From 26add9288c88108e3485ffc57c51ea9bdc0ee719 Mon Sep 17 00:00:00 2001 From: Ben Skeggs Date: Sat, 16 Feb 2008 17:23:12 +1100 Subject: nouveau: match gallium code reorginisation. That was... fun.. --- src/gallium/drivers/nv30/nv30_fragprog.c | 835 +++++++++++++++++++++++++++++++ 1 file changed, 835 insertions(+) create mode 100644 src/gallium/drivers/nv30/nv30_fragprog.c (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c new file mode 100644 index 0000000000..09ad555c32 --- /dev/null +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -0,0 +1,835 @@ +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_state.h" +#include "pipe/p_util.h" + +#include "pipe/p_shader_tokens.h" +#include "tgsi/util/tgsi_parse.h" +#include "tgsi/util/tgsi_util.h" + +#include "nv30_context.h" + +#define SWZ_X 0 +#define SWZ_Y 1 +#define SWZ_Z 2 +#define SWZ_W 3 +#define MASK_X 1 +#define MASK_Y 2 +#define MASK_Z 4 +#define MASK_W 8 +#define MASK_ALL (MASK_X|MASK_Y|MASK_Z|MASK_W) +#define DEF_SCALE NV30_FP_OP_DST_SCALE_1X +#define DEF_CTEST NV30_FP_OP_COND_TR +#include "nv30_shader.h" + +#define swz(s,x,y,z,w) nv30_sr_swz((s), SWZ_##x, SWZ_##y, SWZ_##z, SWZ_##w) +#define neg(s) nv30_sr_neg((s)) +#define abs(s) nv30_sr_abs((s)) +#define scale(s,v) nv30_sr_scale((s), NV30_FP_OP_DST_SCALE_##v) + +#define MAX_CONSTS 128 +#define MAX_IMM 32 +struct nv30_fpc { + struct nv30_fragment_program *fp; + + uint attrib_map[PIPE_MAX_SHADER_INPUTS]; + + int high_temp; + int temp_temp_count; + int num_regs; + + uint depth_id; + uint colour_id; + + unsigned inst_offset; + + struct { + int pipe; + float vals[4]; + } consts[MAX_CONSTS]; + int nr_consts; + + struct nv30_sreg imm[MAX_IMM]; + unsigned nr_imm; +}; + +static INLINE struct nv30_sreg +temp(struct nv30_fpc *fpc) +{ + int idx; + + idx = fpc->temp_temp_count++; + idx += fpc->high_temp + 1; + return nv30_sr(NV30SR_TEMP, idx); +} + +static INLINE struct nv30_sreg +constant(struct nv30_fpc *fpc, int pipe, float vals[4]) +{ + int idx; + + if (fpc->nr_consts == MAX_CONSTS) + assert(0); + idx = fpc->nr_consts++; + + fpc->consts[idx].pipe = pipe; + if (pipe == -1) + memcpy(fpc->consts[idx].vals, vals, 4 * sizeof(float)); + return nv30_sr(NV30SR_CONST, idx); +} + +#define arith(cc,s,o,d,m,s0,s1,s2) \ + nv30_fp_arith((cc), (s), NV30_FP_OP_OPCODE_##o, \ + (d), (m), (s0), (s1), (s2)) +#define tex(cc,s,o,u,d,m,s0,s1,s2) \ + nv30_fp_tex((cc), (s), NV30_FP_OP_OPCODE_##o, (u), \ + (d), (m), (s0), none, none) + +static void +grow_insns(struct nv30_fpc *fpc, int size) +{ + struct nv30_fragment_program *fp = fpc->fp; + + fp->insn_len += size; + fp->insn = realloc(fp->insn, sizeof(uint32_t) * fp->insn_len); +} + +static void +emit_src(struct nv30_fpc *fpc, int pos, struct nv30_sreg src) +{ + struct nv30_fragment_program *fp = fpc->fp; + uint32_t *hw = &fp->insn[fpc->inst_offset]; + uint32_t sr = 0; + + switch (src.type) { + case NV30SR_INPUT: + sr |= (NV30_FP_REG_TYPE_INPUT << NV30_FP_REG_TYPE_SHIFT); + hw[0] |= (src.index << NV30_FP_OP_INPUT_SRC_SHIFT); + break; + case NV30SR_OUTPUT: + sr |= NV30_FP_REG_SRC_HALF; + /* fall-through */ + case NV30SR_TEMP: + sr |= (NV30_FP_REG_TYPE_TEMP << NV30_FP_REG_TYPE_SHIFT); + sr |= (src.index << NV30_FP_REG_SRC_SHIFT); + break; + case NV30SR_CONST: + grow_insns(fpc, 4); + hw = &fp->insn[fpc->inst_offset]; + if (fpc->consts[src.index].pipe >= 0) { + struct nv30_fragment_program_data *fpd; + + fp->consts = realloc(fp->consts, ++fp->nr_consts * + sizeof(*fpd)); + fpd = &fp->consts[fp->nr_consts - 1]; + fpd->offset = fpc->inst_offset + 4; + fpd->index = fpc->consts[src.index].pipe; + memset(&fp->insn[fpd->offset], 0, sizeof(uint32_t) * 4); + } else { + memcpy(&fp->insn[fpc->inst_offset + 4], + fpc->consts[src.index].vals, + sizeof(uint32_t) * 4); + } + + sr |= (NV30_FP_REG_TYPE_CONST << NV30_FP_REG_TYPE_SHIFT); + break; + case NV30SR_NONE: + sr |= (NV30_FP_REG_TYPE_INPUT << NV30_FP_REG_TYPE_SHIFT); + break; + default: + assert(0); + } + + if (src.negate) + sr |= NV30_FP_REG_NEGATE; + + if (src.abs) + hw[1] |= (1 << (29 + pos)); + + sr |= ((src.swz[0] << NV30_FP_REG_SWZ_X_SHIFT) | + (src.swz[1] << NV30_FP_REG_SWZ_Y_SHIFT) | + (src.swz[2] << NV30_FP_REG_SWZ_Z_SHIFT) | + (src.swz[3] << NV30_FP_REG_SWZ_W_SHIFT)); + + hw[pos + 1] |= sr; +} + +static void +emit_dst(struct nv30_fpc *fpc, struct nv30_sreg dst) +{ + struct nv30_fragment_program *fp = fpc->fp; + uint32_t *hw = &fp->insn[fpc->inst_offset]; + + switch (dst.type) { + case NV30SR_TEMP: + if (fpc->num_regs < (dst.index + 1)) + fpc->num_regs = dst.index + 1; + break; + case NV30SR_OUTPUT: + if (dst.index == 1) { + fp->fp_control |= 0xe; + } else { + hw[0] |= NV30_FP_OP_OUT_REG_HALF; + } + break; + case NV30SR_NONE: + hw[0] |= (1 << 30); + break; + default: + assert(0); + } + + hw[0] |= (dst.index << NV30_FP_OP_OUT_REG_SHIFT); +} + +static void +nv30_fp_arith(struct nv30_fpc *fpc, int sat, int op, + struct nv30_sreg dst, int mask, + struct nv30_sreg s0, struct nv30_sreg s1, struct nv30_sreg s2) +{ + struct nv30_fragment_program *fp = fpc->fp; + uint32_t *hw; + + fpc->inst_offset = fp->insn_len; + grow_insns(fpc, 4); + hw = &fp->insn[fpc->inst_offset]; + memset(hw, 0, sizeof(uint32_t) * 4); + + if (op == NV30_FP_OP_OPCODE_KIL) + fp->fp_control |= NV34TCL_FP_CONTROL_USES_KIL; + hw[0] |= (op << NV30_FP_OP_OPCODE_SHIFT); + hw[0] |= (mask << NV30_FP_OP_OUTMASK_SHIFT); + hw[2] |= (dst.dst_scale << NV30_FP_OP_DST_SCALE_SHIFT); + + if (sat) + hw[0] |= NV30_FP_OP_OUT_SAT; + + if (dst.cc_update) + hw[0] |= NV30_FP_OP_COND_WRITE_ENABLE; + hw[1] |= (dst.cc_test << NV30_FP_OP_COND_SHIFT); + hw[1] |= ((dst.cc_swz[0] << NV30_FP_OP_COND_SWZ_X_SHIFT) | + (dst.cc_swz[1] << NV30_FP_OP_COND_SWZ_Y_SHIFT) | + (dst.cc_swz[2] << NV30_FP_OP_COND_SWZ_Z_SHIFT) | + (dst.cc_swz[3] << NV30_FP_OP_COND_SWZ_W_SHIFT)); + + emit_dst(fpc, dst); + emit_src(fpc, 0, s0); + emit_src(fpc, 1, s1); + emit_src(fpc, 2, s2); +} + +static void +nv30_fp_tex(struct nv30_fpc *fpc, int sat, int op, int unit, + struct nv30_sreg dst, int mask, + struct nv30_sreg s0, struct nv30_sreg s1, struct nv30_sreg s2) +{ + struct nv30_fragment_program *fp = fpc->fp; + + nv30_fp_arith(fpc, sat, op, dst, mask, s0, s1, s2); + + fp->insn[fpc->inst_offset] |= (unit << NV30_FP_OP_TEX_UNIT_SHIFT); + fp->samplers |= (1 << unit); +} + +static INLINE struct nv30_sreg +tgsi_src(struct nv30_fpc *fpc, const struct tgsi_full_src_register *fsrc) +{ + struct nv30_sreg src; + + switch (fsrc->SrcRegister.File) { + case TGSI_FILE_INPUT: + src = nv30_sr(NV30SR_INPUT, + fpc->attrib_map[fsrc->SrcRegister.Index]); + break; + case TGSI_FILE_CONSTANT: + src = constant(fpc, fsrc->SrcRegister.Index, NULL); + break; + case TGSI_FILE_IMMEDIATE: + assert(fsrc->SrcRegister.Index < fpc->nr_imm); + src = fpc->imm[fsrc->SrcRegister.Index]; + break; + case TGSI_FILE_TEMPORARY: + src = nv30_sr(NV30SR_TEMP, fsrc->SrcRegister.Index + 1); + if (fpc->high_temp < src.index) + fpc->high_temp = src.index; + break; + /* This is clearly insane, but gallium hands us shaders like this. + * Luckily fragprog results are just temp regs.. + */ + case TGSI_FILE_OUTPUT: + if (fsrc->SrcRegister.Index == fpc->colour_id) + return nv30_sr(NV30SR_OUTPUT, 0); + else + return nv30_sr(NV30SR_OUTPUT, 1); + break; + default: + NOUVEAU_ERR("bad src file\n"); + break; + } + + src.abs = fsrc->SrcRegisterExtMod.Absolute; + src.negate = fsrc->SrcRegister.Negate; + src.swz[0] = fsrc->SrcRegister.SwizzleX; + src.swz[1] = fsrc->SrcRegister.SwizzleY; + src.swz[2] = fsrc->SrcRegister.SwizzleZ; + src.swz[3] = fsrc->SrcRegister.SwizzleW; + return src; +} + +static INLINE struct nv30_sreg +tgsi_dst(struct nv30_fpc *fpc, const struct tgsi_full_dst_register *fdst) { + int idx; + + switch (fdst->DstRegister.File) { + case TGSI_FILE_OUTPUT: + if (fdst->DstRegister.Index == fpc->colour_id) + return nv30_sr(NV30SR_OUTPUT, 0); + else + return nv30_sr(NV30SR_OUTPUT, 1); + break; + case TGSI_FILE_TEMPORARY: + idx = fdst->DstRegister.Index + 1; + if (fpc->high_temp < idx) + fpc->high_temp = idx; + return nv30_sr(NV30SR_TEMP, idx); + case TGSI_FILE_NULL: + return nv30_sr(NV30SR_NONE, 0); + default: + NOUVEAU_ERR("bad dst file %d\n", fdst->DstRegister.File); + return nv30_sr(NV30SR_NONE, 0); + } +} + +static INLINE int +tgsi_mask(uint tgsi) +{ + int mask = 0; + + if (tgsi & TGSI_WRITEMASK_X) mask |= MASK_X; + if (tgsi & TGSI_WRITEMASK_Y) mask |= MASK_Y; + if (tgsi & TGSI_WRITEMASK_Z) mask |= MASK_Z; + if (tgsi & TGSI_WRITEMASK_W) mask |= MASK_W; + return mask; +} + +static boolean +src_native_swz(struct nv30_fpc *fpc, const struct tgsi_full_src_register *fsrc, + struct nv30_sreg *src) +{ + const struct nv30_sreg none = nv30_sr(NV30SR_NONE, 0); + struct nv30_sreg tgsi = tgsi_src(fpc, fsrc); + uint mask = 0, zero_mask = 0, one_mask = 0, neg_mask = 0; + uint neg[4] = { fsrc->SrcRegisterExtSwz.NegateX, + fsrc->SrcRegisterExtSwz.NegateY, + fsrc->SrcRegisterExtSwz.NegateZ, + fsrc->SrcRegisterExtSwz.NegateW }; + uint c; + + for (c = 0; c < 4; c++) { + switch (tgsi_util_get_full_src_register_extswizzle(fsrc, c)) { + case TGSI_EXTSWIZZLE_X: + case TGSI_EXTSWIZZLE_Y: + case TGSI_EXTSWIZZLE_Z: + case TGSI_EXTSWIZZLE_W: + mask |= (1 << c); + break; + case TGSI_EXTSWIZZLE_ZERO: + zero_mask |= (1 << c); + tgsi.swz[c] = SWZ_X; + break; + case TGSI_EXTSWIZZLE_ONE: + one_mask |= (1 << c); + tgsi.swz[c] = SWZ_X; + break; + default: + assert(0); + } + + if (!tgsi.negate && neg[c]) + neg_mask |= (1 << c); + } + + if (mask == MASK_ALL && !neg_mask) + return TRUE; + + *src = temp(fpc); + + if (mask) + arith(fpc, 0, MOV, *src, mask, tgsi, none, none); + + if (zero_mask) + arith(fpc, 0, SFL, *src, zero_mask, *src, none, none); + + if (one_mask) + arith(fpc, 0, STR, *src, one_mask, *src, none, none); + + if (neg_mask) { + struct nv30_sreg one = temp(fpc); + arith(fpc, 0, STR, one, neg_mask, one, none, none); + arith(fpc, 0, MUL, *src, neg_mask, *src, neg(one), none); + } + + return FALSE; +} + +static boolean +nv30_fragprog_parse_instruction(struct nv30_fpc *fpc, + const struct tgsi_full_instruction *finst) +{ + const struct nv30_sreg none = nv30_sr(NV30SR_NONE, 0); + struct nv30_sreg src[3], dst, tmp; + int mask, sat, unit; + int ai = -1, ci = -1; + int i; + + if (finst->Instruction.Opcode == TGSI_OPCODE_END) + return TRUE; + + fpc->temp_temp_count = 0; + for (i = 0; i < finst->Instruction.NumSrcRegs; i++) { + const struct tgsi_full_src_register *fsrc; + + fsrc = &finst->FullSrcRegisters[i]; + if (fsrc->SrcRegister.File == TGSI_FILE_TEMPORARY) { + src[i] = tgsi_src(fpc, fsrc); + } + } + + for (i = 0; i < finst->Instruction.NumSrcRegs; i++) { + const struct tgsi_full_src_register *fsrc; + + fsrc = &finst->FullSrcRegisters[i]; + + switch (fsrc->SrcRegister.File) { + case TGSI_FILE_INPUT: + case TGSI_FILE_CONSTANT: + case TGSI_FILE_TEMPORARY: + if (!src_native_swz(fpc, fsrc, &src[i])) + continue; + break; + default: + break; + } + + switch (fsrc->SrcRegister.File) { + case TGSI_FILE_INPUT: + if (ai == -1 || ai == fsrc->SrcRegister.Index) { + ai = fsrc->SrcRegister.Index; + src[i] = tgsi_src(fpc, fsrc); + } else { + NOUVEAU_MSG("extra src attr %d\n", + fsrc->SrcRegister.Index); + src[i] = temp(fpc); + arith(fpc, 0, MOV, src[i], MASK_ALL, + tgsi_src(fpc, fsrc), none, none); + } + break; + case TGSI_FILE_CONSTANT: + case TGSI_FILE_IMMEDIATE: + if (ci == -1 || ci == fsrc->SrcRegister.Index) { + ci = fsrc->SrcRegister.Index; + src[i] = tgsi_src(fpc, fsrc); + } else { + src[i] = temp(fpc); + arith(fpc, 0, MOV, src[i], MASK_ALL, + tgsi_src(fpc, fsrc), none, none); + } + break; + case TGSI_FILE_TEMPORARY: + /* handled above */ + break; + case TGSI_FILE_SAMPLER: + unit = fsrc->SrcRegister.Index; + break; + case TGSI_FILE_OUTPUT: + break; + default: + NOUVEAU_ERR("bad src file\n"); + return FALSE; + } + } + + dst = tgsi_dst(fpc, &finst->FullDstRegisters[0]); + mask = tgsi_mask(finst->FullDstRegisters[0].DstRegister.WriteMask); + sat = (finst->Instruction.Saturate == TGSI_SAT_ZERO_ONE); + + switch (finst->Instruction.Opcode) { + case TGSI_OPCODE_ABS: + arith(fpc, sat, MOV, dst, mask, abs(src[0]), none, none); + break; + case TGSI_OPCODE_ADD: + arith(fpc, sat, ADD, dst, mask, src[0], src[1], none); + break; + case TGSI_OPCODE_CMP: + tmp = temp(fpc); + arith(fpc, sat, MOV, dst, mask, src[2], none, none); + tmp.cc_update = 1; + arith(fpc, 0, MOV, tmp, 0xf, src[0], none, none); + dst.cc_test = NV30_VP_INST_COND_LT; + arith(fpc, sat, MOV, dst, mask, src[1], none, none); + break; + case TGSI_OPCODE_COS: + arith(fpc, sat, COS, dst, mask, src[0], none, none); + break; + case TGSI_OPCODE_DP3: + arith(fpc, sat, DP3, dst, mask, src[0], src[1], none); + break; + case TGSI_OPCODE_DP4: + arith(fpc, sat, DP4, dst, mask, src[0], src[1], none); + break; + case TGSI_OPCODE_DPH: + tmp = temp(fpc); + arith(fpc, 0, DP3, tmp, MASK_X, src[0], src[1], none); + arith(fpc, sat, ADD, dst, mask, swz(tmp, X, X, X, X), + swz(src[1], W, W, W, W), none); + break; + case TGSI_OPCODE_DST: + arith(fpc, sat, DST, dst, mask, src[0], src[1], none); + break; + case TGSI_OPCODE_EX2: + arith(fpc, sat, EX2, dst, mask, src[0], none, none); + break; + case TGSI_OPCODE_FLR: + arith(fpc, sat, FLR, dst, mask, src[0], none, none); + break; + case TGSI_OPCODE_FRC: + arith(fpc, sat, FRC, dst, mask, src[0], none, none); + break; + case TGSI_OPCODE_KIL: + arith(fpc, 0, KIL, none, 0, none, none, none); + break; + case TGSI_OPCODE_KILP: + dst = nv30_sr(NV30SR_NONE, 0); + dst.cc_update = 1; + arith(fpc, 0, MOV, dst, MASK_ALL, src[0], none, none); + dst.cc_update = 0; dst.cc_test = NV30_FP_OP_COND_LT; + arith(fpc, 0, KIL, dst, 0, none, none, none); + break; + case TGSI_OPCODE_LG2: + arith(fpc, sat, LG2, dst, mask, src[0], none, none); + break; +// case TGSI_OPCODE_LIT: + case TGSI_OPCODE_LRP: + tmp = temp(fpc); + arith(fpc, 0, MAD, tmp, mask, neg(src[0]), src[2], src[2]); + arith(fpc, sat, MAD, dst, mask, src[0], src[1], tmp); + break; + case TGSI_OPCODE_MAD: + arith(fpc, sat, MAD, dst, mask, src[0], src[1], src[2]); + break; + case TGSI_OPCODE_MAX: + arith(fpc, sat, MAX, dst, mask, src[0], src[1], none); + break; + case TGSI_OPCODE_MIN: + arith(fpc, sat, MIN, dst, mask, src[0], src[1], none); + break; + case TGSI_OPCODE_MOV: + arith(fpc, sat, MOV, dst, mask, src[0], none, none); + break; + case TGSI_OPCODE_MUL: + arith(fpc, sat, MUL, dst, mask, src[0], src[1], none); + break; + case TGSI_OPCODE_POW: + tmp = temp(fpc); + arith(fpc, 0, LG2, tmp, MASK_X, + swz(src[0], X, X, X, X), none, none); + arith(fpc, 0, MUL, tmp, MASK_X, swz(tmp, X, X, X, X), + swz(src[1], X, X, X, X), none); + arith(fpc, sat, EX2, dst, mask, + swz(tmp, X, X, X, X), none, none); + break; + case TGSI_OPCODE_RCP: + arith(fpc, sat, RCP, dst, mask, src[0], none, none); + break; + case TGSI_OPCODE_RET: + assert(0); + break; + case TGSI_OPCODE_RFL: + tmp = temp(fpc); + arith(fpc, 0, DP3, tmp, MASK_X, src[0], src[0], none); + arith(fpc, 0, DP3, tmp, MASK_Y, src[0], src[1], none); + arith(fpc, 0, DIV, scale(tmp, 2X), MASK_Z, + swz(tmp, Y, Y, Y, Y), swz(tmp, X, X, X, X), none); + arith(fpc, sat, MAD, dst, mask, + swz(tmp, Z, Z, Z, Z), src[0], neg(src[1])); + break; + case TGSI_OPCODE_RSQ: + tmp = temp(fpc); + arith(fpc, 0, LG2, scale(tmp, INV_2X), MASK_X, + abs(swz(src[0], X, X, X, X)), none, none); + arith(fpc, sat, EX2, dst, mask, + neg(swz(tmp, X, X, X, X)), none, none); + break; + case TGSI_OPCODE_SCS: + if (mask & MASK_X) { + arith(fpc, sat, COS, dst, MASK_X, + swz(src[0], X, X, X, X), none, none); + } + if (mask & MASK_Y) { + arith(fpc, sat, SIN, dst, MASK_Y, + swz(src[0], X, X, X, X), none, none); + } + break; + case TGSI_OPCODE_SIN: + arith(fpc, sat, SIN, dst, mask, src[0], none, none); + break; + case TGSI_OPCODE_SGE: + arith(fpc, sat, SGE, dst, mask, src[0], src[1], none); + break; + case TGSI_OPCODE_SLT: + arith(fpc, sat, SLT, dst, mask, src[0], src[1], none); + break; + case TGSI_OPCODE_SUB: + arith(fpc, sat, ADD, dst, mask, src[0], neg(src[1]), none); + break; + case TGSI_OPCODE_TEX: + if (finst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide == + TGSI_EXTSWIZZLE_W) { + tex(fpc, sat, TXP, unit, dst, mask, src[0], none, none); + } else + tex(fpc, sat, TEX, unit, dst, mask, src[0], none, none); + break; + case TGSI_OPCODE_TXB: + tex(fpc, sat, TXB, unit, dst, mask, src[0], none, none); + break; + case TGSI_OPCODE_XPD: + tmp = temp(fpc); + arith(fpc, 0, MUL, tmp, mask, + swz(src[0], Z, X, Y, Y), swz(src[1], Y, Z, X, X), none); + arith(fpc, sat, MAD, dst, (mask & ~MASK_W), + swz(src[0], Y, Z, X, X), swz(src[1], Z, X, Y, Y), + neg(tmp)); + break; + default: + NOUVEAU_ERR("invalid opcode %d\n", finst->Instruction.Opcode); + return FALSE; + } + + return TRUE; +} + +static boolean +nv30_fragprog_parse_decl_attrib(struct nv30_fpc *fpc, + const struct tgsi_full_declaration *fdec) +{ + int hw; + + switch (fdec->Semantic.SemanticName) { + case TGSI_SEMANTIC_POSITION: + hw = NV30_FP_OP_INPUT_SRC_POSITION; + break; + case TGSI_SEMANTIC_COLOR: + if (fdec->Semantic.SemanticIndex == 0) { + hw = NV30_FP_OP_INPUT_SRC_COL0; + } else + if (fdec->Semantic.SemanticIndex == 1) { + hw = NV30_FP_OP_INPUT_SRC_COL1; + } else { + NOUVEAU_ERR("bad colour semantic index\n"); + return FALSE; + } + break; + case TGSI_SEMANTIC_FOG: + hw = NV30_FP_OP_INPUT_SRC_FOGC; + break; + case TGSI_SEMANTIC_GENERIC: + if (fdec->Semantic.SemanticIndex <= 7) { + hw = NV30_FP_OP_INPUT_SRC_TC(fdec->Semantic. + SemanticIndex); + } else { + NOUVEAU_ERR("bad generic semantic index\n"); + return FALSE; + } + break; + default: + NOUVEAU_ERR("bad input semantic\n"); + return FALSE; + } + + fpc->attrib_map[fdec->u.DeclarationRange.First] = hw; + return TRUE; +} + +static boolean +nv30_fragprog_parse_decl_output(struct nv30_fpc *fpc, + const struct tgsi_full_declaration *fdec) +{ + switch (fdec->Semantic.SemanticName) { + case TGSI_SEMANTIC_POSITION: + fpc->depth_id = fdec->u.DeclarationRange.First; + break; + case TGSI_SEMANTIC_COLOR: + fpc->colour_id = fdec->u.DeclarationRange.First; + break; + default: + NOUVEAU_ERR("bad output semantic\n"); + return FALSE; + } + + return TRUE; +} + +void +nv30_fragprog_translate(struct nv30_context *nv30, + struct nv30_fragment_program *fp) +{ + struct tgsi_parse_context parse; + struct nv30_fpc *fpc = NULL; + + fpc = CALLOC(1, sizeof(struct nv30_fpc)); + if (!fpc) + return; + fpc->fp = fp; + fpc->high_temp = -1; + fpc->num_regs = 2; + + tgsi_parse_init(&parse, fp->pipe->tokens); + + while (!tgsi_parse_end_of_tokens(&parse)) { + tgsi_parse_token(&parse); + + switch (parse.FullToken.Token.Type) { + case TGSI_TOKEN_TYPE_DECLARATION: + { + const struct tgsi_full_declaration *fdec; + fdec = &parse.FullToken.FullDeclaration; + switch (fdec->Declaration.File) { + case TGSI_FILE_INPUT: + if (!nv30_fragprog_parse_decl_attrib(fpc, fdec)) + goto out_err; + break; + case TGSI_FILE_OUTPUT: + if (!nv30_fragprog_parse_decl_output(fpc, fdec)) + goto out_err; + break; + default: + break; + } + } + break; + case TGSI_TOKEN_TYPE_IMMEDIATE: + { + struct tgsi_full_immediate *imm; + float vals[4]; + int i; + + imm = &parse.FullToken.FullImmediate; + assert(imm->Immediate.DataType == TGSI_IMM_FLOAT32); + assert(fpc->nr_imm < MAX_IMM); + + for (i = 0; i < 4; i++) + vals[i] = imm->u.ImmediateFloat32[i].Float; + fpc->imm[fpc->nr_imm++] = constant(fpc, -1, vals); + } + break; + case TGSI_TOKEN_TYPE_INSTRUCTION: + { + const struct tgsi_full_instruction *finst; + + finst = &parse.FullToken.FullInstruction; + if (!nv30_fragprog_parse_instruction(fpc, finst)) + goto out_err; + } + break; + default: + break; + } + } + + fp->fp_control |= (fpc->num_regs-1)/2; + fp->fp_reg_control = (1<<16)|0x4; + + /* Terminate final instruction */ + fp->insn[fpc->inst_offset] |= 0x00000001; + + /* Append NOP + END instruction, may or may not be necessary. */ + fpc->inst_offset = fp->insn_len; + grow_insns(fpc, 4); + fp->insn[fpc->inst_offset + 0] = 0x00000001; + fp->insn[fpc->inst_offset + 1] = 0x00000000; + fp->insn[fpc->inst_offset + 2] = 0x00000000; + fp->insn[fpc->inst_offset + 3] = 0x00000000; + + fp->translated = TRUE; + fp->on_hw = FALSE; +out_err: + tgsi_parse_free(&parse); + free(fpc); +} + +void +nv30_fragprog_bind(struct nv30_context *nv30, struct nv30_fragment_program *fp) +{ + struct pipe_winsys *ws = nv30->pipe.winsys; + int i; + + if (!fp->translated) { + nv30_fragprog_translate(nv30, fp); + if (!fp->translated) + assert(0); + } + + if (fp->nr_consts) { + float *map = ws->buffer_map(ws, nv30->fragprog.constant_buf, + PIPE_BUFFER_USAGE_CPU_READ); + for (i = 0; i < fp->nr_consts; i++) { + struct nv30_fragment_program_data *fpd = &fp->consts[i]; + uint32_t *p = &fp->insn[fpd->offset]; + uint32_t *cb = (uint32_t *)&map[fpd->index * 4]; + + if (!memcmp(p, cb, 4 * sizeof(float))) + continue; + memcpy(p, cb, 4 * sizeof(float)); + fp->on_hw = 0; + } + ws->buffer_unmap(ws, nv30->fragprog.constant_buf); + } + + if (!fp->on_hw) { + const uint32_t le = 1; + uint32_t *map; + + if (!fp->buffer) + fp->buffer = ws->buffer_create(ws, 0x100, 0, + fp->insn_len * 4); + map = ws->buffer_map(ws, fp->buffer, + PIPE_BUFFER_USAGE_CPU_WRITE); + +#if 0 + for (i = 0; i < fp->insn_len; i++) { + NOUVEAU_ERR("%d 0x%08x\n", i, fp->insn[i]); + } +#endif + + if ((*(const uint8_t *)&le)) { + for (i = 0; i < fp->insn_len; i++) { + map[i] = fp->insn[i]; + } + } else { + /* Weird swapping for big-endian chips */ + for (i = 0; i < fp->insn_len; i++) { + map[i] = ((fp->insn[i] & 