summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/nv50
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/drivers/nv50')
-rw-r--r--src/gallium/drivers/nv50/nv50_context.c8
-rw-r--r--src/gallium/drivers/nv50/nv50_context.h3
-rw-r--r--src/gallium/drivers/nv50/nv50_miptree.c1
-rw-r--r--src/gallium/drivers/nv50/nv50_program.c1437
-rw-r--r--src/gallium/drivers/nv50/nv50_program.h35
-rw-r--r--src/gallium/drivers/nv50/nv50_screen.c4
-rw-r--r--src/gallium/drivers/nv50/nv50_state.c6
-rw-r--r--src/gallium/drivers/nv50/nv50_state_validate.c17
-rw-r--r--src/gallium/drivers/nv50/nv50_surface.c6
-rw-r--r--src/gallium/drivers/nv50/nv50_transfer.c4
10 files changed, 979 insertions, 542 deletions
diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c
index 6e8f4f9750..fca078b174 100644
--- a/src/gallium/drivers/nv50/nv50_context.c
+++ b/src/gallium/drivers/nv50/nv50_context.c
@@ -37,11 +37,12 @@ nv50_flush(struct pipe_context *pipe, unsigned flags,
/* We need this in the ddx for reliable composite, not sure what we're
* actually flushing. We generate all our own flushes with flags = 0. */
- WAIT_RING(chan, 3);
+ WAIT_RING(chan, 2);
BEGIN_RING(chan, eng2d, 0x0110, 1);
OUT_RING (chan, 0);
- FIRE_RING(chan);
+ if (flags & PIPE_FLUSH_FRAME)
+ FIRE_RING(chan);
}
static void
@@ -110,6 +111,9 @@ nv50_create(struct pipe_screen *pscreen, unsigned pctx_id)
nv50->pipe.is_texture_referenced = nv50_is_texture_referenced;
nv50->pipe.is_buffer_referenced = nv50_is_buffer_referenced;
+ screen->base.channel->user_private = nv50;
+ screen->base.channel->flush_notify = nv50_state_flush_notify;
+
nv50_init_surface_functions(nv50);
nv50_init_state_functions(nv50);
nv50_init_query_functions(nv50);
diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h
index 1e9e8e49bf..4608854d71 100644
--- a/src/gallium/drivers/nv50/nv50_context.h
+++ b/src/gallium/drivers/nv50/nv50_context.h
@@ -116,6 +116,7 @@ struct nv50_state {
unsigned miptree_nr;
struct nouveau_stateobj *vertprog;
struct nouveau_stateobj *fragprog;
+ struct nouveau_stateobj *programs;
struct nouveau_stateobj *vtxfmt;
struct nouveau_stateobj *vtxbuf;
struct nouveau_stateobj *vtxattr;
@@ -190,10 +191,12 @@ extern void nv50_clear(struct pipe_context *pipe, unsigned buffers,
/* nv50_program.c */
extern void nv50_vertprog_validate(struct nv50_context *nv50);
extern void nv50_fragprog_validate(struct nv50_context *nv50);
+extern void nv50_linkage_validate(struct nv50_context *nv50);
extern void nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p);
/* nv50_state_validate.c */
extern boolean nv50_state_validate(struct nv50_context *nv50);
+extern void nv50_state_flush_notify(struct nouveau_channel *chan);
/* nv50_tex.c */
extern void nv50_tex_validate(struct nv50_context *);
diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c
index 03b9243b82..93479a0314 100644
--- a/src/gallium/drivers/nv50/nv50_miptree.c
+++ b/src/gallium/drivers/nv50/nv50_miptree.c
@@ -148,6 +148,7 @@ nv50_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt,
mt->image_nr = 1;
mt->level[0].pitch = *stride;
mt->level[0].image_offset = CALLOC(1, sizeof(unsigned));
+ mt->level[0].tile_mode = bo->tile_mode;
nouveau_bo_ref(bo, &mt->base.bo);
return &mt->base.base;
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index 4a838529de..eb90d5e66f 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -112,6 +112,10 @@ struct nv50_pc {
struct nv50_reg *temp_temp[16];
unsigned temp_temp_nr;
+ /* broadcast and destination replacement regs */
+ struct nv50_reg *r_brdc;
+ struct nv50_reg *r_dst[4];
+
unsigned interp_mode[32];
/* perspective interpolation registers */
struct nv50_reg *iv_p;
@@ -124,6 +128,25 @@ struct nv50_pc {
boolean allow32;
};
+static INLINE void
+ctor_reg(struct nv50_reg *reg, unsigned type, int index, int hw)
+{
+ reg->type = type;
+ reg->index = index;
+ reg->hw = hw;
+ reg->neg = 0;
+ reg->rhw = -1;
+ reg->acc = 0;
+}
+
+static INLINE unsigned
+popcnt4(uint32_t val)
+{
+ static const unsigned cnt[16]
+ = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4 };
+ return cnt[val & 0xf];
+}
+
static void
alloc_reg(struct nv50_pc *pc, struct nv50_reg *reg)
{
@@ -184,11 +207,8 @@ alloc_temp(struct nv50_pc *pc, struct nv50_reg *dst)
for (i = 0; i < NV50_SU_MAX_TEMP; i++) {
if (!pc->r_temp[i]) {
- r = CALLOC_STRUCT(nv50_reg);
- r->type = P_TEMP;
- r->index = -1;
- r->hw = i;
- r->rhw = -1;
+ r = MALLOC_STRUCT(nv50_reg);
+ ctor_reg(r, P_TEMP, -1, i);
pc->r_temp[i] = r;
return r;
}
@@ -254,10 +274,8 @@ alloc_temp4(struct nv50_pc *pc, struct nv50_reg *dst[4], int idx)
return alloc_temp4(pc, dst, idx + 4);
for (i = 0; i < 4; i++) {
- dst[i] = CALLOC_STRUCT(nv50_reg);
- dst[i]->type = P_TEMP;
- dst[i]->index = -1;
- dst[i]->hw = idx + i;
+ dst[i] = MALLOC_STRUCT(nv50_reg);
+ ctor_reg(dst[i], P_TEMP, -1, idx + i);
pc->r_temp[idx + i] = dst[i];
}
@@ -309,7 +327,7 @@ ctor_immd(struct nv50_pc *pc, float x, float y, float z, float w)
static struct nv50_reg *
alloc_immd(struct nv50_pc *pc, float f)
{
- struct nv50_reg *r = CALLOC_STRUCT(nv50_reg);
+ struct nv50_reg *r = MALLOC_STRUCT(nv50_reg);
unsigned hw;
for (hw = 0; hw < pc->immd_nr * 4; hw++)
@@ -319,9 +337,7 @@ alloc_immd(struct nv50_pc *pc, float f)
if (hw == pc->immd_nr * 4)
hw = ctor_immd(pc, f, -f, 0.5 * f, 0) * 4;
- r->type = P_IMMD;
- r->hw = hw;
- r->index = -1;
+ ctor_reg(r, P_IMMD, -1, hw);
return r;
}
@@ -786,6 +802,9 @@ emit_precossin(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
#define CVTOP_SAT 0x08
#define CVTOP_ABS 0x10
+/* 0x04 == 32 bit */
+/* 0x40 == dst is float */
+/* 0x80 == src is float */
#define CVT_F32_F32 0xc4
#define CVT_F32_S32 0x44
#define CVT_F32_U32 0x64
@@ -795,7 +814,7 @@ emit_precossin(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
static void
emit_cvt(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src,
- int wp, unsigned cop, unsigned fmt)
+ int wp, unsigned cvn, unsigned fmt)
{
struct nv50_program_exec *e;
@@ -804,7 +823,7 @@ emit_cvt(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src,
e->inst[0] |= 0xa0000000;
e->inst[1] |= 0x00004000;
- e->inst[1] |= (cop << 16);
+ e->inst[1] |= (cvn << 16);
e->inst[1] |= (fmt << 24);
set_src_0(pc, src, e);
@@ -821,49 +840,80 @@ emit_cvt(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src,
emit(pc, e);
}
+/* nv50 Condition codes:
+ * 0x1 = LT
+ * 0x2 = EQ
+ * 0x3 = LE
+ * 0x4 = GT
+ * 0x5 = NE
+ * 0x6 = GE
+ * 0x7 = set condition code ? (used before bra.lt/le/gt/ge)
+ * 0x8 = unordered bit (allows NaN)
+ */
static void
-emit_set(struct nv50_pc *pc, unsigned c_op, struct nv50_reg *dst,
+emit_set(struct nv50_pc *pc, unsigned ccode, struct nv50_reg *dst, int wp,
struct nv50_reg *src0, struct nv50_reg *src1)
{
+ static const unsigned cc_swapped[8] = { 0, 4, 2, 6, 1, 5, 3, 7 };
+
struct nv50_program_exec *e = exec(pc);
- unsigned inv_cop[8] = { 0, 4, 2, 6, 1, 5, 3, 7 };
struct nv50_reg *rdst;
- assert(c_op <= 7);
+ assert(ccode < 16);
if (check_swap_src_0_1(pc, &src0, &src1))
- c_op = inv_cop[c_op];
+ ccode = cc_swapped[ccode & 7] | (ccode & 8);
rdst = dst;
- if (dst->type != P_TEMP)
+ if (dst && dst->type != P_TEMP)
dst = alloc_temp(pc, NULL);
/* set.u32 */
set_long(pc, e);
e->inst[0] |= 0xb0000000;
- e->inst[1] |= (3 << 29);
- e->inst[1] |= (c_op << 14);
- /*XXX: breaks things, .u32 by default?
- * decuda will disasm as .u16 and use .lo/.hi regs, but this
- * doesn't seem to match what the hw actually does.
- inst[1] |= 0x04000000; << breaks things.. .u32 by default?
+ e->inst[1] |= 0x60000000 | (ccode << 14);
+
+ /* XXX: decuda will disasm as .u16 and use .lo/.hi regs, but
+ * that doesn't seem to match what the hw actually does
+ e->inst[1] |= 0x04000000; << breaks things, u32 by default ?
