRAG_ATTRIB_FOGC:
src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
src = swizzle(src, W, ZERO, ZERO, ONE);
break;
case FRAG_ATTRIB_TEX0:
case FRAG_ATTRIB_TEX1:
case FRAG_ATTRIB_TEX2:
case FRAG_ATTRIB_TEX3:
case FRAG_ATTRIB_TEX4:
case FRAG_ATTRIB_TEX5:
case FRAG_ATTRIB_TEX6:
case FRAG_ATTRIB_TEX7:
src = i915_emit_decl(p, REG_TYPE_T,
T_TEX0 + (source->Index - FRAG_ATTRIB_TEX0),
D0_CHANNEL_ALL);
break;
default:
i915_program_error(p, "Bad source->Index");
return 0;
}
break;
/* Various paramters and env values. All emitted to
* hardware as program constants.
*/
case PROGRAM_LOCAL_PARAM:
src = i915_emit_param4fv(p, program->Base.LocalParams[source->Index]);
break;
case PROGRAM_ENV_PARAM:
src =
i915_emit_param4fv(p,
p->ctx->FragmentProgram.Parameters[source->
Index]);
break;
case PROGRAM_CONSTANT:
case PROGRAM_STATE_VAR:
case PROGRAM_NAMED_PARAM:
src =
i915_emit_param4fv(p,
program->Base.Parameters->ParameterValues[source->
Index]);
break;
default:
i915_program_error(p, "Bad source->File");
return 0;
}
src = swizzle(src,
GET_SWZ(source->Swizzle, 0),
GET_SWZ(source->Swizzle, 1),
GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
if (source->NegateBase)
src = negate(src,
GET_BIT(source->NegateBase, 0),
GET_BIT(source->NegateBase, 1),
GET_BIT(source->NegateBase, 2),
GET_BIT(source->NegateBase, 3));
return src;
}
/* Possible concerns:
*
* SIN, COS -- could use another taylor step?
* LIT -- results seem a little different to sw mesa
* LOG -- different to mesa on negative numbers, but this is conformant.
*
* Parse failures -- Mesa doesn't currently give a good indication
* internally whether a particular program string parsed or not. This
* can lead to confusion -- hopefully we cope with it ok now.
*
*/
static void
upload_program(struct i915_fragment_program *p)
{
const struct gl_fragment_program *program =
p->ctx->FragmentProgram._Current;
const struct prog_instruction *inst = program->Base.Instructions;
/* _mesa_debug_fp_inst(program->Base.NumInstructions, inst); */
/* Is this a parse-failed program? Ensure a valid program is
* loaded, as the flagging of an error isn't sufficient to stop
* this being uploaded to hardware.
*/
if (inst[0].Opcode == OPCODE_END) {
GLuint tmp = i915_get_utemp(p);
i915_emit_arith(p,
A0_MOV,
UREG(REG_TYPE_OC, 0),
A0_DEST_CHANNEL_ALL, 0,
swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
return;
}
if (program->Base.NumInstructions > I915_MAX_INSN) {
i915_program_error( p, "Exceeded max instructions" );
return;
}
/* Not always needed:
*/
calc_live_regs(p);
while (1) {
GLuint src0, src1, src2, flags;
GLuint tmp = 0, consts0 = 0, consts1 = 0;
switch (inst->Opcode) {
case OPCODE_ABS:
src0 = src_vector(p, &inst->SrcReg[0], program);
i915_emit_arith(p,
A0_MAX,
get_result_vector(p, inst),
get_result_flags(inst), 0,
src0, negate(src0, 1, 1, 1, 1), 0);
break;
case OPCODE_ADD:
EMIT_2ARG_ARITH(A0_ADD);
break;
case OPCODE_CMP:
src0 = src_vector(p, &inst->SrcReg[0], program);
src1 = src_vector(p, &inst->SrcReg[1], program);
src2 = src_vector(p, &inst->SrcReg[2], program);
i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */
break;
case OPCODE_COS:
src0 = src_vector(p, &inst->SrcReg[0], program);
tmp = i915_get_utemp(p);
consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
/* Reduce range from repeating about [-pi,pi] to [-1,1] */
i915_emit_arith(p,
A0_MAD,
tmp, A0_DEST_CHANNEL_X, 0,
src0,
swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
i915_emit_arith(p,
A0_MAD,
tmp, A0_DEST_CHANNEL_X, 0,
tmp,
swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
/* Compute COS with the same calculation used for SIN, but a
* different source range has been mapped to [-1,1] this time.
*/
/* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
i915_emit_arith(p,
A0_MAX,
tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, ZERO, X, ZERO, ZERO),
negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
0);
/* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
i915_emit_arith(p,
A0_MUL,
tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, ZERO, X, ZERO, ZERO),
tmp,
0);
/* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
i915_emit_arith(p,
A0_DP3,
tmp, A0_DEST_CHANNEL_X, 0,
tmp,
swizzle(consts1, X, Y, ZERO, ZERO),
0);
/* tmp.x now contains a first approximation (y). Now, weight it
* against tmp.y**2 to get closer.
