#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_dump.h"
#include "tgsi/tgsi_util.h"
+#include "tgsi/tgsi_ureg.h"
#include "draw/draw_context.h"
struct nvfx_vpc {
struct nvfx_context* nvfx;
+ struct pipe_shader_state pipe;
struct nvfx_vertex_program *vp;
+ struct tgsi_shader_info* info;
struct nvfx_vertex_program_exec *vpi;
struct nvfx_reg *r_address;
struct nvfx_reg *r_temp;
struct nvfx_reg *r_const;
+ struct nvfx_reg r_0_1;
struct nvfx_reg *imm;
unsigned nr_imm;
return nvfx_reg(NVFXSR_CONST, idx);
}
-#define arith(s,o,d,m,s0,s1,s2) \
- nvfx_insn(0, (NVFX_VP_INST_SLOT_##s << 7) | NVFX_VP_INST_##s##_OP_##o, -1, (d), (m), (s0), (s1), (s2))
+#define arith(s,t,o,d,m,s0,s1,s2) \
+ nvfx_insn((s), (NVFX_VP_INST_SLOT_##t << 7) | NVFX_VP_INST_##t##_OP_##o, -1, (d), (m), (s0), (s1), (s2))
static void
emit_src(struct nvfx_context* nvfx, struct nvfx_vpc *vpc, uint32_t *hw, int pos, struct nvfx_src src)
if(insn.cc_update)
hw[0] |= NVFX_VP(INST_COND_UPDATE_ENABLE);
+ if(insn.sat)
+ {
+ assert(nvfx->use_nv4x);
+ if(nvfx->use_nv4x)
+ hw[0] |= NV40_VP_INST_SATURATE;
+ }
+
if(!nvfx->is_nv4x) {
if(slot == 0)
hw[1] |= (op << NV30_VP_INST_VEC_OPCODE_SHIFT);
src.swz[2] = fsrc->Register.SwizzleZ;
src.swz[3] = fsrc->Register.SwizzleW;
src.indirect = 0;
+ src.indirect_reg = 0;
+ src.indirect_swz = 0;
if(fsrc->Register.Indirect) {
if(fsrc->Indirect.File == TGSI_FILE_ADDRESS &&
{
struct nvfx_src src[3], tmp;
struct nvfx_reg dst;
+ struct nvfx_reg final_dst;
struct nvfx_src none = nvfx_src(nvfx_reg(NVFXSR_NONE, 0));
struct nvfx_insn insn;
struct nvfx_relocation reloc;
struct nvfx_loop_entry loop;
+ boolean sat = FALSE;
int mask;
int ai = -1, ci = -1, ii = -1;
int i;
-
- if (finst->Instruction.Opcode == TGSI_OPCODE_END)
- return TRUE;
+ unsigned sub_depth = 0;
for (i = 0; i < finst->Instruction.NumSrcRegs; i++) {
const struct tgsi_full_src_register *fsrc;
src[i] = tgsi_src(vpc, fsrc);
} else {
src[i] = nvfx_src(temp(vpc));
- nvfx_vp_emit(vpc, arith(VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, tgsi_src(vpc, fsrc), none, none));
+ nvfx_vp_emit(vpc, arith(0, VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, tgsi_src(vpc, fsrc), none, none));
}
break;
case TGSI_FILE_CONSTANT:
src[i] = tgsi_src(vpc, fsrc);
} else {
src[i] = nvfx_src(temp(vpc));
- nvfx_vp_emit(vpc, arith(VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, tgsi_src(vpc, fsrc), none, none));
+ nvfx_vp_emit(vpc, arith(0, VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, tgsi_src(vpc, fsrc), none, none));
}
break;
case TGSI_FILE_IMMEDIATE:
src[i] = tgsi_src(vpc, fsrc);
} else {
src[i] = nvfx_src(temp(vpc));
- nvfx_vp_emit(vpc, arith(VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, tgsi_src(vpc, fsrc), none, none));
+ nvfx_vp_emit(vpc, arith(0, VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, tgsi_src(vpc, fsrc), none, none));
}
break;
case TGSI_FILE_TEMPORARY:
finst->Instruction.Opcode != TGSI_OPCODE_ARL)
return FALSE;
- dst = tgsi_dst(vpc, &finst->Dst[0]);
+ final_dst = dst = tgsi_dst(vpc, &finst->Dst[0]);
mask = tgsi_mask(finst->Dst[0].Register.WriteMask);
+ if(finst->Instruction.Saturate == TGSI_SAT_ZERO_ONE)
