SUB_NOISE1, SUB_NOISE2, SUB_NOISE3, SUB_NOISE4
};
-/* Only guess, need a flag in gl_fragment_program later */
+
+/**
+ * Determine if the given fragment program uses GLSL features such
+ * as flow conditionals, loops, subroutines.
+ * Some GLSL shaders may use these features, others might not.
+ */
GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp)
{
int i;
for (i = 0; i < fp->Base.NumInstructions; i++) {
- struct prog_instruction *inst = &fp->Base.Instructions[i];
+ const struct prog_instruction *inst = &fp->Base.Instructions[i];
switch (inst->Opcode) {
case OPCODE_IF:
case OPCODE_TRUNC:
return GL_FALSE;
}
+
+/**
+ * Record the mapping of a Mesa register to a hardware register.
+ */
static void set_reg(struct brw_wm_compile *c, int file, int index,
int component, struct brw_reg reg)
{
c->wm_regs[file][index][component].inited = GL_TRUE;
}
+/**
+ * Examine instruction's write mask to find index of first component
+ * enabled for writing.
+ */
static int get_scalar_dst_index(struct prog_instruction *inst)
{
int i;
return reg;
}
+/**
+ * Save current temp register info.
+ * There must be a matching call to release_tmps().
+ */
static int mark_tmps(struct brw_wm_compile *c)
{
return c->tmp_index;
c->tmp_index = mark;
}
+/**
+ * Convert Mesa src register to brw register.
+ *
+ * Since we're running in SOA mode each Mesa register corresponds to four
+ * hardware registers. We allocate the hardware registers as needed here.
+ *
+ * \param file register file, one of PROGRAM_x
+ * \param index register number
+ * \param component src component (X=0, Y=1, Z=2, W=3)
+ * \param nr not used?!?
+ * \param neg negate value?
+ * \param abs take absolute value?
+ */
static struct brw_reg
-get_reg(struct brw_wm_compile *c, int file, int index, int component, int nr, GLuint neg, GLuint abs)
+get_reg(struct brw_wm_compile *c, int file, int index, int component,
+ int nr, GLuint neg, GLuint abs)
{
struct brw_reg reg;
switch (file) {
break;
case PROGRAM_UNDEFINED:
return brw_null_reg();
- default:
+ case PROGRAM_TEMPORARY:
+ case PROGRAM_INPUT:
+ case PROGRAM_OUTPUT:
+ case PROGRAM_PAYLOAD:
break;
+ default:
+ _mesa_problem(NULL, "Unexpected file in get_reg()");
+ return brw_null_reg();
}
- if(c->wm_regs[file][index][component].inited)
+ /* see if we've already allocated a HW register for this Mesa register */
+ if (c->wm_regs[file][index][component].inited) {
+ /* yes, re-use */
reg = c->wm_regs[file][index][component].reg;
- else
+ }
+ else {
+ /* no, allocate new register */
reg = brw_vec8_grf(c->reg_index, 0);
+ }
- if(!c->wm_regs[file][index][component].inited) {
+ /* if this is a new register allocation, record it in the table */
+ if (!c->wm_regs[file][index][component].inited) {
set_reg(c, file, index, component, reg);
c->reg_index++;
}
- if (neg & (1<< component)) {
+ if (c->reg_index >= BRW_WM_MAX_GRF - 12) {
+ /* ran out of temporary registers! */
+#if 1
+ /* This is a big hack for now.
+ * Return bad register index, just don't hang the GPU.
+ */
+ _mesa_fprintf(stderr, "out of regs %d\n", c->reg_index);
+ c->reg_index = BRW_WM_MAX_GRF - 13;
+#else
+ return brw_null_reg();
+#endif
+ }
+
+ if (neg & (1 << component)) {
reg = negate(reg);
}
if (abs)
return reg;
}
+
+/**
+ * Preallocate registers. This sets up the Mesa to hardware register
+ * mapping for certain registers, such as constants (uniforms/state vars)
+ * and shader inputs.
+ */
static void prealloc_reg(struct brw_wm_compile *c)
{
int i, j;
GLuint inputs = FRAG_BIT_WPOS | c->fp_interp_emitted | c->fp_deriv_emitted;
for (i = 0; i < 4; i++) {
- reg = (i < c->key.nr_depth_regs)
- ? brw_vec8_grf(i*2, 0) : brw_vec8_grf(0, 0);
+ if (i < c->key.nr_depth_regs)
+ reg = brw_vec8_grf(i * 2, 0);
+ else
+ reg = brw_vec8_grf(0, 0);
set_reg(c, PROGRAM_PAYLOAD, PAYLOAD_DEPTH, i, reg);
}
- c->reg_index += 2*c->key.nr_depth_regs;
+ c->reg_index += 2 * c->key.nr_depth_regs;
+
+ /* constants */
{
- int nr_params = c->fp->program.Base.Parameters->NumParameters;
- struct gl_program_parameter_list *plist =
- c->fp->program.Base.Parameters;
- int index = 0;
- c->prog_data.nr_params = 4*nr_params;
- for (i = 0; i < nr_params; i++) {
- for (j = 0; j < 4; j++, index++) {
- reg = brw_vec1_grf(c->reg_index + index/8,
- index%8);
- c->prog_data.param[index] =
- &plist->ParameterValues[i][j];
- set_reg(c, PROGRAM_STATE_VAR, i, j, reg);
- }
- }
- c->nr_creg = 2*((4*nr_params+15)/16);
- c->reg_index += c->nr_creg;
+ const int nr_params = c->fp->program.Base.Parameters->NumParameters;
+
+ if (1 /* XXX threshold: nr_params <= 8 */) {
+ const struct gl_program_parameter_list *plist =
+ c->fp->program.Base.Parameters;
+ int index = 0;
+
+ /* number of float constants in CURBE */
+ c->prog_data.nr_params = 4 * nr_params;
+
+ /* loop over program constants (float[4]) */
+ for (i = 0; i < nr_params; i++) {
+ /* loop over XYZW channels */
+ for (j = 0; j < 4; j++, index++) {
+ reg = brw_vec1_grf(c->reg_index + index / 8, index % 8);
+ /* Save pointer to parameter/constant value.