*/
i915_emit_arith(p,
A0_MAX,
tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, ZERO, X, ZERO, ZERO),
negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
0);
/* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
i915_emit_arith(p,
A0_MAD,
tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, ZERO, X, ZERO, ZERO),
swizzle(tmp, ZERO, Y, ZERO, ZERO),
negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
/* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
i915_emit_arith(p,
A0_MAD,
get_result_vector(p, inst),
get_result_flags(inst), 0,
swizzle(consts1, W, W, W, W),
swizzle(tmp, Y, Y, Y, Y),
swizzle(tmp, X, X, X, X));
break;
case OPCODE_DP3:
EMIT_2ARG_ARITH(A0_DP3);
break;
case OPCODE_DP4:
EMIT_2ARG_ARITH(A0_DP4);
break;
case OPCODE_DPH:
src0 = src_vector(p, &inst->SrcReg[0], program);
src1 = src_vector(p, &inst->SrcReg[1], program);
i915_emit_arith(p,
A0_DP4,
get_result_vector(p, inst),
get_result_flags(inst), 0,
swizzle(src0, X, Y, Z, ONE), src1, 0);
break;
case OPCODE_DST:
src0 = src_vector(p, &inst->SrcReg[0], program);
src1 = src_vector(p, &inst->SrcReg[1], program);
/* result[0] = 1 * 1;
* result[1] = a[1] * b[1];
* result[2] = a[2] * 1;
* result[3] = 1 * b[3];
*/
i915_emit_arith(p,
A0_MUL,
get_result_vector(p, inst),
get_result_flags(inst), 0,
swizzle(src0, ONE, Y, Z, ONE),
swizzle(src1, ONE, Y, ONE, W), 0);
break;
case OPCODE_EX2:
src0 = src_vector(p, &inst->SrcReg[0], program);
i915_emit_arith(p,
A0_EXP,
get_result_vector(p, inst),
get_result_flags(inst), 0,
swizzle(src0, X, X, X, X), 0, 0);
break;
case OPCODE_FLR:
EMIT_1ARG_ARITH(A0_FLR);
break;
case OPCODE_FRC:
EMIT_1ARG_ARITH(A0_FRC);
break;
case OPCODE_KIL:
src0 = src_vector(p, &inst->SrcReg[0], program);
tmp = i915_get_utemp(p);
i915_emit_texld(p, get_live_regs(p, inst),
tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
0, src0, T0_TEXKILL);
break;
case OPCODE_LG2:
src0 = src_vector(p, &inst->SrcReg[0], program);
i915_emit_arith(p,
A0_LOG,
get_result_vector(p, inst),
get_result_flags(inst), 0,
swizzle(src0, X, X, X, X), 0, 0);
break;
case OPCODE_LIT:
src0 = src_vector(p, &inst->SrcReg[0], program);
tmp = i915_get_utemp(p);
/* tmp = max( a.xyzw, a.00zw )
* XXX: Clamp tmp.w to -128..128
* tmp.y = log(tmp.y)
* tmp.y = tmp.w * tmp.y
* tmp.y = exp(tmp.y)
* result = cmp (a.11-x1, a.1x01, a.1xy1 )
*/
i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, Y, Y, Y, Y), 0, 0);
i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, ZERO, Y, ZERO, ZERO),
swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, Y, Y, Y, Y), 0, 0);
i915_emit_arith(p, A0_CMP,
get_result_vector(p, inst),
get_result_flags(inst), 0,
negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
swizzle(tmp, ONE, X, ZERO, ONE),
swizzle(tmp, ONE, X, Y, ONE));
break;
case OPCODE_LRP:
src0 = src_vector(p, &inst->SrcReg[0], program);
src1 = src_vector(p, &inst->SrcReg[1], program);
src2 = src_vector(p, &inst->SrcReg[2], program);
flags = get_result_flags(inst);
tmp = i915_get_utemp(p);
/* b*a + c*(1-a)
*
* b*a + c - ca
*
* tmp = b*a + c,
* result = (-c)*a + tmp
*/
i915_emit_arith(p, A0_MAD, tmp,
flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
i915_emit_arith(p, A0_MAD,
get_result_vector(p, inst),
flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
break;
case OPCODE_MAD:
EMIT_3ARG_ARITH(A0_MAD);
break;
case OPCODE_MAX:
EMIT_2ARG_ARITH(A0_MAX);
break;
case OPCODE_MIN:
src0 = src_vector(p, &inst->SrcReg[0], program);
src1 = src_vector(p, &inst->SrcReg[1], program);
tmp = i915_get_utemp(p);
flags = get_result_flags(inst);
i915_emit_arith(p,
A0_MAX,
tmp, flags & A0_DEST_CHANNEL_ALL, 0,
negate(src0, 1, 1, 1, 1),
negate(src1, 1, 1, 1, 1), 0);
i915_emit_arith(p,
A0_MOV,
get_result_vector(p, inst),
flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
break;
case OPCODE_MOV:
EMIT_1ARG_ARITH(A0_MOV);
break;
case OPCODE_MUL:
EMIT_2ARG_ARITH(A0_MUL);
break;
case OPCODE_POW:
src0 = src_vector(p, &inst->SrcReg[0], program);
src1 = src_vector(p, &inst->SrcReg[1], program);
tmp = i915_get_utemp(p);
flags = get_result_flags(inst);
/* XXX: masking on intermediate values, here and elsewhere.