+ {
+ assert(finst->Instruction.Opcode != TGSI_OPCODE_ARL);
+ if(nvfx->use_nv4x)
+ sat = TRUE;
+ else if(dst.type != NVFXSR_TEMP)
+ dst = temp(vpc);
+ }
switch (finst->Instruction.Opcode) {
case TGSI_OPCODE_ABS:
- nvfx_vp_emit(vpc, arith(VEC, MOV, dst, mask, abs(src[0]), none, none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, MOV, dst, mask, abs(src[0]), none, none));
break;
case TGSI_OPCODE_ADD:
- nvfx_vp_emit(vpc, arith(VEC, ADD, dst, mask, src[0], none, src[1]));
+ nvfx_vp_emit(vpc, arith(sat, VEC, ADD, dst, mask, src[0], none, src[1]));
break;
case TGSI_OPCODE_ARL:
- nvfx_vp_emit(vpc, arith(VEC, ARL, dst, mask, src[0], none, none));
+ nvfx_vp_emit(vpc, arith(0, VEC, ARL, dst, mask, src[0], none, none));
break;
case TGSI_OPCODE_CMP:
- insn = arith(VEC, MOV, none.reg, mask, src[0], none, none);
+ insn = arith(0, VEC, MOV, none.reg, mask, src[0], none, none);
insn.cc_update = 1;
nvfx_vp_emit(vpc, insn);
- insn = arith(VEC, MOV, dst, mask, src[2], none, none);
+ insn = arith(sat, VEC, MOV, dst, mask, src[2], none, none);
insn.cc_test = NVFX_COND_GE;
nvfx_vp_emit(vpc, insn);
- insn = arith(VEC, MOV, dst, mask, src[1], none, none);
+ insn = arith(sat, VEC, MOV, dst, mask, src[1], none, none);
insn.cc_test = NVFX_COND_LT;
nvfx_vp_emit(vpc, insn);
break;
case TGSI_OPCODE_COS:
- nvfx_vp_emit(vpc, arith(SCA, COS, dst, mask, none, none, src[0]));
+ nvfx_vp_emit(vpc, arith(sat, SCA, COS, dst, mask, none, none, src[0]));
break;
case TGSI_OPCODE_DP2:
tmp = nvfx_src(temp(vpc));
- nvfx_vp_emit(vpc, arith(VEC, MUL, tmp.reg, NVFX_VP_MASK_X | NVFX_VP_MASK_Y, src[0], src[1], none));
- nvfx_vp_emit(vpc, arith(VEC, ADD, dst, mask, swz(tmp, X, X, X, X), none, swz(tmp, Y, Y, Y, Y)));
+ nvfx_vp_emit(vpc, arith(0, VEC, MUL, tmp.reg, NVFX_VP_MASK_X | NVFX_VP_MASK_Y, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, ADD, dst, mask, swz(tmp, X, X, X, X), none, swz(tmp, Y, Y, Y, Y)));
break;
case TGSI_OPCODE_DP3:
- nvfx_vp_emit(vpc, arith(VEC, DP3, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, DP3, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_DP4:
- nvfx_vp_emit(vpc, arith(VEC, DP4, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, DP4, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_DPH:
- nvfx_vp_emit(vpc, arith(VEC, DPH, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, DPH, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_DST:
- nvfx_vp_emit(vpc, arith(VEC, DST, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, DST, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_EX2:
- nvfx_vp_emit(vpc, arith(SCA, EX2, dst, mask, none, none, src[0]));
+ nvfx_vp_emit(vpc, arith(sat, SCA, EX2, dst, mask, none, none, src[0]));
break;
case TGSI_OPCODE_EXP:
- nvfx_vp_emit(vpc, arith(SCA, EXP, dst, mask, none, none, src[0]));
+ nvfx_vp_emit(vpc, arith(sat, SCA, EXP, dst, mask, none, none, src[0]));
break;
case TGSI_OPCODE_FLR:
- nvfx_vp_emit(vpc, arith(VEC, FLR, dst, mask, src[0], none, none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, FLR, dst, mask, src[0], none, none));
break;
case TGSI_OPCODE_FRC:
- nvfx_vp_emit(vpc, arith(VEC, FRC, dst, mask, src[0], none, none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, FRC, dst, mask, src[0], none, none));
break;
case TGSI_OPCODE_LG2:
- nvfx_vp_emit(vpc, arith(SCA, LG2, dst, mask, none, none, src[0]));
+ nvfx_vp_emit(vpc, arith(sat, SCA, LG2, dst, mask, none, none, src[0]));
break;
case TGSI_OPCODE_LIT:
- nvfx_vp_emit(vpc, arith(SCA, LIT, dst, mask, none, none, src[0]));
+ nvfx_vp_emit(vpc, arith(sat, SCA, LIT, dst, mask, none, none, src[0]));
break;
case TGSI_OPCODE_LOG:
- nvfx_vp_emit(vpc, arith(SCA, LOG, dst, mask, none, none, src[0]));
+ nvfx_vp_emit(vpc, arith(sat, SCA, LOG, dst, mask, none, none, src[0]));
break;
case TGSI_OPCODE_LRP:
tmp = nvfx_src(temp(vpc));
- nvfx_vp_emit(vpc, arith(VEC, MAD, tmp.reg, mask, neg(src[0]), src[2], src[2]));
- nvfx_vp_emit(vpc, arith(VEC, MAD, dst, mask, src[0], src[1], tmp));
+ nvfx_vp_emit(vpc, arith(0, VEC, MAD, tmp.reg, mask, neg(src[0]), src[2], src[2]));
+ nvfx_vp_emit(vpc, arith(sat, VEC, MAD, dst, mask, src[0], src[1], tmp));
break;
case TGSI_OPCODE_MAD:
- nvfx_vp_emit(vpc, arith(VEC, MAD, dst, mask, src[0], src[1], src[2]));
+ nvfx_vp_emit(vpc, arith(sat, VEC, MAD, dst, mask, src[0], src[1], src[2]));
break;
case TGSI_OPCODE_MAX:
- nvfx_vp_emit(vpc, arith(VEC, MAX, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, MAX, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_MIN:
- nvfx_vp_emit(vpc, arith(VEC, MIN, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, MIN, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_MOV:
- nvfx_vp_emit(vpc, arith(VEC, MOV, dst, mask, src[0], none, none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, MOV, dst, mask, src[0], none, none));
break;
case TGSI_OPCODE_MUL:
- nvfx_vp_emit(vpc, arith(VEC, MUL, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, MUL, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_NOP:
break;
case TGSI_OPCODE_POW:
tmp = nvfx_src(temp(vpc));
- nvfx_vp_emit(vpc, arith(SCA, LG2, tmp.reg, NVFX_VP_MASK_X, none, none, swz(src[0], X, X, X, X)));
- nvfx_vp_emit(vpc, arith(VEC, MUL, tmp.reg, NVFX_VP_MASK_X, swz(tmp, X, X, X, X), swz(src[1], X, X, X, X), none));
- nvfx_vp_emit(vpc, arith(SCA, EX2, dst, mask, none, none, swz(tmp, X, X, X, X)));
+ nvfx_vp_emit(vpc, arith(0, SCA, LG2, tmp.reg, NVFX_VP_MASK_X, none, none, swz(src[0], X, X, X, X)));
+ nvfx_vp_emit(vpc, arith(0, VEC, MUL, tmp.reg, NVFX_VP_MASK_X, swz(tmp, X, X, X, X), swz(src[1], X, X, X, X), none));
+ nvfx_vp_emit(vpc, arith(sat, SCA, EX2, dst, mask, none, none, swz(tmp, X, X, X, X)));
break;
case TGSI_OPCODE_RCP:
- nvfx_vp_emit(vpc, arith(SCA, RCP, dst, mask, none, none, src[0]));
+ nvfx_vp_emit(vpc, arith(sat, SCA, RCP, dst, mask, none, none, src[0]));
break;
case TGSI_OPCODE_RSQ:
- nvfx_vp_emit(vpc, arith(SCA, RSQ, dst, mask, none, none, abs(src[0])));
+ nvfx_vp_emit(vpc, arith(sat, SCA, RSQ, dst, mask, none, none, abs(src[0])));
break;
case TGSI_OPCODE_SEQ:
- nvfx_vp_emit(vpc, arith(VEC, SEQ, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, SEQ, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_SFL:
- nvfx_vp_emit(vpc, arith(VEC, SFL, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, SFL, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_SGE:
- nvfx_vp_emit(vpc, arith(VEC, SGE, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, SGE, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_SGT:
- nvfx_vp_emit(vpc, arith(VEC, SGT, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, SGT, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_SIN:
- nvfx_vp_emit(vpc, arith(SCA, SIN, dst, mask, none, none, src[0]));
+ nvfx_vp_emit(vpc, arith(sat, SCA, SIN, dst, mask, none, none, src[0]));
break;
case TGSI_OPCODE_SLE:
- nvfx_vp_emit(vpc, arith(VEC, SLE, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, SLE, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_SLT:
- nvfx_vp_emit(vpc, arith(VEC, SLT, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, SLT, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_SNE:
- nvfx_vp_emit(vpc, arith(VEC, SNE, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, SNE, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_SSG:
- nvfx_vp_emit(vpc, arith(VEC, SSG, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, SSG, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_STR:
- nvfx_vp_emit(vpc, arith(VEC, STR, dst, mask, src[0], src[1], none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, STR, dst, mask, src[0], src[1], none));
break;
case TGSI_OPCODE_SUB:
- nvfx_vp_emit(vpc, arith(VEC, ADD, dst, mask, src[0], none, neg(src[1])));
+ nvfx_vp_emit(vpc, arith(sat, VEC, ADD, dst, mask, src[0], none, neg(src[1])));
break;
case TGSI_OPCODE_TRUNC:
tmp = nvfx_src(temp(vpc));
- insn = arith(VEC, MOV, none.reg, mask, src[0], none, none);
+ insn = arith(0, VEC, MOV, none.reg, mask, src[0], none, none);
insn.cc_update = 1;
nvfx_vp_emit(vpc, insn);
- nvfx_vp_emit(vpc, arith(VEC, FLR, tmp.reg, mask, abs(src[0]), none, none));
- nvfx_vp_emit(vpc, arith(VEC, MOV, dst, mask, tmp, none, none));
+ nvfx_vp_emit(vpc, arith(0, VEC, FLR, tmp.reg, mask, abs(src[0]), none, none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, MOV, dst, mask, tmp, none, none));
- insn = arith(VEC, MOV, dst, mask, neg(tmp), none, none);
+ insn = arith(sat, VEC, MOV, dst, mask, neg(tmp), none, none);
insn.cc_test = NVFX_COND_LT;
nvfx_vp_emit(vpc, insn);
break;
case TGSI_OPCODE_XPD:
tmp = nvfx_src(temp(vpc));
- nvfx_vp_emit(vpc, arith(VEC, MUL, tmp.reg, mask, swz(src[0], Z, X, Y, Y), swz(src[1], Y, Z, X, X), none));
- nvfx_vp_emit(vpc, arith(VEC, MAD, dst, (mask & ~NVFX_VP_MASK_W), swz(src[0], Y, Z, X, X), swz(src[1], Z, X, Y, Y), neg(tmp)));
+ nvfx_vp_emit(vpc, arith(0, VEC, MUL, tmp.reg, mask, swz(src[0], Z, X, Y, Y), swz(src[1], Y, Z, X, X), none));
+ nvfx_vp_emit(vpc, arith(sat, VEC, MAD, dst, (mask & ~NVFX_VP_MASK_W), swz(src[0], Y, Z, X, X), swz(src[1], Z, X, Y, Y), neg(tmp)));
break;
case TGSI_OPCODE_IF:
- insn = arith(VEC, MOV, none.reg, NVFX_VP_MASK_X, src[0], none, none);
+ insn = arith(0, VEC, MOV, none.reg, NVFX_VP_MASK_X, src[0], none, none);
insn.cc_update = 1;
nvfx_vp_emit(vpc, insn);
reloc.target = finst->Label.Label + 1;
util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc);
- insn = arith(SCA, BRA, none.reg, 0, none, none, none);
+ insn = arith(0, SCA, BRA, none.reg, 0, none, none, none);