+ * Constants will be copied in prepare_constant_buffer()
+ */
+ c->prog_data.param[index] = &plist->ParameterValues[i][j];
+ set_reg(c, PROGRAM_STATE_VAR, i, j, reg);
+ }
+ }
+ /* number of constant regs used (each reg is float[8]) */
+ c->nr_creg = 2 * ((4 * nr_params + 15) / 16);
+ c->reg_index += c->nr_creg;
+ }
+ else {
+ /* number of float constants in CURBE */
+ c->prog_data.nr_params = 0;
+
+ /* When there's a lot of FP constanst we'll store them in a
+ * texture-like buffer instead of using the CURBE buffer.
+ * This means we won't use GRF registers for constants and we'll
+ * have to fetch constants with a dataport read.
+ */
+ }
}
+
+ /* fragment shader inputs */
for (i = 0; i < FRAG_ATTRIB_MAX; i++) {
if (inputs & (1<<i)) {
nr_interp_regs++;
for (j = 0; j < 4; j++)
set_reg(c, PROGRAM_PAYLOAD, i, j, reg);
c->reg_index += 2;
-
}
}
+
c->prog_data.first_curbe_grf = c->key.nr_depth_regs * 2;
c->prog_data.urb_read_length = nr_interp_regs * 2;
c->prog_data.curb_read_length = c->nr_creg;
c->reg_index += 2;
}
+
+/**
+ * Convert Mesa dst register to brw register.
+ */
static struct brw_reg get_dst_reg(struct brw_wm_compile *c,
- struct prog_instruction *inst, int component, int nr)
+ const struct prog_instruction *inst,
+ GLuint component)
{
+ const int nr = 1;
return get_reg(c, inst->DstReg.File, inst->DstReg.Index, component, nr,
0, 0);
}
+
+/**
+ * Convert Mesa src register to brw register.
+ */
static struct brw_reg get_src_reg(struct brw_wm_compile *c,
- struct prog_src_register *src, int index, int nr)
+ const struct prog_instruction *inst,
+ GLuint srcRegIndex, GLuint channel)
{
- int component = GET_SWZ(src->Swizzle, index);
+ const struct prog_src_register *src = &inst->SrcReg[srcRegIndex];
+ const GLuint nr = 1;
+ const GLuint component = GET_SWZ(src->Swizzle, channel);
+
return get_reg(c, src->File, src->Index, component, nr,
- src->NegateBase, src->Abs);
+ src->NegateBase, src->Abs);
+}
+
+
+/**
+ * Same as \sa get_src_reg() but if the register is a literal, emit
+ * a brw_reg encoding the literal.
+ * Note that a brw instruction only allows one src operand to be a literal.
+ * For instructions with more than one operand, only the second can be a literal.
+ */
+static struct brw_reg get_src_reg_imm(struct brw_wm_compile *c,
+ const struct prog_instruction *inst,
+ GLuint srcRegIndex, GLuint channel)
+{
+ const struct prog_src_register *src = &inst->SrcReg[srcRegIndex];
+ if (src->File == PROGRAM_CONSTANT) {
+ /* a literal */
+ const int component = GET_SWZ(src->Swizzle, channel);
+ const GLfloat *param =
+ c->fp->program.Base.Parameters->ParameterValues[src->Index];
+ GLfloat value = param[component];
+ if (src->NegateBase)
+ value = -value;
+ if (src->Abs)
+ value = FABSF(value);
+ return brw_imm_f(value);
+ }
+ else {
+ return get_src_reg(c, inst, srcRegIndex, channel);
+ }
}
-/* Subroutines are minimal support for resusable instruction sequences.
- They are implemented as simply as possible to minimise overhead: there
- is no explicit support for communication between the caller and callee
- other than saving the return address in a temporary register, nor is
- there any automatic local storage. This implies that great care is
- required before attempting reentrancy or any kind of nested
- subroutine invocations. */
+
+/**
+ * Subroutines are minimal support for resusable instruction sequences.
+ * They are implemented as simply as possible to minimise overhead: there
+ * is no explicit support for communication between the caller and callee
+ * other than saving the return address in a temporary register, nor is
+ * there any automatic local storage. This implies that great care is
+ * required before attempting reentrancy or any kind of nested
+ * subroutine invocations.
+ */
static void invoke_subroutine( struct brw_wm_compile *c,
enum _subroutine subroutine,
void (*emit)( struct brw_wm_compile * ) )
for (i = 0; i < 4; i++) {
if (inst->DstReg.WriteMask & (1<<i)) {
struct brw_reg src, dst;
- dst = get_dst_reg(c, inst, i, 1);
- src = get_src_reg(c, &inst->SrcReg[0], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src = get_src_reg(c, inst, 0, i);
brw_MOV(p, dst, brw_abs(src));
}
}
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
struct brw_reg src, dst;
- dst = get_dst_reg(c, inst, i, 1) ;
- src = get_src_reg(c, &inst->SrcReg[0], i, 1);
- brw_RNDD(p, dst, src);
+ dst = get_dst_reg(c, inst, i);
+ src = get_src_reg(c, inst, 0, i);
+ brw_RNDZ(p, dst, src);
}
}
brw_set_saturate(p, 0);
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
struct brw_reg src, dst;
- dst = get_dst_reg(c, inst, i, 1);
- src = get_src_reg(c, &inst->SrcReg[0], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src = get_src_reg_imm(c, inst, 0, i);
brw_MOV(p, dst, src);
}
}
struct brw_compile *p = &c->func;
GLuint mask = inst->DstReg.WriteMask;
- dst0 = get_dst_reg(c, inst, 0, 1);
- dst1 = get_dst_reg(c, inst, 1, 1);
+ dst0 = get_dst_reg(c, inst, 0);
+ dst1 = get_dst_reg(c, inst, 1);
/* Calculate pixel centers by adding 1 or 0 to each of the
* micro-tile coordinates passed in r1.