*/
i915_emit_arith(p,
A0_LOG,
tmp, A0_DEST_CHANNEL_X, 0,
swizzle(src0, X, X, X, X), 0, 0);
i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
i915_emit_arith(p,
A0_EXP,
get_result_vector(p, inst),
flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
break;
case OPCODE_RCP:
src0 = src_vector(p, &inst->SrcReg[0], program);
i915_emit_arith(p,
A0_RCP,
get_result_vector(p, inst),
get_result_flags(inst), 0,
swizzle(src0, X, X, X, X), 0, 0);
break;
case OPCODE_RSQ:
src0 = src_vector(p, &inst->SrcReg[0], program);
i915_emit_arith(p,
A0_RSQ,
get_result_vector(p, inst),
get_result_flags(inst), 0,
swizzle(src0, X, X, X, X), 0, 0);
break;
case OPCODE_SCS:
src0 = src_vector(p, &inst->SrcReg[0], program);
tmp = i915_get_utemp(p);
/*
* t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
* t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
* t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
* scs.x = DP4 t1, sin_constants
* t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
* scs.y = DP4 t1, cos_constants
*/
i915_emit_arith(p,
A0_MUL,
tmp, A0_DEST_CHANNEL_XY, 0,
swizzle(src0, X, X, ONE, ONE),
swizzle(src0, X, ONE, ONE, ONE), 0);
i915_emit_arith(p,
A0_MUL,
tmp, A0_DEST_CHANNEL_ALL, 0,
swizzle(tmp, X, Y, X, Y),
swizzle(tmp, X, X, ONE, ONE), 0);
if (inst->DstReg.WriteMask & WRITEMASK_Y) {
GLuint tmp1;
if (inst->DstReg.WriteMask & WRITEMASK_X)
tmp1 = i915_get_utemp(p);
else
tmp1 = tmp;
i915_emit_arith(p,
A0_MUL,
tmp1, A0_DEST_CHANNEL_ALL, 0,
swizzle(tmp, X, Y, Y, W),
swizzle(tmp, X, Z, ONE, ONE), 0);
i915_emit_arith(p,
A0_DP4,
get_result_vector(p, inst),
A0_DEST_CHANNEL_Y, 0,
swizzle(tmp1, W, Z, Y, X),
i915_emit_const4fv(p, sin_constants), 0);
}
if (inst->DstReg.WriteMask & WRITEMASK_X) {
i915_emit_arith(p,
A0_MUL,
tmp, A0_DEST_CHANNEL_XYZ, 0,
swizzle(tmp, X, X, Z, ONE),
swizzle(tmp, Z, ONE, ONE, ONE), 0);
i915_emit_arith(p,
A0_DP4,
get_result_vector(p, inst),
A0_DEST_CHANNEL_X, 0,
swizzle(tmp, ONE, Z, Y, X),
i915_emit_const4fv(p, cos_constants), 0);
}
break;
case OPCODE_SGE:
EMIT_2ARG_ARITH(A0_SGE);
break;
case OPCODE_SIN:
src0 = src_vector(p, &inst->SrcReg[0], program);
tmp = i915_get_utemp(p);
consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
/* Reduce range from repeating about [-pi,pi] to [-1,1] */
i915_emit_arith(p,
A0_MAD,
tmp, A0_DEST_CHANNEL_X, 0,
src0,
swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
i915_emit_arith(p,
A0_MAD,
tmp, A0_DEST_CHANNEL_X, 0,
tmp,
swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
/* Compute sin using a quadratic and quartic. It gives continuity
* that repeating the Taylor series lacks every 2*pi, and has
* reduced error.
*
* The idea was described at:
* http://www.devmaster.net/forums/showthread.php?t=5784
*/
/* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
i915_emit_arith(p,
A0_MAX,
tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, ZERO, X, ZERO, ZERO),
negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
0);
/* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
i915_emit_arith(p,
A0_MUL,
tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, ZERO, X, ZERO, ZERO),
tmp,
0);
/* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
i915_emit_arith(p,
A0_DP3,
tmp, A0_DEST_CHANNEL_X, 0,
tmp,
swizzle(consts1, X, Y, ZERO, ZERO),
0);
/* tmp.x now contains a first approximation (y). Now, weight it
* against tmp.y**2 to get closer.