*/
- set_dst(pc, dst, e);
+
+ if (wp >= 0)
+ set_pred_wr(pc, 1, wp, e);
+ if (dst)
+ set_dst(pc, dst, e);
+ else {
+ e->inst[0] |= 0x000001fc;
+ e->inst[1] |= 0x00000008;
+ }
+
set_src_0(pc, src0, e);
set_src_1(pc, src1, e);
- emit(pc, e);
- /* cvt.f32.u32 */
- e = exec(pc);
- e->inst[0] = 0xa0000001;
- e->inst[1] = 0x64014780;
- set_dst(pc, rdst, e);
- set_src_0(pc, dst, e);
emit(pc, e);
- if (dst != rdst)
+ /* cvt.f32.u32/s32 (?) if we didn't only write the predicate */
+ if (rdst)
+ emit_cvt(pc, rdst, dst, -1, CVTOP_ABS | CVTOP_RN, CVT_F32_S32);
+ if (rdst && rdst != dst)
free_temp(pc, dst);
}
+static INLINE unsigned
+map_tgsi_setop_cc(unsigned op)
+{
+ switch (op) {
+ case TGSI_OPCODE_SLT: return 0x1;
+ case TGSI_OPCODE_SGE: return 0x6;
+ case TGSI_OPCODE_SEQ: return 0x2;
+ case TGSI_OPCODE_SGT: return 0x4;
+ case TGSI_OPCODE_SLE: return 0x3;
+ case TGSI_OPCODE_SNE: return 0xd;
+ default:
+ assert(0);
+ return 0;
+ }
+}
+
static INLINE void
emit_flr(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
{
@@ -890,6 +940,12 @@ emit_abs(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
emit_cvt(pc, dst, src, -1, CVTOP_ABS, CVT_F32_F32);
}
+static INLINE void
+emit_sat(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
+{
+ emit_cvt(pc, dst, src, -1, CVTOP_SAT, CVT_F32_F32);
+}
+
static void
emit_lit(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask,
struct nv50_reg **src)
@@ -1159,6 +1215,70 @@ negate_supported(const struct tgsi_full_instruction *insn, int i)
}
}
+/* Return a read mask for source registers deduced from opcode & write mask. */
+static unsigned
+nv50_tgsi_src_mask(const struct tgsi_full_instruction *insn, int c)
+{
+ unsigned x, mask = insn->FullDstRegisters[0].DstRegister.WriteMask;
+
+ switch (insn->Instruction.Opcode) {
+ case TGSI_OPCODE_COS:
+ case TGSI_OPCODE_SIN:
+ return (mask & 0x8) | ((mask & 0x7) ? 0x1 : 0x0);
+ case TGSI_OPCODE_DP3:
+ return 0x7;
+ case TGSI_OPCODE_DP4:
+ case TGSI_OPCODE_DPH:
+ case TGSI_OPCODE_KIL: /* WriteMask ignored */
+ return 0xf;
+ case TGSI_OPCODE_DST:
+ return mask & (c ? 0xa : 0x6);
+ case TGSI_OPCODE_EX2:
+ case TGSI_OPCODE_LG2:
+ case TGSI_OPCODE_POW:
+ case TGSI_OPCODE_RCP:
+ case TGSI_OPCODE_RSQ:
+ case TGSI_OPCODE_SCS:
+ return 0x1;
+ case TGSI_OPCODE_LIT:
+ return 0xb;
+ case TGSI_OPCODE_TEX:
+ case TGSI_OPCODE_TXP:
+ {
+ const struct tgsi_instruction_ext_texture *tex;
+
+ assert(insn->Instruction.Extended);
+ tex = &insn->InstructionExtTexture;
+
+ mask = 0x7;
+ if (insn->Instruction.Opcode == TGSI_OPCODE_TXP)
+ mask |= 0x8;
+
+ switch (tex->Texture) {
+ case TGSI_TEXTURE_1D:
+ mask &= 0x9;
+ break;
+ case TGSI_TEXTURE_2D:
+ mask &= 0xb;
+ break;
+ default:
+ break;
+ }
+ }
+ return mask;
+ case TGSI_OPCODE_XPD:
+ x = 0;
+ if (mask & 1) x |= 0x6;
+ if (mask & 2) x |= 0x5;
+ if (mask & 4) x |= 0x3;
+ return x;
+ default:
+ break;
+ }
+
+ return mask;
+}
+
static struct nv50_reg *
tgsi_dst(struct nv50_pc *pc, int c, const struct tgsi_full_dst_register *dst)
{
@@ -1258,93 +1378,126 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src,
return r;
}
-/* returns TRUE if instruction can overwrite sources before they're read */
+/* return TRUE for ops that produce only a single result */
static boolean
-direct2dest_op(const struct tgsi_full_instruction *insn)
+is_scalar_op(unsigned op)
{
- if (insn->Instruction.Saturate)
- return FALSE;
-
- switch (insn->Instruction.Opcode) {
+ switch (op) {
case TGSI_OPCODE_COS:
+ case TGSI_OPCODE_DP2:
case TGSI_OPCODE_DP3:
case TGSI_OPCODE_DP4:
case TGSI_OPCODE_DPH:
- case TGSI_OPCODE_KIL:
- case TGSI_OPCODE_LIT:
+ case TGSI_OPCODE_EX2:
+ case TGSI_OPCODE_LG2:
case TGSI_OPCODE_POW:
case TGSI_OPCODE_RCP:
case TGSI_OPCODE_RSQ:
- case TGSI_OPCODE_SCS:
case TGSI_OPCODE_SIN:
+ /*
+ case TGSI_OPCODE_KIL:
+ case TGSI_OPCODE_LIT:
+ case TGSI_OPCODE_SCS:
+ */
+ return TRUE;
+ default:
+ return FALSE;
+ }
+}
+
+/* Returns a bitmask indicating which dst components depend
+ * on source s, component c (reverse of nv50_tgsi_src_mask).
+ */
+static unsigned
+nv50_tgsi_dst_revdep(unsigned op, int s, int c)
+{
+ if (is_scalar_op(op))
+ return 0x1;
+
+ switch (op) {
+ case TGSI_OPCODE_DST:
+ return (1 << c) & (s ? 0xa : 0x6);
+ case TGSI_OPCODE_XPD:
+ switch (c) {
+ case 0: return 0x6;
+ case 1: return 0x5;
+ case 2: return 0x3;
+ case 3: return 0x0;
+ default:
+ assert(0);
+ return 0x0;
+ }
+ case TGSI_OPCODE_LIT:
+ case TGSI_OPCODE_SCS:
case TGSI_OPCODE_TEX:
case TGSI_OPCODE_TXP:
- return FALSE;
+ /* these take care of dangerous swizzles themselves */
+ return 0x0;
+ case TGSI_OPCODE_IF:
+ case TGSI_OPCODE_KIL:
+ /* don't call this function for these ops */
+ assert(0);
+ return 0;
default:
- return TRUE;
+ /* linear vector instruction */
+ return (1 << c);
}
}
static boolean
-nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
+nv50_program_tx_insn(struct nv50_pc *pc,
+ const struct tgsi_full_instruction *inst)
{
- const struct tgsi_full_instruction *inst = &tok->FullInstruction;
- struct nv50_reg *rdst[4], *dst[4], *src[3][4], *temp;
+ struct nv50_reg *rdst[4], *dst[4], *brdc, *src[3][4], *temp;
unsigned mask, sat, unit;
- boolean assimilate = FALSE;
int i, c;
mask = inst->FullDstRegisters[0].DstRegister.WriteMask;
sat = inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE;
+ memset(src, 0, sizeof(src));
+
for (c = 0; c < 4; c++) {
- if (mask & (1 << c))
+ if ((mask & (1 << c)) && !pc->r_dst[c])
dst[c] = tgsi_dst(pc, c, &inst->FullDstRegisters[0]);
else
- dst[c] = NULL;
- rdst[c] = NULL;
- src[0][c] = NULL;
- src[1][c] = NULL;
- src[2][c] = NULL;
+ dst[c] = pc->r_dst[c];
+ rdst[c] = dst[c];
}
for (i = 0; i < inst->Instruction.NumSrcRegs; i++) {
const struct tgsi_full_src_register *fs = &inst->FullSrcRegisters[i];
+ unsigned src_mask;
+ boolean neg_supp;
+
+ src_mask = nv50_tgsi_src_mask(inst, i);
+ neg_supp = negate_supported(inst, i);
if (fs->SrcRegister.File == TGSI_FILE_SAMPLER)
unit = fs->SrcRegister.Index;
for (c = 0; c < 4; c++)
- src[i][c] = tgsi_src(pc, c, fs,
- negate_supported(inst, i));
+ if (src_mask & (1 << c))
+ src[i][c] = tgsi_src(pc, c, fs, neg_supp);
}
- if (sat) {
- for (c = 0; c < 4; c++) {
- rdst[c] = dst[c];
- dst[c] = temp_temp(pc);
- }
+ brdc = temp = pc->r_brdc;
+ if (brdc && brdc->type != P_TEMP) {
+ temp = temp_temp(pc);
+ if (sat)
+ brdc = temp;
} else
- if (direct2dest_op(inst)) {
+ if (sat) {
for (c = 0; c < 4; c++) {
- if (!dst[c] || dst[c]->type != P_TEMP)
+ if (!(mask & (1 << c)) || dst[c]->type == P_TEMP)
continue;
-
- for (i = c + 1; i < 4; i++) {
- if (dst[c] == src[0][i] ||
- dst[c] == src[1][i] ||
- dst[c] == src[2][i])
- break;
- }
- if (i == 4)
- continue;
-
- assimilate = TRUE;
rdst[c] = dst[c];
- dst[c] = alloc_temp(pc, NULL);
+ dst[c] = temp_temp(pc);
}
}