*/
stride(suboffset(r1_uw, 5), 2, 4, 0),
brw_imm_v(0x11001100));
}
-
}
static void emit_delta_xy(struct brw_wm_compile *c,
- struct prog_instruction *inst)
+ struct prog_instruction *inst)
{
struct brw_reg r1 = brw_vec1_grf(1, 0);
struct brw_reg dst0, dst1, src0, src1;
struct brw_compile *p = &c->func;
GLuint mask = inst->DstReg.WriteMask;
- dst0 = get_dst_reg(c, inst, 0, 1);
- dst1 = get_dst_reg(c, inst, 1, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], 0, 1);
- src1 = get_src_reg(c, &inst->SrcReg[0], 1, 1);
+ dst0 = get_dst_reg(c, inst, 0);
+ dst1 = get_dst_reg(c, inst, 1);
+ src0 = get_src_reg(c, inst, 0, 0);
+ src1 = get_src_reg(c, inst, 0, 1);
/* Calc delta X,Y by subtracting origin in r1 from the pixel
* centers.
*/
negate(suboffset(r1,1)));
}
-
}
-
static void fire_fb_write( struct brw_wm_compile *c,
GLuint base_reg,
GLuint nr,
*/
if (c->key.aa_dest_stencil_reg)
nr += 1;
- {
- brw_push_insn_state(p);
- for (channel = 0; channel < 4; channel++) {
- src0 = get_src_reg(c, &inst->SrcReg[0], channel, 1);
- /* mov (8) m2.0<1>:ud r28.0<8;8,1>:ud { Align1 } */
- /* mov (8) m6.0<1>:ud r29.0<8;8,1>:ud { Align1 SecHalf } */
- brw_MOV(p, brw_message_reg(nr + channel), src0);
- }
- /* skip over the regs populated above: */
- nr += 8;
- brw_pop_insn_state(p);
+
+ brw_push_insn_state(p);
+ for (channel = 0; channel < 4; channel++) {
+ src0 = get_src_reg(c, inst, 0, channel);
+ /* mov (8) m2.0<1>:ud r28.0<8;8,1>:ud { Align1 } */
+ /* mov (8) m6.0<1>:ud r29.0<8;8,1>:ud { Align1 SecHalf } */
+ brw_MOV(p, brw_message_reg(nr + channel), src0);
}
+ /* skip over the regs populated above: */
+ nr += 8;
+ brw_pop_insn_state(p);
- if (c->key.source_depth_to_render_target)
- {
- if (c->key.computes_depth) {
- src0 = get_src_reg(c, &inst->SrcReg[2], 2, 1);
- brw_MOV(p, brw_message_reg(nr), src0);
- } else {
- src0 = get_src_reg(c, &inst->SrcReg[1], 1, 1);
- brw_MOV(p, brw_message_reg(nr), src0);
- }
+ if (c->key.source_depth_to_render_target) {
+ if (c->key.computes_depth) {
+ src0 = get_src_reg(c, inst, 2, 2);
+ brw_MOV(p, brw_message_reg(nr), src0);
+ }
+ else {
+ src0 = get_src_reg(c, inst, 1, 1);
+ brw_MOV(p, brw_message_reg(nr), src0);
+ }
+
+ nr += 2;
+ }
- nr += 2;
+ if (c->key.dest_depth_reg) {
+ GLuint comp = c->key.dest_depth_reg / 2;
+ GLuint off = c->key.dest_depth_reg % 2;
+
+ assert(comp == 1);
+ assert(off == 0);
+#if 0
+ /* XXX do we need this code? comp always 1, off always 0, it seems */
+ if (off != 0) {
+ brw_push_insn_state(p);
+ brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+
+ brw_MOV(p, brw_message_reg(nr), offset(arg1[comp],1));
+ /* 2nd half? */
+ brw_MOV(p, brw_message_reg(nr+1), arg1[comp+1]);
+ brw_pop_insn_state(p);
+ }
+ else
+#endif
+ {
+ struct brw_reg src = get_src_reg(c, inst, 1, 1);
+ brw_MOV(p, brw_message_reg(nr), src);
+ }
+ nr += 2;
}
- target = inst->Sampler >> 1;
- eot = inst->Sampler & 1;
+
+ target = inst->Aux >> 1;
+ eot = inst->Aux & 1;
fire_fb_write(c, 0, nr, target, eot);
}
struct brw_reg dst, src0, delta0, delta1;
struct brw_reg interp3;
- dst = get_dst_reg(c, inst, 3, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], 0, 1);
- delta0 = get_src_reg(c, &inst->SrcReg[1], 0, 1);
- delta1 = get_src_reg(c, &inst->SrcReg[1], 1, 1);
+ dst = get_dst_reg(c, inst, 3);
+ src0 = get_src_reg(c, inst, 0, 0);
+ delta0 = get_src_reg(c, inst, 1, 0);
+ delta1 = get_src_reg(c, inst, 1, 1);
interp3 = brw_vec1_grf(src0.nr+1, 4);
/* Calc 1/w - just linterp wpos[3] optimized by putting the
struct brw_reg interp[4];
struct brw_reg dst, delta0, delta1;
struct brw_reg src0;
+ GLuint nr, i;
- src0 = get_src_reg(c, &inst->SrcReg[0], 0, 1);
- delta0 = get_src_reg(c, &inst->SrcReg[1], 0, 1);
- delta1 = get_src_reg(c, &inst->SrcReg[1], 1, 1);
- GLuint nr = src0.nr;
- int i;
+ src0 = get_src_reg(c, inst, 0, 0);
+ delta0 = get_src_reg(c, inst, 1, 0);
+ delta1 = get_src_reg(c, inst, 1, 1);
+ nr = src0.nr;
interp[0] = brw_vec1_grf(nr, 0);
interp[1] = brw_vec1_grf(nr, 4);
for(i = 0; i < 4; i++ ) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
+ dst = get_dst_reg(c, inst, i);
brw_LINE(p, brw_null_reg(), interp[i], delta0);
brw_MAC(p, dst, suboffset(interp[i],1), delta1);
}
struct brw_reg interp[4];
struct brw_reg dst, src0;
+ GLuint nr, i;
- src0 = get_src_reg(c, &inst->SrcReg[0], 0, 1);
- GLuint nr = src0.nr;
- int i;
+ src0 = get_src_reg(c, inst, 0, 0);
+ nr = src0.nr;
interp[0] = brw_vec1_grf(nr, 0);
interp[1] = brw_vec1_grf(nr, 4);