*/
i915_emit_arith(p,
A0_MAX,
tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, ZERO, X, ZERO, ZERO),
negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
0);
/* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
i915_emit_arith(p,
A0_MAD,
tmp, A0_DEST_CHANNEL_Y, 0,
swizzle(tmp, ZERO, X, ZERO, ZERO),
swizzle(tmp, ZERO, Y, ZERO, ZERO),
negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
/* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
i915_emit_arith(p,
A0_MAD,
get_result_vector(p, inst),
get_result_flags(inst), 0,
swizzle(consts1, W, W, W, W),
swizzle(tmp, Y, Y, Y, Y),
swizzle(tmp, X, X, X, X));
break;
case OPCODE_SLT:
EMIT_2ARG_ARITH(A0_SLT);
break;
case OPCODE_SUB:
src0 = src_vector(p, &inst->SrcReg[0], program);
src1 = src_vector(p, &inst->SrcReg[1], program);
i915_emit_arith(p,
A0_ADD,
get_result_vector(p, inst),
get_result_flags(inst), 0,
src0, negate(src1, 1, 1, 1, 1), 0);
break;
case OPCODE_SWZ:
EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
break;
case OPCODE_TEX:
EMIT_TEX(T0_TEXLD);
break;
case OPCODE_TXB:
EMIT_TEX(T0_TEXLDB);
break;
case OPCODE_TXP:
EMIT_TEX(T0_TEXLDP);
break;
case OPCODE_XPD:
/* Cross product:
* result.x = src0.y * src1.z - src0.z * src1.y;
* result.y = src0.z * src1.x - src0.x * src1.z;
* result.z = src0.x * src1.y - src0.y * src1.x;
* result.w = undef;
*/
src0 = src_vector(p, &inst->SrcReg[0], program);
src1 = src_vector(p, &inst->SrcReg[1], program);
tmp = i915_get_utemp(p);
i915_emit_arith(p,
A0_MUL,
tmp, A0_DEST_CHANNEL_ALL, 0,
swizzle(src0, Z, X, Y, ONE),
swizzle(src1, Y, Z, X, ONE), 0);
i915_emit_arith(p,
A0_MAD,
get_result_vector(p, inst),
get_result_flags(inst), 0,
swizzle(src0, Y, Z, X, ONE),
swizzle(src1, Z, X, Y, ONE),
negate(tmp, 1, 1, 1, 0));
break;
case OPCODE_END:
return;
default:
i915_program_error(p, "bad opcode");
return;
}
inst++;
i915_release_utemps(p);
}
}
static bool
try_constant_propagate(const struct gen_device_info *devinfo,
vec4_instruction *inst,
int arg, const copy_entry *entry)
{
/* For constant propagation, we only handle the same constant
* across all 4 channels. Some day, we should handle the 8-bit
* float vector format, which would let us constant propagate
* vectors better.
* We could be more aggressive here -- some channels might not get used
* based on the destination writemask.
*/
src_reg value =
get_copy_value(*entry,
brw_apply_inv_swizzle_to_mask(inst->src[arg].swizzle,
WRITEMASK_XYZW));
if (value.file != IMM)
return false;
if (value.type == BRW_REGISTER_TYPE_VF) {
/* The result of bit-casting the component values of a vector float
* cannot in general be represented as an immediate.
*/
if (inst->src[arg].type != BRW_REGISTER_TYPE_F)
return false;
} else {
value.type = inst->src[arg].type;
}
if (inst->src[arg].abs) {
if ((devinfo->gen >= 8 && is_logic_op(inst->opcode)) ||
!brw_abs_immediate(value.type, &value.as_brw_reg())) {
return false;
}
}
if (inst->src[arg].negate) {
if ((devinfo->gen >= 8 && is_logic_op(inst->opcode)) ||
!brw_negate_immediate(value.type, &value.as_brw_reg())) {
return false;
}
}
value = swizzle(value, inst->src[arg].swizzle);
switch (inst->opcode) {
case BRW_OPCODE_MOV:
case SHADER_OPCODE_BROADCAST:
inst->src[arg] = value;
return true;
case SHADER_OPCODE_POW:
case SHADER_OPCODE_INT_QUOTIENT:
case SHADER_OPCODE_INT_REMAINDER:
if (devinfo->gen < 8)
break;
/* fallthrough */
case BRW_OPCODE_DP2:
case BRW_OPCODE_DP3:
case BRW_OPCODE_DP4:
case BRW_OPCODE_DPH:
case BRW_OPCODE_BFI1:
case BRW_OPCODE_ASR:
case BRW_OPCODE_SHL:
case BRW_OPCODE_SHR:
case BRW_OPCODE_SUBB:
if (arg == 1) {
inst->src[arg] = value;
return true;
}
break;
case BRW_OPCODE_MACH:
case BRW_OPCODE_MUL:
case SHADER_OPCODE_MULH:
case BRW_OPCODE_ADD:
case BRW_OPCODE_OR:
case BRW_OPCODE_AND:
case BRW_OPCODE_XOR:
case BRW_OPCODE_ADDC:
if (arg == 1) {
inst->src[arg] = value;
return true;
} else if (arg == 0 && inst->src[1].file != IMM) {
/* Fit this constant in by commuting the operands. Exception: we
* can't do this for 32-bit integer MUL/MACH because it's asymmetric.
*/
if ((inst->opcode == BRW_OPCODE_MUL ||
inst->opcode == BRW_OPCODE_MACH) &&
(inst->src[1].type == BRW_REGISTER_TYPE_D ||
inst->src[1].type == BRW_REGISTER_TYPE_UD))
break;
inst->src[0] = inst->src[1];
inst->src[1] = value;
return true;
}
break;
case GS_OPCODE_SET_WRITE_OFFSET:
/* This is just a multiply by a constant with special strides.
* The generator will handle immediates in both arguments (generating
* a single MOV of the product). So feel free to propagate in src0.
*/
inst->src[arg] = value;
return true;
case BRW_OPCODE_CMP:
if (arg == 1) {
inst->src[arg] = value;
return true;
} else if (arg == 0 && inst->src[1].file != IMM) {
enum brw_conditional_mod new_cmod;
new_cmod = brw_swap_cmod(inst->conditional_mod);
if (new_cmod != BRW_CONDITIONAL_NONE) {
/* Fit this constant in by swapping the operands and
* flipping the test.
*/
inst->src[0] = inst->src[1];
inst->src[1] = value;
inst->conditional_mod = new_cmod;
return true;
}
}
break;
case BRW_OPCODE_SEL:
if (arg == 1) {
inst->src[arg] = value;
return true;
} else if (arg == 0 && inst->src[1].file != IMM) {
inst->src[0] = inst->src[1];
inst->src[1] = value;
/* If this was predicated, flipping operands means
* we also need to flip the predicate.
