struct nv50_reg *r_brdc;
struct nv50_reg *r_dst[4];
+ struct nv50_reg reg_instances[16];
+ unsigned reg_instance_nr;
+
unsigned interp_mode[32];
/* perspective interpolation registers */
struct nv50_reg *iv_p;
boolean allow32;
};
+static INLINE struct nv50_reg *
+reg_instance(struct nv50_pc *pc, struct nv50_reg *reg)
+{
+ struct nv50_reg *dup = NULL;
+ if (reg) {
+ assert(pc->reg_instance_nr < 16);
+ dup = &pc->reg_instances[pc->reg_instance_nr++];
+ *dup = *reg;
+ reg->mod = 0;
+ }
+ return dup;
+}
+
static INLINE void
ctor_reg(struct nv50_reg *reg, unsigned type, int index, int hw)
{
emit_sub(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0,
struct nv50_reg *src1)
{
- assert(src0 != src1);
src1->mod ^= NV50_MOD_NEG;
emit_add(pc, dst, src0, src1);
src1->mod ^= NV50_MOD_NEG;
emit_msb(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0,
struct nv50_reg *src1, struct nv50_reg *src2)
{
- assert(src2 != src0 && src2 != src1);
src2->mod ^= NV50_MOD_NEG;
emit_mad(pc, dst, src0, src1, src2);
src2->mod ^= NV50_MOD_NEG;
static void
load_cube_tex_coords(struct nv50_pc *pc, struct nv50_reg *t[4],
- struct nv50_reg **src, boolean proj)
+ struct nv50_reg **src, unsigned arg, boolean proj)
{
int mod[3] = { src[0]->mod, src[1]->mod, src[2]->mod };
if (proj && 0 /* looks more correct without this */)
emit_mul(pc, t[2], t[2], src[3]);
+ else
+ if (arg == 4) /* there is no textureProj(samplerCubeShadow) */
+ emit_mov(pc, t[3], src[3]);
+
emit_flop(pc, 0, t[2], t[2]);
emit_mul(pc, t[0], src[0], t[2]);
}
static void
-emit_tex(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask,
- struct nv50_reg **src, unsigned unit, unsigned type, boolean proj)
+load_proj_tex_coords(struct nv50_pc *pc, struct nv50_reg *t[4],
+ struct nv50_reg **src, unsigned dim, unsigned arg)
{
- struct nv50_reg *t[4];
- struct nv50_program_exec *e;
+ unsigned c, mode;
+
+ if (src[0]->type == P_TEMP && src[0]->rhw != -1) {
+ mode = pc->interp_mode[src[0]->index] | INTERP_PERSPECTIVE;
+
+ t[3]->rhw = src[3]->rhw;
+ emit_interp(pc, t[3], NULL, (mode & INTERP_CENTROID));
+ emit_flop(pc, 0, t[3], t[3]);
- unsigned c, mode, dim;
+ for (c = 0; c < dim; ++c) {
+ t[c]->rhw = src[c]->rhw;
+ emit_interp(pc, t[c], t[3], mode);
+ }
+ if (arg != dim) { /* depth reference value */
+ t[dim]->rhw = src[2]->rhw;
+ emit_interp(pc, t[dim], t[3], mode);
+ }
+ } else {
+ /* XXX: for some reason the blob sometimes uses MAD
+ * (mad f32 $rX $rY $rZ neg $r63)
+ */
+ emit_flop(pc, 0, t[3], src[3]);
+ for (c = 0; c < dim; ++c)
+ emit_mul(pc, t[c], src[c], t[3]);
+ if (arg != dim) /* depth reference value */
+ emit_mul(pc, t[dim], src[2], t[3]);
+ }
+}
+static INLINE void
+get_tex_dim(unsigned type, unsigned *dim, unsigned *arg)
+{
switch (type) {
case TGSI_TEXTURE_1D:
- dim = 1;
+ *arg = *dim = 1;
+ break;
+ case TGSI_TEXTURE_SHADOW1D:
+ *dim = 1;
+ *arg = 2;
break;
case TGSI_TEXTURE_UNKNOWN:
case TGSI_TEXTURE_2D:
- case TGSI_TEXTURE_SHADOW1D: /* XXX: x, z */
case TGSI_TEXTURE_RECT:
- dim = 2;
+ *arg = *dim = 2;
+ break;