for(i = 0; i < 4; i++ ) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
+ dst = get_dst_reg(c, inst, i);
brw_MOV(p, dst, suboffset(interp[i],3));
}
}
struct brw_reg interp[4];
struct brw_reg dst, delta0, delta1;
struct brw_reg src0, w;
+ GLuint nr, i;
- src0 = get_src_reg(c, &inst->SrcReg[0], 0, 1);
- delta0 = get_src_reg(c, &inst->SrcReg[1], 0, 1);
- delta1 = get_src_reg(c, &inst->SrcReg[1], 1, 1);
- w = get_src_reg(c, &inst->SrcReg[2], 3, 1);
- GLuint nr = src0.nr;
- int i;
+ src0 = get_src_reg(c, inst, 0, 0);
+ delta0 = get_src_reg(c, inst, 1, 0);
+ delta1 = get_src_reg(c, inst, 1, 1);
+ w = get_src_reg(c, inst, 2, 3);
+ nr = src0.nr;
interp[0] = brw_vec1_grf(nr, 0);
interp[1] = brw_vec1_grf(nr, 4);
for(i = 0; i < 4; i++ ) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
+ dst = get_dst_reg(c, inst, i);
brw_LINE(p, brw_null_reg(), interp[i], delta0);
brw_MAC(p, dst, suboffset(interp[i],1),
delta1);
}
}
+/* Sets the destination channels to 1.0 or 0.0 according to glFrontFacing. */
+static void emit_frontfacing(struct brw_wm_compile *c,
+ struct prog_instruction *inst)
+{
+ struct brw_compile *p = &c->func;
+ struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD);
+ struct brw_reg dst;
+ GLuint mask = inst->DstReg.WriteMask;
+ int i;
+
+ for (i = 0; i < 4; i++) {
+ if (mask & (1<<i)) {
+ dst = get_dst_reg(c, inst, i);
+ brw_MOV(p, dst, brw_imm_f(0.0));
+ }
+ }
+
+ /* bit 31 is "primitive is back face", so checking < (1 << 31) gives
+ * us front face
+ */
+ brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, r1_6ud, brw_imm_ud(1 << 31));
+ for (i = 0; i < 4; i++) {
+ if (mask & (1<<i)) {
+ dst = get_dst_reg(c, inst, i);
+ brw_MOV(p, dst, brw_imm_f(1.0));
+ }
+ }
+ brw_set_predicate_control_flag_value(p, 0xff);
+}
+
static void emit_xpd(struct brw_wm_compile *c,
struct prog_instruction *inst)
{
GLuint i1 = (i+1)%3;
if (mask & (1<<i)) {
struct brw_reg src0, src1, dst;
- dst = get_dst_reg(c, inst, i, 1);
- src0 = negate(get_src_reg(c, &inst->SrcReg[0], i2, 1));
- src1 = get_src_reg(c, &inst->SrcReg[1], i1, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = negate(get_src_reg(c, inst, 0, i2));
+ src1 = get_src_reg_imm(c, inst, 1, i1);
brw_MUL(p, brw_null_reg(), src0, src1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i1, 1);
- src1 = get_src_reg(c, &inst->SrcReg[1], i2, 1);
+ src0 = get_src_reg(c, inst, 0, i1);
+ src1 = get_src_reg_imm(c, inst, 1, i2);
brw_set_saturate(p, inst->SaturateMode != SATURATE_OFF);
brw_MAC(p, dst, src0, src1);
brw_set_saturate(p, 0);
int i;
struct brw_compile *p = &c->func;
for (i = 0; i < 3; i++) {
- src0[i] = get_src_reg(c, &inst->SrcReg[0], i, 1);
- src1[i] = get_src_reg(c, &inst->SrcReg[1], i, 1);
+ src0[i] = get_src_reg(c, inst, 0, i);
+ src1[i] = get_src_reg_imm(c, inst, 1, i);
}
- dst = get_dst_reg(c, inst, get_scalar_dst_index(inst), 1);
+ dst = get_dst_reg(c, inst, get_scalar_dst_index(inst));
brw_MUL(p, brw_null_reg(), src0[0], src1[0]);
brw_MAC(p, brw_null_reg(), src0[1], src1[1]);
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
int i;
struct brw_compile *p = &c->func;
for (i = 0; i < 4; i++) {
- src0[i] = get_src_reg(c, &inst->SrcReg[0], i, 1);
- src1[i] = get_src_reg(c, &inst->SrcReg[1], i, 1);
+ src0[i] = get_src_reg(c, inst, 0, i);
+ src1[i] = get_src_reg_imm(c, inst, 1, i);
}
- dst = get_dst_reg(c, inst, get_scalar_dst_index(inst), 1);
+ dst = get_dst_reg(c, inst, get_scalar_dst_index(inst));
brw_MUL(p, brw_null_reg(), src0[0], src1[0]);
brw_MAC(p, brw_null_reg(), src0[1], src1[1]);
brw_MAC(p, brw_null_reg(), src0[2], src1[2]);
int i;
struct brw_compile *p = &c->func;
for (i = 0; i < 4; i++) {
- src0[i] = get_src_reg(c, &inst->SrcReg[0], i, 1);
- src1[i] = get_src_reg(c, &inst->SrcReg[1], i, 1);
+ src0[i] = get_src_reg(c, inst, 0, i);
+ src1[i] = get_src_reg_imm(c, inst, 1, i);
}
- dst = get_dst_reg(c, inst, get_scalar_dst_index(inst), 1);
+ dst = get_dst_reg(c, inst, get_scalar_dst_index(inst));
brw_MUL(p, brw_null_reg(), src0[0], src1[0]);
brw_MAC(p, brw_null_reg(), src0[1], src1[1]);
brw_MAC(p, dst, src0[2], src1[2]);
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
- brw_ADD(p, dst, src0[3], src1[3]);
+ brw_ADD(p, dst, dst, src1[3]);
brw_set_saturate(p, 0);
}
+/**
+ * Emit a scalar instruction, like RCP, RSQ, LOG, EXP.
+ * Note that the result of the function is smeared across the dest
+ * register's X, Y, Z and W channels (subject to writemasking of course).