+ assert(brdc || !is_scalar_op(inst->Instruction.Opcode));
+
switch (inst->Instruction.Opcode) {
case TGSI_OPCODE_ABS:
for (c = 0; c < 4; c++) {
@@ -1360,74 +1513,56 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
emit_add(pc, dst[c], src[0][c], src[1][c]);
}
break;
- case TGSI_OPCODE_COS:
- temp = temp_temp(pc);
- emit_precossin(pc, temp, src[0][0]);
- emit_flop(pc, 5, temp, temp);
+ case TGSI_OPCODE_CEIL:
for (c = 0; c < 4; c++) {
if (!(mask & (1 << c)))
continue;
- emit_mov(pc, dst[c], temp);
+ emit_cvt(pc, dst[c], src[0][c], -1,
+ CVTOP_CEIL, CVT_F32_F32);
+ }
+ break;
+ case TGSI_OPCODE_COS:
+ if (mask & 8) {
+ emit_precossin(pc, temp, src[0][3]);
+ emit_flop(pc, 5, dst[3], temp);
+ if (!(mask &= 7))
+ break;
+ if (temp == dst[3])
+ temp = brdc = temp_temp(pc);
}
+ emit_precossin(pc, temp, src[0][0]);
+ emit_flop(pc, 5, brdc, temp);
break;
case TGSI_OPCODE_DP3:
- temp = temp_temp(pc);
emit_mul(pc, temp, src[0][0], src[1][0]);
emit_mad(pc, temp, src[0][1], src[1][1], temp);
- emit_mad(pc, temp, src[0][2], src[1][2], temp);
- for (c = 0; c < 4; c++) {
- if (!(mask & (1 << c)))
- continue;
- emit_mov(pc, dst[c], temp);
- }
+ emit_mad(pc, brdc, src[0][2], src[1][2], temp);
break;
case TGSI_OPCODE_DP4:
- temp = temp_temp(pc);
emit_mul(pc, temp, src[0][0], src[1][0]);
emit_mad(pc, temp, src[0][1], src[1][1], temp);
emit_mad(pc, temp, src[0][2], src[1][2], temp);
- emit_mad(pc, temp, src[0][3], src[1][3], temp);
- for (c = 0; c < 4; c++) {
- if (!(mask & (1 << c)))
- continue;
- emit_mov(pc, dst[c], temp);
- }
+ emit_mad(pc, brdc, src[0][3], src[1][3], temp);
break;
case TGSI_OPCODE_DPH:
- temp = temp_temp(pc);
emit_mul(pc, temp, src[0][0], src[1][0]);
emit_mad(pc, temp, src[0][1], src[1][1], temp);
emit_mad(pc, temp, src[0][2], src[1][2], temp);
- emit_add(pc, temp, src[1][3], temp);
- for (c = 0; c < 4; c++) {
- if (!(mask & (1 << c)))
- continue;
- emit_mov(pc, dst[c], temp);
- }
+ emit_add(pc, brdc, src[1][3], temp);
break;
case TGSI_OPCODE_DST:
- {
- struct nv50_reg *one = alloc_immd(pc, 1.0);
- if (mask & (1 << 0))
- emit_mov(pc, dst[0], one);
if (mask & (1 << 1))
emit_mul(pc, dst[1], src[0][1], src[1][1]);
if (mask & (1 << 2))
emit_mov(pc, dst[2], src[0][2]);
if (mask & (1 << 3))
emit_mov(pc, dst[3], src[1][3]);
- FREE(one);
- }
+ if (mask & (1 << 0))
+ emit_mov_immdval(pc, dst[0], 1.0f);
break;
case TGSI_OPCODE_EX2:
- temp = temp_temp(pc);
emit_preex2(pc, temp, src[0][0]);
- emit_flop(pc, 6, temp, temp);
- for (c = 0; c < 4; c++) {
- if (!(mask & (1 << c)))
- continue;
- emit_mov(pc, dst[c], temp);
- }
+ emit_flop(pc, 6, brdc, temp);
break;
case TGSI_OPCODE_FLR:
for (c = 0; c < 4; c++) {
@@ -1450,19 +1585,12 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
emit_kil(pc, src[0][1]);
emit_kil(pc, src[0][2]);
emit_kil(pc, src[0][3]);
- pc->p->cfg.fp.regs[2] |= 0x00100000;
break;
case TGSI_OPCODE_LIT:
emit_lit(pc, &dst[0], mask, &src[0][0]);
break;
case TGSI_OPCODE_LG2:
- temp = temp_temp(pc);
- emit_flop(pc, 3, temp, src[0][0]);
- for (c = 0; c < 4; c++) {
- if (!(mask & (1 << c)))
- continue;
- emit_mov(pc, dst[c], temp);
- }
+ emit_flop(pc, 3, brdc, src[0][0]);
break;
case TGSI_OPCODE_LRP:
temp = temp_temp(pc);
@@ -1510,31 +1638,18 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
}
break;
case TGSI_OPCODE_POW:
- temp = temp_temp(pc);
- emit_pow(pc, temp, src[0][0], src[1][0]);
- for (c = 0; c < 4; c++) {
- if (!(mask & (1 << c)))
- continue;
- emit_mov(pc, dst[c], temp);
- }
+ emit_pow(pc, brdc, src[0][0], src[1][0]);
break;
case TGSI_OPCODE_RCP:
- for (c = 3; c >= 0; c--) {
- if (!(mask & (1 << c)))
- continue;
- emit_flop(pc, 0, dst[c], src[0][0]);
- }
+ emit_flop(pc, 0, brdc, src[0][0]);
break;
case TGSI_OPCODE_RSQ:
- for (c = 3; c >= 0; c--) {
- if (!(mask & (1 << c)))
- continue;
- emit_flop(pc, 2, dst[c], src[0][0]);
- }
+ emit_flop(pc, 2, brdc, src[0][0]);
break;
case TGSI_OPCODE_SCS:
temp = temp_temp(pc);
- emit_precossin(pc, temp, src[0][0]);
+ if (mask & 3)
+ emit_precossin(pc, temp, src[0][0]);
if (mask & (1 << 0))
emit_flop(pc, 5, dst[0], temp);
if (mask & (1 << 1))
@@ -1544,28 +1659,29 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
if (mask & (1 << 3))
emit_mov_immdval(pc, dst[3], 1.0);
break;
- case TGSI_OPCODE_SGE:
- for (c = 0; c < 4; c++) {
- if (!(mask & (1 << c)))
- continue;
- emit_set(pc, 6, dst[c], src[0][c], src[1][c]);
- }
- break;
case TGSI_OPCODE_SIN:
- temp = temp_temp(pc);
- emit_precossin(pc, temp, src[0][0]);
- emit_flop(pc, 4, temp, temp);
- for (c = 0; c < 4; c++) {
- if (!(mask & (1 << c)))
- continue;
- emit_mov(pc, dst[c], temp);
+ if (mask & 8) {
+ emit_precossin(pc, temp, src[0][3]);
+ emit_flop(pc, 4, dst[3], temp);
+ if (!(mask &= 7))
+ break;
+ if (temp == dst[3])
+ temp = brdc = temp_temp(pc);
}
+ emit_precossin(pc, temp, src[0][0]);
+ emit_flop(pc, 4, brdc, temp);
break;
case TGSI_OPCODE_SLT:
+ case TGSI_OPCODE_SGE:
+ case TGSI_OPCODE_SEQ:
+ case TGSI_OPCODE_SGT:
+ case TGSI_OPCODE_SLE:
+ case TGSI_OPCODE_SNE:
+ i = map_tgsi_setop_cc(inst->Instruction.Opcode);
for (c = 0; c < 4; c++) {
if (!(mask & (1 << c)))
continue;
- emit_set(pc, 1, dst[c], src[0][c], src[1][c]);
+ emit_set(pc, i, dst[c], -1, src[0][c], src[1][c]);
}
break;
case TGSI_OPCODE_SUB:
@@ -1583,6 +1699,14 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
emit_tex(pc, dst, mask, src[0], unit,
inst->InstructionExtTexture.Texture, TRUE);
break;
+ case TGSI_OPCODE_TRUNC:
+ for (c = 0; c < 4; c++) {
+ if (!(mask & (1 << c)))
+ continue;
+ emit_cvt(pc, dst[c], src[0][c], -1,
+ CVTOP_TRUNC, CVT_F32_F32);
+ }
+ break;
case TGSI_OPCODE_XPD:
temp = temp_temp(pc);
if (mask & (1 << 0)) {
@@ -1607,17 +1731,22 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
return FALSE;
}
+ if (brdc) {
+ if (sat)
+ emit_sat(pc, brdc, brdc);
+ for (c = 0; c < 4; c++)
+ if ((mask & (1 << c)) && dst[c] != brdc)
+ emit_mov(pc, dst[c], brdc);
+ } else
if (sat) {
for (c = 0; c < 4; c++) {
if (!(mask & (1 << c)))
continue;
- emit_cvt(pc, rdst[c], dst[c], -1, CVTOP_SAT,
- CVT_F32_F32);
+ /* in this case we saturate later */
+ if (dst[c]->type == P_TEMP && dst[c]->index < 0)
+ continue;
+ emit_sat(pc, rdst[c], dst[c]);
}
- } else if (assimilate) {
- for (c = 0; c < 4; c++)
- if (rdst[c])
- assimilate_temp(pc, rdst[c], dst[c]);
}
for (i = 0; i < inst->Instruction.NumSrcRegs; i++) {
@@ -1626,9 +1755,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
continue;
if (src[i][c]->index == -1 && src[i][c]->type == P_IMMD)
FREE(src[i][c]);
- else
- if (src[i][c]->acc == pc->insn_cur)
- release_hw(pc, src[i][c]);
}
}
@@ -1636,180 +1762,271 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
return TRUE;
}
-/* Adjust a bitmask that indicates what components of a source are used,
- * we use this in tx_prep so we only load interpolants that are needed.