insn.cc_test = NVFX_COND_EQ;
insn.cc_swz[0] = insn.cc_swz[1] = insn.cc_swz[2] = insn.cc_swz[3] = 0;
nvfx_vp_emit(vpc, insn);
util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc);
if(finst->Instruction.Opcode == TGSI_OPCODE_CAL)
- insn = arith(SCA, CAL, none.reg, 0, none, none, none);
+ insn = arith(0, SCA, CAL, none.reg, 0, none, none, none);
else
- insn = arith(SCA, BRA, none.reg, 0, none, none, none);
+ insn = arith(0, SCA, BRA, none.reg, 0, none, none, none);
nvfx_vp_emit(vpc, insn);
break;
case TGSI_OPCODE_RET:
- tmp = none;
- tmp.swz[0] = tmp.swz[1] = tmp.swz[2] = tmp.swz[3] = 0;
- nvfx_vp_emit(vpc, arith(SCA, RET, none.reg, 0, none, none, tmp));
+ if(sub_depth || !nvfx->use_vp_clipping) {
+ tmp = none;
+ tmp.swz[0] = tmp.swz[1] = tmp.swz[2] = tmp.swz[3] = 0;
+ nvfx_vp_emit(vpc, arith(0, SCA, RET, none.reg, 0, none, none, tmp));
+ } else {
+ reloc.location = vpc->vp->nr_insns;
+ reloc.target = vpc->info->num_instructions;
+ util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc);
+ nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none));
+ }
break;
case TGSI_OPCODE_BGNSUB:
+ ++sub_depth;
+ break;
case TGSI_OPCODE_ENDSUB:
+ --sub_depth;
+ break;
case TGSI_OPCODE_ENDIF:
/* nothing to do here */
break;
reloc.target = loop.cont_target;
util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc);
- nvfx_vp_emit(vpc, arith(SCA, BRA, none.reg, 0, none, none, none));
+ nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none));
break;
case TGSI_OPCODE_CONT:
reloc.target = loop.cont_target;
util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc);
- nvfx_vp_emit(vpc, arith(SCA, BRA, none.reg, 0, none, none, none));
+ nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none));
break;
case TGSI_OPCODE_BRK:
reloc.target = loop.brk_target;
util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc);
- nvfx_vp_emit(vpc, arith(SCA, BRA, none.reg, 0, none, none, none));
+ nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none));
+ break;
+
+ case TGSI_OPCODE_END:
+ assert(!sub_depth);
+ if(nvfx->use_vp_clipping) {
+ if(idx != (vpc->info->num_instructions - 1)) {
+ reloc.location = vpc->vp->nr_insns;
+ reloc.target = vpc->info->num_instructions;
+ util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc);
+ nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none));
+ }
+ } else {
+ if(vpc->vp->nr_insns)
+ vpc->vp->insns[vpc->vp->nr_insns - 1].data[3] |= NVFX_VP_INST_LAST;
+ nvfx_vp_emit(vpc, arith(0, VEC, NOP, none.reg, 0, none, none, none));
+ vpc->vp->insns[vpc->vp->nr_insns - 1].data[3] |= NVFX_VP_INST_LAST;
+ }
break;
default:
return FALSE;
}
+ if(finst->Instruction.Saturate == TGSI_SAT_ZERO_ONE && !nvfx->use_nv4x)
+ {
+ if(!vpc->r_0_1.type)
+ vpc->r_0_1 = constant(vpc, -1, 0, 1, 0, 0);
+ nvfx_vp_emit(vpc, arith(0, VEC, MAX, dst, mask, nvfx_src(dst), swz(nvfx_src(vpc->r_0_1), X, X, X, X), none));
+ nvfx_vp_emit(vpc, arith(0, VEC, MIN, final_dst, mask, nvfx_src(dst), swz(nvfx_src(vpc->r_0_1), Y, Y, Y, Y), none));
+ }
+
release_temps(vpc);
return TRUE;
}
unsigned num_outputs;
unsigned num_texcoords = nvfx->is_nv4x ? 10 : 8;
