diff options
Diffstat (limited to 'dist/Mesa/src/mesa/drivers/dri/nouveau/nouveau_shader_0.c')
| -rw-r--r-- | dist/Mesa/src/mesa/drivers/dri/nouveau/nouveau_shader_0.c | 1050 |
1 files changed, 1050 insertions, 0 deletions
diff --git a/dist/Mesa/src/mesa/drivers/dri/nouveau/nouveau_shader_0.c b/dist/Mesa/src/mesa/drivers/dri/nouveau/nouveau_shader_0.c new file mode 100644 index 000000000..8c203cc66 --- /dev/null +++ b/dist/Mesa/src/mesa/drivers/dri/nouveau/nouveau_shader_0.c @@ -0,0 +1,1050 @@ +/* + * Copyright (C) 2006 Ben Skeggs. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +/* + * Authors: + * Ben Skeggs <darktama@iinet.net.au> + */ + +#include "glheader.h" +#include "macros.h" +#include "enums.h" + +#include "shader/prog_instruction.h" +#include "shader/prog_parameter.h" +#include "shader/prog_statevars.h" +#include "shader/programopt.h" + +#include "nouveau_context.h" +#include "nouveau_shader.h" +#include "nouveau_msg.h" + +static nvsFixedReg _tx_mesa_vp_dst_reg[VERT_RESULT_MAX] = { + NVS_FR_POSITION, NVS_FR_COL0, NVS_FR_COL1, NVS_FR_FOGCOORD, + NVS_FR_TEXCOORD0, NVS_FR_TEXCOORD1, NVS_FR_TEXCOORD2, NVS_FR_TEXCOORD3, + NVS_FR_TEXCOORD4, NVS_FR_TEXCOORD5, NVS_FR_TEXCOORD6, NVS_FR_TEXCOORD7, + NVS_FR_POINTSZ, NVS_FR_BFC0, NVS_FR_BFC1, NVS_FR_UNKNOWN /* EDGE */ +}; + +static nvsFixedReg _tx_mesa_fp_dst_reg[FRAG_RESULT_MAX] = { + NVS_FR_FRAGDATA0 /* COLR */, NVS_FR_FRAGDATA0 /* COLH */, + NVS_FR_UNKNOWN /* DEPR */ +}; + +static nvsFixedReg _tx_mesa_fp_src_reg[FRAG_ATTRIB_MAX] = { + NVS_FR_POSITION, NVS_FR_COL0, NVS_FR_COL1, NVS_FR_FOGCOORD, + NVS_FR_TEXCOORD0, NVS_FR_TEXCOORD1, NVS_FR_TEXCOORD2, NVS_FR_TEXCOORD3, + NVS_FR_TEXCOORD4, NVS_FR_TEXCOORD5, NVS_FR_TEXCOORD6, NVS_FR_TEXCOORD7 +}; + +static nvsSwzComp _tx_mesa_swizzle[4] = { + NVS_SWZ_X, NVS_SWZ_Y, NVS_SWZ_Z, NVS_SWZ_W +}; + +static nvsOpcode _tx_mesa_opcode[] = { + [OPCODE_ABS] = NVS_OP_ABS, [OPCODE_ADD] = NVS_OP_ADD, + [OPCODE_ARA] = NVS_OP_ARA, [OPCODE_ARL] = NVS_OP_ARL, + [OPCODE_ARL_NV] = NVS_OP_ARL, [OPCODE_ARR] = NVS_OP_ARR, + [OPCODE_CMP] = NVS_OP_CMP, [OPCODE_COS] = NVS_OP_COS, + [OPCODE_DDX] = NVS_OP_DDX, [OPCODE_DDY] = NVS_OP_DDY, + [OPCODE_DP3] = NVS_OP_DP3, [OPCODE_DP4] = NVS_OP_DP4, + [OPCODE_DPH] = NVS_OP_DPH, [OPCODE_DST] = NVS_OP_DST, + [OPCODE_EX2] = NVS_OP_EX2, [OPCODE_EXP] = NVS_OP_EXP, + [OPCODE_FLR] = NVS_OP_FLR, [OPCODE_FRC] = NVS_OP_FRC, + [OPCODE_KIL] = NVS_OP_EMUL, [OPCODE_KIL_NV] = NVS_OP_KIL, + [OPCODE_LG2] = NVS_OP_LG2, [OPCODE_LIT] = NVS_OP_LIT, + [OPCODE_LOG] = NVS_OP_LOG, + [OPCODE_LRP] = NVS_OP_LRP, + [OPCODE_MAD] = NVS_OP_MAD, [OPCODE_MAX] = NVS_OP_MAX, + [OPCODE_MIN] = NVS_OP_MIN, [OPCODE_MOV] = NVS_OP_MOV, + [OPCODE_MUL] = NVS_OP_MUL, + [OPCODE_PK2H] = NVS_OP_PK2H, [OPCODE_PK2US] = NVS_OP_PK2US, + [OPCODE_PK4B] = NVS_OP_PK4B, [OPCODE_PK4UB] = NVS_OP_PK4UB, + [OPCODE_POW] = NVS_OP_POW, [OPCODE_POPA] = NVS_OP_POPA, + [OPCODE_PUSHA] = NVS_OP_PUSHA, + [OPCODE_RCC] = NVS_OP_RCC, [OPCODE_RCP] = NVS_OP_RCP, + [OPCODE_RFL] = NVS_OP_RFL, [OPCODE_RSQ] = NVS_OP_RSQ, + [OPCODE_SCS] = NVS_OP_SCS, [OPCODE_SEQ] = NVS_OP_SEQ, + [OPCODE_SFL] = NVS_OP_SFL, [OPCODE_SGE] = NVS_OP_SGE, + [OPCODE_SGT] = NVS_OP_SGT, [OPCODE_SIN] = NVS_OP_SIN, + [OPCODE_SLE] = NVS_OP_SLE, [OPCODE_SLT] = NVS_OP_SLT, + [OPCODE_SNE] = NVS_OP_SNE, [OPCODE_SSG] = NVS_OP_SSG, + [OPCODE_STR] = NVS_OP_STR, [OPCODE_SUB] = NVS_OP_SUB, + [OPCODE_SWZ] = NVS_OP_MOV, + [OPCODE_TEX] = NVS_OP_TEX, [OPCODE_TXB] = NVS_OP_TXB, + [OPCODE_TXD] = NVS_OP_TXD, + [OPCODE_TXL] = NVS_OP_TXL, [OPCODE_TXP] = NVS_OP_TXP, + [OPCODE_TXP_NV] = NVS_OP_TXP, + [OPCODE_UP2H] = NVS_OP_UP2H, [OPCODE_UP2US] = NVS_OP_UP2US, + [OPCODE_UP4B] = NVS_OP_UP4B, [OPCODE_UP4UB] = NVS_OP_UP4UB, + [OPCODE_X2D] = NVS_OP_X2D, + [OPCODE_XPD] = NVS_OP_XPD +}; + +static nvsCond _tx_mesa_condmask[] = { + NVS_COND_TR, /* workaround mesa not filling a valid value */ + NVS_COND_GT, NVS_COND_LT, NVS_COND_UN, NVS_COND_GE, + NVS_COND_LE, NVS_COND_NE, NVS_COND_NE, NVS_COND_TR, NVS_COND_FL +}; + +struct pass0_rec { + int nvs_ipos; + int next_temp; + + int mesa_const_base; + int mesa_const_last; + + int swzconst_done; + int swzconst_id; + nvsRegister const_half; +}; + +#define X NVS_SWZ_X +#define Y NVS_SWZ_Y +#define Z NVS_SWZ_Z +#define W NVS_SWZ_W + +#define FILL_CONDITION_FLAGS(fragment) do { \ + (fragment)->cond = \ + pass0_make_condmask(inst->DstReg.CondMask); \ + if ((fragment)->cond != NVS_COND_TR) \ + (fragment)->cond_test = 1; \ + (fragment)->cond_reg = inst->CondDst; \ + pass0_make_swizzle((fragment)->cond_swizzle, inst->DstReg.CondSwizzle);\ +} while(0) + +#define ARITH(op,dest,mask,sat,s0,s1,s2) do { \ + nvsinst = pass0_emit(nvs, parent, fpos, (op), \ + (dest), (mask), (sat), (s0), (s1), (s2));\ + FILL_CONDITION_FLAGS(nvsinst); \ +} while(0) + +#define ARITHu(op,dest,mask,sat,s0,s1,s2) do { \ + nvsinst = pass0_emit(nvs, parent, fpos, (op), \ + (dest), (mask), (sat), (s0), (s1), (s2));\ +} while(0) + +static void +pass0_append_fragment(nvsFragmentHeader *parent, + nvsFragmentHeader *fragment, + int pos) +{ + nvsFragmentHeader **head, **tail; + assert(parent && fragment); + + switch (parent->type) { + case NVS_BRANCH: + if (pos == 0) { + head = &((nvsBranch *)parent)->target_head; + tail = &((nvsBranch *)parent)->target_tail; + } else { + head = &((nvsBranch *)parent)->else_head; + tail = &((nvsBranch *)parent)->else_tail; + } + break; + case NVS_LOOP: + head = &((nvsLoop *)parent)->insn_head; + tail = &((nvsLoop *)parent)->insn_tail; + break; + case NVS_SUBROUTINE: + head = &((nvsSubroutine *)parent)->insn_head; + tail = &((nvsSubroutine *)parent)->insn_tail; + break; + default: + assert(0); + break; + } + + fragment->parent = parent; + fragment->prev = *tail; + fragment->next = NULL; + if (!(*head)) + *head = fragment; + else + (*tail)->next = fragment; + *tail = fragment; + +} + +static nvsSubroutine * +pass0_create_subroutine(nouveauShader *nvs, const char *label) +{ + nvsSubroutine *sub; + + sub = CALLOC_STRUCT(nvs_subroutine); + if (sub) { + sub->header.type = NVS_SUBROUTINE; + sub->label = strdup(label); + if (!nvs->program_tree) + nvs->program_tree = &sub->header; + else + pass0_append_fragment(nvs->program_tree, + &sub->header, 0); + } + + return sub; +} + +static void +pass0_make_reg(nouveauShader *nvs, nvsRegister *reg, + nvsRegFile file, unsigned int index) +{ + struct pass0_rec *rec = nvs->pass_rec; + + /* defaults */ + *reg = nvr_unused; + /* -1 == quick-and-dirty temp alloc */ + if (file == NVS_FILE_TEMP && index == -1) { + index = rec->next_temp++; + assert(index < NVS_MAX_TEMPS); + } + reg->file = file; + reg->index = index; +} + +static void +pass0_make_swizzle(nvsSwzComp *swz, unsigned int mesa) +{ + int i; + + for (i=0;i<4;i++) + swz[i] = _tx_mesa_swizzle[GET_SWZ(mesa, i)]; +} + +static nvsOpcode +pass0_make_opcode(enum prog_opcode op) +{ + if (op > MAX_OPCODE) + return NVS_OP_UNKNOWN; + return _tx_mesa_opcode[op]; +} + +static nvsCond +pass0_make_condmask(GLuint mesa) +{ + if (mesa > COND_FL) + return NVS_COND_UNKNOWN; + return _tx_mesa_condmask[mesa]; +} + +static unsigned int +pass0_make_mask(GLuint mesa_mask) +{ + unsigned int mask = 0; + + if (mesa_mask & WRITEMASK_X) mask |= SMASK_X; + if (mesa_mask & WRITEMASK_Y) mask |= SMASK_Y; + if (mesa_mask & WRITEMASK_Z) mask |= SMASK_Z; + if (mesa_mask & WRITEMASK_W) mask |= SMASK_W; + + return mask; +} + +static GLboolean +pass0_opcode_is_tex(enum prog_opcode op) +{ + switch (op) { + case OPCODE_TEX: + case OPCODE_TXB: + case OPCODE_TXD: + case OPCODE_TXL: + case OPCODE_TXP: + return GL_TRUE; + default: + break; + } + + return GL_FALSE; +} + +static nvsTexTarget +pass0_make_tex_target(GLuint mesa) +{ + switch (mesa) { + case TEXTURE_1D_INDEX: return NVS_TEX_TARGET_1D; + case TEXTURE_2D_INDEX: return NVS_TEX_TARGET_2D; + case TEXTURE_3D_INDEX: return NVS_TEX_TARGET_3D; + case TEXTURE_CUBE_INDEX: return NVS_TEX_TARGET_CUBE; + case TEXTURE_RECT_INDEX: return NVS_TEX_TARGET_RECT; + default: + return NVS_TEX_TARGET_UNKNOWN; + } +} + +static void +pass0_make_dst_reg(nvsPtr nvs, nvsRegister *reg, + struct prog_dst_register *dst) +{ + struct gl_program *mesa = (struct gl_program*)&nvs->mesa.vp; + nvsFixedReg sfr; + + switch (dst->File) { + case PROGRAM_OUTPUT: + if (mesa->Target == GL_VERTEX_PROGRAM_ARB) { + sfr = (dst->Index < VERT_RESULT_MAX) ? + _tx_mesa_vp_dst_reg[dst->Index] : + NVS_FR_UNKNOWN; + } else { + sfr = (dst->Index < FRAG_RESULT_MAX) ? + _tx_mesa_fp_dst_reg[dst->Index] : + NVS_FR_UNKNOWN; + } + pass0_make_reg(nvs, reg, NVS_FILE_RESULT, sfr); + break; + case PROGRAM_TEMPORARY: + pass0_make_reg(nvs, reg, NVS_FILE_TEMP, dst->Index); + break; + case PROGRAM_ADDRESS: + pass0_make_reg(nvs, reg, NVS_FILE_ADDRESS, dst->Index); + break; + default: + fprintf(stderr, "Unknown dest file %d\n", dst->File); + assert(0); + } +} + +static void +pass0_make_src_reg(nvsPtr nvs, nvsRegister *reg, struct prog_src_register *src) +{ + struct pass0_rec *rec = nvs->pass_rec; + struct gl_program *mesa = (struct gl_program *)&nvs->mesa.vp.Base; + int i; + + *reg = nvr_unused; + + switch (src->File) { + case PROGRAM_INPUT: + reg->file = NVS_FILE_ATTRIB; + if (mesa->Target == GL_VERTEX_PROGRAM_ARB) { + for (i=0; i<NVS_MAX_ATTRIBS; i++) { + if (nvs->vp_attrib_map[i] == src->Index) { + reg->index = i; + break; + } + } + if (i==NVS_MAX_ATTRIBS) + reg->index = NVS_FR_UNKNOWN; + } else { + reg->index = (src->Index < FRAG_ATTRIB_MAX) ? + _tx_mesa_fp_src_reg[src->Index] : + NVS_FR_UNKNOWN; + } + break; + case PROGRAM_STATE_VAR: + case PROGRAM_NAMED_PARAM: + case PROGRAM_CONSTANT: + reg->file = NVS_FILE_CONST; + reg->index = src->Index + rec->mesa_const_base; + reg->indexed = src->RelAddr; + if (reg->indexed) { + reg->addr_reg = 0; + reg->addr_comp = NVS_SWZ_X; + } + break; + case PROGRAM_TEMPORARY: + reg->file = NVS_FILE_TEMP; + reg->index = src->Index; + break; + default: + fprintf(stderr, "Unknown source type %d\n", src->File); + assert(0); + } + + /* per-component negate handled elsewhere */ + reg->negate = src->NegateBase != 0; + reg->abs = src->Abs; + pass0_make_swizzle(reg->swizzle, src->Swizzle); +} + +static nvsInstruction * +pass0_emit(nouveauShader *nvs, nvsFragmentHeader *parent, int fpos, + nvsOpcode op, nvsRegister dst, + unsigned int mask, int saturate, + nvsRegister src0, nvsRegister src1, nvsRegister src2) +{ + nvsInstruction *sif; + + sif = CALLOC_STRUCT(nvs_instruction); + if (!sif) + return NULL; + + /* Seems mesa doesn't explicitly 0 this.. */ + if (nvs->mesa.vp.Base.Target == GL_VERTEX_PROGRAM_ARB) + saturate = 0; + + sif->op = op; + sif->saturate = saturate; + sif->dest = dst; + sif->mask = mask; + sif->dest_scale = NVS_SCALE_1X; + sif->src[0] = src0; + sif->src[1] = src1; + sif->src[2] = src2; + sif->cond = COND_TR; + sif->cond_reg = 0; + sif->cond_test = 0; + sif->cond_update= 