- */
static void
-insn_adjust_mask(const struct tgsi_full_instruction *insn, unsigned *mask)
+prep_inspect_insn(struct nv50_pc *pc, const struct tgsi_full_instruction *insn)
{
- const struct tgsi_instruction_ext_texture *tex;
-
- switch (insn->Instruction.Opcode) {
- case TGSI_OPCODE_DP3:
- *mask = 0x7;
- break;
- case TGSI_OPCODE_DP4:
- case TGSI_OPCODE_DPH:
- *mask = 0xF;
- break;
- case TGSI_OPCODE_LIT:
- *mask = 0xB;
- break;
- case TGSI_OPCODE_RCP:
- case TGSI_OPCODE_RSQ:
- *mask = 0x1;
- break;
- case TGSI_OPCODE_TEX:
- case TGSI_OPCODE_TXP:
- assert(insn->Instruction.Extended);
- tex = &insn->InstructionExtTexture;
-
- *mask = 0x7;
- if (tex->Texture == TGSI_TEXTURE_1D)
- *mask = 0x1;
- else
- if (tex->Texture == TGSI_TEXTURE_2D)
- *mask = 0x3;
-
- if (insn->Instruction.Opcode == TGSI_OPCODE_TXP)
- *mask |= 0x8;
- break;
- default:
- break;
- }
-}
-
-static void
-prep_inspect_insn(struct nv50_pc *pc, const union tgsi_full_token *tok,
- unsigned *r_usage[2])
-{
- const struct tgsi_full_instruction *insn;
+ struct nv50_reg *reg = NULL;
const struct tgsi_full_src_register *src;
const struct tgsi_dst_register *dst;
+ unsigned i, c, k, mask;
- unsigned i, c, k, n, mask, *acc_p;
-
- insn = &tok->FullInstruction;
dst = &insn->FullDstRegisters[0].DstRegister;
mask = dst->WriteMask;
- if (!r_usage[0])
- r_usage[0] = CALLOC(pc->temp_nr * 4, sizeof(unsigned));
- if (!r_usage[1])
- r_usage[1] = CALLOC(pc->attr_nr * 4, sizeof(unsigned));
+ if (dst->File == TGSI_FILE_TEMPORARY)
+ reg = pc->temp;
+ else
+ if (dst->File == TGSI_FILE_OUTPUT)
+ reg = pc->result;
- if (dst->File == TGSI_FILE_TEMPORARY) {
+ if (reg) {
for (c = 0; c < 4; c++) {
if (!(mask & (1 << c)))
continue;
- r_usage[0][dst->Index * 4 + c] = pc->insn_nr;
+ reg[dst->Index * 4 + c].acc = pc->insn_nr;
}
}
for (i = 0; i < insn->Instruction.NumSrcRegs; i++) {
src = &insn->FullSrcRegisters[i];
- switch (src->SrcRegister.File) {
- case TGSI_FILE_TEMPORARY:
- acc_p = r_usage[0];
- break;
- case TGSI_FILE_INPUT:
- acc_p = r_usage[1];
- break;
- default:
+ if (src->SrcRegister.File == TGSI_FILE_TEMPORARY)
+ reg = pc->temp;
+ else
+ if (src->SrcRegister.File == TGSI_FILE_INPUT)
+ reg = pc->attr;
+ else
continue;
- }
- insn_adjust_mask(insn, &mask);
+ mask = nv50_tgsi_src_mask(insn, i);
for (c = 0; c < 4; c++) {
if (!(mask & (1 << c)))
continue;
-
k = tgsi_util_get_full_src_register_extswizzle(src, c);
- switch (k) {
- case TGSI_EXTSWIZZLE_X:
- case TGSI_EXTSWIZZLE_Y:
- case TGSI_EXTSWIZZLE_Z:
- case TGSI_EXTSWIZZLE_W:
- n = src->SrcRegister.Index * 4 + k;
- acc_p[n] = pc->insn_nr;
- break;
- default:
- break;
- }
+
+ if (k > TGSI_EXTSWIZZLE_W)
+ continue;
+
+ reg[src->SrcRegister.Index * 4 + k].acc = pc->insn_nr;
}
}
}
+/* Returns a bitmask indicating which dst components need to be
+ * written to temporaries first to avoid 'corrupting' sources.
+ *
+ * m[i] (out) indicate component to write in the i-th position
+ * rdep[c] (in) bitmasks of dst[i] that require dst[c] as source
+ */
static unsigned
-load_fp_attrib(struct nv50_pc *pc, int i, unsigned *acc, int *mid,
- int *aid, int *p_oid)
+nv50_revdep_reorder(unsigned m[4], unsigned rdep[4])
{
- struct nv50_reg *iv;
- int oid, c, n;
- unsigned mask = 0;
+ unsigned i, c, x, unsafe;
- iv = (pc->interp_mode[i] & INTERP_CENTROID) ? pc->iv_c : pc->iv_p;
+ for (c = 0; c < 4; c++)
+ m[c] = c;
- for (c = 0, n = i * 4; c < 4; c++, n++) {
- oid = (*p_oid)++;
- pc->attr[n].type = P_TEMP;
- pc->attr[n].index = i;
+ /* Swap as long as a dst component written earlier is depended on
+ * by one written later, but the next one isn't depended on by it.
+ */
+ for (c = 0; c < 3; c++) {
+ if (rdep[m[c + 1]] & (1 << m[c]))
+ continue; /* if next one is depended on by us */
+ for (i = c + 1; i < 4; i++)
+ /* if we are depended on by a later one */
+ if (rdep[m[c]] & (1 << m[i]))
+ break;
+ if (i == 4)
+ continue;
+ /* now, swap */
+ x = m[c];
+ m[c] = m[c + 1];
+ m[c + 1] = x;
+
+ /* restart */
+ c = 0;
+ }
+
+ /* mark dependencies that could not be resolved by reordering */
+ for (i = 0; i < 3; ++i)
+ for (c = i + 1; c < 4; ++c)
+ if (rdep[m[i]] & (1 << m[c]))
+ unsafe |= (1 << i);
+
+ /* NOTE: $unsafe is with respect to order, not component */
+ return unsafe;
+}
- if (pc->attr[n].acc == acc[n])
+/* Select a suitable dst register for broadcasting scalar results,
+ * or return NULL if we have to allocate an extra TEMP.
+ *
+ * If e.g. only 1 component is written, we may also emit the final
+ * result to a write-only register.
+ */
+static struct nv50_reg *
+tgsi_broadcast_dst(struct nv50_pc *pc,
+ const struct tgsi_full_dst_register *fd, unsigned mask)
+{
+ if (fd->DstRegister.File == TGSI_FILE_TEMPORARY) {
+ int c = ffs(~mask & fd->DstRegister.WriteMask);
+ if (c)
+ return tgsi_dst(pc, c - 1, fd);
+ } else {
+ int c = ffs(fd->DstRegister.WriteMask) - 1;
+ if ((1 << c) == fd->DstRegister.WriteMask)
+ return tgsi_dst(pc, c, fd);
+ }
+
+ return NULL;
+}
+
+/* Scan source swizzles and return a bitmask indicating dst regs that
+ * also occur among the src regs, and fill rdep for nv50_revdep_reoder.
+ */
+static unsigned
+nv50_tgsi_scan_swizzle(const struct tgsi_full_instruction *insn,
+ unsigned rdep[4])
+{
+ const struct tgsi_full_dst_register *fd = &insn->FullDstRegisters[0];
+ const struct tgsi_full_src_register *fs;
+ unsigned i, deqs = 0;
+
+ for (i = 0; i < 4; ++i)
+ rdep[i] = 0;
+
+ for (i = 0; i < insn->Instruction.NumSrcRegs; i++) {
+ unsigned chn, mask = nv50_tgsi_src_mask(insn, i);
+ boolean neg_supp = negate_supported(insn, i);
+
+ fs = &insn->FullSrcRegisters[i];
+ if (fs->SrcRegister.File != fd->DstRegister.File ||
+ fs->SrcRegister.Index != fd->DstRegister.Index)
continue;
- mask |= (1 << c);
- pc->attr[n].acc = acc[n];
- pc->attr[n].rhw = pc->attr[n].hw = -1;
- alloc_reg(pc, &pc->attr[n]);
+ for (chn = 0; chn < 4; ++chn) {
+ unsigned s, c;
+
+ if (!(mask & (1 << chn))) /* src is not read */
+ continue;
+ c = tgsi_util_get_full_src_register_extswizzle(fs, chn);
+ s = tgsi_util_get_full_src_register_sign_mode(fs, chn);
- pc->attr[n].rhw = (*aid)++;
- emit_interp(pc, &pc->attr[n], iv, pc->interp_mode[i]);
+ if (c > TGSI_EXTSWIZZLE_W ||
+ !(fd->DstRegister.WriteMask & (1 << c)))
+ continue;
- pc->p->cfg.fp.map[(*mid) / 4] |= oid << (8 * ((*mid) % 4));
- (*mid)++;
- pc->p->cfg.fp.regs[1] += 0x00010001;
+ /* no danger if src is copied to TEMP first */
+ if ((s != TGSI_UTIL_SIGN_KEEP) &&
+ (s != TGSI_UTIL_SIGN_TOGGLE || !neg_supp))
+ continue;
+
+ rdep[c] |= nv50_tgsi_dst_revdep(
+ insn->Instruction.Opcode, i, chn);
+ deqs |= (1 << c);
+ }
}
- return mask;
+ return deqs;
}
static boolean
-nv50_program_tx_prep(struct nv50_pc *pc)
+nv50_tgsi_insn(struct nv50_pc *pc, const union tgsi_full_token *tok)
{
- struct tgsi_parse_context p;
- boolean ret = FALSE;
- unsigned i, c;
- unsigned fcol, bcol, fcrd, depr;
+ struct tgsi_full_instruction insn = tok->FullInstruction;
+ const struct tgsi_full_dst_register *fd;
+ unsigned i, deqs, rdep[4], m[4];
+
+ fd = &tok->FullInstruction.FullDstRegisters[0];
+ deqs = nv50_tgsi_scan_swizzle(&insn, rdep);
+
+ if (is_scalar_op(insn.Instruction.Opcode)) {
+ pc->r_brdc = tgsi_broadcast_dst(pc, fd, deqs);
+ if (!pc->r_brdc)
+ pc->r_brdc = temp_temp(pc);
+ return nv50_program_tx_insn(pc, &insn);
+ }
+ pc->r_brdc = NULL;
+
+ if (!deqs)
+ return nv50_program_tx_insn(pc, &insn);
+
+ deqs = nv50_revdep_reorder(m, rdep);
- /* count (centroid) perspective interpolations */
- unsigned centroid_loads = 0;
- unsigned perspect_loads = 0;
+ for (i = 0; i < 4; ++i) {
+ assert(pc->r_dst[m[i]] == NULL);
- /* track register access for temps and attrs */
- unsigned *r_usage[2];
- r_usage[0] = NULL;
- r_usage[1] = NULL;
+ insn.FullDstRegisters[0].DstRegister.WriteMask =
+ fd->DstRegister.WriteMask & (1 << m[i]);
- depr = fcol = bcol = fcrd = 0xffff;
+ if (!insn.FullDstRegisters[0].DstRegister.WriteMask)
+ continue;
+
+ if (deqs & (1 << i))
+ pc->r_dst[m[i]] = alloc_temp(pc, NULL);
- if (pc->p->type == PIPE_SHADER_FRAGMENT) {
- pc->p->cfg.fp.regs[0] = 0x01000404;
- pc->p->cfg.fp.regs[1] = 0x00000400;
+ if (!nv50_program_tx_insn(pc, &insn))
+ return FALSE;
}
- tgsi_parse_init(&p, pc->p->pipe.tokens);
- while (!tgsi_parse_end_of_tokens(&p)) {
- const union tgsi_full_token *tok = &p.FullToken;
+ for (i = 0; i < 4; i++) {
+ struct nv50_reg *reg = pc->r_dst[i];
+ if (!reg)
+ continue;
+ pc->r_dst[i] = NULL;
+
+ if (insn.Instruction.Saturate == TGSI_SAT_ZERO_ONE)
+ emit_sat(pc, tgsi_dst(pc, i, fd), reg);
+ else
+ emit_mov(pc, tgsi_dst(pc, i, fd), reg);
+ free_temp(pc, reg);
+ }
- tgsi_parse_token(&p);
+ return TRUE;
+}
+
+static void
+load_interpolant(struct nv50_pc *pc, struct nv50_reg *reg)
+{
+ struct nv50_reg *iv, **ppiv;
+ unsigned mode = pc->interp_mode[reg->index];
+
+ ppiv = (mode & INTERP_CENTROID) ? &pc->iv_c : &pc->iv_p;
+ iv = *ppiv;
+
+ if ((mode & INTERP_PERSPECTIVE) && !iv) {
+ iv = *ppiv = alloc_temp(pc, NULL);
+ iv->rhw = popcnt4(pc->p->cfg.regs[1] >> 24) - 1;
+
+ emit_interp(pc, iv, NULL, mode & INTERP_CENTROID);
+ emit_flop(pc, 0, iv, iv);
+
+ /* XXX: when loading interpolants dynamically, move these
+ * to the program head, or make sure it can't be skipped.