- num_outputs = util_semantic_set_from_program_file(&set, vpc->vp->pipe.tokens, TGSI_FILE_OUTPUT);
+ num_outputs = util_semantic_set_from_program_file(&set, vpc->pipe.tokens, TGSI_FILE_OUTPUT);
if(num_outputs > num_texcoords) {
NOUVEAU_ERR("too many vertex program outputs: %i\n", num_outputs);
/* hope 0xf is (0, 0, 0, 1) initialized; otherwise, we are _probably_ not required to do this */
memset(vpc->vp->generic_to_fp_input, 0x0f, sizeof(vpc->vp->generic_to_fp_input));
- for(int i = 0; i < 10; ++i) {
+ for(int i = 0; i < num_texcoords; ++i) {
if(sem_layout[i] == 0xff)
continue;
//printf("vp: GENERIC[%i] to fpreg %i\n", sem_layout[i], NVFX_FP_OP_INPUT_SRC_TC(0) + i);
}
vpc->vp->sprite_fp_input = -1;
- for(int i = 0; i < 10; ++i)
+ for(int i = 0; i < num_texcoords; ++i)
{
if(sem_layout[i] == 0xff)
{
}
}
- tgsi_parse_init(&p, vpc->vp->pipe.tokens);
+ tgsi_parse_init(&p, vpc->pipe.tokens);
while (!tgsi_parse_end_of_tokens(&p)) {
const union tgsi_full_token *tok = &p.FullToken;
DEBUG_GET_ONCE_BOOL_OPTION(nvfx_dump_vp, "NVFX_DUMP_VP", FALSE)
-static void
-nvfx_vertprog_translate(struct nvfx_context *nvfx,
- struct nvfx_vertex_program *vp)
+static struct nvfx_vertex_program*
+nvfx_vertprog_translate(struct nvfx_context *nvfx, const struct pipe_shader_state* vps, struct tgsi_shader_info* info)
{
struct tgsi_parse_context parse;
+ struct nvfx_vertex_program* vp = NULL;
struct nvfx_vpc *vpc = NULL;
struct nvfx_src none = nvfx_src(nvfx_reg(NVFXSR_NONE, 0));
struct util_dynarray insns;
int i;
- vpc = CALLOC(1, sizeof(struct nvfx_vpc));
+ tgsi_parse_init(&parse, vps->tokens);
+
+ vp = CALLOC_STRUCT(nvfx_vertex_program);
+ if(!vp)
+ goto out_err;
+
+ vpc = CALLOC_STRUCT(nvfx_vpc);
if (!vpc)
- return;
+ goto out_err;
+
vpc->nvfx = nvfx;
vpc->vp = vp;
+ vpc->pipe = *vps;
+ vpc->info = info;
+
+ {
+ // TODO: use a 64-bit atomic here!
+ static unsigned long long id = 0;
+ vp->id = ++id;
+ }
/* reserve space for ucps */
if(nvfx->use_vp_clipping)
if (!nvfx_vertprog_prepare(nvfx, vpc)) {
FREE(vpc);
- return;
+ return NULL;
}
/* Redirect post-transform vertex position to a temp if user clip
vpc->r_temps_discard = 0;
}
- tgsi_parse_init(&parse, vp->pipe.tokens);
-
util_dynarray_init(&insns);
while (!tgsi_parse_end_of_tokens(&parse)) {
tgsi_parse_token(&parse);
NVFX_VP(INST_DEST_POS));
struct nvfx_src htmp = nvfx_src(vpc->r_result[vpc->hpos_idx]);
- nvfx_vp_emit(vpc, arith(VEC, MOV, hpos, NVFX_VP_MASK_ALL, htmp, none, none));
+ nvfx_vp_emit(vpc, arith(0, VEC, MOV, hpos, NVFX_VP_MASK_ALL, htmp, none, none));
}
/* Insert code to handle user clip planes */
else
mask = NVFX_VP_MASK_X;
- nvfx_vp_emit(vpc, arith(VEC, DP4, cdst, mask, htmp, ceqn, none));
+ nvfx_vp_emit(vpc, arith(0, VEC, DP4, cdst, mask, htmp, ceqn, none));
}
}
- else
- {
- if(vp->nr_insns)
- vp->insns[vp->nr_insns - 1].data[3] |= NVFX_VP_INST_LAST;
-
- nvfx_vp_emit(vpc, arith(VEC, NOP, none.reg, 0, none, none, none));
- vp->insns[vp->nr_insns - 1].data[3] |= NVFX_VP_INST_LAST;
- }
if(debug_get_option_nvfx_dump_vp())
{
debug_printf("\n");
- tgsi_dump(vp->pipe.tokens, 0);
+ tgsi_dump(vpc->pipe.tokens, 0);
debug_printf("\n%s vertex program:\n", nvfx->is_nv4x ? "nv4x" : "nv3x");