+ */
static void emit_math1(struct brw_wm_compile *c,
struct prog_instruction *inst, GLuint func)
{
struct brw_compile *p = &c->func;
- struct brw_reg src0, dst;
+ struct brw_reg src0, dst, tmp;
+ const int mark = mark_tmps( c );
+ int i;
- src0 = get_src_reg(c, &inst->SrcReg[0], 0, 1);
- dst = get_dst_reg(c, inst, get_scalar_dst_index(inst), 1);
+ tmp = alloc_tmp(c);
+
+ /* Get first component of source register */
+ src0 = get_src_reg(c, inst, 0, 0);
+
+ /* tmp = func(src0) */
brw_MOV(p, brw_message_reg(2), src0);
brw_math(p,
- dst,
- func,
- (inst->SaturateMode != SATURATE_OFF) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE,
- 2,
- brw_null_reg(),
- BRW_MATH_DATA_VECTOR,
- BRW_MATH_PRECISION_FULL);
+ tmp,
+ func,
+ (inst->SaturateMode != SATURATE_OFF) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE,
+ 2,
+ brw_null_reg(),
+ BRW_MATH_DATA_VECTOR,
+ BRW_MATH_PRECISION_FULL);
+
+ /*tmp.dw1.bits.swizzle = SWIZZLE_XXXX;*/
+
+ /* replicate tmp value across enabled dest channels */
+ for (i = 0; i < 4; i++) {
+ if (inst->DstReg.WriteMask & (1 << i)) {
+ dst = get_dst_reg(c, inst, i);
+ brw_MOV(p, dst, tmp);
+ }
+ }
+
+ release_tmps(c, mark);
}
static void emit_rcp(struct brw_wm_compile *c,
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
for (i = 0 ; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
- src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = get_src_reg(c, inst, 0, i);
+ src1 = get_src_reg_imm(c, inst, 1, i);
brw_ADD(p, dst, src0, src1);
}
}
brw_set_saturate(p, 0);
}
+static void emit_arl(struct brw_wm_compile *c,
+ struct prog_instruction *inst)
+{
+ struct brw_compile *p = &c->func;
+ struct brw_reg src0, addr_reg;
+ brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
+ addr_reg = brw_uw8_reg(BRW_ARCHITECTURE_REGISTER_FILE,
+ BRW_ARF_ADDRESS, 0);
+ src0 = get_src_reg(c, inst, 0, 0); /* channel 0 */
+ brw_MOV(p, addr_reg, src0);
+ brw_set_saturate(p, 0);
+}
+
static void emit_sub(struct brw_wm_compile *c,
struct prog_instruction *inst)
{
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
for (i = 0 ; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
- src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = get_src_reg(c, inst, 0, i);
+ src1 = get_src_reg_imm(c, inst, 1, i);
brw_ADD(p, dst, src0, negate(src1));
}
}
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
for (i = 0 ; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
- src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = get_src_reg(c, inst, 0, i);
+ src1 = get_src_reg_imm(c, inst, 1, i);
brw_MUL(p, dst, src0, src1);
}
}
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
for (i = 0 ; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = get_src_reg_imm(c, inst, 0, i);
brw_FRC(p, dst, src0);
}
}
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
for (i = 0 ; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = get_src_reg_imm(c, inst, 0, i);
brw_RNDD(p, dst, src0);
}
}
brw_push_insn_state(p);
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
- src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = get_src_reg(c, inst, 0, i);
+ src1 = get_src_reg_imm(c, inst, 1, i);
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
brw_MOV(p, dst, src0);
brw_set_saturate(p, 0);
brw_push_insn_state(p);
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
- src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = get_src_reg_imm(c, inst, 0, i);
+ src1 = get_src_reg(c, inst, 1, i);
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
brw_MOV(p, dst, src0);
brw_set_saturate(p, 0);
{
struct brw_compile *p = &c->func;
struct brw_reg dst, src0, src1;
- dst = get_dst_reg(c, inst, get_scalar_dst_index(inst), 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], 0, 1);
- src1 = get_src_reg(c, &inst->SrcReg[1], 0, 1);
+ dst = get_dst_reg(c, inst, get_scalar_dst_index(inst));
+ src0 = get_src_reg_imm(c, inst, 0, 0);
+ src1 = get_src_reg_imm(c, inst, 1, 0);
brw_MOV(p, brw_message_reg(2), src0);
brw_MOV(p, brw_message_reg(3), src1);
int mark = mark_tmps(c);
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = get_src_reg(c, inst, 0, i);
- src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
+ src1 = get_src_reg_imm(c, inst, 1, i);
if (src1.nr == dst.nr) {
tmp1 = alloc_tmp(c);
} else
tmp1 = src1;
- src2 = get_src_reg(c, &inst->SrcReg[2], i, 1);
+ src2 = get_src_reg(c, inst, 2, i);
if (src2.nr == dst.nr) {
tmp2 = alloc_tmp(c);
brw_MOV(p, tmp2, src2);
}
}
+/**
+ * For GLSL shaders, this KIL will be unconditional.
+ * It may be contained inside an IF/ENDIF structure of course.