*/
if (inst->conditional_mod == BRW_CONDITIONAL_NONE) {
inst->predicate_inverse = !inst->predicate_inverse;
}
return true;
}
break;
default:
break;
}
return false;
}
static struct ureg swizzle1( struct ureg reg, int x )
{
return swizzle(reg, x, x, x, x);
}
/* Need to add some addtional parameters to allow lighting in object
* space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
* space lighting.
*/
static void build_lighting( struct tnl_program *p )
{
const GLboolean twoside = p->state->light_twoside;
const GLboolean separate = p->state->separate_specular;
GLuint nr_lights = 0, count = 0;
struct ureg normal = get_transformed_normal(p);
struct ureg lit = get_temp(p);
struct ureg dots = get_temp(p);
struct ureg _col0 = undef, _col1 = undef;
struct ureg _bfc0 = undef, _bfc1 = undef;
GLuint i;
/*
* NOTE:
* dots.x = dot(normal, VPpli)
* dots.y = dot(normal, halfAngle)
* dots.z = back.shininess
* dots.w = front.shininess
*/
for (i = 0; i < MAX_LIGHTS; i++)
if (p->state->unit[i].light_enabled)
nr_lights++;
set_material_flags(p);
{
if (!p->state->material_shininess_is_zero) {
struct ureg shininess = get_material(p, 0, STATE_SHININESS);
emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X));
release_temp(p, shininess);
}
_col0 = make_temp(p, get_scenecolor(p, 0));
if (separate)
_col1 = make_temp(p, get_identity_param(p));
else
_col1 = _col0;
}
if (twoside) {
if (!p->state->material_shininess_is_zero) {
/* Note that we negate the back-face specular exponent here.
* The negation will be un-done later in the back-face code below.
*/
struct ureg shininess = get_material(p, 1, STATE_SHININESS);
emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
negate(swizzle1(shininess,X)));
release_temp(p, shininess);
}
_bfc0 = make_temp(p, get_scenecolor(p, 1));
if (separate)
_bfc1 = make_temp(p, get_identity_param(p));
else
_bfc1 = _bfc0;
}
/* If no lights, still need to emit the scenecolor.
*/
{
struct ureg res0 = register_output( p, VERT_RESULT_COL0 );
emit_op1(p, OPCODE_MOV, res0, 0, _col0);
}
if (separate) {
struct ureg res1 = register_output( p, VERT_RESULT_COL1 );
emit_op1(p, OPCODE_MOV, res1, 0, _col1);
}
if (twoside) {
struct ureg res0 = register_output( p, VERT_RESULT_BFC0 );
emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
}
if (twoside && separate) {
struct ureg res1 = register_output( p, VERT_RESULT_BFC1 );
emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
}
if (nr_lights == 0) {
release_temps(p);
return;
}
for (i = 0; i < MAX_LIGHTS; i++) {
if (p->state->unit[i].light_enabled) {
struct ureg half = undef;
struct ureg att = undef, VPpli = undef;
count++;
if (p->state->unit[i].light_eyepos3_is_zero) {
/* Can used precomputed constants in this case.
* Attenuation never applies to infinite lights.
*/
VPpli = register_param3(p, STATE_INTERNAL,
STATE_LIGHT_POSITION_NORMALIZED, i);
if (!p->state->material_shininess_is_zero) {
if (p->state->light_local_viewer) {
struct ureg eye_hat = get_eye_position_normalized(p);
half = get_temp(p);
emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
emit_normalize_vec3(p, half, half);
}
else {
half = register_param3(p, STATE_INTERNAL,
STATE_LIGHT_HALF_VECTOR, i);
}
}
}
else {
struct ureg Ppli = register_param3(p, STATE_INTERNAL,
STATE_LIGHT_POSITION, i);
struct ureg V = get_eye_position(p);
struct ureg dist = get_temp(p);
VPpli = get_temp(p);
/* Calculate VPpli vector
*/
emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
/* Normalize VPpli. The dist value also used in
* attenuation below.