+ case TGSI_TEXTURE_SHADOW2D:
+ case TGSI_TEXTURE_SHADOWRECT:
+ *dim = 2;
+ *arg = 3;
break;
case TGSI_TEXTURE_3D:
case TGSI_TEXTURE_CUBE:
- case TGSI_TEXTURE_SHADOW2D:
- case TGSI_TEXTURE_SHADOWRECT: /* XXX */
- dim = 3;
+ *dim = *arg = 3;
break;
default:
assert(0);
break;
}
+}
+
+static void
+emit_tex(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask,
+ struct nv50_reg **src, unsigned unit, unsigned type,
+ boolean proj, int bias_lod)
+{
+ struct nv50_reg *t[4];
+ struct nv50_program_exec *e;
+ unsigned c, dim, arg;
- /* some cards need t[0]'s hw index to be a multiple of 4 */
+ /* t[i] must be within a single 128 bit super-reg */
alloc_temp4(pc, t, 0);
+ e = exec(pc);
+ e->inst[0] = 0xf0000000;
+ set_long(pc, e);
+ set_dst(pc, t[0], e);
+
+ /* TIC and TSC binding indices (TSC is ignored as TSC_LINKED = TRUE): */
+ e->inst[0] |= (unit << 9) /* | (unit << 17) */;
+
+ /* live flag (don't set if TEX results affect input to another TEX): */
+ /* e->inst[0] |= 0x00000004; */
+
+ get_tex_dim(type, &dim, &arg);
+
if (type == TGSI_TEXTURE_CUBE) {
- load_cube_tex_coords(pc, t, src, proj);
+ e->inst[0] |= 0x08000000;
+ load_cube_tex_coords(pc, t, src, arg, proj);
} else
- if (proj) {
- if (src[0]->type == P_TEMP && src[0]->rhw != -1) {
- mode = pc->interp_mode[src[0]->index];
-
- t[3]->rhw = src[3]->rhw;
- emit_interp(pc, t[3], NULL, (mode & INTERP_CENTROID));
- emit_flop(pc, 0, t[3], t[3]);
-
- for (c = 0; c < dim; c++) {
- t[c]->rhw = src[c]->rhw;
- emit_interp(pc, t[c], t[3],
- (mode | INTERP_PERSPECTIVE));
- }
- } else {
- emit_flop(pc, 0, t[3], src[3]);
- for (c = 0; c < dim; c++)
- emit_mul(pc, t[c], src[c], t[3]);
-
- /* XXX: for some reason the blob sometimes uses MAD:
- * emit_mad(pc, t[c], src[0][c], t[3], t[3])
- * pc->p->exec_tail->inst[1] |= 0x080fc000;
- */
- }
- } else {
+ if (proj)
+ load_proj_tex_coords(pc, t, src, dim, arg);
+ else {
for (c = 0; c < dim; c++)
emit_mov(pc, t[c], src[c]);
+ if (arg != dim) /* depth reference value (always src.z here) */
+ emit_mov(pc, t[dim], src[2]);
}
- e = exec(pc);
- set_long(pc, e);
- e->inst[0] |= 0xf0000000;
- e->inst[1] |= 0x00000004;
- set_dst(pc, t[0], e);
- e->inst[0] |= (unit << 9);
-
- if (dim == 2)
- e->inst[0] |= 0x00400000;
- else
- if (dim == 3) {
- e->inst[0] |= 0x00800000;
- if (type == TGSI_TEXTURE_CUBE)
- e->inst[0] |= 0x08000000;
+ if (bias_lod) {
+ assert(arg < 4);
+ emit_mov(pc, t[arg++], src[3]);
+ e->inst[1] |= (bias_lod < 0) ? 0x20000000 : 0x40000000;
}
+ e->inst[0] |= (arg - 1) << 22;
+
e->inst[0] |= (mask & 0x3) << 25;
e->inst[1] |= (mask & 0xc) << 12;
static boolean
negate_supported(const struct tgsi_full_instruction *insn, int i)
{
- int s;
-
switch (insn->Instruction.Opcode) {
case TGSI_OPCODE_DDY:
case TGSI_OPCODE_DP3:
case TGSI_OPCODE_ADD:
case TGSI_OPCODE_SUB:
case TGSI_OPCODE_MAD:
- break;
+ return TRUE;
case TGSI_OPCODE_POW:
if (i == 1)
- break;
+ return TRUE;
return FALSE;
default:
return FALSE;
}
-
- /* Watch out for possible multiple uses of an nv50_reg, we
- * can't use nv50_reg::neg in these cases.