+ */
static void emit_kil(struct brw_wm_compile *c)
{
- struct brw_compile *p = &c->func;
- struct brw_reg depth = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
- brw_push_insn_state(p);
- brw_set_mask_control(p, BRW_MASK_DISABLE);
- brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); //IMASK
- brw_AND(p, depth, c->emit_mask_reg, depth);
- brw_pop_insn_state(p);
+ struct brw_compile *p = &c->func;
+ struct brw_reg depth = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
+ brw_push_insn_state(p);
+ brw_set_mask_control(p, BRW_MASK_DISABLE);
+ brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); //IMASK
+ brw_AND(p, depth, c->emit_mask_reg, depth);
+ brw_pop_insn_state(p);
}
static void emit_mad(struct brw_wm_compile *c,
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
- src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
- src2 = get_src_reg(c, &inst->SrcReg[2], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = get_src_reg(c, inst, 0, i);
+ src1 = get_src_reg_imm(c, inst, 1, i);
+ src2 = get_src_reg_imm(c, inst, 2, i);
brw_MUL(p, dst, src0, src1);
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
- src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
- src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
+ dst = get_dst_reg(c, inst, i);
+ src0 = get_src_reg(c, inst, 0, i);
+ src1 = get_src_reg_imm(c, inst, 1, i);
brw_push_insn_state(p);
brw_CMP(p, brw_null_reg(), cond, src0, src1);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
struct brw_reg dst;
struct brw_reg src0, w;
GLuint nr, i;
- src0 = get_src_reg(c, &inst->SrcReg[0], 0, 1);
- w = get_src_reg(c, &inst->SrcReg[1], 3, 1);
+ src0 = get_src_reg(c, inst, 0, 0);
+ w = get_src_reg(c, inst, 1, 3);
nr = src0.nr;
interp[0] = brw_vec1_grf(nr, 0);
interp[1] = brw_vec1_grf(nr, 4);
brw_set_saturate(p, inst->SaturateMode != SATURATE_OFF);
for(i = 0; i < 4; i++ ) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
+ dst = get_dst_reg(c, inst, i);
brw_MOV(p, dst, interp[i]);
brw_MUL(p, dst, dst, w);
}
struct brw_reg src0, w;
GLuint nr, i;
- src0 = get_src_reg(c, &inst->SrcReg[0], 0, 1);
+ src0 = get_src_reg(c, inst, 0, 0);
nr = src0.nr;
- w = get_src_reg(c, &inst->SrcReg[1], 3, 1);
+ w = get_src_reg(c, inst, 1, 3);
interp[0] = brw_vec1_grf(nr, 0);
interp[1] = brw_vec1_grf(nr, 4);
interp[2] = brw_vec1_grf(nr+1, 0);
brw_set_saturate(p, inst->SaturateMode != SATURATE_OFF);
for(i = 0; i < 4; i++ ) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
+ dst = get_dst_reg(c, inst, i);
brw_MOV(p, dst, suboffset(interp[i], 1));
brw_MUL(p, dst, dst, w);
}
brw_set_saturate(p, 0);
}
-static __inline struct brw_reg high_words( struct brw_reg reg )
+static INLINE struct brw_reg high_words( struct brw_reg reg )
{
return stride( suboffset( retype( reg, BRW_REGISTER_TYPE_W ), 1 ),
0, 8, 2 );
}
-static __inline struct brw_reg low_words( struct brw_reg reg )
+static INLINE struct brw_reg low_words( struct brw_reg reg )
{
return stride( retype( reg, BRW_REGISTER_TYPE_W ), 0, 8, 2 );
}
-static __inline struct brw_reg even_bytes( struct brw_reg reg )
+static INLINE struct brw_reg even_bytes( struct brw_reg reg )
{
return stride( retype( reg, BRW_REGISTER_TYPE_B ), 0, 16, 2 );
}
-static __inline struct brw_reg odd_bytes( struct brw_reg reg )
+static INLINE struct brw_reg odd_bytes( struct brw_reg reg )
{
return stride( suboffset( retype( reg, BRW_REGISTER_TYPE_B ), 1 ),
0, 16, 2 );
assert( mark == 0 );
- src = get_src_reg( c, inst->SrcReg, 0, 1 );
+ src = get_src_reg( c, inst, 0, 0 );
param = alloc_tmp( c );
brw_set_saturate( p, inst->SaturateMode == SATURATE_ZERO_ONE );
for (i = 0 ; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
+ dst = get_dst_reg(c, inst, i);
brw_MOV( p, dst, param );
}
}
assert( mark == 0 );
- src0 = get_src_reg( c, inst->SrcReg, 0, 1 );
- src1 = get_src_reg( c, inst->SrcReg, 1, 1 );
+ src0 = get_src_reg( c, inst, 0, 0 );
+ src1 = get_src_reg( c, inst, 0, 1 );
param0 = alloc_tmp( c );
param1 = alloc_tmp( c );
brw_set_saturate( p, inst->SaturateMode == SATURATE_ZERO_ONE );
for (i = 0 ; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
+ dst = get_dst_reg(c, inst, i);
brw_MOV( p, dst, param0 );
}
}
release_tmps( c, mark );
}
-/* The three-dimensional case is much like the one- and two- versions above,
- but since the number of corners is rapidly growing we now pack 16 16-bit
- hashes into each register to extract more parallelism from the EUs. */
+/**
+ * The three-dimensional case is much like the one- and two- versions above,
+ * but since the number of corners is rapidly growing we now pack 16 16-bit
+ * hashes into each register to extract more parallelism from the EUs.
+ */
static void noise3_sub( struct brw_wm_compile *c ) {
struct brw_compile *p = &c->func;
assert( mark == 0 );
- src0 = get_src_reg( c, inst->SrcReg, 0, 1 );
- src1 = get_src_reg( c, inst->SrcReg, 1, 1 );
- src2 = get_src_reg( c, inst->SrcReg, 2, 1 );
+ src0 = get_src_reg( c, inst, 0, 0 );
+ src1 = get_src_reg( c, inst, 0, 1 );
+ src2 = get_src_reg( c, inst, 0, 2 );
param0 = alloc_tmp( c );
param1 = alloc_tmp( c );
brw_set_saturate( p, inst->SaturateMode == SATURATE_ZERO_ONE );
for (i = 0 ; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
+ dst = get_dst_reg(c, inst, i);
brw_MOV( p, dst, param0 );
}
}
release_tmps( c, mark );
}
-/* For the four-dimensional case, the little micro-optimisation benefits
- we obtain by unrolling all the loops aren't worth the massive bloat it
- now causes. Instead, we loop twice around performing a similar operation
- to noise3, once for the w=0 cube and once for the w=1, with a bit more
- code to glue it all together. */
-static void noise4_sub( struct brw_wm_compile *c ) {
-
+/**
+ * For the four-dimensional case, the little micro-optimisation benefits
+ * we obtain by unrolling all the loops aren't worth the massive bloat it
+ * now causes. Instead, we loop twice around performing a similar operation
+ * to noise3, once for the w=0 cube and once for the w=1, with a bit more
+ * code to glue it all together.