0xffff) << 16) | + ((fp->insn[i] >> 16) & 0xffff); + } + } + + ws->buffer_unmap(ws, fp->buffer); + fp->on_hw = TRUE; + } + + BEGIN_RING(rankine, NV34TCL_FP_CONTROL, 1); + OUT_RING (fp->fp_control); + BEGIN_RING(rankine, NV34TCL_FP_REG_CONTROL, 1); + OUT_RING (fp->fp_reg_control); + + nv30->fragprog.active = fp; +} + +void +nv30_fragprog_destroy(struct nv30_context *nv30, + struct nv30_fragment_program *fp) +{ + if (fp->insn_len) + free(fp->insn); +} + -- cgit v1.2.3 From 169912b71a4242389301890ef303046d49ce71df Mon Sep 17 00:00:00 2001 From: Patrice Mandin Date: Tue, 11 Mar 2008 19:22:02 +0100 Subject: nv30: silence some warnings --- src/gallium/drivers/nv30/nv30_fragprog.c | 2 +- src/gallium/drivers/nv30/nv30_state.c | 2 +- src/gallium/drivers/nv30/nv30_vbo.c | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 09ad555c32..0a2ba04f95 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -378,7 +378,7 @@ nv30_fragprog_parse_instruction(struct nv30_fpc *fpc, { const struct nv30_sreg none = nv30_sr(NV30SR_NONE, 0); struct nv30_sreg src[3], dst, tmp; - int mask, sat, unit; + int mask, sat, unit = 0; int ai = -1, ci = -1; int i; diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c index aa3fe7837e..319d53fcca 100644 --- a/src/gallium/drivers/nv30/nv30_state.c +++ b/src/gallium/drivers/nv30/nv30_state.c @@ -538,7 +538,7 @@ nv30_set_framebuffer_state(struct pipe_context *pipe, { struct nv30_context *nv30 = nv30_context(pipe); struct pipe_surface *rt[4], *zeta = NULL; - uint32_t rt_enable, rt_format, w, h; + uint32_t rt_enable, rt_format, w = 0, h = 0; int i, colour_format = 0, zeta_format = 0; rt_enable = 0; diff --git a/src/gallium/drivers/nv30/nv30_vbo.c b/src/gallium/drivers/nv30/nv30_vbo.c index 9e00cdac3f..a62462f7bc 100644 --- a/src/gallium/drivers/nv30/nv30_vbo.c +++ b/src/gallium/drivers/nv30/nv30_vbo.c @@ -102,7 +102,7 @@ nv30_vbo_arrays_update(struct nv30_context *nv30) { struct nv30_vertex_program *vp = nv30->vertprog.active; uint32_t inputs, vtxfmt[16]; - int hw, num_hw; + int hw, num_hw = 0; nv30->vb_enable = 0; -- cgit v1.2.3 From 9a4938d7033101122b627786273ff37229b5558a Mon Sep 17 00:00:00 2001 From: Ben Skeggs Date: Thu, 13 Mar 2008 12:36:35 +1100 Subject: nouveau: match interface changes --- src/gallium/drivers/nv30/nv30_fragprog.c | 11 ++++---- src/gallium/drivers/nv30/nv30_state.c | 30 +++++++++++--------- src/gallium/drivers/nv30/nv30_state.h | 4 +-- src/gallium/drivers/nv30/nv30_vertprog.c | 4 +-- src/gallium/drivers/nv40/nv40_context.h | 2 ++ src/gallium/drivers/nv40/nv40_fragprog.c | 11 ++++---- src/gallium/drivers/nv40/nv40_state.c | 47 +++++++++++++++++++++++--------- src/gallium/drivers/nv40/nv40_state.h | 4 +-- src/gallium/drivers/nv40/nv40_vertprog.c | 4 +-- src/gallium/drivers/nv50/nv50_state.c | 11 ++++---- 10 files changed, 76 insertions(+), 52 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 0a2ba04f95..6f61d36f4e 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -583,15 +583,14 @@ nv30_fragprog_parse_instruction(struct nv30_fpc *fpc, arith(fpc, sat, ADD, dst, mask, src[0], neg(src[1]), none); break; case TGSI_OPCODE_TEX: - if (finst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide == - TGSI_EXTSWIZZLE_W) { - tex(fpc, sat, TXP, unit, dst, mask, src[0], none, none); - } else - tex(fpc, sat, TEX, unit, dst, mask, src[0], none, none); + tex(fpc, sat, TEX, unit, dst, mask, src[0], none, none); break; case TGSI_OPCODE_TXB: tex(fpc, sat, TXB, unit, dst, mask, src[0], none, none); break; + case TGSI_OPCODE_TXP: + tex(fpc, sat, TXP, unit, dst, mask, src[0], none, none); + break; case TGSI_OPCODE_XPD: tmp = temp(fpc); arith(fpc, 0, MUL, tmp, mask, @@ -683,7 +682,7 @@ nv30_fragprog_translate(struct nv30_context *nv30, fpc->high_temp = -1; fpc->num_regs = 2; - tgsi_parse_init(&parse, fp->pipe->tokens); + tgsi_parse_init(&parse, fp->pipe.tokens); while (!tgsi_parse_end_of_tokens(&parse)) { tgsi_parse_token(&parse); diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c index 319d53fcca..951a32bc81 100644 --- a/src/gallium/drivers/nv30/nv30_state.c +++ b/src/gallium/drivers/nv30/nv30_state.c @@ -231,14 +231,15 @@ nv30_sampler_state_create(struct pipe_context *pipe, } static void -nv30_sampler_state_bind(struct pipe_context *pipe, unsigned unit, - void *hwcso) +nv30_sampler_state_bind(struct pipe_context *pipe, unsigned nr, void **sampler) { struct nv30_context *nv30 = nv30_context(pipe); - struct nv30_sampler_state *ps = hwcso; + unsigned unit; - nv30->tex_sampler[unit] = ps; - nv30->dirty_samplers |= (1 << unit); + for (unit = 0; unit < nr; unit++) { + nv30->tex_sampler[unit] = sampler[unit]; + nv30->dirty_samplers |= (1 << unit); + } } static void @@ -248,13 +249,16 @@ nv30_sampler_state_delete(struct pipe_context *pipe, void *hwcso) } static void -nv30_set_sampler_texture(struct pipe_context *pipe, unsigned unit, - struct pipe_texture *miptree) +nv30_set_sampler_texture(struct pipe_context *pipe, unsigned nr, + struct pipe_texture **miptree) { struct nv30_context *nv30 = nv30_context(pipe); + unsigned unit; - nv30->tex_miptree[unit] = (struct nv30_miptree *)miptree; - nv30->dirty_samplers |= (1 << unit); + for (unit = 0; unit < nr; unit++) { + nv30->tex_miptree[unit] = (struct nv30_miptree *)miptree[unit]; + nv30->dirty_samplers |= (1 << unit); + } } static void * @@ -440,7 +444,7 @@ nv30_vp_state_create(struct pipe_context *pipe, struct nv30_vertex_program *vp; vp = CALLOC(1, sizeof(struct nv30_vertex_program)); - vp->pipe = cso; + vp->pipe = *cso; return (void *)vp; } @@ -472,7 +476,7 @@ nv30_fp_state_create(struct pipe_context *pipe, struct nv30_fragment_program *fp; fp = CALLOC(1, sizeof(struct nv30_fragment_program)); - fp->pipe = cso; + fp->pipe = *cso; return (void *)fp; } @@ -709,9 +713,9 @@ nv30_init_state_functions(struct nv30_context *nv30) nv30->pipe.delete_blend_state = nv30_blend_state_delete; nv30->pipe.create_sampler_state = nv30_sampler_state_create; - nv30->pipe.bind_sampler_state = nv30_sampler_state_bind; + nv30->pipe.bind_sampler_states = nv30_sampler_state_bind; nv30->pipe.delete_sampler_state = nv30_sampler_state_delete; - nv30->pipe.set_sampler_texture = nv30_set_sampler_texture; + nv30->pipe.set_sampler_textures = nv30_set_sampler_texture; nv30->pipe.create_rasterizer_state = nv30_rasterizer_state_create; nv30->pipe.bind_rasterizer_state = nv30_rasterizer_state_bind; diff --git a/src/gallium/drivers/nv30/nv30_state.h b/src/gallium/drivers/nv30/nv30_state.h index 233600f69a..117520dd13 100644 --- a/src/gallium/drivers/nv30/nv30_state.h +++ b/src/gallium/drivers/nv30/nv30_state.h @@ -60,7 +60,7 @@ struct nv30_vertex_program_data { }; struct nv30_vertex_program { - const struct pipe_shader_state *pipe; + struct pipe_shader_state pipe; boolean translated; struct nv30_vertex_program_exec *insns; @@ -84,7 +84,7 @@ struct nv30_fragment_program_data { }; struct nv30_fragment_program { - const struct pipe_shader_state *pipe; + struct pipe_shader_state pipe; boolean translated; boolean on_hw; diff --git a/src/gallium/drivers/nv30/nv30_vertprog.c b/src/gallium/drivers/nv30/nv30_vertprog.c index 75f7351261..e9b62ff48b 100644 --- a/src/gallium/drivers/nv30/nv30_vertprog.c +++ b/src/gallium/drivers/nv30/nv30_vertprog.c @@ -541,7 +541,7 @@ nv30_vertprog_prepare(struct nv30_vpc *vpc) struct tgsi_parse_context p; int nr_imm = 0; - tgsi_parse_init(&p, vpc->vp->pipe->tokens); + tgsi_parse_init(&p, vpc->vp->pipe.tokens); while (!tgsi_parse_end_of_tokens(&p)) { const union tgsi_full_token *tok = &p.FullToken; @@ -582,7 +582,7 @@ nv30_vertprog_translate(struct nv30_context *nv30, return; } - tgsi_parse_init(&parse, vp->pipe->tokens); + tgsi_parse_init(&parse, vp->pipe.tokens); while (!tgsi_parse_end_of_tokens(&parse)) { tgsi_parse_token(&parse); diff --git a/src/gallium/drivers/nv40/nv40_context.h b/src/gallium/drivers/nv40/nv40_context.h index e118776306..100c678187 100644 --- a/src/gallium/drivers/nv40/nv40_context.h +++ b/src/gallium/drivers/nv40/nv40_context.h @@ -136,6 +136,8 @@ struct nv40_context { unsigned idxbuf_format; struct nv40_sampler_state *tex_sampler[PIPE_MAX_SAMPLERS]; struct nv40_miptree *tex_miptree[PIPE_MAX_SAMPLERS]; + unsigned nr_samplers; + unsigned nr_textures; unsigned dirty_samplers; struct pipe_vertex_buffer vtxbuf[PIPE_ATTRIB_MAX]; struct pipe_vertex_element vtxelt[PIPE_ATTRIB_MAX]; diff --git a/src/gallium/drivers/nv40/nv40_fragprog.c b/src/gallium/drivers/nv40/nv40_fragprog.c index d981a02a63..953f9cd908 100644 --- a/src/gallium/drivers/nv40/nv40_fragprog.c +++ b/src/gallium/drivers/nv40/nv40_fragprog.c @@ -580,15 +580,14 @@ nv40_fragprog_parse_instruction(struct nv40_fpc *fpc, arith(fpc, sat, ADD, dst, mask, src[0], neg(src[1]), none); break; case TGSI_OPCODE_TEX: - if (finst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide == - TGSI_EXTSWIZZLE_W) { - tex(fpc, sat, TXP, unit, dst, mask, src[0], none, none); - } else - tex(fpc, sat, TEX, unit, dst, mask, src[0], none, none); + tex(fpc, sat, TEX, unit, dst, mask, src[0], none, none); break; case TGSI_OPCODE_TXB: tex(fpc, sat, TXB, unit, dst, mask, src[0], none, none); break; + case TGSI_OPCODE_TXP: + tex(fpc, sat, TXP, unit, dst, mask, src[0], none, none); + break; case TGSI_OPCODE_XPD: tmp = temp(fpc); arith(fpc, 0, MUL, tmp, mask, @@ -680,7 +679,7 @@ nv40_fragprog_translate(struct nv40_context *nv40, fpc->high_temp = -1; fpc->num_regs = 2; - tgsi_parse_init(&parse, fp->pipe->tokens); + tgsi_parse_init(&parse, fp->pipe.tokens); while (!tgsi_parse_end_of_tokens(&parse)) { tgsi_parse_token(&parse); diff --git a/src/gallium/drivers/nv40/nv40_state.c b/src/gallium/drivers/nv40/nv40_state.c index caa2f9df0c..321d5de041 100644 --- a/src/gallium/drivers/nv40/nv40_state.c +++ b/src/gallium/drivers/nv40/nv40_state.c @@ -1,6 +1,7 @@ #include "pipe/p_state.h" #include "pipe/p_defines.h" #include "pipe/p_util.h" +#include "pipe/p_inlines.h" #include "nv40_context.h" #include "nv40_state.h" @@ -251,14 +252,22 @@ nv40_sampler_state_create(struct pipe_context *pipe, } static void -nv40_sampler_state_bind(struct pipe_context *pipe, unsigned unit, - void *hwcso) +nv40_sampler_state_bind(struct pipe_context *pipe, unsigned nr, void **sampler) { struct nv40_context *nv40 = nv40_context(pipe); - struct nv40_sampler_state *ps = hwcso; + unsigned unit; - nv40->tex_sampler[unit] = ps; - nv40->dirty_samplers |= (1 << unit); + for (unit = 0; unit < nr; unit++) { + nv40->tex_sampler[unit] = sampler[unit]; + nv40->dirty_samplers |= (1 << unit); + } + + for (unit = nr; unit < nv40->nr_samplers; unit++) { + nv40->tex_sampler[unit] = NULL; + nv40->dirty_samplers |= (1 << unit); + } + + nv40->nr_samplers = nr; nv40->dirty |= NV40_NEW_SAMPLER; } @@ -269,13 +278,25 @@ nv40_sampler_state_delete(struct pipe_context *pipe, void *hwcso) } static void -nv40_set_sampler_texture(struct pipe_context *pipe, unsigned unit, - struct pipe_texture *miptree) +nv40_set_sampler_texture(struct pipe_context *pipe, unsigned nr, + struct pipe_texture **miptree) { struct nv40_context *nv40 = nv40_context(pipe); + unsigned unit; + + for (unit = 0; unit < nr; unit++) { + pipe_texture_reference((struct pipe_texture **) + &nv40->tex_miptree[unit], miptree[unit]); + nv40->dirty_samplers |= (1 << unit); + } + + for (unit = nr; unit < nv40->nr_textures; unit++) { + pipe_texture_reference((struct pipe_texture **) + &nv40->tex_miptree[unit], NULL); + nv40->dirty_samplers |= (1 << unit); + } - nv40->tex_miptree[unit] = (struct nv40_miptree *)miptree; - nv40->dirty_samplers |= (1 << unit); + nv40->nr_textures = nr; nv40->dirty |= NV40_NEW_SAMPLER; } @@ -490,7 +511,7 @@ nv40_vp_state_create(struct pipe_context *pipe, struct nv40_vertex_program *vp; vp = CALLOC(1, sizeof(struct nv40_vertex_program)); - vp->pipe = cso; + vp->pipe = *cso; return (void *)vp; } @@ -521,7 +542,7 @@ nv40_fp_state_create(struct pipe_context *pipe, struct nv40_fragment_program *fp; fp = CALLOC(1, sizeof(struct nv40_fragment_program)); - fp->pipe = cso; + fp->pipe = *cso; return (void *)fp; } @@ -649,9 +670,9 @@ nv40_init_state_functions(struct nv40_context *nv40) nv40->pipe.delete_blend_state = nv40_blend_state_delete; nv40->pipe.create_sampler_state = nv40_sampler_state_create; - nv40->pipe.bind_sampler_state = nv40_sampler_state_bind; + nv40->pipe.bind_sampler_states = nv40_sampler_state_bind; nv40->pipe.delete_sampler_state = nv40_sampler_state_delete; - nv40->pipe.set_sampler_texture = nv40_set_sampler_texture; + nv40->pipe.set_sampler_textures = nv40_set_sampler_texture; nv40->pipe.create_rasterizer_state = nv40_rasterizer_state_create; nv40->pipe.bind_rasterizer_state = nv40_rasterizer_state_bind; diff --git a/src/gallium/drivers/nv40/nv40_state.h b/src/gallium/drivers/nv40/nv40_state.h index e5217fe91c..a02ea0c878 100644 --- a/src/gallium/drivers/nv40/nv40_state.h +++ b/src/gallium/drivers/nv40/nv40_state.h @@ -23,7 +23,7 @@ struct nv40_vertex_program_data { }; struct nv40_vertex_program { - const struct pipe_shader_state *pipe; + struct pipe_shader_state pipe; boolean translated; struct nv40_vertex_program_exec *insns; @@ -48,7 +48,7 @@ struct nv40_fragment_program_data { }; struct nv40_fragment_program { - const struct pipe_shader_state *pipe; + struct pipe_shader_state pipe; boolean translated; unsigned samplers; diff --git a/src/gallium/drivers/nv40/nv40_vertprog.c b/src/gallium/drivers/nv40/nv40_vertprog.c index 5b7a343e55..3d730c1a32 100644 --- a/src/gallium/drivers/nv40/nv40_vertprog.c +++ b/src/gallium/drivers/nv40/nv40_vertprog.c @@ -535,7 +535,7 @@ nv40_vertprog_prepare(struct nv40_vpc *vpc) struct tgsi_parse_context p; int nr_imm = 0; - tgsi_parse_init(&p, vpc->vp->pipe->tokens); + tgsi_parse_init(&p, vpc->vp->pipe.tokens); while (!tgsi_parse_end_of_tokens(&p)) { const union tgsi_full_token *tok = &p.FullToken; @@ -576,7 +576,7 @@ nv40_vertprog_translate(struct nv40_context *nv40, return; } - tgsi_parse_init(&parse, vp->pipe->tokens); + tgsi_parse_init(&parse, vp->pipe.tokens); while (!tgsi_parse_end_of_tokens(&parse)) { tgsi_parse_token(&parse); diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c index aa65fd482e..b096a2583d 100644 --- a/src/gallium/drivers/nv50/nv50_state.c +++ b/src/gallium/drivers/nv50/nv50_state.c @@ -90,8 +90,7 @@ nv50_sampler_state_create(struct pipe_context *pipe, } static void -nv50_sampler_state_bind(struct pipe_context *pipe, unsigned unit, - void *hwcso) +nv50_sampler_state_bind(struct pipe_context *pipe, unsigned nr, void **sampler) { } @@ -101,8 +100,8 @@ nv50_sampler_state_delete(struct pipe_context *pipe, void *hwcso) } static void -nv50_set_sampler_texture(struct pipe_context *pipe, unsigned unit, - struct pipe_texture *pt) +nv50_set_sampler_texture(struct pipe_context *pipe, unsigned nr, + struct pipe_texture **pt) { } @@ -442,9 +441,9 @@ nv50_init_state_functions(struct nv50_context *nv50) nv50->pipe.delete_blend_state = nv50_blend_state_delete; nv50->pipe.create_sampler_state = nv50_sampler_state_create; - nv50->pipe.bind_sampler_state = nv50_sampler_state_bind; + nv50->pipe.bind_sampler_states = nv50_sampler_state_bind; nv50->pipe.delete_sampler_state = nv50_sampler_state_delete; - nv50->pipe.set_sampler_texture = nv50_set_sampler_texture; + nv50->pipe.set_sampler_textures = nv50_set_sampler_texture; nv50->pipe.create_rasterizer_state = nv50_rasterizer_state_create; nv50->pipe.bind_rasterizer_state = nv50_rasterizer_state_bind; -- cgit v1.2.3 From 7b389f8d2f307fa0714494f2a43e9141cc04ed3e Mon Sep 17 00:00:00 2001 From: Patrice Mandin Date: Sun, 30 Mar 2008 21:52:36 +0200 Subject: nv30: use FREE macro --- src/gallium/drivers/nv30/nv30_context.c | 2 +- src/gallium/drivers/nv30/nv30_draw.c | 2 +- src/gallium/drivers/nv30/nv30_fragprog.c | 4 ++-- src/gallium/drivers/nv30/nv30_miptree.c | 6 +++--- src/gallium/drivers/nv30/nv30_query.c | 2 +- src/gallium/drivers/nv30/nv30_state.c | 4 ++-- src/gallium/drivers/nv30/nv30_vertprog.c | 8 ++++---- 7 files changed, 14 insertions(+), 14 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_context.c b/src/gallium/drivers/nv30/nv30_context.c index 1e729d789b..b8e8b86d99 100644 --- a/src/gallium/drivers/nv30/nv30_context.c +++ b/src/gallium/drivers/nv30/nv30_context.c @@ -41,7 +41,7 @@ nv30_destroy(struct pipe_context *pipe) nvws->grobj_free(&nv30->rankine); - free(nv30); + FREE(nv30); } static boolean diff --git a/src/gallium/drivers/nv30/nv30_draw.c b/src/gallium/drivers/nv30/nv30_draw.c index 59a7265799..8ec0835225 100644 --- a/src/gallium/drivers/nv30/nv30_draw.c +++ b/src/gallium/drivers/nv30/nv30_draw.c @@ -40,7 +40,7 @@ nv30_draw_reset_stipple_counter(struct draw_stage *draw) static void nv30_draw_destroy(struct draw_stage *draw) { - free(draw); + FREE(draw); } struct draw_stage * diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 6f61d36f4e..51000bd6fc 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -753,7 +753,7 @@ nv30_fragprog_translate(struct nv30_context *nv30, fp->on_hw = FALSE; out_err: tgsi_parse_free(&parse); - free(fpc); + FREE(fpc); } void @@ -829,6 +829,6 @@ nv30_fragprog_destroy(struct nv30_context *nv30, struct nv30_fragment_program *fp) { if (fp->insn_len) - free(fp->insn); + FREE(fp->insn); } diff --git a/src/gallium/drivers/nv30/nv30_miptree.c b/src/gallium/drivers/nv30/nv30_miptree.c index 6fdcf42a24..afb05fdd2b 100644 --- a/src/gallium/drivers/nv30/nv30_miptree.c +++ b/src/gallium/drivers/nv30/nv30_miptree.c @@ -72,7 +72,7 @@ nv30_miptree_create(struct pipe_screen *screen, const struct pipe_texture *pt) mt->buffer = ws->buffer_create(ws, 256, PIPE_BUFFER_USAGE_PIXEL, mt->total_size); if (!mt->buffer) { - free(mt); + FREE(mt); return NULL; } @@ -93,9 +93,9 @@ nv30_miptree_release(struct pipe_screen *screen, struct pipe_texture **pt) pipe_buffer_reference(ws, &nv30mt->buffer, NULL); for (l = 0; l <= mt->last_level; l++) { if (nv30mt->level[l].image_offset) - free(nv30mt->level[l].image_offset); + FREE(nv30mt->level[l].image_offset); } - free(nv30mt); + FREE(nv30mt); } } diff --git a/src/gallium/drivers/nv30/nv30_query.c b/src/gallium/drivers/nv30/nv30_query.c index e19cb455dc..0c2d941562 100644 --- a/src/gallium/drivers/nv30/nv30_query.c +++ b/src/gallium/drivers/nv30/nv30_query.c @@ -35,7 +35,7 @@ nv30_query_destroy(struct pipe_context *pipe, struct pipe_query *pq) if (q->object) nv30->nvws->res_free(&q->object); - free(q); + FREE(q); } static void diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c index 983638adcc..620038fa64 100644 --- a/src/gallium/drivers/nv30/nv30_state.c +++ b/src/gallium/drivers/nv30/nv30_state.c @@ -486,7 +486,7 @@ nv30_vp_state_delete(struct pipe_context *pipe, void *hwcso) struct nv30_vertex_program *vp = hwcso; nv30_vertprog_destroy(nv30, vp); - free(vp); + FREE(vp); } static void * @@ -522,7 +522,7 @@ nv30_fp_state_delete(struct pipe_context *pipe, void *hwcso) struct nv30_fragment_program *fp = hwcso; nv30_fragprog_destroy(nv30, fp); - free(fp); + FREE(fp); } static void diff --git a/src/gallium/drivers/nv30/nv30_vertprog.c b/src/gallium/drivers/nv30/nv30_vertprog.c index e9b62ff48b..fe1a467565 100644 --- a/src/gallium/drivers/nv30/nv30_vertprog.c +++ b/src/gallium/drivers/nv30/nv30_vertprog.c @@ -578,7 +578,7 @@ nv30_vertprog_translate(struct nv30_context *nv30, vpc->high_temp = -1; if (!nv30_vertprog_prepare(vpc)) { - free(vpc); + FREE(vpc); return; } @@ -634,7 +634,7 @@ nv30_vertprog_translate(struct nv30_context *nv30, vp->translated = TRUE; out_err: tgsi_parse_free(&parse); - free(vpc); + FREE(vpc); } void @@ -790,8 +790,8 @@ void nv30_vertprog_destroy(struct nv30_context *nv30, struct nv30_vertex_program *vp) { if (vp->nr_consts) - free(vp->consts); + FREE(vp->consts); if (vp->nr_insns) - free(vp->insns); + FREE(vp->insns); } -- cgit v1.2.3 From 5a01060eb95cb2cb168cb7224ecc805020584c91 Mon Sep 17 00:00:00 2001 From: Ben Skeggs Date: Mon, 23 Jun 2008 00:14:21 +1000 Subject: nouveau: update for interface changes + hack around gallium x86_64 bustage --- src/gallium/auxiliary/draw/draw_context.c | 2 ++ src/gallium/drivers/nv10/nv10_context.h | 1 + src/gallium/drivers/nv10/nv10_state.c | 2 ++ src/gallium/drivers/nv10/nv10_vbo.c | 4 +++- src/gallium/drivers/nv30/nv30_fragprog.c | 6 +++--- src/gallium/drivers/nv30/nv30_vertprog.c | 2 +- src/gallium/drivers/nv40/nv40_context.h | 1 + src/gallium/drivers/nv40/nv40_draw.c | 4 +++- src/gallium/drivers/nv40/nv40_fragprog.c | 8 ++++---- src/gallium/drivers/nv40/nv40_state.c | 5 +++-- src/gallium/drivers/nv40/nv40_vertprog.c | 10 +++++----- 11 files changed, 28 insertions(+), 17 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c index 2f263cf06a..0e6f55a928 100644 --- a/src/gallium/auxiliary/draw/draw_context.c +++ b/src/gallium/auxiliary/draw/draw_context.c @@ -63,8 +63,10 @@ struct draw_context *draw_create( void ) if (!draw_pt_init( draw )) goto fail; +#ifndef PIPE_ARCH_X86 if (!draw_vs_init( draw )) goto fail; +#endif return draw; diff --git a/src/gallium/drivers/nv10/nv10_context.h b/src/gallium/drivers/nv10/nv10_context.h index 5636dfc9d2..2bdba53db8 100644 --- a/src/gallium/drivers/nv10/nv10_context.h +++ b/src/gallium/drivers/nv10/nv10_context.h @@ -66,6 +66,7 @@ struct nv10_context { //struct pipe_buffer *constbuf[PIPE_SHADER_TYPES]; float *constbuf[PIPE_SHADER_TYPES][32][4]; + unsigned constbuf_nr[PIPE_SHADER_TYPES]; struct vertex_info vertex_info; diff --git a/src/gallium/drivers/nv10/nv10_state.c b/src/gallium/drivers/nv10/nv10_state.c index 4dcb9a31ab..11664fae2a 