*/
emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
emit_op1(p, OPCODE_RSQ, dist, 0, dist);
emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
/* Calculate attenuation:
*/
if (!p->state->unit[i].light_spotcutoff_is_180 ||
p->state->unit[i].light_attenuated) {
att = calculate_light_attenuation(p, i, VPpli, dist);
}
/* Calculate viewer direction, or use infinite viewer:
*/
if (!p->state->material_shininess_is_zero) {
half = get_temp(p);
if (p->state->light_local_viewer) {
struct ureg eye_hat = get_eye_position_normalized(p);
emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
}
else {
struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
}
emit_normalize_vec3(p, half, half);
}
release_temp(p, dist);
}
/* Calculate dot products:
*/
if (p->state->material_shininess_is_zero) {
emit_op2(p, OPCODE_DP3, dots, 0, normal, VPpli);
}
else {
emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
}
/* Front face lighting:
*/
{
struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
struct ureg res0, res1;
GLuint mask0, mask1;
if (count == nr_lights) {
if (separate) {
mask0 = WRITEMASK_XYZ;
mask1 = WRITEMASK_XYZ;
res0 = register_output( p, VERT_RESULT_COL0 );
res1 = register_output( p, VERT_RESULT_COL1 );
}
else {
mask0 = 0;
mask1 = WRITEMASK_XYZ;
res0 = _col0;
res1 = register_output( p, VERT_RESULT_COL0 );
}
}
else {
mask0 = 0;
mask1 = 0;
res0 = _col0;
res1 = _col1;
}
if (!is_undef(att)) {
/* light is attenuated by distance */
emit_op1(p, OPCODE_LIT, lit, 0, dots);
emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
}
else if (!p->state->material_shininess_is_zero) {
/* there's a non-zero specular term */
emit_op1(p, OPCODE_LIT, lit, 0, dots);
emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
}
else {
/* no attenutation, no specular */
emit_degenerate_lit(p, lit, dots);
emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
}
emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
release_temp(p, ambient);
release_temp(p, diffuse);
release_temp(p, specular);
}
/* Back face lighting:
*/
if (twoside) {
struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
struct ureg res0, res1;
GLuint mask0, mask1;
if (count == nr_lights) {
if (separate) {
mask0 = WRITEMASK_XYZ;
mask1 = WRITEMASK_XYZ;
res0 = register_output( p, VERT_RESULT_BFC0 );
res1 = register_output( p, VERT_RESULT_BFC1 );
}
else {
mask0 = 0;
mask1 = WRITEMASK_XYZ;
res0 = _bfc0;
res1 = register_output( p, VERT_RESULT_BFC0 );
}
}
else {
res0 = _bfc0;
res1 = _bfc1;
mask0 = 0;
mask1 = 0;
}
/* For the back face we need to negate the X and Y component
* dot products. dots.Z has the negated back-face specular
* exponent. We swizzle that into the W position. This
* negation makes the back-face specular term positive again.
*/
dots = negate(swizzle(dots,X,Y,W,Z));
if (!is_undef(att)) {
emit_op1(p, OPCODE_LIT, lit, 0, dots);
emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
}
else if (!p->state->material_shininess_is_zero) {
emit_op1(p, OPCODE_LIT, lit, 0, dots);
emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0); /**/
}
else {
emit_degenerate_lit(p, lit, dots);
emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0);
}
emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
/* restore dots to its original state for subsequent lights
* by negating and swizzling again.
*/
dots = negate(swizzle(dots,X,Y,W,Z));
release_temp(p, ambient);
release_temp(p, diffuse);
release_temp(p, specular);
}
release_temp(p, half);
release_temp(p, VPpli);
release_temp(p, att);
}
}
release_temps( p );
}
ir_swizzle *
swizzle_yyyy(operand a)
{
return swizzle(a, SWIZZLE_YYYY, 4);
}
ir_swizzle *
swizzle_xxxx(operand a)
{
return swizzle(a, SWIZZLE_XXXX, 4);
}
void BVH4iIntersector1::occluded(BVH4i* bvh, Ray& ray)
{
/* near and node stack */
__aligned(64) NodeRef stack_node[3*BVH4i::maxDepth+1];
/* setup */
const mic3f rdir16 = rcp_safe(mic3f(ray.dir.x,ray.dir.y,ray.dir.z));
const mic_f inf = mic_f(pos_inf);
const mic_f zero = mic_f::zero();
const Node * __restrict__ nodes = (Node *)bvh->nodePtr();
const Triangle1 * __restrict__ accel = (Triangle1*)bvh->triPtr();
stack_node[0] = BVH4i::invalidNode;
stack_node[1] = bvh->root;
size_t sindex = 2;
const mic_f org_xyz = loadAOS4to16f(ray.org.x,ray.org.y,ray.org.z);
const mic_f dir_xyz = loadAOS4to16f(ray.dir.x,ray.dir.y,ray.dir.z);