- */
- for (s = 0; s < insn->Instruction.NumSrcRegs; ++s) {
- if (s == i)
- continue;
- if ((insn->Src[s].Register.Index ==
- insn->Src[i].Register.Index) &&
- (insn->Src[s].Register.File ==
- insn->Src[i].Register.File))
- return FALSE;
- }
-
- return TRUE;
}
/* Return a read mask for source registers deduced from opcode & write mask. */
case TGSI_OPCODE_RSQ:
case TGSI_OPCODE_SCS:
return 0x1;
+ case TGSI_OPCODE_IF:
+ return 0x1;
case TGSI_OPCODE_LIT:
return 0xb;
case TGSI_OPCODE_TEX:
+ case TGSI_OPCODE_TXB:
+ case TGSI_OPCODE_TXL:
case TGSI_OPCODE_TXP:
{
const struct tgsi_instruction_texture *tex;
tex = &insn->Texture;
mask = 0x7;
- if (insn->Instruction.Opcode == TGSI_OPCODE_TXP)
- mask |= 0x8;
+ if (insn->Instruction.Opcode != TGSI_OPCODE_TEX &&
+ insn->Instruction.Opcode != TGSI_OPCODE_TXD)
+ mask |= 0x8; /* bias, lod or proj */
switch (tex->Texture) {
case TGSI_TEXTURE_1D:
mask &= 0x9;
break;
+ case TGSI_TEXTURE_SHADOW1D:
+ mask &= 0x5;
+ break;
case TGSI_TEXTURE_2D:
mask &= 0xb;
break;
case TGSI_OPCODE_LIT:
case TGSI_OPCODE_SCS:
case TGSI_OPCODE_TEX:
+ case TGSI_OPCODE_TXB:
+ case TGSI_OPCODE_TXL:
case TGSI_OPCODE_TXP:
/* these take care of dangerous swizzles themselves */
return 0x0;
for (c = 0; c < 4; c++)
if (src_mask & (1 << c))
- src[i][c] = tgsi_src(pc, c, fs, neg_supp);
+ src[i][c] = reg_instance(pc,
+ tgsi_src(pc, c, fs, neg_supp));
}
brdc = temp = pc->r_brdc;
break;
case TGSI_OPCODE_TEX:
emit_tex(pc, dst, mask, src[0], unit,
- inst->Texture.Texture, FALSE);
+ inst->Texture.Texture, FALSE, 0);
+ break;
+ case TGSI_OPCODE_TXB:
+ emit_tex(pc, dst, mask, src[0], unit,
+ inst->Texture.Texture, FALSE, -1);
+ break;
+ case TGSI_OPCODE_TXL:
+ emit_tex(pc, dst, mask, src[0], unit,
+ inst->Texture.Texture, FALSE, 1);
break;
case TGSI_OPCODE_TXP:
emit_tex(pc, dst, mask, src[0], unit,
- inst->Texture.Texture, TRUE);
+ inst->Texture.Texture, TRUE, 0);
break;
case TGSI_OPCODE_TRUNC:
for (c = 0; c < 4; c++) {
for (c = 0; c < 4; c++) {
if (!src[i][c])
continue;
- src[i][c]->mod = 0;
- if (src[i][c]->index == -1 && src[i][c]->type == P_IMMD)
- FREE(src[i][c]);
- else
if (src[i][c]->acc < 0 && src[i][c]->type == P_CONST)
FREE(src[i][c]); /* indirect constant */
}
}
kill_temp_temp(pc);
+ pc->reg_instance_nr = 0;
+
return TRUE;
}