for (i = 0; i < vp->nr_insns; i++)
vp->clip_nr = -1;
vp->exec_start = -1;
- vp->translated = TRUE;
-out_err:
+
+out:
tgsi_parse_free(&parse);
- util_dynarray_fini(&vpc->label_relocs);
- util_dynarray_fini(&vpc->loop_stack);
- if (vpc->r_temp)
+ if(vpc) {
+ util_dynarray_fini(&vpc->label_relocs);
+ util_dynarray_fini(&vpc->loop_stack);
FREE(vpc->r_temp);
- if (vpc->r_address)
FREE(vpc->r_address);
- if (vpc->r_const)
FREE(vpc->r_const);
- if (vpc->imm)
FREE(vpc->imm);
- FREE(vpc);
+ FREE(vpc);
+ }
+ return vp;
+
+out_err:
+ FREE(vp);
+ vp = NULL;
+ goto out;
+}
+
+static struct nvfx_vertex_program*
+nvfx_vertprog_translate_draw_vp(struct nvfx_context *nvfx, struct nvfx_pipe_vertex_program* pvp)
+{
+ struct nvfx_vertex_program* vp = NULL;
+ struct pipe_shader_state vps;
+ struct tgsi_shader_info info;
+ struct ureg_program *ureg = NULL;
+ unsigned num_outputs = MIN2(pvp->info.num_outputs, 16);
+
+ ureg = ureg_create( TGSI_PROCESSOR_VERTEX );
+ if(ureg == NULL)
+ return 0;
+
+ for (unsigned i = 0; i < num_outputs; i++)
+ ureg_MOV(ureg, ureg_DECL_output(ureg, pvp->info.output_semantic_name[i], pvp->info.output_semantic_index[i]), ureg_DECL_vs_input(ureg, i));
+
+ ureg_END( ureg );
+
+ vps.tokens = ureg_get_tokens(ureg, 0);
+ tgsi_scan_shader(vps.tokens, &info);
+ vp = nvfx_vertprog_translate(nvfx, &vps, &info);
+ ureg_free_tokens(vps.tokens);
+ ureg_destroy(ureg);
+
+ return vp;
}
boolean
struct nvfx_screen *screen = nvfx->screen;
struct nouveau_channel *chan = screen->base.channel;
struct nouveau_grobj *eng3d = screen->eng3d;
- struct nvfx_vertex_program *vp;
+ struct nvfx_pipe_vertex_program *pvp = nvfx->vertprog;
+ struct nvfx_vertex_program* vp;
struct pipe_resource *constbuf;
boolean upload_code = FALSE, upload_data = FALSE;
int i;
if (nvfx->render_mode == HW) {
- vp = nvfx->vertprog;
- constbuf = nvfx->constbuf[PIPE_SHADER_VERTEX];
- } else {
- vp = nvfx->swtnl.vertprog;
- constbuf = NULL;
- }
-
- /* Translate TGSI shader into hw bytecode */
- if (!vp->translated)
- {
nvfx->fallback_swtnl &= ~NVFX_NEW_VERTPROG;
- nvfx_vertprog_translate(nvfx, vp);
- if (!vp->translated) {
+ vp = pvp->vp;
+
+ if(!vp) {
+ vp = nvfx_vertprog_translate(nvfx, &pvp->pipe, &pvp->info);
+ if(!vp)
+ vp = NVFX_VP_FAILED;
+ pvp->vp = vp;
+ }
+
+ if(vp == NVFX_VP_FAILED) {
nvfx->fallback_swtnl |= NVFX_NEW_VERTPROG;
return FALSE;
}
+
+ constbuf = nvfx->constbuf[PIPE_SHADER_VERTEX];
+ } else {
+ vp = pvp->draw_vp;
+ if(!vp)
+ {
+ pvp->draw_vp = vp = nvfx_vertprog_translate_draw_vp(nvfx, pvp);
+ if(!vp) {
+ _debug_printf("Error: unable to create a swtnl passthrough vertex shader: aborting.");
+ abort();
+ }
+ }
+ constbuf = NULL;
}
+ nvfx->hw_vertprog = vp;
+
/* Allocate hw vtxprog exec slots */
if (!vp->exec) {
struct nouveau_resource *heap = nvfx->screen->vp_exec_heap;
map = (float*)nvfx_buffer(constbuf)->data;
/*
+ * WAIT_RING(chan, 512 * 6);
for (i = 0; i < 512; i++) {
float v[4] = {0.1, 0,2, 0.3, 0.4};
- BEGIN_RING(chan, eng3d, NV34TCL_VP_UPLOAD_CONST_ID, 5);
- OUT_RING (chan, i);
- OUT_RINGp (chan, (uint32_t *)v, 4);
+ OUT_RING(chan, RING_3D(NV30_3D_VP_UPLOAD_CONST_ID, 5));
+ OUT_RING(chan, i);