0; + pass0_make_swizzle(sif->cond_swizzle, SWIZZLE_NOOP); + pass0_append_fragment(parent, &sif->header, fpos); + + return sif; +} + +static void +pass0_fixup_swizzle(nvsPtr nvs, nvsFragmentHeader *parent, int fpos, + struct prog_src_register *src, + unsigned int sm1, + unsigned int sm2) +{ + static const float sc[4] = { 1.0, 0.0, -1.0, 0.0 }; + struct pass0_rec *rec = nvs->pass_rec; + int fixup_1, fixup_2; + nvsInstruction *nvsinst; + nvsRegister sr, dr = nvr_unused; + nvsRegister sm1const, sm2const; + + if (!rec->swzconst_done) { + struct gl_program *prog = &nvs->mesa.vp.Base; + GLuint swizzle; + rec->swzconst_id = _mesa_add_unnamed_constant(prog->Parameters, + sc, 4, &swizzle); + /* XXX what about swizzle? */ + rec->swzconst_done = 1; + COPY_4V(nvs->params[rec->swzconst_id].val, sc); + } + + fixup_1 = (sm1 != MAKE_SWIZZLE4(0,0,0,0) && + sm2 != MAKE_SWIZZLE4(2,2,2,2)); + fixup_2 = (sm2 != MAKE_SWIZZLE4(2,2,2,2)); + + if (src->File != PROGRAM_TEMPORARY && src->File != PROGRAM_INPUT) { + /* We can't use more than one const in an instruction, + * so move the const into a temp, and swizzle from there. + * + * TODO: should just emit the swizzled const, instead of + * swizzling it in the shader.. would need to reswizzle + * any state params when they change however.. + */ + pass0_make_reg(nvs, &dr, NVS_FILE_TEMP, -1); + pass0_make_src_reg(nvs, &sr, src); + ARITHu(NVS_OP_MOV, dr, SMASK_ALL, 0, + sr, nvr_unused, nvr_unused); + pass0_make_reg(nvs, &sr, NVS_FILE_TEMP, dr.index); + } else { + if (fixup_1) + src->NegateBase = 0; + pass0_make_src_reg(nvs, &sr, src); + pass0_make_reg(nvs, &dr, NVS_FILE_TEMP, -1); + } + + pass0_make_reg(nvs, &sm1const, NVS_FILE_CONST, rec->swzconst_id); + pass0_make_swizzle(sm1const.swizzle, sm1); + if (fixup_1 && fixup_2) { + /* Any combination with SWIZZLE_ONE */ + pass0_make_reg(nvs, &sm2const, + NVS_FILE_CONST, rec->swzconst_id); + pass0_make_swizzle(sm2const.swizzle, sm2); + ARITHu(NVS_OP_MAD, dr, SMASK_ALL, 0, sr, sm1const, sm2const); + } else { + /* SWIZZLE_ZERO || arbitrary negate */ + ARITHu(NVS_OP_MUL, dr, SMASK_ALL, 0, sr, sm1const, nvr_unused); + } + + src->File = PROGRAM_TEMPORARY; + src->Index = dr.index; + src->Swizzle = SWIZZLE_NOOP; +} + +#define SET_SWZ(fs, cp, c) fs = (fs & ~(0x7<<(cp*3))) | (c<<(cp*3)) +static void +pass0_check_sources(nvsPtr nvs, nvsFragmentHeader *parent, int fpos, + struct prog_instruction *inst) +{ + unsigned int insrc = -1, constsrc = -1; + int i; + + for (i=0;i<_mesa_num_inst_src_regs(inst->Opcode);i++) { + struct prog_src_register *src = &inst->SrcReg[i]; + unsigned int sm_1 = 0, sm_2 = 0; + nvsRegister sr, dr; + int do_mov = 0, c; + + /* Build up swizzle masks as if we were going to use + * "MAD new, src, const1, const2" to support arbitrary negation + * and SWIZZLE_ZERO/SWIZZLE_ONE. + */ + for (c=0;c<4;c++) { + if (GET_SWZ(src->Swizzle, c) == SWIZZLE_ZERO) { + SET_SWZ(sm_1, c, SWIZZLE_Y); /* 0.0 */ + SET_SWZ(sm_2, c, SWIZZLE_Y); + SET_SWZ(src->Swizzle, c, SWIZZLE_X); + } else if (GET_SWZ(src->Swizzle, c) == SWIZZLE_ONE) { + SET_SWZ(sm_1, c, SWIZZLE_Y); + if (src->NegateBase & (1<<c)) + SET_SWZ(sm_2, c, SWIZZLE_Z); /* -1.0 */ + else + SET_SWZ(sm_2, c, SWIZZLE_X); /* 1.0 */ + SET_SWZ(src->Swizzle, c, SWIZZLE_X); + } else { + if (src->NegateBase & (1<<c)) + SET_SWZ(sm_1, c, SWIZZLE_Z); /* -[xyzw] */ + else + SET_SWZ(sm_1, c, SWIZZLE_X); /*[xyzw]*/ + SET_SWZ(sm_2, c, SWIZZLE_Y); + } + } + + /* Unless we're multiplying by 1.0 or -1.0 on all components, + * and we're adding nothing to any component we have to + * emulate the swizzle. + */ + if ((sm_1 != MAKE_SWIZZLE4(0,0,0,0) && + sm_1 != MAKE_SWIZZLE4(2,2,2,2)) || + sm_2 != MAKE_SWIZZLE4(1,1,1,1)) { + pass0_fixup_swizzle(nvs, parent, fpos, src, sm_1, sm_2); + /* The source is definitely in a temp now, so don't + * bother checking for multiple ATTRIB/CONST regs. + */ + continue; + } + + /* HW can't use more than one ATTRIB or PARAM in a single + * instruction */ + switch (src->File) { + case PROGRAM_INPUT: + if (insrc != -1 && insrc != src->Index) + do_mov = 1; + else insrc = src->Index; + break; + case PROGRAM_STATE_VAR: + if (constsrc != -1 && constsrc != src->Index) + do_mov = 1; + else constsrc = src->Index; + break; + default: + break; + } + + /* Emit any extra ATTRIB/CONST to a temp, and modify the Mesa + * instruction to point at the temp. + */ + if (do_mov) { + pass0_make_src_reg(nvs, &sr, src); + pass0_make_reg(nvs, &dr, NVS_FILE_TEMP, -1); + pass0_emit(nvs, parent, fpos, NVS_OP_MOV, + dr, SMASK_ALL, 0, + sr, nvr_unused, nvr_unused); + + src->File = PROGRAM_TEMPORARY; + src->Index = dr.index; + src->Swizzle= SWIZZLE_NOOP; + } + } +} + +static GLboolean +pass0_emulate_instruction(nouveauShader *nvs, + nvsFragmentHeader *parent, int fpos, + struct prog_instruction *inst) +{ + nvsFunc *shader = nvs->func; + nvsRegister src[3], dest, temp; + nvsInstruction *nvsinst; + unsigned int mask = pass0_make_mask(inst->DstReg.WriteMask); + int i, sat; + + sat = (inst->SaturateMode == SATURATE_ZERO_ONE); + + /* Build all the "real" regs for the instruction */ + for (i=0; i<_mesa_num_inst_src_regs(inst->Opcode); i++) + pass0_make_src_reg(nvs, &src[i], &inst->SrcReg[i]); + if (inst->Opcode != OPCODE_KIL) + pass0_make_dst_reg(nvs, &dest, &inst->DstReg); + + switch (inst->Opcode) { + case OPCODE_ABS: + if (shader->caps & SCAP_SRC_ABS) + ARITH(NVS_OP_MOV, dest, mask, sat, + nvsAbs(src[0]), nvr_unused, nvr_unused); + else + ARITH(NVS_OP_MAX, dest, mask, sat, + src[0], nvsNegate(src[0]), nvr_unused); + break; + case OPCODE_CMP: + /*XXX: this will clobber CC0... */ + ARITH (NVS_OP_MOV, dest, mask, sat, + src[2], nvr_unused, nvr_unused); + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + ARITHu(NVS_OP_MOV, temp, SMASK_ALL, 0, + src[0], nvr_unused, nvr_unused); + nvsinst->cond_update = 1; + nvsinst->cond_reg = 0; + ARITH (NVS_OP_MOV, dest, mask, sat, + src[1], nvr_unused, nvr_unused); + nvsinst->cond = COND_LT; + nvsinst->cond_reg = 0; + nvsinst->cond_test = 1; + break; + case OPCODE_DPH: + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + ARITHu(NVS_OP_DP3, temp, SMASK_X, 0, + src[0], src[1], nvr_unused); + ARITH (NVS_OP_ADD, dest, mask, sat, + nvsSwizzle(temp, X, X, X, X), + nvsSwizzle(src[1], W, W, W, W), + nvr_unused); + break; + case OPCODE_KIL: + /* This is only in ARB shaders, so we don't have to worry + * about clobbering a CC reg as they aren't supported anyway. + *XXX: might have to worry with GLSL however... + */ + /* MOVC0 temp, src */ + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + ARITHu(NVS_OP_MOV, temp, SMASK_ALL, 0, + src[0], nvr_unused, nvr_unused); + nvsinst->cond_update = 1; + nvsinst->cond_reg = 0; + /* KIL_NV (LT0.xyzw) temp */ + ARITHu(NVS_OP_KIL, nvr_unused, 0, 0, + nvr_unused, nvr_unused, nvr_unused); + nvsinst->cond = COND_LT; + nvsinst->cond_reg = 0; + nvsinst->cond_test = 1; + pass0_make_swizzle(nvsinst->cond_swizzle, SWIZZLE_NOOP); + break; + case OPCODE_LRP: + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + ARITHu(NVS_OP_MAD, temp, mask, 0, + nvsNegate(src[0]), src[2], src[2]); + ARITH (NVS_OP_MAD, dest, mask, sat, src[0], src[1], temp); + break; + case OPCODE_POW: + if (shader->SupportsOpcode(shader, NVS_OP_LG2) && + shader->SupportsOpcode(shader, NVS_OP_EX2)) { + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + /* LG2 temp.x, src0.c */ + ARITHu(NVS_OP_LG2, temp, SMASK_X, 0, + nvsSwizzle(src[0], X, X, X, X), + nvr_unused, nvr_unused); + /* MUL temp.x, temp.x, src1.c */ + ARITHu(NVS_OP_MUL, temp, SMASK_X, 0, + nvsSwizzle(temp, X, X, X, X), + nvsSwizzle(src[1], X, X, X, X), + nvr_unused); + /* EX2 dest, temp.x */ + ARITH (NVS_OP_EX2, dest, mask, sat, + nvsSwizzle(temp, X, X, X, X), + nvr_unused, nvr_unused); + } else { + /* can we use EXP/LOG instead of EX2/LG2?? */ + fprintf(stderr, "Implement POW for NV20 vtxprog!\n"); + return GL_FALSE; + } + break; + case OPCODE_RSQ: + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + ARITHu(NVS_OP_LG2, temp, SMASK_X, 0, + nvsAbs(nvsSwizzle(src[0], X, X, X, X)), + nvr_unused, nvr_unused); + nvsinst->dest_scale = NVS_SCALE_INV_2X; + ARITH (NVS_OP_EX2, dest, mask, sat, + nvsNegate(nvsSwizzle(temp, X, X, X, X)), + nvr_unused, nvr_unused); + break; + case OPCODE_SCS: + if (mask & SMASK_X) + ARITH(NVS_OP_COS, dest, SMASK_X, sat, + nvsSwizzle(src[0], X, X, X, X), + nvr_unused, nvr_unused); + if (mask & SMASK_Y) + ARITH(NVS_OP_SIN, dest, SMASK_Y, sat, + nvsSwizzle(src[0], X, X, X, X), + nvr_unused, nvr_unused); + break; + case OPCODE_SUB: + ARITH(NVS_OP_ADD, dest, mask, sat, + src[0], nvsNegate(src[1]), nvr_unused); + break; + case OPCODE_XPD: + pass0_make_reg(nvs, &temp, NVS_FILE_TEMP, -1); + ARITHu(NVS_OP_MUL, temp, SMASK_ALL, 0, + nvsSwizzle(src[0], Z, X, Y, Y), + nvsSwizzle(src[1], Y, Z, X, X), + nvr_unused); + ARITH (NVS_OP_MAD, dest, (mask & ~SMASK_W), sat, + nvsSwizzle(src[0], Y, Z, X, X), + nvsSwizzle(src[1], Z, X, Y, Y), + nvsNegate(temp)); + break; + default: + WARN_ONCE("hw doesn't support opcode \"%s\"," + "and no emulation found\n", + _mesa_opcode_string(inst->Opcode)); + return GL_FALSE; + } + + return GL_TRUE; +} + +static GLboolean +pass0_translate_arith(nouveauShader *nvs, struct gl_program *prog, + int ipos, int fpos, + nvsFragmentHeader *parent) +{ + struct prog_instruction *inst = &prog->Instructions[ipos]; + nvsFunc *shader = nvs->func; + nvsInstruction *nvsinst; + GLboolean ret; + + /* Deal with multiple ATTRIB/PARAM in a single instruction */ + pass0_check_sources(nvs, parent, fpos, inst); + + /* Now it's safe to do the prog_instruction->nvsInstruction + * conversion + */ + if (shader->SupportsOpcode(shader, + pass0_make_opcode(inst->Opcode))) { + nvsRegister src[3], dest; + int i; + + for (i=0; i<_mesa_num_inst_src_regs(inst->Opcode); i++) + pass0_make_src_reg(nvs, &src[i], &inst->SrcReg[i]); + pass0_make_dst_reg(nvs, &dest, &inst->DstReg); + + ARITH(pass0_make_opcode(inst->Opcode), dest, + pass0_make_mask(inst->DstReg.WriteMask), + (inst->SaturateMode != SATURATE_OFF), + src[0], src[1], src[2]); + nvsinst->tex_unit = inst->TexSrcUnit; + if (pass0_opcode_is_tex(inst->Opcode)) + nvsinst->tex_target = + pass0_make_tex_target(inst->TexSrcTarget); + else + nvsinst->tex_target = NVS_TEX_TARGET_UNKNOWN; + + ret = GL_TRUE; + } else + ret = pass0_emulate_instruction(nvs, parent, fpos, inst); + + return ret; +} + +static GLboolean +pass0_translate_instructions(nouveauShader *nvs, int ipos, int fpos, + nvsFragmentHeader *parent) +{ + struct gl_program *prog = (struct gl_program *)&nvs->mesa.vp; + + while (1) { + struct prog_instruction *inst = &prog->Instructions[ipos]; + + switch (inst->Opcode) { + case OPCODE_END: + return GL_TRUE; + case OPCODE_BRA: + case OPCODE_CAL: + case OPCODE_RET: + //case OPCODE_LOOP: + //case OPCODE_ENDLOOP: + //case OPCODE_IF: + //case OPCODE_ELSE: + //case OPCODE_ENDIF: + WARN_ONCE("branch ops unimplemented\n"); + return