+ */
+ }
+
+ emit_interp(pc, reg, iv, mode);
+}
+
+static boolean
+nv50_program_tx_prep(struct nv50_pc *pc)
+{
+ struct tgsi_parse_context tp;
+ struct nv50_program *p = pc->p;
+ boolean ret = FALSE;
+ unsigned i, c, flat_nr = 0;
+
+ tgsi_parse_init(&tp, pc->p->pipe.tokens);
+ while (!tgsi_parse_end_of_tokens(&tp)) {
+ const union tgsi_full_token *tok = &tp.FullToken;
+
+ tgsi_parse_token(&tp);
switch (tok->Token.Type) {
case TGSI_TOKEN_TYPE_IMMEDIATE:
{
const struct tgsi_full_immediate *imm =
- &p.FullToken.FullImmediate;
+ &tp.FullToken.FullImmediate;
ctor_immd(pc, imm->u[0].Float,
imm->u[1].Float,
@@ -1820,78 +2037,61 @@ nv50_program_tx_prep(struct nv50_pc *pc)
case TGSI_TOKEN_TYPE_DECLARATION:
{
const struct tgsi_full_declaration *d;
- unsigned last, first, mode;
+ unsigned si, last, first, mode;
- d = &p.FullToken.FullDeclaration;
+ d = &tp.FullToken.FullDeclaration;
first = d->DeclarationRange.First;
last = d->DeclarationRange.Last;
switch (d->Declaration.File) {
case TGSI_FILE_TEMPORARY:
- if (pc->temp_nr < (last + 1))
- pc->temp_nr = last + 1;
break;
case TGSI_FILE_OUTPUT:
- if (pc->result_nr < (last + 1))
- pc->result_nr = last + 1;
-
- if (!d->Declaration.Semantic)
+ if (!d->Declaration.Semantic ||
+ p->type == PIPE_SHADER_FRAGMENT)
break;
+ si = d->Semantic.SemanticIndex;
switch (d->Semantic.SemanticName) {
- case TGSI_SEMANTIC_POSITION:
- depr = first;
- pc->p->cfg.fp.regs[2] |= 0x00000100;
- pc->p->cfg.fp.regs[3] |= 0x00000011;
+ case TGSI_SEMANTIC_BCOLOR:
+ p->cfg.two_side[si].hw = first;
+ if (p->cfg.io_nr > first)
+ p->cfg.io_nr = first;
+ break;
+ case TGSI_SEMANTIC_PSIZE:
+ p->cfg.psiz = first;
+ if (p->cfg.io_nr > first)
+ p->cfg.io_nr = first;
+ break;
+ /*
+ case TGSI_SEMANTIC_CLIP_DISTANCE:
+ p->cfg.clpd = MIN2(p->cfg.clpd, first);
break;
+ */
default:
break;
}
-
break;
case TGSI_FILE_INPUT:
{
- if (pc->attr_nr < (last + 1))
- pc->attr_nr = last + 1;
-
- if (pc->p->type != PIPE_SHADER_FRAGMENT)
+ if (p->type != PIPE_SHADER_FRAGMENT)
break;
switch (d->Declaration.Interpolate) {
case TGSI_INTERPOLATE_CONSTANT:
mode = INTERP_FLAT;
+ flat_nr++;
break;
case TGSI_INTERPOLATE_PERSPECTIVE:
mode = INTERP_PERSPECTIVE;
+ p->cfg.regs[1] |= 0x08 << 24;
break;
default:
mode = INTERP_LINEAR;
break;
}
-
- if (d->Declaration.Semantic) {
- switch (d->Semantic.SemanticName) {
- case TGSI_SEMANTIC_POSITION:
- fcrd = first;
- break;
- case TGSI_SEMANTIC_COLOR:
- fcol = first;
- mode = INTERP_PERSPECTIVE;
- break;
- case TGSI_SEMANTIC_BCOLOR:
- bcol = first;
- mode = INTERP_PERSPECTIVE;
- break;
- }
- }
-
- if (d->Declaration.Centroid) {
+ if (d->Declaration.Centroid)
mode |= INTERP_CENTROID;
- if (mode & INTERP_PERSPECTIVE)
- centroid_loads++;
- } else
- if (mode & INTERP_PERSPECTIVE)
- perspect_loads++;
assert(last < 32);
for (i = first; i <= last; i++)
@@ -1899,8 +2099,6 @@ nv50_program_tx_prep(struct nv50_pc *pc)
}
break;
case TGSI_FILE_CONSTANT:
- if (pc->param_nr < (last + 1))
- pc->param_nr = last + 1;
break;
case TGSI_FILE_SAMPLER:
break;
@@ -1913,182 +2111,155 @@ nv50_program_tx_prep(struct nv50_pc *pc)
break;
case TGSI_TOKEN_TYPE_INSTRUCTION:
pc->insn_nr++;
- prep_inspect_insn(pc, tok, r_usage);
+ prep_inspect_insn(pc, &tok->FullInstruction);
break;
default:
break;
}
}
- if (pc->temp_nr) {
- pc->temp = CALLOC(pc->temp_nr * 4, sizeof(struct nv50_reg));
- if (!pc->temp)
- goto out_err;
+ if (p->type == PIPE_SHADER_VERTEX) {
+ int rid = 0;
- for (i = 0; i < pc->temp_nr; i++) {
- for (c = 0; c < 4; c++) {
- pc->temp[i*4+c].type = P_TEMP;
- pc->temp[i*4+c].hw = -1;
- pc->temp[i*4+c].rhw = -1;
- pc->temp[i*4+c].index = i;
- pc->temp[i*4+c].acc = r_usage[0][i*4+c];
+ for (i = 0; i < pc->attr_nr * 4; ++i) {
+ if (pc->attr[i].acc) {
+ pc->attr[i].hw = rid++;
+ p->cfg.attr[i / 32] |= 1 << (i % 32);
}
}
- }
-
- if (pc->attr_nr) {
- int oid = 4, mid = 4, aid = 0;
- /* oid = VP output id
- * aid = FP attribute/interpolant id
- * mid = VP output mapping field ID
- */
- pc->attr = CALLOC(pc->attr_nr * 4, sizeof(struct nv50_reg));
- if (!pc->attr)
- goto out_err;
+ for (i = 0, rid = 0; i < pc->result_nr; ++i) {
+ p->cfg.io[i].hw = rid;
+ p->cfg.io[i].id_vp = i;
- if (pc->p->type == PIPE_SHADER_FRAGMENT) {
- /* position should be loaded first */
- if (fcrd != 0xffff) {
- unsigned mask;
- mid = 0;
- mask = load_fp_attrib(pc, fcrd, r_usage[1],
- &mid, &aid, &oid);
- oid = 0;
- pc->p->cfg.fp.regs[1] |= (mask << 24);
- pc->p->cfg.fp.map[0] = 0x04040404 * fcrd;
- }
- pc->p->cfg.fp.map[0] += 0x03020100;
-
- /* should do MAD fcrd.xy, fcrd, SOME_CONST, fcrd */
-
- if (perspect_loads) {
- pc->iv_p = alloc_temp(pc, NULL);
-
- if (!(pc->p->cfg.fp.regs[1] & 0x08000000)) {
- pc->p->cfg.fp.regs[1] |= 0x08000000;
- pc->iv_p->rhw = aid++;
- emit_interp(pc, pc->iv_p, NULL,
- INTERP_LINEAR);
- emit_flop(pc, 0, pc->iv_p, pc->iv_p);
- } else {
- pc->iv_p->rhw = aid - 1;
- emit_flop(pc, 0, pc->iv_p,
- &pc->attr[fcrd * 4 + 3]);
- }
+ for (c = 0; c < 4; ++c) {
+ int n = i * 4 + c;
+ if (!pc->result[n].acc)
+ continue;
+ pc->result[n].hw = rid++;
+ p->cfg.io[i].mask |= 1 << c;
}
+ }
- if (centroid_loads) {
- pc->iv_c = alloc_temp(pc, NULL);
- pc->iv_c->rhw = pc->iv_p ? aid - 1 : aid++;
- emit_interp(pc, pc->iv_c, NULL,
- INTERP_CENTROID);
- emit_flop(pc, 0, pc->iv_c, pc->iv_c);
- pc->p->cfg.fp.regs[1] |= 0x08000000;
- }
+ for (c = 0; c < 2; ++c)
+ if (p->cfg.two_side[c].hw < 0x40)
+ p->cfg.two_side[c] = p->cfg.io[
+ p->cfg.two_side[c].hw];
- for (c = 0; c < 4; c++) {
- /* I don't know what these values do, but
- * let's set them like the blob does:
- */
- if (fcol != 0xffff && r_usage[1][fcol * 4 + c])
- pc->p->cfg.fp.regs[0] += 0x00010000;
- if (bcol != 0xffff && r_usage[1][bcol * 4 + c])
- pc->p->cfg.fp.regs[0] += 0x00010000;
- }
-
- for (i = 0; i < pc->attr_nr; i++)
- load_fp_attrib(pc, i, r_usage[1],
- &mid, &aid, &oid);
+ if (p->cfg.psiz < 0x40)
+ p->cfg.psiz = p->cfg.io[p->cfg.psiz].hw;
+ } else
+ if (p->type == PIPE_SHADER_FRAGMENT) {
+ int rid, aid;
+ unsigned n = 0, m = pc->attr_nr - flat_nr;
- if (pc->iv_p)
- free_temp(pc, pc->iv_p);
- if (pc->iv_c)
- free_temp(pc, pc->iv_c);
+ int base = (TGSI_SEMANTIC_POSITION ==
+ p->info.input_semantic_name[0]) ? 0 : 1;
- pc->p->cfg.fp.high_map = (mid / 4);
- pc->p->cfg.fp.high_map += ((mid % 4) ? 1 : 0);
- } else {
- /* vertex program */
- for (i = 0; i < pc->attr_nr * 4; i++) {
- pc->p->cfg.vp.attr[aid / 32] |=
- (1 << (aid % 32));
- pc->attr[i].type = P_ATTR;
- pc->attr[i].hw = aid++;
- pc->attr[i].index = i / 4;
+ /* non-flat interpolants have to be mapped to
+ * the lower hardware IDs, so sort them:
+ */
+ for (i = 0; i < pc->attr_nr; i++) {
+ if (pc->interp_mode[i] == INTERP_FLAT) {
+ p->cfg.io[m].id_vp = i + base;
+ p->cfg.io[m++].id_fp = i;
+ } else {
+ if (!(pc->interp_mode[i] & INTERP_PERSPECTIVE))
+ p->cfg.io[n].linear = TRUE;
+ p->cfg.io[n].id_vp = i + base;
+ p->cfg.io[n++].id_fp = i;
}
}
- }
- if (pc->result_nr) {
- int rid = 0;
+ if (!base) /* set w-coordinate mask from perspective interp */
+ p->cfg.io[0].mask |= p->cfg.regs[1] >> 24;
- pc->result = CALLOC(pc->result_nr * 4, sizeof(struct nv50_reg));
- if (!pc->result)
- goto out_err;
+ aid = popcnt4( /* if fcrd isn't contained in cfg.io */
+ base ? (p->cfg.regs[1] >> 24) : p->cfg.io[0].mask);
- for (i = 0; i < pc->result_nr; i++) {
- for (c = 0; c < 4; c++) {
- if (pc->p->type == PIPE_SHADER_FRAGMENT) {
- pc->result[i*4+c].type = P_TEMP;
- pc->result[i*4+c].hw = -1;
- pc->result[i*4+c].rhw = (i == depr) ?