100644 --- a/src/gallium/drivers/nv10/nv10_state.c +++ b/src/gallium/drivers/nv10/nv10_state.c @@ -469,6 +469,8 @@ nv10_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, if (buf->size && (mapped = ws->buffer_map(ws, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) { memcpy(nv10->constbuf[shader], mapped, buf->size); + nv10->constbuf_nr[shader] = + buf->size / (4 * sizeof(float)); ws->buffer_unmap(ws, buf->buffer); } } diff --git a/src/gallium/drivers/nv10/nv10_vbo.c b/src/gallium/drivers/nv10/nv10_vbo.c index 2a334e137d..f024f53420 100644 --- a/src/gallium/drivers/nv10/nv10_vbo.c +++ b/src/gallium/drivers/nv10/nv10_vbo.c @@ -44,7 +44,9 @@ boolean nv10_draw_elements( struct pipe_context *pipe, draw_set_mapped_element_buffer(draw, 0, NULL); } - draw_set_mapped_constant_buffer(draw, nv10->constbuf[PIPE_SHADER_VERTEX]); + draw_set_mapped_constant_buffer(draw, + nv10->constbuf[PIPE_SHADER_VERTEX], + nv10->constbuf_nr[PIPE_SHADER_VERTEX]); /* draw! */ draw_arrays(nv10->draw, prim, start, count); diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 51000bd6fc..54d6bea55f 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -645,7 +645,7 @@ nv30_fragprog_parse_decl_attrib(struct nv30_fpc *fpc, return FALSE; } - fpc->attrib_map[fdec->u.DeclarationRange.First] = hw; + fpc->attrib_map[fdec->DeclarationRange.First] = hw; return TRUE; } @@ -655,10 +655,10 @@ nv30_fragprog_parse_decl_output(struct nv30_fpc *fpc, { switch (fdec->Semantic.SemanticName) { case TGSI_SEMANTIC_POSITION: - fpc->depth_id = fdec->u.DeclarationRange.First; + fpc->depth_id = fdec->DeclarationRange.First; break; case TGSI_SEMANTIC_COLOR: - fpc->colour_id = fdec->u.DeclarationRange.First; + fpc->colour_id = fdec->DeclarationRange.First; break; default: NOUVEAU_ERR("bad output semantic\n"); diff --git a/src/gallium/drivers/nv30/nv30_vertprog.c b/src/gallium/drivers/nv30/nv30_vertprog.c index 71aea3a59c..1b34d2e4b2 100644 --- a/src/gallium/drivers/nv30/nv30_vertprog.c +++ b/src/gallium/drivers/nv30/nv30_vertprog.c @@ -530,7 +530,7 @@ nv30_vertprog_parse_decl_output(struct nv30_vpc *vpc, return FALSE; } - vpc->output_map[fdec->u.DeclarationRange.First] = hw; + vpc->output_map[fdec->DeclarationRange.First] = hw; return TRUE; } diff --git a/src/gallium/drivers/nv40/nv40_context.h b/src/gallium/drivers/nv40/nv40_context.h index d8d1891dff..8e60a81e68 100644 --- a/src/gallium/drivers/nv40/nv40_context.h +++ b/src/gallium/drivers/nv40/nv40_context.h @@ -136,6 +136,7 @@ struct nv40_context { struct nv40_vertex_program *vertprog; struct nv40_fragment_program *fragprog; struct pipe_buffer *constbuf[PIPE_SHADER_TYPES]; + unsigned constbuf_nr[PIPE_SHADER_TYPES]; struct nv40_rasterizer_state *rasterizer; struct nv40_zsa_state *zsa; struct nv40_blend_state *blend; diff --git a/src/gallium/drivers/nv40/nv40_draw.c b/src/gallium/drivers/nv40/nv40_draw.c index 1d78324dda..2cf58e2950 100644 --- a/src/gallium/drivers/nv40/nv40_draw.c +++ b/src/gallium/drivers/nv40/nv40_draw.c @@ -253,9 +253,11 @@ nv40_draw_elements_swtnl(struct pipe_context *pipe, } if (nv40->constbuf[PIPE_SHADER_VERTEX]) { + const unsigned nr = nv40->constbuf_nr[PIPE_SHADER_VERTEX]; + map = ws->buffer_map(ws, nv40->constbuf[PIPE_SHADER_VERTEX], PIPE_BUFFER_USAGE_CPU_READ); - draw_set_mapped_constant_buffer(nv40->draw, map); + draw_set_mapped_constant_buffer(nv40->draw, map, nr); } draw_arrays(nv40->draw, mode, start, count); diff --git a/src/gallium/drivers/nv40/nv40_fragprog.c b/src/gallium/drivers/nv40/nv40_fragprog.c index 4b7667e038..428348c338 100644 --- a/src/gallium/drivers/nv40/nv40_fragprog.c +++ b/src/gallium/drivers/nv40/nv40_fragprog.c @@ -708,7 +708,7 @@ nv40_fragprog_parse_decl_attrib(struct nv40_fpc *fpc, return FALSE; } - fpc->attrib_map[fdec->u.DeclarationRange.First] = hw; + fpc->attrib_map[fdec->DeclarationRange.First] = hw; return TRUE; } @@ -716,7 +716,7 @@ static boolean nv40_fragprog_parse_decl_output(struct nv40_fpc *fpc, const struct tgsi_full_declaration *fdec) { - unsigned idx = fdec->u.DeclarationRange.First; + unsigned idx = fdec->DeclarationRange.First; unsigned hw; switch (fdec->Semantic.SemanticName) { @@ -770,9 +770,9 @@ nv40_fragprog_prepare(struct nv40_fpc *fpc) goto out_err; break; case TGSI_FILE_TEMPORARY: - if (fdec->u.DeclarationRange.Last > high_temp) { + if (fdec->DeclarationRange.Last > high_temp) { high_temp = - fdec->u.DeclarationRange.Last; + fdec->DeclarationRange.Last; } break; default: diff --git a/src/gallium/drivers/nv40/nv40_state.c b/src/gallium/drivers/nv40/nv40_state.c index 2d921d2b8a..afcf336a65 100644 --- a/src/gallium/drivers/nv40/nv40_state.c +++ b/src/gallium/drivers/nv40/nv40_state.c @@ -603,12 +603,13 @@ nv40_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, { struct nv40_context *nv40 = nv40_context(pipe); + nv40->constbuf[shader] = buf->buffer; + nv40->constbuf_nr[shader] = buf->size / (4 * sizeof(float)); + if (shader == PIPE_SHADER_VERTEX) { - nv40->constbuf[PIPE_SHADER_VERTEX] = buf->buffer; nv40->dirty |= NV40_NEW_VERTPROG; } else if (shader == PIPE_SHADER_FRAGMENT) { - nv40->constbuf[PIPE_SHADER_FRAGMENT] = buf->buffer; nv40->dirty |= NV40_NEW_FRAGPROG; } } diff --git a/src/gallium/drivers/nv40/nv40_vertprog.c b/src/gallium/drivers/nv40/nv40_vertprog.c index eb14869bfe..1e486a66ef 100644 --- a/src/gallium/drivers/nv40/nv40_vertprog.c +++ b/src/gallium/drivers/nv40/nv40_vertprog.c @@ -603,7 +603,7 @@ static boolean nv40_vertprog_parse_decl_output(struct nv40_vpc *vpc, const struct tgsi_full_declaration *fdec) { - unsigned idx = fdec->u.DeclarationRange.First; + unsigned idx = fdec->DeclarationRange.First; int hw; switch (fdec->Semantic.SemanticName) { @@ -678,16 +678,16 @@ nv40_vertprog_prepare(struct nv40_vpc *vpc) fdec = &p.FullToken.FullDeclaration; switch (fdec->Declaration.File) { case TGSI_FILE_TEMPORARY: - if (fdec->u.DeclarationRange.Last > high_temp) { + if (fdec->DeclarationRange.Last > high_temp) { high_temp = - fdec->u.DeclarationRange.Last; + fdec->DeclarationRange.Last; } break; #if 0 /* this would be nice.. except gallium doesn't track it */ case TGSI_FILE_ADDRESS: - if (fdec->u.DeclarationRange.Last > high_addr) { + if (fdec->DeclarationRange.Last > high_addr) { high_addr = - fdec->u.DeclarationRange.Last; + fdec->DeclarationRange.Last; } break; #endif -- cgit v1.2.3 From 28268f7b22905974a2cc3c099b1a842d20a7712a Mon Sep 17 00:00:00 2001 From: Stephane Marchesin Date: Tue, 8 Jul 2008 16:38:27 +0200 Subject: nv30: use native instructions. --- src/gallium/drivers/nv30/nv30_fragprog.c | 26 ++++---------------------- 1 file changed, 4 insertions(+), 22 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 54d6bea55f..740d3ab430 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -510,9 +510,7 @@ nv30_fragprog_parse_instruction(struct nv30_fpc *fpc, break; // case TGSI_OPCODE_LIT: case TGSI_OPCODE_LRP: - tmp = temp(fpc); - arith(fpc, 0, MAD, tmp, mask, neg(src[0]), src[2], src[2]); - arith(fpc, sat, MAD, dst, mask, src[0], src[1], tmp); + arith(fpc, sat, LRP, dst, mask, src[0], src[1], src[2]); break; case TGSI_OPCODE_MAD: arith(fpc, sat, MAD, dst, mask, src[0], src[1], src[2]); @@ -530,13 +528,7 @@ nv30_fragprog_parse_instruction(struct nv30_fpc *fpc, arith(fpc, sat, MUL, dst, mask, src[0], src[1], none); break; case TGSI_OPCODE_POW: - tmp = temp(fpc); - arith(fpc, 0, LG2, tmp, MASK_X, - swz(src[0], X, X, X, X), none, none); - arith(fpc, 0, MUL, tmp, MASK_X, swz(tmp, X, X, X, X), - swz(src[1], X, X, X, X), none); - arith(fpc, sat, EX2, dst, mask, - swz(tmp, X, X, X, X), none, none); + arith(fpc, sat, POW, dst, mask, src[0], src[1], none); break; case TGSI_OPCODE_RCP: arith(fpc, sat, RCP, dst, mask, src[0], none, none); @@ -545,20 +537,10 @@ nv30_fragprog_parse_instruction(struct nv30_fpc *fpc, assert(0); break; case TGSI_OPCODE_RFL: - tmp = temp(fpc); - arith(fpc, 0, DP3, tmp, MASK_X, src[0], src[0], none); - arith(fpc, 0, DP3, tmp, MASK_Y, src[0], src[1], none); - arith(fpc, 0, DIV, scale(tmp, 2X), MASK_Z, - swz(tmp, Y, Y, Y, Y), swz(tmp, X, X, X, X), none); - arith(fpc, sat, MAD, dst, mask, - swz(tmp, Z, Z, Z, Z), src[0], neg(src[1])); + arith(fpc, 0, RFL, dst, mask, src[0], src[1], none); break; case TGSI_OPCODE_RSQ: - tmp = temp(fpc); - arith(fpc, 0, LG2, scale(tmp, INV_2X), MASK_X, - abs(swz(src[0], X, X, X, X)), none, none); - arith(fpc, sat, EX2, dst, mask, - neg(swz(tmp, X, X, X, X)), none, none); + arith(fpc, sat, RSQ, dst, mask, abs(swz(src[0], X, X, X, X)), none, none); break; case TGSI_OPCODE_SCS: if (mask & MASK_X) { -- cgit v1.2.3 From 37a418b3b05db51aea0a1010bad8118b92e0e9d6 Mon Sep 17 00:00:00 2001 From: Patrice Mandin Date: Fri, 11 Jul 2008 21:31:24 +0200 Subject: nv30: split fragprog_prepare from fragprog_translate --- src/gallium/drivers/nv30/nv30_fragprog.c | 91 ++++++++++++++++++++++++-------- 1 file changed, 68 insertions(+), 23 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 740d3ab430..7278bf5886 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -1,7 +1,6 @@ #include "pipe/p_context.h" #include "pipe/p_defines.h" #include "pipe/p_state.h" -#include "pipe/p_util.h" #include "pipe/p_shader_tokens.h" #include "tgsi/util/tgsi_parse.h" @@ -650,30 +649,22 @@ nv30_fragprog_parse_decl_output(struct nv30_fpc *fpc, return TRUE; } -void -nv30_fragprog_translate(struct nv30_context *nv30, - struct nv30_fragment_program *fp) +static boolean +nv30_fragprog_prepare(struct nv30_fpc *fpc) { - struct tgsi_parse_context parse; - struct nv30_fpc *fpc = NULL; - - fpc = CALLOC(1, sizeof(struct nv30_fpc)); - if (!fpc) - return; - fpc->fp = fp; - fpc->high_temp = -1; - fpc->num_regs = 2; + struct tgsi_parse_context p; + /*int high_temp = -1, i;*/ - tgsi_parse_init(&parse, fp->pipe.tokens); + tgsi_parse_init(&p, fpc->fp->pipe.tokens); + while (!tgsi_parse_end_of_tokens(&p)) { + const union tgsi_full_token *tok = &p.FullToken; - while (!tgsi_parse_end_of_tokens(&parse)) { - tgsi_parse_token(&parse); - - switch (parse.FullToken.Token.Type) { + tgsi_parse_token(&p); + switch(tok->Token.Type) { case TGSI_TOKEN_TYPE_DECLARATION: { const struct tgsi_full_declaration *fdec; - fdec = &parse.FullToken.FullDeclaration; + fdec = &p.FullToken.FullDeclaration; switch (fdec->Declaration.File) { case TGSI_FILE_INPUT: if (!nv30_fragprog_parse_decl_attrib(fpc, fdec)) @@ -683,6 +674,12 @@ nv30_fragprog_translate(struct nv30_context *nv30, if (!nv30_fragprog_parse_decl_output(fpc, fdec)) goto out_err; break; + /*case TGSI_FILE_TEMPORARY: + if (fdec->DeclarationRange.Last > high_temp) { + high_temp = + fdec->DeclarationRange.Last; + } + break;*/ default: break; } @@ -692,17 +689,65 @@ nv30_fragprog_translate(struct nv30_context *nv30, { struct tgsi_full_immediate *imm; float vals[4]; - int i; - imm = &parse.FullToken.FullImmediate; + imm = &p.FullToken.FullImmediate; assert(imm->Immediate.DataType == TGSI_IMM_FLOAT32); assert(fpc->nr_imm < MAX_IMM); - for (i = 0; i < 4; i++) - vals[i] = imm->u.ImmediateFloat32[i].Float; + vals[0] = imm->u.ImmediateFloat32[0].Float; + vals[1] = imm->u.ImmediateFloat32[1].Float; + vals[2] = imm->u.ImmediateFloat32[2].Float; + vals[3] = imm->u.ImmediateFloat32[3].Float; fpc->imm[fpc->nr_imm++] = constant(fpc, -1, vals); } break; + default: + break; + } + } + tgsi_parse_free(&p); + + /*if (++high_temp) { + fpc->r_temp = CALLOC(high_temp, sizeof(struct nv30_sreg)); + for (i = 0; i < high_temp; i++) + fpc->r_temp[i] = temp(fpc); + fpc->r_temps_discard = 0; + }*/ + + return TRUE; + +out_err: + /*if (fpc->r_temp) + FREE(fpc->r_temp);*/ + tgsi_parse_free(&p); + return FALSE; +} + +void +nv30_fragprog_translate(struct nv30_context *nv30, + struct nv30_fragment_program *fp) +{ + struct tgsi_parse_context parse; + struct nv30_fpc *fpc = NULL; + + fpc = CALLOC(1, sizeof(struct nv30_fpc)); + if (!fpc) + return; + fpc->fp = fp; + fpc->high_temp = -1; + fpc->num_regs = 2; + + if (!nv30_fragprog_prepare(fpc)) { + FREE(fpc); + return; + } + + tgsi_parse_init(&parse, fp->pipe.tokens); + + while (!tgsi_parse_end_of_tokens(&parse)) { + tgsi_parse_token(&parse); + + switch (parse.FullToken.Token.Type) { case TGSI_TOKEN_TYPE_INSTRUCTION: { const struct tgsi_full_instruction *finst; -- cgit v1.2.3 From b327b064058aaab1debb1af08f4b9abe1385adc6 Mon Sep 17 00:00:00 2001 From: Patrice Mandin Date: Fri, 11 Jul 2008 22:06:12 +0200 Subject: nv30: split fragprog_upload from fragprog_bind --- src/gallium/drivers/nv30/nv30_fragprog.c | 57 ++++++++++++++++++++------------ 1 file changed, 35 insertions(+), 22 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 7278bf5886..6af86e4f4f 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -783,6 +783,40 @@ out_err: FREE(fpc); } +static void +nv30_fragprog_upload(struct nv30_context *nv30, + struct nv30_fragment_program *fp) +{ + struct pipe_winsys *ws = nv30->pipe.winsys; + const uint32_t le = 1; + uint32_t *map; + int i; + + map = ws->buffer_map(ws, fp->buffer, PIPE_BUFFER_USAGE_CPU_WRITE); + +#if 0 + for (i = 0; i < fp->insn_len; i++) { + fflush(stdout); fflush(stderr); + NOUVEAU_ERR("%d 0x%08x\n", i, fp->insn[i]); + fflush(stdout); fflush(stderr); + } +#endif + + if ((*(const uint8_t *)&le)) { + for (i = 0; i < fp->insn_len; i++) { + map[i] = fp->insn[i]; + } + } else { + /* Weird swapping for big-endian chips */ + for (i = 0; i < fp->insn_len; i++) { + map[i] = ((fp->insn[i] & 0xffff) << 16) | + ((fp->insn[i] >> 16) & 0xffff); + } + } + + ws->buffer_unmap(ws, fp->buffer); +} + void nv30_fragprog_bind(struct nv30_context *nv30, struct nv30_fragment_program *fp) { @@ -818,28 +852,7 @@ nv30_fragprog_bind(struct nv30_context *nv30, struct nv30_fragment_program *fp) if (!fp->buffer) fp->buffer = ws->buffer_create(ws, 0x100, 0, fp->insn_len * 4); - map = ws->buffer_map(ws, fp->buffer, - PIPE_BUFFER_USAGE_CPU_WRITE); - -#if 0 - for (i = 0; i < fp->insn_len; i++) { - NOUVEAU_ERR("%d 0x%08x\n", i, fp->insn[i]); - } -#endif - - if ((*(const uint8_t *)&le)) { - for (i = 0; i < fp->insn_len; i++) { - map[i] = fp->insn[i]; - } - } else { - /* Weird swapping for big-endian chips */ - for (i = 0; i < fp->insn_len; i++) { - map[i] = ((fp->insn[i] & 0xffff) << 16) | - ((fp->insn[i] >> 16) & 0xffff); - } - } - - ws->buffer_unmap(ws, fp->buffer); + nv30_fragprog_upload(nv30, fp); fp->on_hw = TRUE; } -- cgit v1.2.3 From c7086277546d065eb94ba8dbeca1620605f167ea Mon Sep 17 00:00:00 2001 From: Patrice Mandin Date: Fri, 11 Jul 2008 23:17:47 +0200 Subject: nv30: Move constant buffers out of vert/frag prog structures --- src/gallium/drivers/nv30/nv30_context.h | 6 ++---- src/gallium/drivers/nv30/nv30_fragprog.c | 12 ++++++------ src/gallium/drivers/nv30/nv30_state.c | 5 +++-- src/gallium/drivers/nv30/nv30_vertprog.c | 18 ++++++++---------- 4 files changed, 19 insertions(+), 22 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_context.h b/src/gallium/drivers/nv30/nv30_context.h index 72f803c80a..c3c8b73309 100644 --- a/src/gallium/drivers/nv30/nv30_context.h +++ b/src/gallium/drivers/nv30/nv30_context.h @@ -124,6 +124,8 @@ struct nv30_context { unsigned dirty; struct pipe_scissor_state scissor; unsigned stipple[32]; + struct pipe_buffer *constbuf[PIPE_SHADER_TYPES]; + unsigned constbuf_nr[PIPE_SHADER_TYPES]; struct nv30_rasterizer_state *rasterizer; struct nv30_zsa_state *zsa; struct nv30_blend_state *blend; @@ -150,16 +152,12 @@ struct nv30_context { struct { struct nv30_vertex_program *active; - struct nv30_vertex_program *current; - struct pipe_buffer *constant_buf; } vertprog; struct { struct nv30_fragment_program *active; - struct nv30_fragment_program *current; - struct pipe_buffer *constant_buf; } fragprog; struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS]; diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 6af86e4f4f..b560455a02 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -820,6 +820,8 @@ nv30_fragprog_upload(struct nv30_context *nv30, void nv30_fragprog_bind(struct nv30_context *nv30, struct nv30_fragment_program *fp) { + struct pipe_buffer *constbuf = + nv30->constbuf[PIPE_SHADER_FRAGMENT]; struct pipe_winsys *ws = nv30->pipe.winsys; int i; @@ -830,8 +832,9 @@ nv30_fragprog_bind(struct nv30_context *nv30, struct nv30_fragment_program *fp) } if (fp->nr_consts) { - float *map = ws->buffer_map(ws, nv30->fragprog.constant_buf, - PIPE_BUFFER_USAGE_CPU_READ); + float *map; + + map = ws->buffer_map(ws, constbuf, PIPE_BUFFER_USAGE_CPU_READ); for (i = 0; i < fp->nr_consts; i++) { struct nv30_fragment_program_data *fpd = &fp->consts[i]; uint32_t *p = &fp->insn[fpd->offset]; @@ -842,13 +845,10 @@ nv30_fragprog_bind(struct nv30_context *nv30, struct nv30_fragment_program *fp) memcpy(p, cb, 4 * sizeof(float)); fp->on_hw = 0; } - ws->buffer_unmap(ws, nv30->fragprog.constant_buf); + ws->buffer_unmap(ws, constbuf); } if (!fp->on_hw) { - const uint32_t le = 1; - uint32_t *map; - if (!fp->buffer) fp->buffer = ws->buffer_create(ws, 0x100, 0, fp->insn_len * 4); diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c index 72b9515eb1..aec6fa2b15 100644 --- a/src/gallium/drivers/nv30/nv30_state.c +++ b/src/gallium/drivers/nv30/nv30_state.c @@ -589,12 +589,13 @@ nv30_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, { struct nv30_context *nv30 = nv30_context(pipe); + nv30->constbuf[shader] = buf->buffer; + nv30->constbuf_nr[shader] = buf->size / (4 * sizeof(float)); + if (shader == PIPE_SHADER_VERTEX) { - nv30->vertprog.constant_buf = buf->buffer; nv30->dirty |= NV30_NEW_VERTPROG; } else if (shader == PIPE_SHADER_FRAGMENT) { - nv30->fragprog.constant_buf = buf->buffer; nv30->dirty |= NV30_NEW_FRAGPROG; } } diff --git a/src/gallium/drivers/nv30/nv30_vertprog.c b/src/gallium/drivers/nv30/nv30_vertprog.c index 1b34d2e4b2..90756febd2 100644 --- a/src/gallium/drivers/nv30/nv30_vertprog.c +++ b/src/gallium/drivers/nv30/nv30_vertprog.c @@ -135,7 +135,7 @@ emit_src(struct nv30_vpc *vpc, uint32_t *hw, int pos, struct nv30_sreg src) /* * |VVV| - * d°.°b + * d�.�b * \u/ * */ @@ -641,9 +641,12 @@ nv30_vertprog_bind(struct nv30_context *nv30, struct nv30_vertex_program *vp) { struct nouveau_winsys *nvws = nv30->nvws; struct pipe_winsys *ws = nv30->pipe.winsys; + struct pipe_buffer *constbuf; boolean upload_code = FALSE, upload_data = FALSE; int i; + constbuf = nv30->constbuf[PIPE_SHADER_VERTEX]; + /* Translate TGSI shader into hw bytecode */ if (!vp->translated) { nv30_vertprog_translate(nv30, vp); @@ -730,8 +733,8 @@ nv30_vertprog_bind(struct nv30_context *nv30, struct nv30_vertex_program *vp) if (vp->nr_consts) { float *map = NULL; - if (nv30->vertprog.constant_buf) { - map = ws->buffer_map(ws, nv30->vertprog.constant_buf, + if (constbuf) { + map = ws->buffer_map(ws, constbuf, PIPE_BUFFER_USAGE_CPU_READ); } @@ -750,15 +753,10 @@ nv30_vertprog_bind(struct nv30_context *nv30, struct nv30_vertex_program *vp) BEGIN_RING(rankine, NV34TCL_VP_UPLOAD_CONST_ID, 5); OUT_RING (i + vp->data->start); OUT_RINGp ((uint32_t *)vpd->value, 4); -#if 0 - NOUVEAU_MSG("VP const %d: %f %f %f %f\n", - i, vpd->value[0], vpd->value[1], - vpd->value[2], vpd->value[3]); -#endif } - if (map) { - ws->buffer_unmap(ws, nv30->vertprog.constant_buf); + if (constbuf) { + ws->buffer_unmap(ws, constbuf); } } -- cgit v1.2.3 From 58737dc87575625438d288fe2f816b6a9d2086f1 Mon Sep 17 00:00:00 2001 From: Patrice Mandin Date: Sat, 12 Jul 2008 00:48:26 +0200 Subject: nv30: Emit fragment program using state objects --- src/gallium/drivers/nv30/nv30_context.h | 11 +---- src/gallium/drivers/nv30/nv30_fragprog.c | 65 +++++++++++++++++++++--------- src/gallium/drivers/nv30/nv30_fragtex.c | 2 +- src/gallium/drivers/nv30/nv30_state.c | 9 ++--- src/gallium/drivers/nv30/nv30_state.h | 3 ++ src/gallium/drivers/nv30/nv30_state_emit.c | 14 +------ 6 files changed, 56 insertions(+), 48 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_context.h b/src/gallium/drivers/nv30/nv30_context.h index 5d7080a555..0c900c5598 100644 --- a/src/gallium/drivers/nv30/nv30_context.h +++ b/src/gallium/drivers/nv30/nv30_context.h @@ -125,6 +125,7 @@ struct nv30_context { struct pipe_scissor_state scissor; unsigned stipple[32]; struct nv30_vertex_program *vertprog; + struct nv30_fragment_program *fragprog; struct pipe_buffer *constbuf[PIPE_SHADER_TYPES]; unsigned constbuf_nr[PIPE_SHADER_TYPES]; struct nv30_rasterizer_state *rasterizer; @@ -151,11 +152,6 @@ struct nv30_context { unsigned delta; } vb[16]; - struct { - struct nv30_fragment_program *active; - struct nv30_fragment_program *current; - } fragprog; - struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS]; struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS]; }; @@ -188,10 +184,6 @@ extern void nv30_vertprog_destroy(struct nv30_context *, struct nv30_vertex_program *); /* nv30_fragprog.c */ -extern void nv30_fragprog_translate(struct nv30_context *, - struct nv30_fragment_program *); -extern void nv30_fragprog_bind(struct nv30_context *, - struct nv30_fragment_program *); extern void nv30_fragprog_destroy(struct nv30_context *, struct nv30_fragment_program *); @@ -205,6 +197,7 @@ extern void nv30_state_tex_update(struct nv30_context *nv30); extern struct nv30_state_entry nv30_state_rasterizer; extern struct nv30_state_entry nv30_state_scissor; extern struct nv30_state_entry nv30_state_stipple; +extern struct nv30_state_entry nv30_state_fragprog; extern struct nv30_state_entry nv30_state_vertprog; extern struct nv30_state_entry nv30_state_blend; extern struct nv30_state_entry