const mic_f rdir_xyz = loadAOS4to16f(rdir16.x[0],rdir16.y[0],rdir16.z[0]);
const mic_f org_rdir_xyz = org_xyz * rdir_xyz;
const mic_f min_dist_xyz = broadcast1to16f(&ray.tnear);
const mic_f max_dist_xyz = broadcast1to16f(&ray.tfar);
const unsigned int leaf_mask = BVH4I_LEAF_MASK;
while (1)
{
NodeRef curNode = stack_node[sindex-1];
sindex--;
while (1)
{
/* test if this is a leaf node */
if (unlikely(curNode.isLeaf(leaf_mask))) break;
const Node* __restrict__ const node = curNode.node(nodes);
const float* __restrict const plower = (float*)node->lower;
const float* __restrict const pupper = (float*)node->upper;
prefetch<PFHINT_L1>((char*)node + 0);
prefetch<PFHINT_L1>((char*)node + 64);
/* intersect single ray with 4 bounding boxes */
const mic_f tLowerXYZ = load16f(plower) * rdir_xyz - org_rdir_xyz;
const mic_f tUpperXYZ = load16f(pupper) * rdir_xyz - org_rdir_xyz;
const mic_f tLower = mask_min(0x7777,min_dist_xyz,tLowerXYZ,tUpperXYZ);
const mic_f tUpper = mask_max(0x7777,max_dist_xyz,tLowerXYZ,tUpperXYZ);
sindex--;
curNode = stack_node[sindex];
const Node* __restrict__ const next = curNode.node(nodes);
prefetch<PFHINT_L2>((char*)next + 0);
prefetch<PFHINT_L2>((char*)next + 64);
const mic_f tNear = vreduce_max4(tLower);
const mic_f tFar = vreduce_min4(tUpper);
const mic_m hitm = le(0x8888,tNear,tFar);
const mic_f tNear_pos = select(hitm,tNear,inf);
/* if no child is hit, continue with early popped child */
if (unlikely(none(hitm))) continue;
sindex++;
const unsigned long hiti = toInt(hitm);
const unsigned long pos_first = bitscan64(hiti);
const unsigned long num_hitm = countbits(hiti);
/* if a single child is hit, continue with that child */
curNode = ((unsigned int *)plower)[pos_first];
if (likely(num_hitm == 1)) continue;
/* if two children are hit, push in correct order */
const unsigned long pos_second = bitscan64(pos_first,hiti);
if (likely(num_hitm == 2))
{
const unsigned int dist_first = ((unsigned int*)&tNear)[pos_first];
const unsigned int dist_second = ((unsigned int*)&tNear)[pos_second];
const unsigned int node_first = curNode;
const unsigned int node_second = ((unsigned int*)plower)[pos_second];
if (dist_first <= dist_second)
{
stack_node[sindex] = node_second;
sindex++;
assert(sindex < 3*BVH4i::maxDepth+1);
continue;
}
else
{
stack_node[sindex] = curNode;
curNode = node_second;
sindex++;
assert(sindex < 3*BVH4i::maxDepth+1);
continue;
}
}
/* continue with closest child and push all others */
const mic_f min_dist = set_min_lanes(tNear_pos);
const unsigned old_sindex = sindex;
sindex += countbits(hiti) - 1;
assert(sindex < 3*BVH4i::maxDepth+1);
const mic_m closest_child = eq(hitm,min_dist,tNear);
const unsigned long closest_child_pos = bitscan64(closest_child);
const mic_m m_pos = andn(hitm,andn(closest_child,(mic_m)((unsigned int)closest_child - 1)));
const mic_i plower_node = load16i((int*)plower);
curNode = ((unsigned int*)plower)[closest_child_pos];
compactustore16i(m_pos,&stack_node[old_sindex],plower_node);
}
/* return if stack is empty */
if (unlikely(curNode == BVH4i::invalidNode)) break;
/* intersect one ray against four triangles */
//////////////////////////////////////////////////////////////////////////////////////////////////
const Triangle1* tptr = (Triangle1*) curNode.leaf(accel);
prefetch<PFHINT_L1>(tptr + 3);
prefetch<PFHINT_L1>(tptr + 2);
prefetch<PFHINT_L1>(tptr + 1);
prefetch<PFHINT_L1>(tptr + 0);
const mic_i and_mask = broadcast4to16i(zlc4);
const mic_f v0 = gather_4f_zlc(and_mask,
(float*)&tptr[0].v0,
(float*)&tptr[1].v0,
(float*)&tptr[2].v0,
(float*)&tptr[3].v0);
const mic_f v1 = gather_4f_zlc(and_mask,
(float*)&tptr[0].v1,
(float*)&tptr[1].v1,
(float*)&tptr[2].v1,
(float*)&tptr[3].v1);
const mic_f v2 = gather_4f_zlc(and_mask,
(float*)&tptr[0].v2,
(float*)&tptr[1].v2,
(float*)&tptr[2].v2,
(float*)&tptr[3].v2);
const mic_f e1 = v1 - v0;
const mic_f e2 = v0 - v2;
const mic_f normal = lcross_zxy(e1,e2);
const mic_f org = v0 - org_xyz;
const mic_f odzxy = msubr231(org * swizzle(dir_xyz,_MM_SWIZ_REG_DACB), dir_xyz, swizzle(org,_MM_SWIZ_REG_DACB));
const mic_f den = ldot3_zxy(dir_xyz,normal);
const mic_f rcp_den = rcp(den);
const mic_f uu = ldot3_zxy(e2,odzxy);
const mic_f vv = ldot3_zxy(e1,odzxy);
const mic_f u = uu * rcp_den;
const mic_f v = vv * rcp_den;
#if defined(__BACKFACE_CULLING__)
const mic_m m_init = (mic_m)0x1111 & (den > zero);
#else
const mic_m m_init = 0x1111;