+ */
+static void noise4_sub( struct brw_wm_compile *c )
+{
struct brw_compile *p = &c->func;
struct brw_reg param[ 4 ],
x0y0, x0y1, x1y0, x1y1, /* gradients at four of the corners */
assert( mark == 0 );
- src0 = get_src_reg( c, inst->SrcReg, 0, 1 );
- src1 = get_src_reg( c, inst->SrcReg, 1, 1 );
- src2 = get_src_reg( c, inst->SrcReg, 2, 1 );
- src3 = get_src_reg( c, inst->SrcReg, 3, 1 );
+ src0 = get_src_reg( c, inst, 0, 0 );
+ src1 = get_src_reg( c, inst, 0, 1 );
+ src2 = get_src_reg( c, inst, 0, 2 );
+ src3 = get_src_reg( c, inst, 0, 3 );
param0 = alloc_tmp( c );
param1 = alloc_tmp( c );
brw_set_saturate( p, inst->SaturateMode == SATURATE_ZERO_ONE );
for (i = 0 ; i < 4; i++) {
if (mask & (1<<i)) {
- dst = get_dst_reg(c, inst, i, 1);
+ dst = get_dst_reg(c, inst, i);
brw_MOV( p, dst, param0 );
}
}
GLuint mask = inst->DstReg.WriteMask;
struct brw_reg src0[2], dst[2];
- dst[0] = get_dst_reg(c, inst, 0, 1);
- dst[1] = get_dst_reg(c, inst, 1, 1);
+ dst[0] = get_dst_reg(c, inst, 0);
+ dst[1] = get_dst_reg(c, inst, 1);
- src0[0] = get_src_reg(c, &inst->SrcReg[0], 0, 1);
- src0[1] = get_src_reg(c, &inst->SrcReg[0], 1, 1);
+ src0[0] = get_src_reg(c, inst, 0, 0);
+ src0[1] = get_src_reg(c, inst, 0, 1);
/* Calculate the pixel offset from window bottom left into destination
* X and Y channels.
}
/* TODO
- BIAS on SIMD8 not workind yet...
+ BIAS on SIMD8 not working yet...
*/
static void emit_txb(struct brw_wm_compile *c,
struct prog_instruction *inst)
struct brw_compile *p = &c->func;
struct brw_reg dst[4], src[4], payload_reg;
GLuint unit = c->fp->program.Base.SamplerUnits[inst->TexSrcUnit];
-
GLuint i;
+
payload_reg = get_reg(c, PROGRAM_PAYLOAD, PAYLOAD_DEPTH, 0, 1, 0, 0);
+
for (i = 0; i < 4; i++)
- dst[i] = get_dst_reg(c, inst, i, 1);
+ dst[i] = get_dst_reg(c, inst, i);
for (i = 0; i < 4; i++)
- src[i] = get_src_reg(c, &inst->SrcReg[0], i, 1);
+ src[i] = get_src_reg(c, inst, 0, i);
switch (inst->TexSrcTarget) {
case TEXTURE_1D_INDEX:
- brw_MOV(p, brw_message_reg(2), src[0]);
- brw_MOV(p, brw_message_reg(3), brw_imm_f(0));
- brw_MOV(p, brw_message_reg(4), brw_imm_f(0));
+ brw_MOV(p, brw_message_reg(2), src[0]); /* s coord */
+ brw_MOV(p, brw_message_reg(3), brw_imm_f(0)); /* t coord */
+ brw_MOV(p, brw_message_reg(4), brw_imm_f(0)); /* r coord */
break;
case TEXTURE_2D_INDEX:
case TEXTURE_RECT_INDEX:
brw_MOV(p, brw_message_reg(4), src[2]);
break;
}
- brw_MOV(p, brw_message_reg(5), src[3]);
- brw_MOV(p, brw_message_reg(6), brw_imm_f(0));
+ brw_MOV(p, brw_message_reg(5), src[3]); /* bias */
+ brw_MOV(p, brw_message_reg(6), brw_imm_f(0)); /* ref (unused?) */
brw_SAMPLE(p,
- retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW),
- 1,
- retype(payload_reg, BRW_REGISTER_TYPE_UW),
- unit + MAX_DRAW_BUFFERS, /* surface */
- unit, /* sampler */
- inst->DstReg.WriteMask,
- BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS,
- 4,
- 4,
- 0);
+ retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW), /* dest */
+ 1, /* msg_reg_nr */
+ retype(payload_reg, BRW_REGISTER_TYPE_UW), /* src0 */
+ unit + MAX_DRAW_BUFFERS, /* surface */
+ unit, /* sampler */
+ inst->DstReg.WriteMask, /* writemask */
+ BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS, /* msg_type */
+ 4, /* response_length */
+ 4, /* msg_length */
+ 0); /* eot */
}
+
static void emit_tex(struct brw_wm_compile *c,
struct prog_instruction *inst)
{
struct brw_compile *p = &c->func;
struct brw_reg dst[4], src[4], payload_reg;
GLuint unit = c->fp->program.Base.SamplerUnits[inst->TexSrcUnit];
-
GLuint msg_len;
GLuint i, nr;
GLuint emit;
payload_reg = get_reg(c, PROGRAM_PAYLOAD, PAYLOAD_DEPTH, 0, 1, 0, 0);
for (i = 0; i < 4; i++)
- dst[i] = get_dst_reg(c, inst, i, 1);
+ dst[i] = get_dst_reg(c, inst, i);
for (i = 0; i < 4; i++)
- src[i] = get_src_reg(c, &inst->SrcReg[0], i, 1);
-
+ src[i] = get_src_reg(c, inst, 0, i);
switch (inst->TexSrcTarget) {
case TEXTURE_1D_INDEX:
}
msg_len = 1;
+ /* move/load S, T, R coords */
for (i = 0; i < nr; i++) {
static const GLuint swz[4] = {0,1,2,2};
if (emit & (1<<i))
}
if (shadow) {
- brw_MOV(p, brw_message_reg(5), brw_imm_f(0));
- brw_MOV(p, brw_message_reg(6), src[2]);
+ brw_MOV(p, brw_message_reg(5), brw_imm_f(0)); /* lod / bias */
+ brw_MOV(p, brw_message_reg(6), src[2]); /* ref value / R coord */
}
brw_SAMPLE(p,
- retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW),
- 1,
- retype(payload_reg, BRW_REGISTER_TYPE_UW),
- unit + MAX_DRAW_BUFFERS, /* surface */
- unit, /* sampler */
- inst->DstReg.WriteMask,
- BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE,
- 4,
- shadow ? 6 : 4,
- 0);
+ retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW), /* dest */