nv30_state_blend_colour; diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index b560455a02..06ed04cbfe 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -723,7 +723,7 @@ out_err: return FALSE; } -void +static void nv30_fragprog_translate(struct nv30_context *nv30, struct nv30_fragment_program *fp) { @@ -817,23 +817,46 @@ nv30_fragprog_upload(struct nv30_context *nv30, ws->buffer_unmap(ws, fp->buffer); } -void -nv30_fragprog_bind(struct nv30_context *nv30, struct nv30_fragment_program *fp) +static boolean +nv30_fragprog_validate(struct nv30_context *nv30) { + struct nv30_fragment_program *fp = nv30->fragprog; struct pipe_buffer *constbuf = nv30->constbuf[PIPE_SHADER_FRAGMENT]; struct pipe_winsys *ws = nv30->pipe.winsys; + struct nouveau_stateobj *so; + boolean new_consts = FALSE; int i; + if (fp->translated) + goto update_constants; + + /*nv30->fallback_swrast &= ~NV30_NEW_FRAGPROG;*/ + nv30_fragprog_translate(nv30, fp); if (!fp->translated) { - nv30_fragprog_translate(nv30, fp); - if (!fp->translated) - assert(0); + /*nv30->fallback_swrast |= NV30_NEW_FRAGPROG;*/ + return FALSE; } + fp->buffer = ws->buffer_create(ws, 0x100, 0, fp->insn_len * 4); + nv30_fragprog_upload(nv30, fp); + + so = so_new(4, 1); + so_method(so, nv30->screen->rankine, NV34TCL_FP_ACTIVE_PROGRAM, 1); + so_reloc (so, fp->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | + NOUVEAU_BO_RD | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, + NV34TCL_FP_ACTIVE_PROGRAM_DMA0, NV34TCL_FP_ACTIVE_PROGRAM_DMA1); + so_method(so, nv30->screen->rankine, NV34TCL_FP_CONTROL, 1); + so_data (so, fp->fp_control); + /* FIXME: Add these, and you'll have big slowdown */ + /*so_method(so, nv30->screen->rankine, NV34TCL_FP_REG_CONTROL, 1); + so_data (so, fp->fp_control);*/ + so_ref(so, &fp->so); + +update_constants: if (fp->nr_consts) { float *map; - + map = ws->buffer_map(ws, constbuf, PIPE_BUFFER_USAGE_CPU_READ); for (i = 0; i < fp->nr_consts; i++) { struct nv30_fragment_program_data *fpd = &fp->consts[i]; @@ -843,25 +866,20 @@ nv30_fragprog_bind(struct nv30_context *nv30, struct nv30_fragment_program *fp) if (!memcmp(p, cb, 4 * sizeof(float))) continue; memcpy(p, cb, 4 * sizeof(float)); - fp->on_hw = 0; + new_consts = TRUE; } ws->buffer_unmap(ws, constbuf); - } - if (!fp->on_hw) { - if (!fp->buffer) - fp->buffer = ws->buffer_create(ws, 0x100, 0, - fp->insn_len * 4); - nv30_fragprog_upload(nv30, fp); - fp->on_hw = TRUE; + if (new_consts) + nv30_fragprog_upload(nv30, fp); } - BEGIN_RING(rankine, NV34TCL_FP_CONTROL, 1); - OUT_RING (fp->fp_control); - BEGIN_RING(rankine, NV34TCL_FP_REG_CONTROL, 1); - OUT_RING (fp->fp_reg_control); + if (new_consts || fp->so != nv30->state.hw[NV30_STATE_FRAGPROG]) { + so_ref(fp->so, &nv30->state.hw[NV30_STATE_FRAGPROG]); + return TRUE; + } - nv30->fragprog.active = fp; + return FALSE; } void @@ -872,3 +890,10 @@ nv30_fragprog_destroy(struct nv30_context *nv30, FREE(fp->insn); } +struct nv30_state_entry nv30_state_fragprog = { + .validate = nv30_fragprog_validate, + .dirty = { + .pipe = NV30_NEW_FRAGPROG, + .hw = NV30_STATE_FRAGPROG + } +}; diff --git a/src/gallium/drivers/nv30/nv30_fragtex.c b/src/gallium/drivers/nv30/nv30_fragtex.c index abe77b51df..8c5e88ea1d 100644 --- a/src/gallium/drivers/nv30/nv30_fragtex.c +++ b/src/gallium/drivers/nv30/nv30_fragtex.c @@ -143,7 +143,7 @@ nv30_fragtex_build(struct nv30_context *nv30, int unit) static boolean nv30_fragtex_validate(struct nv30_context *nv30) { - struct nv30_fragment_program *fp = nv30->fragprog.current; + struct nv30_fragment_program *fp = nv30->fragprog; struct nv30_state *state = &nv30->state; struct nouveau_stateobj *so; unsigned samplers, unit; diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c index 92695ce23c..8dc16d361d 100644 --- a/src/gallium/drivers/nv30/nv30_state.c +++ b/src/gallium/drivers/nv30/nv30_state.c @@ -539,6 +539,8 @@ nv30_fp_state_create(struct pipe_context *pipe, fp = CALLOC(1, sizeof(struct nv30_fragment_program)); fp->pipe = *cso; + tgsi_scan_shader(fp->pipe.tokens, &fp->info); + return (void *)fp; } @@ -546,13 +548,8 @@ static void nv30_fp_state_bind(struct pipe_context *pipe, void *hwcso) { struct nv30_context *nv30 = nv30_context(pipe); - struct nv30_fragment_program *fp = hwcso; - - if (!hwcso) { - return; - } - nv30->fragprog.current = fp; + nv30->fragprog = hwcso; nv30->dirty |= NV30_NEW_FRAGPROG; } diff --git a/src/gallium/drivers/nv30/nv30_state.h b/src/gallium/drivers/nv30/nv30_state.h index a897bc5068..20c5176160 100644 --- a/src/gallium/drivers/nv30/nv30_state.h +++ b/src/gallium/drivers/nv30/nv30_state.h @@ -2,6 +2,7 @@ #define __NV30_STATE_H__ #include "pipe/p_state.h" +#include "tgsi/util/tgsi_scan.h" struct nv30_sampler_state { uint32_t fmt; @@ -50,6 +51,7 @@ struct nv30_fragment_program_data { struct nv30_fragment_program { struct pipe_shader_state pipe; + struct tgsi_shader_info info; boolean translated; boolean on_hw; @@ -65,6 +67,7 @@ struct nv30_fragment_program { uint32_t fp_control; uint32_t fp_reg_control; + struct nouveau_stateobj *so; }; struct nv30_miptree { diff --git a/src/gallium/drivers/nv30/nv30_state_emit.c b/src/gallium/drivers/nv30/nv30_state_emit.c index 4ab62ddc01..c4ccc9422b 100644 --- a/src/gallium/drivers/nv30/nv30_state_emit.c +++ b/src/gallium/drivers/nv30/nv30_state_emit.c @@ -6,6 +6,7 @@ static struct nv30_state_entry *render_states[] = { &nv30_state_rasterizer, &nv30_state_scissor, &nv30_state_stipple, + &nv30_state_fragprog, &nv30_state_fragtex, &nv30_state_vertprog, &nv30_state_blend, @@ -61,11 +62,6 @@ nv30_emit_hw_state(struct nv30_context *nv30) screen->cur_pctx = nv30->pctx_id; } - if (nv30->dirty & NV30_NEW_FRAGPROG) { - nv30_fragprog_bind(nv30, nv30->fragprog.current); - /*XXX: clear NV30_NEW_FRAGPROG if no new program uploaded */ - } - for (i = 0, states = state->dirty; states; i++) { if (!(states & (1ULL << i))) continue; @@ -83,13 +79,7 @@ nv30_emit_hw_state(struct nv30_context *nv30) state->hw[NV30_STATE_FRAGTEX0+i]); samplers &= ~(1ULL << i); } - - /* Fragment program */ - BEGIN_RING(rankine, NV34TCL_FP_ACTIVE_PROGRAM, 1); - OUT_RELOC (nv30->fragprog.active->buffer, 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_LOW | - NOUVEAU_BO_OR, NV34TCL_FP_ACTIVE_PROGRAM_DMA0, - NV34TCL_FP_ACTIVE_PROGRAM_DMA1); + so_emit_reloc_markers(nv30->nvws, state->hw[NV30_STATE_FRAGPROG]); } boolean -- cgit v1.2.3 From 9489de99802e635271c1ae84630fc02892af1699 Mon Sep 17 00:00:00 2001 From: Patrice Mandin Date: Sat, 12 Jul 2008 10:13:58 +0200 Subject: nv30: was setting wrong register --- src/gallium/drivers/nv30/nv30_fragprog.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 06ed04cbfe..59e79d66f2 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -841,16 +841,15 @@ nv30_fragprog_validate(struct nv30_context *nv30) fp->buffer = ws->buffer_create(ws, 0x100, 0, fp->insn_len * 4); nv30_fragprog_upload(nv30, fp); - so = so_new(4, 1); + so = so_new(6, 1); so_method(so, nv30->screen->rankine, NV34TCL_FP_ACTIVE_PROGRAM, 1); so_reloc (so, fp->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, NV34TCL_FP_ACTIVE_PROGRAM_DMA0, NV34TCL_FP_ACTIVE_PROGRAM_DMA1); so_method(so, nv30->screen->rankine, NV34TCL_FP_CONTROL, 1); so_data (so, fp->fp_control); - /* FIXME: Add these, and you'll have big slowdown */ - /*so_method(so, nv30->screen->rankine, NV34TCL_FP_REG_CONTROL, 1); - so_data (so, fp->fp_control);*/ + so_method(so, nv30->screen->rankine, NV34TCL_FP_REG_CONTROL, 1); + so_data (so, fp->fp_reg_control); so_ref(so, &fp->so); update_constants: -- cgit v1.2.3 From fda01b584715c05696a0e6768fda669ef1eb5f3b Mon Sep 17 00:00:00 2001 From: Ben Skeggs Date: Thu, 7 Aug 2008 11:26:17 +1000 Subject: nouveau: fix build --- src/gallium/drivers/nv04/nv04_fragprog.c | 4 ++-- src/gallium/drivers/nv04/nv04_state.c | 2 +- src/gallium/drivers/nv04/nv04_state.h | 2 +- src/gallium/drivers/nv10/nv10_fragprog.c | 4 ++-- src/gallium/drivers/nv10/nv10_state.c | 2 +- src/gallium/drivers/nv10/nv10_state.h | 2 +- src/gallium/drivers/nv30/nv30_fragprog.c | 4 ++-- src/gallium/drivers/nv30/nv30_state.c | 2 +- src/gallium/drivers/nv30/nv30_state.h | 2 +- src/gallium/drivers/nv30/nv30_vertprog.c | 2 +- src/gallium/drivers/nv40/nv40_fragprog.c | 4 ++-- src/gallium/drivers/nv40/nv40_state.c | 2 +- src/gallium/drivers/nv40/nv40_state.h | 2 +- src/gallium/drivers/nv40/nv40_vertprog.c | 4 ++-- src/gallium/drivers/nv50/nv50_program.c | 4 ++-- src/gallium/drivers/nv50/nv50_program.h | 2 +- src/gallium/drivers/nv50/nv50_state.c | 2 +- 17 files changed, 23 insertions(+), 23 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv04/nv04_fragprog.c b/src/gallium/drivers/nv04/nv04_fragprog.c index 11f4360ed8..215974eec0 100644 --- a/src/gallium/drivers/nv04/nv04_fragprog.c +++ b/src/gallium/drivers/nv04/nv04_fragprog.c @@ -4,8 +4,8 @@ #include "pipe/p_util.h" #include "pipe/p_shader_tokens.h" -#include "tgsi/util/tgsi_parse.h" -#include "tgsi/util/tgsi_util.h" +#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_util.h" #include "nv04_context.h" diff --git a/src/gallium/drivers/nv04/nv04_state.c b/src/gallium/drivers/nv04/nv04_state.c index 7e71dee7b5..668d875671 100644 --- a/src/gallium/drivers/nv04/nv04_state.c +++ b/src/gallium/drivers/nv04/nv04_state.c @@ -4,7 +4,7 @@ #include "pipe/p_util.h" #include "pipe/p_shader_tokens.h" -#include "tgsi/util/tgsi_parse.h" +#include "tgsi/tgsi_parse.h" #include "nv04_context.h" #include "nv04_state.h" diff --git a/src/gallium/drivers/nv04/nv04_state.h b/src/gallium/drivers/nv04/nv04_state.h index 39f7cd17b3..399f750dbe 100644 --- a/src/gallium/drivers/nv04/nv04_state.h +++ b/src/gallium/drivers/nv04/nv04_state.h @@ -2,7 +2,7 @@ #define __NV04_STATE_H__ #include "pipe/p_state.h" -#include "tgsi/util/tgsi_scan.h" +#include "tgsi/tgsi_scan.h" struct nv04_blend_state { uint32_t b_enable; diff --git a/src/gallium/drivers/nv10/nv10_fragprog.c b/src/gallium/drivers/nv10/nv10_fragprog.c index 2a63c8a704..137de9d53e 100644 --- a/src/gallium/drivers/nv10/nv10_fragprog.c +++ b/src/gallium/drivers/nv10/nv10_fragprog.c @@ -4,8 +4,8 @@ #include "pipe/p_util.h" #include "pipe/p_shader_tokens.h" -#include "tgsi/util/tgsi_parse.h" -#include "tgsi/util/tgsi_util.h" +#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_util.h" #include "nv10_context.h" diff --git a/src/gallium/drivers/nv10/nv10_state.c b/src/gallium/drivers/nv10/nv10_state.c index 43b9c32f12..f902fd54b6 100644 --- a/src/gallium/drivers/nv10/nv10_state.c +++ b/src/gallium/drivers/nv10/nv10_state.c @@ -4,7 +4,7 @@ #include "pipe/p_util.h" #include "pipe/p_shader_tokens.h" -#include "tgsi/util/tgsi_parse.h" +#include "tgsi/tgsi_parse.h" #include "nv10_context.h" #include "nv10_state.h" diff --git