- -1 : rid++;
- } else {
- pc->result[i*4+c].type = P_RESULT;
- pc->result[i*4+c].hw = rid++;
- }
- pc->result[i*4+c].index = i;
- }
+ for (n = 0; n < pc->attr_nr; ++n) {
+ p->cfg.io[n].hw = rid = aid;
+ i = p->cfg.io[n].id_fp;
+
+ for (c = 0; c < 4; ++c) {
+ if (!pc->attr[i * 4 + c].acc)
+ continue;
+ pc->attr[i * 4 + c].rhw = rid++;
+ p->cfg.io[n].mask |= 1 << c;
- if (pc->p->type == PIPE_SHADER_FRAGMENT &&
- depr != 0xffff) {
- pc->result[depr * 4 + 2].rhw =
- (pc->result_nr - 1) * 4;
+ load_interpolant(pc, &pc->attr[i * 4 + c]);
}
+ aid += popcnt4(p->cfg.io[n].mask);
}
- }
- if (pc->param_nr) {
- int rid = 0;
+ if (!base)
+ p->cfg.regs[1] |= p->cfg.io[0].mask << 24;
- pc->param = CALLOC(pc->param_nr * 4, sizeof(struct nv50_reg));
- if (!pc->param)
- goto out_err;
+ m = popcnt4(p->cfg.regs[1] >> 24);
- for (i = 0; i < pc->param_nr; i++) {
- for (c = 0; c < 4; c++) {
- pc->param[i*4+c].type = P_CONST;
- pc->param[i*4+c].hw = rid++;
- pc->param[i*4+c].index = i;
+ /* set count of non-position inputs and of non-flat
+ * non-position inputs for FP_INTERPOLANT_CTRL
+ */
+ p->cfg.regs[1] |= aid - m;
+
+ if (flat_nr) {
+ i = p->cfg.io[pc->attr_nr - flat_nr].hw;
+ p->cfg.regs[1] |= (i - m) << 16;
+ } else
+ p->cfg.regs[1] |= p->cfg.regs[1] << 16;
+
+ /* mark color semantic for light-twoside */
+ n = 0x40;
+ for (i = 0; i < pc->attr_nr; i++) {
+ ubyte si, sn;
+
+ sn = p->info.input_semantic_name[p->cfg.io[i].id_fp];
+ si = p->info.input_semantic_index[p->cfg.io[i].id_fp];
+
+ if (sn == TGSI_SEMANTIC_COLOR) {
+ p->cfg.two_side[si] = p->cfg.io[i];
+
+ /* increase colour count */
+ p->cfg.regs[0] += popcnt4(
+ p->cfg.two_side[si].mask) << 16;
+
+ n = MIN2(n, p->cfg.io[i].hw - m);
}
}
+ if (n < 0x40)
+ p->cfg.regs[0] += n;
+
+ /* Initialize FP results:
+ * FragDepth is always first TGSI and last hw output
+ */
+ i = p->info.writes_z ? 4 : 0;
+ for (rid = 0; i < pc->result_nr * 4; i++)
+ pc->result[i].rhw = rid++;
+ if (p->info.writes_z)
+ pc->result[2].rhw = rid;
}
if (pc->immd_nr) {
int rid = 0;
- pc->immd = CALLOC(pc->immd_nr * 4, sizeof(struct nv50_reg));
+ pc->immd = MALLOC(pc->immd_nr * 4 * sizeof(struct nv50_reg));
if (!pc->immd)
goto out_err;
for (i = 0; i < pc->immd_nr; i++) {
- for (c = 0; c < 4; c++) {
- pc->immd[i*4+c].type = P_IMMD;
- pc->immd[i*4+c].hw = rid++;
- pc->immd[i*4+c].index = i;
- }
+ for (c = 0; c < 4; c++, rid++)
+ ctor_reg(&pc->immd[rid], P_IMMD, i, rid);
}
}
ret = TRUE;
out_err:
- if (r_usage[0])
- FREE(r_usage[0]);
- if (r_usage[1])
- FREE(r_usage[1]);
+ if (pc->iv_p)
+ free_temp(pc, pc->iv_p);
+ if (pc->iv_c)
+ free_temp(pc, pc->iv_c);
- tgsi_parse_free(&p);
+ tgsi_parse_free(&tp);
return ret;
}
@@ -2110,6 +2281,88 @@ free_nv50_pc(struct nv50_pc *pc)
}
static boolean
+ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p)
+{
+ int i, c;
+ unsigned rtype[2] = { P_ATTR, P_RESULT };
+
+ pc->p = p;
+ pc->temp_nr = p->info.file_max[TGSI_FILE_TEMPORARY] + 1;
+ pc->attr_nr = p->info.file_max[TGSI_FILE_INPUT] + 1;
+ pc->result_nr = p->info.file_max[TGSI_FILE_OUTPUT] + 1;
+ pc->param_nr = p->info.file_max[TGSI_FILE_CONSTANT] + 1;
+
+ p->cfg.high_temp = 4;
+
+ p->cfg.two_side[0].hw = 0x40;
+ p->cfg.two_side[1].hw = 0x40;
+
+ switch (p->type) {
+ case PIPE_SHADER_VERTEX:
+ p->cfg.psiz = 0x40;
+ p->cfg.clpd = 0x40;
+ p->cfg.io_nr = pc->result_nr;
+ break;
+ case PIPE_SHADER_FRAGMENT:
+ rtype[0] = rtype[1] = P_TEMP;
+
+ p->cfg.regs[0] = 0x01000004;
+ p->cfg.io_nr = pc->attr_nr;
+
+ if (p->info.writes_z) {
+ p->cfg.regs[2] |= 0x00000100;
+ p->cfg.regs[3] |= 0x00000011;
+ }
+ if (p->info.uses_kill)
+ p->cfg.regs[2] |= 0x00100000;
+ break;
+ }
+
+ if (pc->temp_nr) {
+ pc->temp = MALLOC(pc->temp_nr * 4 * sizeof(struct nv50_reg));
+ if (!pc->temp)
+ return FALSE;
+
+ for (i = 0; i < pc->temp_nr * 4; ++i)
+ ctor_reg(&pc->temp[i], P_TEMP, i / 4, -1);
+ }
+
+ if (pc->attr_nr) {
+ pc->attr = MALLOC(pc->attr_nr * 4 * sizeof(struct nv50_reg));
+ if (!pc->attr)
+ return FALSE;
+
+ for (i = 0; i < pc->attr_nr * 4; ++i)
+ ctor_reg(&pc->attr[i], rtype[0], i / 4, -1);
+ }
+
+ if (pc->result_nr) {
+ unsigned nr = pc->result_nr * 4;
+
+ pc->result = MALLOC(nr * sizeof(struct nv50_reg));
+ if (!pc->result)
+ return FALSE;
+
+ for (i = 0; i < nr; ++i)
+ ctor_reg(&pc->result[i], rtype[1], i / 4, -1);
+ }
+
+ if (pc->param_nr) {
+ int rid = 0;
+
+ pc->param = MALLOC(pc->param_nr * 4 * sizeof(struct nv50_reg));
+ if (!pc->param)
+ return FALSE;
+
+ for (i = 0; i < pc->param_nr; ++i)
+ for (c = 0; c < 4; ++c, ++rid)
+ ctor_reg(&pc->param[rid], P_CONST, i, rid);
+ }
+
+ return TRUE;
+}
+
+static boolean
nv50_program_tx(struct nv50_program *p)
{
struct tgsi_parse_context parse;
@@ -2120,8 +2373,10 @@ nv50_program_tx(struct nv50_program *p)
pc = CALLOC_STRUCT(nv50_pc);
if (!pc)
return FALSE;
- pc->p = p;
- pc->p->cfg.high_temp = 4;
+
+ ret = ctor_nv50_pc(pc, p);
+ if (ret == FALSE)
+ goto out_cleanup;
ret = nv50_program_tx_prep(pc);
if (ret == FALSE)
@@ -2141,7 +2396,7 @@ nv50_program_tx(struct nv50_program *p)
switch (tok->Token.Type) {
case TGSI_TOKEN_TYPE_INSTRUCTION:
++pc->insn_cur;
- ret = nv50_program_tx_insn(pc, tok);
+ ret = nv50_tgsi_insn(pc, tok);
if (ret == FALSE)
goto out_err;
break;
@@ -2152,8 +2407,8 @@ nv50_program_tx(struct nv50_program *p)
if (p->type == PIPE_SHADER_FRAGMENT) {
struct nv50_reg out;
+ ctor_reg(&out, P_TEMP, -1, -1);
- out.type = P_TEMP;
for (k = 0; k < pc->result_nr * 4; k++) {
if (pc->result[k].rhw == -1)
continue;
@@ -2258,30 +2513,19 @@ nv50_program_validate_data(struct nv50_context *nv50, struct nv50_program *p)
p->immd_nr, NV50_CB_PMISC);
}
- if (!p->data[1] && p->param_nr) {
- struct nouveau_resource *heap =
- nv50->screen->parm_heap[p->type];
-
- if (nouveau_resource_alloc(heap, p->param_nr, p, &p->data[1])) {
- while (heap->next && heap->size < p->param_nr) {
- struct nv50_program *evict = heap->next->priv;
- nouveau_resource_free(&evict->data[1]);
- }
-
- if (nouveau_resource_alloc(heap, p->param_nr, p,
- &p->data[1]))
- assert(0);
- }
- }
+ assert(p->param_nr <= 128);
if (p->param_nr) {
- unsigned cbuf = NV50_CB_PVP;
+ unsigned cb;
float *map = pipe_buffer_map(pscreen, nv50->constbuf[p->type],
PIPE_BUFFER_USAGE_CPU_READ);
- if (p->type == PIPE_SHADER_FRAGMENT)
- cbuf = NV50_CB_PFP;
- nv50_program_upload_data(nv50, map, p->data[1]->start,
- p->param_nr, cbuf);
+
+ if (p->type == PIPE_SHADER_VERTEX)
+ cb = NV50_CB_PVP;
+ else
+ cb = NV50_CB_PFP;
+
+ nv50_program_upload_data(nv50, map, 0, p->param_nr, cb);
pipe_buffer_unmap(pscreen, nv50->constbuf[p->type]);
}
}
@@ -2303,32 +2547,30 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p)
upload = TRUE;
}
- if ((p->data[0] && p->data[0]->start != p->data_start[0]) ||
- (p->data[1] && p->data[1]->start != p->data_start[1])) {
- for (e = p->exec_head; e; e = e->next) {
- unsigned ei, ci, bs;
-
- if (e->param.index < 0)
- continue;
- bs = (e->inst[1] >> 22) & 0x07;
- assert(bs < 2);