+ OUT_RINGp(chan, (uint32_t *)v, 4);
printf("frob %i\n", i);
}
*/
//printf("upload into %i + %i: %f %f %f %f\n", vp->data->start, i, vpd->value[0], vpd->value[1], vpd->value[2], vpd->value[3]);
- BEGIN_RING(chan, eng3d, NV34TCL_VP_UPLOAD_CONST_ID, 5);
- OUT_RING (chan, i + vp->data->start);
- OUT_RINGp (chan, (uint32_t *)vpd->value, 4);
+ BEGIN_RING(chan, eng3d, NV30_3D_VP_UPLOAD_CONST_ID, 5);
+ OUT_RING(chan, i + vp->data->start);
+ OUT_RINGp(chan, (uint32_t *)vpd->value, 4);
}
}
/* Upload vtxprog */
if (upload_code) {
- BEGIN_RING(chan, eng3d, NV34TCL_VP_UPLOAD_FROM_ID, 1);
- OUT_RING (chan, vp->exec->start);
+ BEGIN_RING(chan, eng3d, NV30_3D_VP_UPLOAD_FROM_ID, 1);
+ OUT_RING(chan, vp->exec->start);
for (i = 0; i < vp->nr_insns; i++) {
- BEGIN_RING(chan, eng3d, NV34TCL_VP_UPLOAD_INST(0), 4);
+ BEGIN_RING(chan, eng3d, NV30_3D_VP_UPLOAD_INST(0), 4);
//printf("%08x %08x %08x %08x\n", vp->insns[i].data[0], vp->insns[i].data[1], vp->insns[i].data[2], vp->insns[i].data[3]);
- OUT_RINGp (chan, vp->insns[i].data, 4);
+ OUT_RINGp(chan, vp->insns[i].data, 4);
}
vp->clip_nr = -1;
}
if(nvfx->dirty & (NVFX_NEW_VERTPROG))
{
- WAIT_RING(chan, 6);
- OUT_RING(chan, RING_3D(NV34TCL_VP_START_FROM_ID, 1));
+ BEGIN_RING(chan, eng3d, NV30_3D_VP_START_FROM_ID, 1);
OUT_RING(chan, vp->exec->start);
if(nvfx->is_nv4x) {
- OUT_RING(chan, RING_3D(NV40TCL_VP_ATTRIB_EN, 1));
+ BEGIN_RING(chan, eng3d, NV40_3D_VP_ATTRIB_EN, 1);
OUT_RING(chan, vp->ir);
}
}
util_dynarray_fini(&vp->branch_relocs);
util_dynarray_fini(&vp->const_relocs);
+ FREE(vp);
}
static void *
-nvfx_vp_state_create(struct pipe_context *pipe,
- const struct pipe_shader_state *cso)
+nvfx_vp_state_create(struct pipe_context *pipe, const struct pipe_shader_state *cso)
{
- struct nvfx_context *nvfx = nvfx_context(pipe);
- struct nvfx_vertex_program *vp;
-
- // TODO: use a 64-bit atomic here!
- static unsigned long long id = 0;
+ struct nvfx_pipe_vertex_program *pvp;
- vp = CALLOC(1, sizeof(struct nvfx_vertex_program));
- vp->pipe.tokens = tgsi_dup_tokens(cso->tokens);
- vp->draw = draw_create_vertex_shader(nvfx->draw, &vp->pipe);
- vp->id = ++id;
+ pvp = CALLOC(1, sizeof(struct nvfx_pipe_vertex_program));
+ pvp->pipe.tokens = tgsi_dup_tokens(cso->tokens);
+ tgsi_scan_shader(pvp->pipe.tokens, &pvp->info);
+ pvp->draw_elements = MAX2(1, MIN2(pvp->info.num_outputs, 16));
+ pvp->draw_no_elements = pvp->info.num_outputs == 0;
- return (void *)vp;
+ return (void *)pvp;
}
static void
static void
nvfx_vp_state_delete(struct pipe_context *pipe, void *hwcso)
{
- struct nvfx_context *nvfx = nvfx_context(pipe);
- struct nvfx_vertex_program *vp = hwcso;
-
- draw_delete_vertex_shader(nvfx->draw, vp->draw);
- nvfx_vertprog_destroy(nvfx, vp);
- FREE((void*)vp->pipe.tokens);
- FREE(vp);
+ struct nvfx_context *nvfx = nvfx_context(pipe);
+ struct nvfx_pipe_vertex_program *pvp = hwcso;
+
+ if(pvp->draw_vs)
+ draw_delete_vertex_shader(nvfx->draw, pvp->draw_vs);
+ if(pvp->vp && pvp->vp != NVFX_VP_FAILED)
+ nvfx_vertprog_destroy(nvfx, pvp->vp);
+ if(pvp->draw_vp)
+ nvfx_vertprog_destroy(nvfx, pvp->draw_vp);
+ FREE((void*)pvp->pipe.tokens);
+ FREE(pvp);
}
void
projects / mesa.git / blobdiff