#endif
const mic_m valid_u = ge(m_init,u,zero);
const mic_m valid_v = ge(valid_u,v,zero);
const mic_m m_aperture = le(valid_v,u+v,mic_f::one());
const mic_f nom = ldot3_zxy(org,normal);
const mic_f t = rcp_den*nom;
if (unlikely(none(m_aperture))) continue;
mic_m m_final = lt(lt(m_aperture,min_dist_xyz,t),t,max_dist_xyz);
#if defined(__USE_RAY_MASK__)
const mic_i rayMask(ray.mask);
const mic_i triMask = swDDDD(gather16i_4i_align(&tptr[0].v2,&tptr[1].v2,&tptr[2].v2,&tptr[3].v2));
const mic_m m_ray_mask = (rayMask & triMask) != mic_i::zero();
m_final &= m_ray_mask;
#endif
#if defined(__INTERSECTION_FILTER__)
/* did the ray hit one of the four triangles? */
while (any(m_final))
{
const mic_f temp_t = select(m_final,t,max_dist_xyz);
const mic_f min_dist = vreduce_min(temp_t);
const mic_m m_dist = eq(min_dist,temp_t);
const size_t vecIndex = bitscan(toInt(m_dist));
const size_t triIndex = vecIndex >> 2;
const Triangle1 *__restrict__ tri_ptr = tptr + triIndex;
const mic_m m_tri = m_dist^(m_dist & (mic_m)((unsigned int)m_dist - 1));
const mic_f gnormalx = mic_f(tri_ptr->Ng.x);
const mic_f gnormaly = mic_f(tri_ptr->Ng.y);
const mic_f gnormalz = mic_f(tri_ptr->Ng.z);
const int geomID = tri_ptr->geomID();
const int primID = tri_ptr->primID();
Geometry* geom = ((Scene*)bvh->geometry)->get(geomID);
if (likely(!geom->hasOcclusionFilter1())) break;
if (runOcclusionFilter1(geom,ray,u,v,min_dist,gnormalx,gnormaly,gnormalz,m_tri,geomID,primID))
break;
m_final ^= m_tri; /* clear bit */
}
#endif
if (unlikely(any(m_final)))
{
ray.geomID = 0;
return;
}
//////////////////////////////////////////////////////////////////////////////////////////////////
}
}
static struct pipe_sampler_view *
nv30_sampler_view_create(struct pipe_context *pipe, struct pipe_resource *pt,
const struct pipe_sampler_view *tmpl)
{
const struct nv30_texfmt *fmt = nv30_texfmt(pipe->screen, tmpl->format);
struct nouveau_object *eng3d = nv30_context(pipe)->screen->eng3d;
struct nv30_miptree *mt = nv30_miptree(pt);
struct nv30_sampler_view *so;
so = MALLOC_STRUCT(nv30_sampler_view);
if (!so)
return NULL;
so->pipe = *tmpl;
so->pipe.reference.count = 1;
so->pipe.texture = NULL;
so->pipe.context = pipe;
pipe_resource_reference(&so->pipe.texture, pt);
so->fmt = NV30_3D_TEX_FORMAT_NO_BORDER;
switch (pt->target) {
case PIPE_TEXTURE_1D:
so->fmt |= NV30_3D_TEX_FORMAT_DIMS_1D;
break;
case PIPE_TEXTURE_CUBE:
so->fmt |= NV30_3D_TEX_FORMAT_CUBIC;
case PIPE_TEXTURE_2D:
case PIPE_TEXTURE_RECT:
so->fmt |= NV30_3D_TEX_FORMAT_DIMS_2D;
break;
case PIPE_TEXTURE_3D:
so->fmt |= NV30_3D_TEX_FORMAT_DIMS_3D;
break;
default:
assert(0);
so->fmt |= NV30_3D_TEX_FORMAT_DIMS_1D;
break;
}
so->filt = fmt->filter;
so->wrap = fmt->wrap;
so->swz = fmt->swizzle;
so->swz |= swizzle(fmt, 3, tmpl->swizzle_a);
so->swz |= swizzle(fmt, 0, tmpl->swizzle_r) << 2;
so->swz |= swizzle(fmt, 1, tmpl->swizzle_g) << 4;
so->swz |= swizzle(fmt, 2, tmpl->swizzle_b) << 6;
/* apparently, we need to ignore the t coordinate for 1D textures to
* fix piglit tex1d-2dborder
*/
so->wrap_mask = ~0;
if (pt->target == PIPE_TEXTURE_1D) {
so->wrap_mask &= ~NV30_3D_TEX_WRAP_T__MASK;
so->wrap |= NV30_3D_TEX_WRAP_T_REPEAT;
}
/* yet more hardware suckage, can't filter 32-bit float formats */
switch (tmpl->format) {
case PIPE_FORMAT_R32_FLOAT:
case PIPE_FORMAT_R32G32B32A32_FLOAT:
so->filt_mask = ~(NV30_3D_TEX_FILTER_MIN__MASK |
NV30_3D_TEX_FILTER_MAG__MASK);
so->filt |= NV30_3D_TEX_FILTER_MIN_NEAREST |
NV30_3D_TEX_FILTER_MAG_NEAREST;
break;
default:
so->filt_mask = ~0;
break;
}
so->npot_size0 = (pt->width0 << 16) | pt->height0;
if (eng3d->oclass >= NV40_3D_CLASS) {
so->npot_size1 = (pt->depth0 << 20) | mt->uniform_pitch;
if (!mt->swizzled)
so->fmt |= NV40_3D_TEX_FORMAT_LINEAR;
so->fmt |= 0x00008000;
so->fmt |= (pt->last_level + 1) << NV40_3D_TEX_FORMAT_MIPMAP_COUNT__SHIFT;
} else {
so->swz |= mt->uniform_pitch << NV30_3D_TEX_SWIZZLE_RECT_PITCH__SHIFT;
if (pt->last_level)
so->fmt |= NV30_3D_TEX_FORMAT_MIPMAP;
so->fmt |= util_logbase2(pt->width0) << 20;
so->fmt |= util_logbase2(pt->height0) << 24;
so->fmt |= util_logbase2(pt->depth0) << 28;
so->fmt |= 0x00010000;
}
so->base_lod = so->pipe.u.tex.first_level << 8;
so->high_lod = MIN2(pt->last_level, so->pipe.u.tex.last_level) << 8;
return &so->pipe;
}
ir_swizzle *
swizzle_xyzw(operand a)
{
return swizzle(a, SWIZZLE_XYZW, 4);
}