- ei = e->param.shift >> 5;
- ci = e->param.index + p->data[bs]->start;
+ if (p->data[0] && p->data[0]->start != p->data_start[0])
+ upload = TRUE;
- e->inst[ei] &= ~e->param.mask;
- e->inst[ei] |= (ci << e->param.shift);
- }
+ if (!upload)
+ return;
- if (p->data[0])
- p->data_start[0] = p->data[0]->start;
- if (p->data[1])
- p->data_start[1] = p->data[1]->start;
+ for (e = p->exec_head; e; e = e->next) {
+ unsigned ei, ci, bs;
- upload = TRUE;
+ if (e->param.index < 0)
+ continue;
+ bs = (e->inst[1] >> 22) & 0x07;
+ assert(bs < 2);
+ ei = e->param.shift >> 5;
+ ci = e->param.index;
+ if (bs == 0)
+ ci += p->data[bs]->start;
+
+ e->inst[ei] &= ~e->param.mask;
+ e->inst[ei] |= (ci << e->param.shift);
}
- if (!upload)
- return;
+ if (p->data[0])
+ p->data_start[0] = p->data[0]->start;
#ifdef NV50_PROGRAM_DUMP
NOUVEAU_ERR("-------\n");
@@ -2402,8 +2644,8 @@ nv50_vertprog_validate(struct nv50_context *nv50)
so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
NOUVEAU_BO_LOW, 0, 0);
so_method(so, tesla, NV50TCL_VP_ATTR_EN_0, 2);
- so_data (so, p->cfg.vp.attr[0]);
- so_data (so, p->cfg.vp.attr[1]);
+ so_data (so, p->cfg.attr[0]);
+ so_data (so, p->cfg.attr[1]);
so_method(so, tesla, NV50TCL_VP_REG_ALLOC_RESULT, 1);
so_data (so, p->cfg.high_result);
so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 2);
@@ -2421,7 +2663,6 @@ nv50_fragprog_validate(struct nv50_context *nv50)
struct nouveau_grobj *tesla = nv50->screen->tesla;
struct nv50_program *p = nv50->fragprog;
struct nouveau_stateobj *so;
- unsigned i;
if (!p->translated) {
nv50_program_validate(nv50, p);
@@ -2438,29 +2679,186 @@ nv50_fragprog_validate(struct nv50_context *nv50)
NOUVEAU_BO_HIGH, 0, 0);
so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
NOUVEAU_BO_LOW, 0, 0);
- so_method(so, tesla, NV50TCL_MAP_SEMANTIC_0, 4);
- so_data (so, p->cfg.fp.regs[0]); /* 0x01000404 / 0x00040404 */
- so_data (so, 0x00000004);
- so_data (so, 0x00000000);
- so_data (so, 0x00000000);
- so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), p->cfg.fp.high_map);
- for (i = 0; i < p->cfg.fp.high_map; i++)
- so_data(so, p->cfg.fp.map[i]);
- so_method(so, tesla, NV50TCL_FP_INTERPOLANT_CTRL, 2);
- so_data (so, p->cfg.fp.regs[1]); /* 0x08040404 / 0x0f000401 */
+ so_method(so, tesla, NV50TCL_FP_REG_ALLOC_TEMP, 1);
so_data (so, p->cfg.high_temp);
so_method(so, tesla, NV50TCL_FP_RESULT_COUNT, 1);
so_data (so, p->cfg.high_result);
so_method(so, tesla, NV50TCL_FP_CTRL_UNK19A8, 1);
- so_data (so, p->cfg.fp.regs[2]);
+ so_data (so, p->cfg.regs[2]);
so_method(so, tesla, NV50TCL_FP_CTRL_UNK196C, 1);
- so_data (so, p->cfg.fp.regs[3]);
+ so_data (so, p->cfg.regs[3]);
so_method(so, tesla, NV50TCL_FP_START_ID, 1);
so_data (so, 0); /* program start offset */
so_ref(so, &nv50->state.fragprog);
so_ref(NULL, &so);
}
+static void
+nv50_pntc_replace(struct nv50_context *nv50, uint32_t pntc[8], unsigned base)
+{
+ struct nv50_program *fp = nv50->fragprog;
+ struct nv50_program *vp = nv50->vertprog;
+ unsigned i, c, m = base;
+
+ /* XXX: This can't work correctly in all cases yet, we either
+ * have to create TGSI_SEMANTIC_PNTC or sprite_coord_mode has
+ * to be per FP input instead of per VP output
+ */
+ memset(pntc, 0, 8 * sizeof(uint32_t));
+
+ for (i = 0; i < fp->cfg.io_nr; i++) {
+ uint8_t sn, si;
+ uint8_t j = fp->cfg.io[i].id_vp, k = fp->cfg.io[i].id_fp;
+ unsigned n = popcnt4(fp->cfg.io[i].mask);
+
+ if (fp->info.input_semantic_name[k] != TGSI_SEMANTIC_GENERIC) {
+ m += n;
+ continue;
+ }
+
+ sn = vp->info.input_semantic_name[j];
+ si = vp->info.input_semantic_index[j];
+
+ if (j < fp->cfg.io_nr && sn == TGSI_SEMANTIC_GENERIC) {
+ ubyte mode =
+ nv50->rasterizer->pipe.sprite_coord_mode[si];
+
+ if (mode == PIPE_SPRITE_COORD_NONE) {
+ m += n;
+ continue;
+ }
+ }
+
+ /* this is either PointCoord or replaced by sprite coords */
+ for (c = 0; c < 4; c++) {
+ if (!(fp->cfg.io[i].mask & (1 << c)))
+ continue;
+ pntc[m / 8] |= (c + 1) << ((m % 8) * 4);
+ ++m;
+ }
+ }
+}
+
+static int
+nv50_sreg4_map(uint32_t *p_map, int mid, uint32_t lin[4],
+ struct nv50_sreg4 *fpi, struct nv50_sreg4 *vpo)
+{
+ int c;
+ uint8_t mv = vpo->mask, mf = fpi->mask, oid = vpo->hw;
+ uint8_t *map = (uint8_t *)p_map;
+
+ for (c = 0; c < 4; ++c) {
+ if (mf & 1) {
+ if (fpi->linear == TRUE)
+ lin[mid / 32] |= 1 << (mid % 32);
+ map[mid++] = (mv & 1) ? oid : ((c == 3) ? 0x41 : 0x40);
+ }
+
+ oid += mv & 1;
+ mf >>= 1;
+ mv >>= 1;
+ }
+
+ return mid;
+}
+
+void
+nv50_linkage_validate(struct nv50_context *nv50)
+{
+ struct nouveau_grobj *tesla = nv50->screen->tesla;
+ struct nv50_program *vp = nv50->vertprog;
+ struct nv50_program *fp = nv50->fragprog;
+ struct nouveau_stateobj *so;
+ struct nv50_sreg4 dummy, *vpo;
+ int i, n, c, m = 0;
+ uint32_t map[16], lin[4], reg[5], pcrd[8];
+
+ memset(map, 0, sizeof(map));
+ memset(lin, 0, sizeof(lin));
+
+ reg[1] = 0x00000004; /* low and high clip distance map ids */
+ reg[2] = 0x00000000; /* layer index map id (disabled, GP only) */
+ reg[3] = 0x00000000; /* point size map id & enable */
+ reg[0] = fp->cfg.regs[0]; /* colour semantic reg */
+ reg[4] = fp->cfg.regs[1]; /* interpolant info */
+
+ dummy.linear = FALSE;
+ dummy.mask = 0xf; /* map all components of HPOS */
+ m = nv50_sreg4_map(map, m, lin, &dummy, &vp->cfg.io[0]);
+
+ dummy.mask = 0x0;
+
+ if (vp->cfg.clpd < 0x40) {
+ for (c = 0; c < vp->cfg.clpd_nr; ++c)
+ map[m++] = vp->cfg.clpd + c;
+ reg[1] = (m << 8);
+ }
+
+ reg[0] |= m << 8; /* adjust BFC0 id */
+
+ /* if light_twoside is active, it seems FFC0_ID == BFC0_ID is bad */
+ if (nv50->rasterizer->pipe.light_twoside) {
+ vpo = &vp->cfg.two_side[0];
+
+ m = nv50_sreg4_map(map, m, lin, &fp->cfg.two_side[0], &vpo[0]);
+ m = nv50_sreg4_map(map, m, lin, &fp->cfg.two_side[1], &vpo[1]);
+ }
+
+ reg[0] += m - 4; /* adjust FFC0 id */
+ reg[4] |= m << 8; /* set mid where 'normal' FP inputs start */
+
+ i = 0;
+ if (fp->info.input_semantic_name[0] == TGSI_SEMANTIC_POSITION)
+ i = 1;
+ for (; i < fp->cfg.io_nr; i++) {
+ ubyte sn = fp->info.input_semantic_name[fp->cfg.io[i].id_fp];
+ ubyte si = fp->info.input_semantic_index[fp->cfg.io[i].id_fp];
+
+ n = fp->cfg.io[i].id_vp;
+ if (n >= vp->cfg.io_nr ||
+ vp->info.output_semantic_name[n] != sn ||
+ vp->info.output_semantic_index[n] != si)
+ vpo = &dummy;
+ else
+ vpo = &vp->cfg.io[n];
+
+ m = nv50_sreg4_map(map, m, lin, &fp->cfg.io[i], vpo);
+ }
+
+ if (nv50->rasterizer->pipe.point_size_per_vertex) {
+ map[m / 4] |= vp->cfg.psiz << ((m % 4) * 8);
+ reg[3] = (m++ << 4) | 1;
+ }
+
+ /* now fill the stateobj */
+ so = so_new(64, 0);
+
+ n = (m + 3) / 4;
+ so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1);
+ so_data (so, m);
+ so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), n);
+ so_datap (so, map, n);
+
+ so_method(so, tesla, NV50TCL_MAP_SEMANTIC_0, 4);
+ so_datap (so, reg, 4);
+
+ so_method(so, tesla, NV50TCL_FP_INTERPOLANT_CTRL, 1);
+ so_data (so, reg[4]);