a/src/gallium/drivers/nv10/nv10_state.h b/src/gallium/drivers/nv10/nv10_state.h index f1f9a12110..3a3fd0d4f4 100644 --- a/src/gallium/drivers/nv10/nv10_state.h +++ b/src/gallium/drivers/nv10/nv10_state.h @@ -2,7 +2,7 @@ #define __NV10_STATE_H__ #include "pipe/p_state.h" -#include "tgsi/util/tgsi_scan.h" +#include "tgsi/tgsi_scan.h" struct nv10_blend_state { uint32_t b_enable; diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 59e79d66f2..68058264e3 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -3,8 +3,8 @@ #include "pipe/p_state.h" #include "pipe/p_shader_tokens.h" -#include "tgsi/util/tgsi_parse.h" -#include "tgsi/util/tgsi_util.h" +#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_util.h" #include "nv30_context.h" diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c index ba02413de5..8d88d6c806 100644 --- a/src/gallium/drivers/nv30/nv30_state.c +++ b/src/gallium/drivers/nv30/nv30_state.c @@ -3,7 +3,7 @@ #include "pipe/p_util.h" #include "pipe/p_inlines.h" -#include "tgsi/util/tgsi_parse.h" +#include "tgsi/tgsi_parse.h" #include "nv30_context.h" #include "nv30_state.h" diff --git a/src/gallium/drivers/nv30/nv30_state.h b/src/gallium/drivers/nv30/nv30_state.h index 20c5176160..e6f23bf166 100644 --- a/src/gallium/drivers/nv30/nv30_state.h +++ b/src/gallium/drivers/nv30/nv30_state.h @@ -2,7 +2,7 @@ #define __NV30_STATE_H__ #include "pipe/p_state.h" -#include "tgsi/util/tgsi_scan.h" +#include "tgsi/tgsi_scan.h" struct nv30_sampler_state { uint32_t fmt; diff --git a/src/gallium/drivers/nv30/nv30_vertprog.c b/src/gallium/drivers/nv30/nv30_vertprog.c index 39852ce983..6c075cdb75 100644 --- a/src/gallium/drivers/nv30/nv30_vertprog.c +++ b/src/gallium/drivers/nv30/nv30_vertprog.c @@ -3,7 +3,7 @@ #include "pipe/p_state.h" #include "pipe/p_shader_tokens.h" -#include "tgsi/util/tgsi_parse.h" +#include "tgsi/tgsi_parse.h" #include "nv30_context.h" #include "nv30_state.h" diff --git a/src/gallium/drivers/nv40/nv40_fragprog.c b/src/gallium/drivers/nv40/nv40_fragprog.c index 428348c338..a361509e8f 100644 --- a/src/gallium/drivers/nv40/nv40_fragprog.c +++ b/src/gallium/drivers/nv40/nv40_fragprog.c @@ -3,8 +3,8 @@ #include "pipe/p_state.h" #include "pipe/p_shader_tokens.h" -#include "tgsi/util/tgsi_parse.h" -#include "tgsi/util/tgsi_util.h" +#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_util.h" #include "nv40_context.h" diff --git a/src/gallium/drivers/nv40/nv40_state.c b/src/gallium/drivers/nv40/nv40_state.c index 5d2c3ab881..63d0ecc915 100644 --- a/src/gallium/drivers/nv40/nv40_state.c +++ b/src/gallium/drivers/nv40/nv40_state.c @@ -5,7 +5,7 @@ #include "draw/draw_context.h" -#include "tgsi/util/tgsi_parse.h" +#include "tgsi/tgsi_parse.h" #include "nv40_context.h" #include "nv40_state.h" diff --git a/src/gallium/drivers/nv40/nv40_state.h b/src/gallium/drivers/nv40/nv40_state.h index 2b4225deb2..8a9d8c8fdf 100644 --- a/src/gallium/drivers/nv40/nv40_state.h +++ b/src/gallium/drivers/nv40/nv40_state.h @@ -2,7 +2,7 @@ #define __NV40_STATE_H__ #include "pipe/p_state.h" -#include "tgsi/util/tgsi_scan.h" +#include "tgsi/tgsi_scan.h" struct nv40_sampler_state { uint32_t fmt; diff --git a/src/gallium/drivers/nv40/nv40_vertprog.c b/src/gallium/drivers/nv40/nv40_vertprog.c index 1e486a66ef..ff988e6a5f 100644 --- a/src/gallium/drivers/nv40/nv40_vertprog.c +++ b/src/gallium/drivers/nv40/nv40_vertprog.c @@ -3,8 +3,8 @@ #include "pipe/p_state.h" #include "pipe/p_shader_tokens.h" -#include "tgsi/util/tgsi_parse.h" -#include "tgsi/util/tgsi_util.h" +#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_util.h" #include "nv40_context.h" #include "nv40_state.h" diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index 0d3ddb8a59..d6fbdd1824 100644 --- a/src/gallium/drivers/nv50/nv50_program.c +++ b/src/gallium/drivers/nv50/nv50_program.c @@ -26,8 +26,8 @@ #include "pipe/p_inlines.h" #include "pipe/p_shader_tokens.h" -#include "tgsi/util/tgsi_parse.h" -#include "tgsi/util/tgsi_util.h" +#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_util.h" #include "nv50_context.h" diff --git a/src/gallium/drivers/nv50/nv50_program.h b/src/gallium/drivers/nv50/nv50_program.h index d643e8db21..78deed6a38 100644 --- a/src/gallium/drivers/nv50/nv50_program.h +++ b/src/gallium/drivers/nv50/nv50_program.h @@ -2,7 +2,7 @@ #define __NV50_PROGRAM_H__ #include "pipe/p_state.h" -#include "tgsi/util/tgsi_scan.h" +#include "tgsi/tgsi_scan.h" struct nv50_program_exec { struct nv50_program_exec *next; diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c index 731409bed4..4055527d9f 100644 --- a/src/gallium/drivers/nv50/nv50_state.c +++ b/src/gallium/drivers/nv50/nv50_state.c @@ -25,7 +25,7 @@ #include "pipe/p_util.h" #include "pipe/p_inlines.h" -#include "tgsi/util/tgsi_parse.h" +#include "tgsi/tgsi_parse.h" #include "nv50_context.h" #include "nv50_texture.h" -- cgit v1.2.3 From dbec107c2549c00120ca6d67282f2a1a4bd8bdbd Mon Sep 17 00:00:00 2001 From: Ben Skeggs Date: Thu, 14 Aug 2008 00:22:49 +1000 Subject: nv30/nv40: KIL/KILP swapped meanings --- src/gallium/drivers/nv30/nv30_fragprog.c | 4 ++-- src/gallium/drivers/nv40/nv40_fragprog.c | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 68058264e3..d3693fdf56 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -494,10 +494,10 @@ nv30_fragprog_parse_instruction(struct nv30_fpc *fpc, case TGSI_OPCODE_FRC: arith(fpc, sat, FRC, dst, mask, src[0], none, none); break; - case TGSI_OPCODE_KIL: + case TGSI_OPCODE_KILP: arith(fpc, 0, KIL, none, 0, none, none, none); break; - case TGSI_OPCODE_KILP: + case TGSI_OPCODE_KIL: dst = nv30_sr(NV30SR_NONE, 0); dst.cc_update = 1; arith(fpc, 0, MOV, dst, MASK_ALL, src[0], none, none); diff --git a/src/gallium/drivers/nv40/nv40_fragprog.c b/src/gallium/drivers/nv40/nv40_fragprog.c index a361509e8f..91dcbebda0 100644 --- a/src/gallium/drivers/nv40/nv40_fragprog.c +++ b/src/gallium/drivers/nv40/nv40_fragprog.c @@ -533,10 +533,10 @@ nv40_fragprog_parse_instruction(struct nv40_fpc *fpc, case TGSI_OPCODE_FRC: arith(fpc, sat, FRC, dst, mask, src[0], none, none); break; - case TGSI_OPCODE_KIL: + case TGSI_OPCODE_KILP: arith(fpc, 0, KIL, none, 0, none, none, none); break; - case TGSI_OPCODE_KILP: + case TGSI_OPCODE_KIL: dst = nv40_sr(NV40SR_NONE, 0); dst.cc_update = 1; arith(fpc, 0, MOV, dst, MASK_ALL, src[0], none, none); -- cgit v1.2.3 From f90e50dff9e5cdbad6e9bb74c0aeaaaa82242b25 Mon Sep 17 00:00:00 2001 From: Stephane Marchesin Date: Mon, 18 Aug 2008 03:00:25 +0200 Subject: nv30: add some opcodes. --- src/gallium/drivers/nv30/nv30_fragprog.c | 11 +++++++++++ src/gallium/drivers/nv30/nv30_vertprog.c | 6 ++++++ 2 files changed, 17 insertions(+) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index d3693fdf56..02f30ad9ba 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -526,6 +526,12 @@ nv30_fragprog_parse_instruction(struct nv30_fpc *fpc, case TGSI_OPCODE_MUL: arith(fpc, sat, MUL, dst, mask, src[0], src[1], none); break; + case TGSI_OPCODE_NOISE1: + case TGSI_OPCODE_NOISE2: + case TGSI_OPCODE_NOISE3: + case TGSI_OPCODE_NOISE4: + arith(fpc, sat, SFL, dst, mask, none, none, none); + break; case TGSI_OPCODE_POW: arith(fpc, sat, POW, dst, mask, src[0], src[1], none); break; @@ -557,6 +563,9 @@ nv30_fragprog_parse_instruction(struct nv30_fpc *fpc, case TGSI_OPCODE_SGE: arith(fpc, sat, SGE, dst, mask, src[0], src[1], none); break; + case TGSI_OPCODE_SGT: + arith(fpc, sat, SGT, dst, mask, src[0], src[1], none); + break; case TGSI_OPCODE_SLT: arith(fpc, sat, SLT, dst, mask, src[0], src[1], none); break; @@ -730,6 +739,8 @@ nv30_fragprog_translate(struct nv30_context *nv30, struct tgsi_parse_context parse; struct nv30_fpc *fpc = NULL; + tgsi_dump(fp->pipe.tokens,0); + fpc = CALLOC(1, sizeof(struct nv30_fpc)); if (!fpc) return; diff --git a/src/gallium/drivers/nv30/nv30_vertprog.c b/src/gallium/drivers/nv30/nv30_vertprog.c index 6c075cdb75..72824559e8 100644 --- a/src/gallium/drivers/nv30/nv30_vertprog.c +++ b/src/gallium/drivers/nv30/nv30_vertprog.c @@ -4,6 +4,7 @@ #include "pipe/p_shader_tokens.h" #include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_dump.h" #include "nv30_context.h" #include "nv30_state.h" @@ -457,6 +458,9 @@ nv30_vertprog_parse_instruction(struct nv30_vpc *vpc, case TGSI_OPCODE_SGE: arith(vpc, 0, OP_SGE, dst, mask, src[0], src[1], none); break; + case TGSI_OPCODE_SGT: + arith(vpc, 0, OP_SGT, dst, mask, src[0], src[1], none); + break; case TGSI_OPCODE_SLT: arith(vpc, 0, OP_SLT, dst, mask, src[0], src[1], none); break; @@ -570,6 +574,8 @@ nv30_vertprog_translate(struct nv30_context *nv30, struct tgsi_parse_context parse; struct nv30_vpc *vpc = NULL; + tgsi_dump(vp->pipe.tokens,0); + vpc = CALLOC(1, sizeof(struct nv30_vpc)); if (!vpc) return; -- cgit v1.2.3 From 0c47bd0374e533f614ca3013f429fc32946e5be7 Mon Sep 17 00:00:00 2001 From: Patrice Mandin Date: Sun, 31 Aug 2008 11:10:08 +0200 Subject: nv30: set fp samplers with fragprog generation --- src/gallium/drivers/nv30/nv30_fragprog.c | 4 +++- src/gallium/drivers/nv30/nv30_state_emit.c | 4 ---- 2 files changed, 3 insertions(+), 5 deletions(-) (limited to 'src/gallium/drivers/nv30/nv30_fragprog.c') diff --git a/src/gallium/drivers/nv30/nv30_fragprog.c b/src/gallium/drivers/nv30/nv30_fragprog.c index 02f30ad9ba..320ba3f4bf 100644 --- a/src/gallium/drivers/nv30/nv30_fragprog.c +++ b/src/gallium/drivers/nv30/nv30_fragprog.c @@ -852,7 +852,7 @@ nv30_fragprog_validate(struct nv30_context *nv30) fp->buffer = ws->buffer_create(ws, 0x100, 0, fp->insn_len * 4); nv30_fragprog_upload(nv30, fp); - so = so_new(6, 1); + so = so_new(8, 1); so_method(so, nv30->screen->rankine, NV34TCL_FP_ACTIVE_PROGRAM, 1); so_reloc (so, fp->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_LOW | NOUVEAU_BO_OR, @@ -861,6 +861,8 @@ nv30_fragprog_validate(struct nv30_context *nv30) so_data (so, fp->fp_control); so_method(so, nv30->screen->rankine, NV34TCL_FP_REG_CONTROL, 1); so_data (so, fp->fp_reg_control); + so_method(so, nv30->screen->rankine, NV34TCL_TX_UNITS_ENABLE, 1); + so_data (so, fp->samplers); so_ref(so, &fp->so); update_constants: diff --git a/src/gallium/drivers/nv30/nv30_state_emit.c b/src/gallium/drivers/nv30/nv30_state_emit.c index 9c96085408..40fed621b2 100644 --- a/src/gallium/drivers/nv30/nv30_state_emit.c +++ b/src/gallium/drivers/nv30/nv30_state_emit.c @@ -71,10 +71,6 @@ nv30_state_emit(struct nv30_context *nv30) state->dirty = 0; - /* FIXME/TODO: Try to find a way to reemit only when changed */ - BEGIN_RING(rankine, NV34TCL_TX_UNITS_ENABLE, 1); - OUT_RING(state->fp_samplers); - so_emit_reloc_markers(nv30->nvws, state->hw[NV30_STATE_FB]); for (i = 0, samplers = state->fp_samplers; i < 16 && samplers; i++) { if (!(samplers & (1 << i))) -- cgit v1.2.3