GL_FALSE; + break; + default: + if (!pass0_translate_arith(nvs, prog, + ipos, fpos, parent)) + return GL_FALSE; + break; + } + + ipos++; + } + + return GL_TRUE; +} + +static void +pass0_build_attrib_map(nouveauShader *nvs, struct gl_vertex_program *vp) +{ + GLuint inputs_read = vp->Base.InputsRead; + GLuint input_alloc = ~0xFFFF; + int i; + + for (i=0; i<NVS_MAX_ATTRIBS; i++) + nvs->vp_attrib_map[i] = -1; + + while (inputs_read) { + int in = ffs(inputs_read) - 1; + int hw; + inputs_read &= ~(1<<in); + + if (vp->IsNVProgram) { + /* NVvp: must alias */ + if (in >= VERT_ATTRIB_GENERIC0) + hw = in - VERT_ATTRIB_GENERIC0; + else + hw = in; + } else { + /* ARBvp: may alias (but we won't) + * GL2.0: must not alias + */ + if (in >= VERT_ATTRIB_GENERIC0) + hw = ffs(~input_alloc) - 1; + else + hw = in; + input_alloc |= (1<<hw); + } + + nvs->vp_attrib_map[hw] = in; + } + + if (NOUVEAU_DEBUG & DEBUG_SHADERS) { + printf("vtxprog attrib map:\n"); + for (i=0; i<NVS_MAX_ATTRIBS; i++) { + printf(" hw:%d = attrib:%d\n", + i, nvs->vp_attrib_map[i]); + } + } +} + +static void +pass0_vp_insert_ff_clip_planes(GLcontext *ctx, nouveauShader *nvs) +{ + struct gl_program *prog = &nvs->mesa.vp.Base; + nvsFragmentHeader *parent = nvs->program_tree; + nvsInstruction *nvsinst; + GLuint fpos = 0; + nvsRegister opos, epos, eqn, mv[4]; + gl_state_index tokens[STATE_LENGTH] + = { STATE_MODELVIEW_MATRIX, 0, 0, 0, 0 }; + GLint id; + int i; + + /* modelview transform */ + pass0_make_reg(nvs, &opos, NVS_FILE_ATTRIB, NVS_FR_POSITION); + pass0_make_reg(nvs, &epos, NVS_FILE_TEMP , -1); + for (i=0; i<4; i++) { + tokens[2] = tokens[3] = i; + id = _mesa_add_state_reference(prog->Parameters, tokens); + pass0_make_reg(nvs, &mv[i], NVS_FILE_CONST, id); + } + ARITHu(NVS_OP_DP4, epos, SMASK_X, 0, opos, mv[0], nvr_unused); + ARITHu(NVS_OP_DP4, epos, SMASK_Y, 0, opos, mv[1], nvr_unused); + ARITHu(NVS_OP_DP4, epos, SMASK_Z, 0, opos, mv[2], nvr_unused); + ARITHu(NVS_OP_DP4, epos, SMASK_W, 0, opos, mv[3], nvr_unused); + + /* Emit code to emulate fixed-function glClipPlane */ + for (i=0; i<6; i++) { + GLuint clipmask = SMASK_X; + nvsRegister clip; + + if (!(ctx->Transform.ClipPlanesEnabled & (1<<i))) + continue; + + /* Point a const at a user clipping plane */ + tokens[0] = STATE_CLIPPLANE; + tokens[1] = i; + id = _mesa_add_state_reference(prog->Parameters, tokens); + pass0_make_reg(nvs, &eqn , NVS_FILE_CONST , id); + pass0_make_reg(nvs, &clip, NVS_FILE_RESULT, NVS_FR_CLIP0 + i); + + /*XXX: something else needs to take care of modifying the + * instructions to write to the correct hw clip register. + */ + switch (i) { + case 0: case 3: clipmask = SMASK_Y; break; + case 1: case 4: clipmask = SMASK_Z; break; + case 2: case 5: clipmask = SMASK_W; break; + } + + /* Emit transform */ + ARITHu(NVS_OP_DP4, clip, clipmask, 0, epos, eqn, nvr_unused); + } +} + +static void +pass0_rebase_mesa_consts(nouveauShader *nvs) +{ + struct pass0_rec *rec = nvs->pass_rec; + struct gl_program *prog = &nvs->mesa.vp.Base; + struct prog_instruction *inst = prog->Instructions; + int i; + + /*XXX: not a good idea, params->hw_index is malloc'd */ + memset(nvs->params, 0x00, sizeof(nvs->params)); + + /* When doing relative addressing on constants, the hardware needs us + * to fill the "const id" field with a positive value. Determine the + * most negative index that is used so that all accesses to a + * mesa-provided constant can be rebased to a positive index. + */ + while (inst->Opcode != OPCODE_END) { + for (i=0; i<_mesa_num_inst_src_regs(inst->Opcode); i++) { + struct prog_src_register *src = &inst->SrcReg[i]; + + switch (src->File) { + case PROGRAM_STATE_VAR: + case PROGRAM_CONSTANT: + case PROGRAM_NAMED_PARAM: + if (src->RelAddr && src->Index < 0) { + int base = src->Index * -1; + if (rec->mesa_const_base < base) + rec->mesa_const_base = base; + } + break; + default: + break; + } + } + + inst++; + } +} + +static GLboolean +pass0_resolve_mesa_consts(nouveauShader *nvs) +{ + struct pass0_rec *rec = nvs->pass_rec; + struct gl_program *prog = &nvs->mesa.vp.Base; + struct gl_program_parameter_list *plist = prog->Parameters; + int i; + + /* Init all const tracking/alloc info from the parameter list, rather + * than doing it as we translate the program. Otherwise: + * 1) we can't get at the correct constant info when relative + * addressing is being used due to src->Index not pointing + * at the exact const; + * 2) as we add extra consts to the program, mesa will call realloc() + * and we get invalid pointers to the const data. + */ + rec->mesa_const_last = plist->NumParameters + rec->mesa_const_base; + nvs->param_high = rec->mesa_const_last; + for (i=0; i<plist->NumParameters; i++) { + int hw = rec->mesa_const_base + i; + + if (hw > NVS_MAX_CONSTS) { + nvsProgramError(nvs, "hw = %d > NVS_MAX_CONSTS!\n", hw); + return GL_FALSE; + } + + switch (plist->Parameters[i].Type) { + case PROGRAM_NAMED_PARAM: + case PROGRAM_STATE_VAR: + nvs->params[hw].in_use = GL_TRUE; + nvs->params[hw].source_val = plist->ParameterValues[i]; + COPY_4V(nvs->params[hw].val, plist->ParameterValues[i]); + break; + case PROGRAM_CONSTANT: + nvs->params[hw].in_use = GL_TRUE; + nvs->params[hw].source_val = NULL; + COPY_4V(nvs->params[hw].val, plist->ParameterValues[i]); + break; + default: + nvsProgramError(nvs, "hit bad type=%d on param %d\n", + plist->Parameters[i].Type, i); + return GL_FALSE; + } + } + + return GL_TRUE; +} + +GLboolean +nouveau_shader_pass0(GLcontext *ctx, nouveauShader *nvs) +{ + nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx); + struct gl_program *prog = (struct gl_program*)nvs; + struct gl_vertex_program *vp = (struct gl_vertex_program *)prog; + struct gl_fragment_program *fp = (struct gl_fragment_program *)prog; + struct pass0_rec *rec; + int ret = GL_FALSE; + + NVSDBG("start: nvs=%p\n", nvs); + + /* Previously detected an error, and haven't recieved new program + * string, so fail immediately. + */ + if (nvs->error) { + NVSDBG("failed previous compile attempt, not retrying\n"); + return GL_FALSE; + } + + rec = CALLOC_STRUCT(pass0_rec); + if (!rec) + return GL_FALSE; + + rec->next_temp = prog->NumTemporaries; + nvs->pass_rec = rec; + + nvs->program_tree = (nvsFragmentHeader*) + pass0_create_subroutine(nvs, "program body"); + if (!nvs->program_tree) { + FREE(rec); + return GL_FALSE; + } + + switch (prog->Target) { + case GL_VERTEX_PROGRAM_ARB: + nvs->func = &nmesa->VPfunc; + + if (vp->IsPositionInvariant) + _mesa_insert_mvp_code(ctx, vp); + pass0_rebase_mesa_consts(nvs); + + if (!prog->String && ctx->Transform.ClipPlanesEnabled) + pass0_vp_insert_ff_clip_planes(ctx, nvs); + + pass0_build_attrib_map(nvs, vp); + break; + case GL_FRAGMENT_PROGRAM_ARB: + nvs->func = &nmesa->FPfunc; + + if (fp->FogOption != GL_NONE) + _mesa_append_fog_code(ctx, fp); + pass0_rebase_mesa_consts(nvs); + break; + default: + fprintf(stderr, "Unknown program type %d", prog->Target); + FREE(rec); + /* DESTROY TREE!! */ + return GL_FALSE; + } + nvs->func->card_priv = &nvs->card_priv; + + ret = pass0_translate_instructions(nvs, 0, 0, nvs->program_tree); + if (ret) + ret = pass0_resolve_mesa_consts(nvs); + + /*XXX: if (!ret) DESTROY TREE!!! */ + + FREE(rec); + return ret; +} + |