+
+ so_method(so, tesla, 0x1540, 4);
+ so_datap (so, lin, 4);
+
+ if (nv50->rasterizer->pipe.point_sprite) {
+ nv50_pntc_replace(nv50, pcrd, (reg[4] >> 8) & 0xff);
+
+ so_method(so, tesla, NV50TCL_POINT_COORD_REPLACE_MAP(0), 8);
+ so_datap (so, pcrd, 8);
+ }
+
+ so_ref(so, &nv50->state.programs);
+ so_ref(NULL, &so);
+}
+
void
nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p)
{
@@ -2476,7 +2874,6 @@ nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p)
nouveau_bo_ref(NULL, &p->bo);
nouveau_resource_free(&p->data[0]);
- nouveau_resource_free(&p->data[1]);
p->translated = 0;
}
diff --git a/src/gallium/drivers/nv50/nv50_program.h b/src/gallium/drivers/nv50/nv50_program.h
index 096e0476aa..d78dee083f 100644
--- a/src/gallium/drivers/nv50/nv50_program.h
+++ b/src/gallium/drivers/nv50/nv50_program.h
@@ -15,6 +15,15 @@ struct nv50_program_exec {
} param;
};
+struct nv50_sreg4 {
+ uint8_t hw;
+ uint8_t id_vp;
+ uint8_t id_fp;
+
+ uint8_t mask;
+ boolean linear;
+};
+
struct nv50_program {
struct pipe_shader_state pipe;
struct tgsi_shader_info info;
@@ -24,8 +33,8 @@ struct nv50_program {
struct nv50_program_exec *exec_head;
struct nv50_program_exec *exec_tail;
unsigned exec_size;
- struct nouveau_resource *data[2];
- unsigned data_start[2];
+ struct nouveau_resource *data[1];
+ unsigned data_start[1];
struct nouveau_bo *bo;
@@ -36,14 +45,20 @@ struct nv50_program {
struct {
unsigned high_temp;
unsigned high_result;
- struct {
- unsigned attr[2];
- } vp;
- struct {
- unsigned regs[4];
- unsigned map[5];
- unsigned high_map;
- } fp;
+
+ uint32_t attr[2];
+ uint32_t regs[4];
+
+ /* for VPs, io_nr doesn't count 'private' results (PSIZ etc.) */
+ unsigned io_nr;
+ struct nv50_sreg4 io[PIPE_MAX_SHADER_OUTPUTS];
+
+ /* FP colour inputs, VP/GP back colour outputs */
+ struct nv50_sreg4 two_side[2];
+
+ /* VP only */
+ uint8_t clpd, clpd_nr;
+ uint8_t psiz;
} cfg;
};
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index c7f80a2203..3b08e1b89f 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -87,12 +87,10 @@ nv50_screen_get_param(struct pipe_screen *pscreen, int param)
return 1;
case PIPE_CAP_GLSL:
return 0;
- case PIPE_CAP_S3TC:
- return 1;
case PIPE_CAP_ANISOTROPIC_FILTER:
return 1;
case PIPE_CAP_POINT_SPRITE:
- return 0;
+ return 1;
case PIPE_CAP_MAX_RENDER_TARGETS:
return 8;
case PIPE_CAP_OCCLUSION_QUERY:
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index 4283808ed9..81fa3e34c5 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -276,6 +276,9 @@ nv50_rasterizer_state_create(struct pipe_context *pipe,
so_method(so, tesla, 0x1684, 1);
so_data (so, cso->flatshade_first ? 0 : 1);
+ so_method(so, tesla, NV50TCL_VERTEX_TWO_SIDE_ENABLE, 1);
+ so_data (so, cso->light_twoside);
+
so_method(so, tesla, NV50TCL_LINE_WIDTH, 1);
so_data (so, fui(cso->line_width));
so_method(so, tesla, NV50TCL_LINE_SMOOTH_ENABLE, 1);
@@ -294,6 +297,9 @@ nv50_rasterizer_state_create(struct pipe_context *pipe,
so_method(so, tesla, NV50TCL_POINT_SIZE, 1);
so_data (so, fui(cso->point_size));
+ so_method(so, tesla, NV50TCL_POINT_SPRITE_ENABLE, 1);
+ so_data (so, cso->point_sprite);
+
so_method(so, tesla, NV50TCL_POLYGON_MODE_FRONT, 3);
if (cso->front_winding == PIPE_WINDING_CCW) {
so_data(so, nvgl_polygon_mode(cso->fill_ccw));
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index 344c2cf6dd..5a3559ed18 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -66,7 +66,8 @@ nv50_state_validate_fb(struct nv50_context *nv50)
so_data(so, NV50TCL_RT_FORMAT_X8R8G8B8_UNORM);
break;
}
- so_data(so, bo->tile_mode << 4);
+ so_data(so, nv50_miptree(pt)->
+ level[fb->cbufs[i]->level].tile_mode << 4);
so_data(so, 0x00000000);
so_method(so, tesla, 0x1224, 1);
@@ -110,7 +111,8 @@ nv50_state_validate_fb(struct nv50_context *nv50)
so_data(so, NV50TCL_ZETA_FORMAT_S8Z24_UNORM);
break;
}
- so_data(so, bo->tile_mode << 4);
+ so_data(so, nv50_miptree(pt)->
+ level[fb->zsbuf->level].tile_mode << 4);
so_data(so, 0x00000000);
so_method(so, tesla, 0x1538, 1);
@@ -187,6 +189,8 @@ nv50_state_emit(struct nv50_context *nv50)
so_emit(chan, nv50->state.vertprog);
if (nv50->state.dirty & NV50_NEW_FRAGPROG)
so_emit(chan, nv50->state.fragprog);
+ if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG))
+ so_emit(chan, nv50->state.programs);
if (nv50->state.dirty & NV50_NEW_RASTERIZER)
so_emit(chan, nv50->state.rast);
if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR)
@@ -208,6 +212,12 @@ nv50_state_emit(struct nv50_context *nv50)
so_emit(chan, nv50->state.vtxattr);
}
nv50->state.dirty = 0;
+}
+
+void
+nv50_state_flush_notify(struct nouveau_channel *chan)
+{
+ struct nv50_context *nv50 = chan->user_private;
so_emit_reloc_markers(chan, nv50->state.fb);
so_emit_reloc_markers(chan, nv50->state.vertprog);
@@ -238,6 +248,9 @@ nv50_state_validate(struct nv50_context *nv50)
if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB))
nv50_fragprog_validate(nv50);
+ if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG))
+ nv50_linkage_validate(nv50);
+
if (nv50->dirty & NV50_NEW_RASTERIZER)
so_ref(nv50->rasterizer->so, &nv50->state.rast);
diff --git a/src/gallium/drivers/nv50/nv50_surface.c b/src/gallium/drivers/nv50/nv50_surface.c
index b266324f58..6bf6f773b0 100644
--- a/src/gallium/drivers/nv50/nv50_surface.c
+++ b/src/gallium/drivers/nv50/nv50_surface.c
@@ -60,13 +60,13 @@ nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst)
format = nv50_format(ps->format);
if (format < 0)
return 1;
-
+
if (!bo->tile_flags) {
BEGIN_RING(chan, eng2d, mthd, 2);
OUT_RING (chan, format);
OUT_RING (chan, 1);
BEGIN_RING(chan, eng2d, mthd + 0x14, 5);
- OUT_RING (chan, mt->level[0].pitch);
+ OUT_RING (chan, mt->level[ps->level].pitch);
OUT_RING (chan, ps->width);
OUT_RING (chan, ps->height);
OUT_RELOCh(chan, bo, ps->offset, flags);
@@ -75,7 +75,7 @@ nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst)
BEGIN_RING(chan, eng2d, mthd, 5);
OUT_RING (chan, format);
OUT_RING (chan, 0);
- OUT_RING (chan, bo->tile_mode << 4);
+ OUT_RING (chan, mt->level[ps->level].tile_mode << 4);
OUT_RING (chan, 1);
OUT_RING (chan, 0);
BEGIN_RING(chan, eng2d, mthd + 0x18, 4);
diff --git a/src/gallium/drivers/nv50/nv50_transfer.c b/src/gallium/drivers/nv50/nv50_transfer.c
index e9c3562194..bb7731855c 100644
--- a/src/gallium/drivers/nv50/nv50_transfer.c
+++ b/src/gallium/drivers/nv50/nv50_transfer.c
@@ -89,14 +89,14 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
if (src_bo->tile_flags) {
BEGIN_RING(chan, m2mf,
NV50_MEMORY_TO_MEMORY_FORMAT_TILING_POSITION_IN, 1);
- OUT_RING (chan, (sy << 16) | sx);
+ OUT_RING (chan, (sy << 16) | (sx * cpp));
} else {
src_offset += (line_count * src_pitch);
}
if (dst_bo->tile_flags) {
BEGIN_RING(chan, m2mf,
NV50_MEMORY_TO_MEMORY_FORMAT_TILING_POSITION_OUT, 1);
- OUT_RING (chan, (dy << 16) | dx);
+ OUT_RING (chan, (dy << 16) | (dx * cpp));
} else {
dst_offset += (line_count * dst_pitch);
}