+ 1, /* msg_reg_nr */
+ retype(payload_reg, BRW_REGISTER_TYPE_UW), /* src0 */
+ unit + MAX_DRAW_BUFFERS, /* surface */
+ unit, /* sampler */
+ inst->DstReg.WriteMask, /* writemask */
+ BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE, /* msg_type */
+ 4, /* response_length */
+ shadow ? 6 : 4, /* msg_length */
+ 0); /* eot */
if (shadow)
brw_MOV(p, dst[3], brw_imm_f(1.0));
}
+
+static void emit_get_constant(struct brw_context *brw,
+ struct brw_wm_compile *c,
+ struct prog_instruction *inst,
+ GLuint constIndex)
+{
+ struct brw_compile *p = &c->func;
+ struct brw_reg dst[4];
+ GLuint i;
+ const int mark = mark_tmps( c );
+ struct brw_reg writeback_reg[4];
+
+ /* XXX only need 1 temp reg??? */
+ for (i = 0; i < 4; i++) {
+ writeback_reg[i] = alloc_tmp(c);
+ }
+
+ for (i = 0; i < 4; i++) {
+ dst[i] = get_dst_reg(c, inst, i);
+ }
+
+ /* Get float[4] vector from constant buffer */
+ brw_dp_READ_4(p,
+ writeback_reg[0], /* first writeback dest */
+ 1, /* msg_reg */
+ GL_FALSE, /* rel addr? */
+ 16 * constIndex, /* byte offset */
+ BRW_WM_MAX_SURF - 1 /* surface, binding table index */
+ );
+
+ /* Extract the four channel values, smear across dest registers */
+ for (i = 0; i < 4; i++) {
+ /* extract 1 float from the writeback reg */
+ struct brw_reg new_src = stride(writeback_reg[0], 0, 1, 0);
+ new_src.subnr = i * 4;
+ /* and smear it into the dest register */
+ brw_MOV(p, dst[i], new_src);
+ }
+
+ release_tmps( c, mark );
+}
+
+
+/**
+ * Resolve subroutine calls after code emit is done.
+ */
static void post_wm_emit( struct brw_wm_compile *c )
{
- GLuint nr_insns = c->fp->program.Base.NumInstructions;
- GLuint insn, target_insn;
- struct prog_instruction *inst1, *inst2;
- struct brw_instruction *brw_inst1, *brw_inst2;
- int offset;
- for (insn = 0; insn < nr_insns; insn++) {
- inst1 = &c->fp->program.Base.Instructions[insn];
- brw_inst1 = inst1->Data;
- switch (inst1->Opcode) {
- case OPCODE_CAL:
- target_insn = inst1->BranchTarget;
- inst2 = &c->fp->program.Base.Instructions[target_insn];
- brw_inst2 = inst2->Data;
- offset = brw_inst2 - brw_inst1;
- brw_set_src1(brw_inst1, brw_imm_d(offset*16));
- break;
- default:
- break;
- }
- }
+ brw_resolve_cals(&c->func);
}
static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
for (i = 0; i < c->nr_fp_insns; i++) {
struct prog_instruction *inst = &c->prog_instructions[i];
- struct prog_instruction *orig_inst;
-
- if ((orig_inst = inst->Data) != 0)
- orig_inst->Data = current_insn(p);
if (inst->CondUpdate)
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
case WM_FB_WRITE:
emit_fb_write(c, inst);
break;
+ case WM_FRONTFACING:
+ emit_frontfacing(c, inst);
+ break;
case OPCODE_ABS:
emit_abs(c, inst);
break;
case OPCODE_ADD:
emit_add(c, inst);
break;
+ case OPCODE_ARL:
+ emit_arl(c, inst);
+ break;
case OPCODE_SUB:
emit_sub(c, inst);
break;
emit_trunc(c, inst);
break;
case OPCODE_MOV:
+#if 0
+ /* test hook for new constant buffer code */
+ if (inst->SrcReg[0].File == PROGRAM_UNIFORM) {
+ emit_get_constant(brw, c, inst, inst->SrcReg[0].Index);
+ }
+ else {
+ emit_mov(c, inst);
+ }
+#else
emit_mov(c, inst);
+#endif
break;
case OPCODE_DP3:
emit_dp3(c, inst);
brw_ENDIF(p, if_inst[--if_insn]);
break;
case OPCODE_BGNSUB:
+ brw_save_label(p, inst->Comment, p->nr_insn);
+ break;
case OPCODE_ENDSUB:
+ /* no-op */
break;
case OPCODE_CAL:
brw_push_insn_state(p);
brw_set_access_mode(p, BRW_ALIGN_16);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(4));
- orig_inst = inst->Data;
- orig_inst->Data = &p->store[p->nr_insn];
+ brw_save_call(&c->func, inst->Comment, p->nr_insn);
brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
brw_pop_insn_state(p);
break;
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
post_wm_emit(c);
- for (i = 0; i < c->fp->program.Base.NumInstructions; i++)
- c->fp->program.Base.Instructions[i].Data = NULL;
+
+ if (c->reg_index >= BRW_WM_MAX_GRF) {
+ _mesa_problem(NULL, "Ran out of registers in brw_wm_emit_glsl()");
+ /* XXX we need to do some proper error recovery here */
+ }
}
+
+/**
+ * Do GPU code generation for shaders that use GLSL features such as
+ * flow control. Other shaders will be compiled with the
+ */
void brw_wm_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c)
{
+ if (INTEL_DEBUG & DEBUG_WM) {
+ _mesa_printf("brw_wm_glsl_emit:\n");
+ }
+
+ /* initial instruction translation/simplification */
brw_wm_pass_fp(c);
+
+ /* actual code generation */
brw_wm_emit_glsl(brw, c);
+
+ if (INTEL_DEBUG & DEBUG_WM) {
+ brw_wm_print_program(c, "brw_wm_glsl_emit done");
+ }
+
c->prog_data.total_grf = c->reg_index;
c->prog_data.total_scratch = 0;
}