/* * Copyright 2010 Christoph Bumiller * * 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 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 AUTHORS 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. */ /* #define NV50PC_DEBUG */ #include "nv50_pc.h" #include "nv50_program.h" #include /* returns TRUE if operands 0 and 1 can be swapped */ boolean nv_op_commutative(uint opcode) { switch (opcode) { case NV_OP_ADD: case NV_OP_MUL: case NV_OP_MAD: case NV_OP_AND: case NV_OP_OR: case NV_OP_XOR: case NV_OP_MIN: case NV_OP_MAX: case NV_OP_SAD: return TRUE; default: return FALSE; } } /* return operand to which the address register applies */ int nv50_indirect_opnd(struct nv_instruction *i) { if (!i->src[4]) return -1; switch (i->opcode) { case NV_OP_MOV: case NV_OP_LDA: case NV_OP_STA: return 0; default: return 1; } } boolean nv50_nvi_can_use_imm(struct nv_instruction *nvi, int s) { if (nvi->flags_src || nvi->flags_def) return FALSE; switch (nvi->opcode) { case NV_OP_ADD: case NV_OP_MUL: case NV_OP_AND: case NV_OP_OR: case NV_OP_XOR: case NV_OP_SHL: case NV_OP_SHR: return (s == 1) && (nvi->src[0]->value->reg.file == NV_FILE_GPR) && (nvi->def[0]->reg.file == NV_FILE_GPR); case NV_OP_MOV: assert(s == 0); return (nvi->def[0]->reg.file == NV_FILE_GPR); default: return FALSE; } } boolean nv50_nvi_can_load(struct nv_instruction *nvi, int s, struct nv_value *value) { int i; for (i = 0; i < 3 && nvi->src[i]; ++i) if (nvi->src[i]->value->reg.file == NV_FILE_IMM) return FALSE; switch (nvi->opcode) { case NV_OP_ABS: case NV_OP_ADD: case NV_OP_CEIL: case NV_OP_FLOOR: case NV_OP_TRUNC: case NV_OP_CVT: case NV_OP_NEG: case NV_OP_MAD: case NV_OP_MUL: case NV_OP_SAT: case NV_OP_SUB: case NV_OP_MAX: case NV_OP_MIN: if (s == 0 && (value->reg.file == NV_FILE_MEM_S || value->reg.file == NV_FILE_MEM_P)) return TRUE; if (value->reg.file < NV_FILE_MEM_C(0) || value->reg.file > NV_FILE_MEM_C(15)) return FALSE; return (s == 1) || ((s == 2) && (nvi->src[1]->value->reg.file == NV_FILE_GPR)); case NV_OP_MOV: assert(s == 0); return /* TRUE */ FALSE; /* don't turn MOVs into loads */ default: return FALSE; } } /* Return whether this instruction can be executed conditionally. */ boolean nv50_nvi_can_predicate(struct nv_instruction *nvi) { int i; if (nvi->flags_src) return FALSE; for (i = 0; i < 4 && nvi->src[i]; ++i) if (nvi->src[i]->value->reg.file == NV_FILE_IMM) return FALSE; return TRUE; } ubyte nv50_supported_src_mods(uint opcode, int s) { switch (opcode) { case NV_OP_ABS: return NV_MOD_NEG | NV_MOD_ABS; /* obviously */ case NV_OP_ADD: case NV_OP_MUL: case NV_OP_MAD: return NV_MOD_NEG; case NV_OP_DFDX: case NV_OP_DFDY: assert(s == 0); return NV_MOD_NEG; case NV_OP_MAX: case NV_OP_MIN: return NV_MOD_ABS; case NV_OP_CVT: case NV_OP_LG2: case NV_OP_NEG: case NV_OP_PREEX2: case NV_OP_PRESIN: case NV_OP_RCP: case NV_OP_RSQ: return NV_MOD_ABS | NV_MOD_NEG; default: return 0; } } /* We may want an opcode table. */ boolean nv50_op_can_write_flags(uint opcode) { if (nv_is_vector_op(opcode)) return FALSE; switch (opcode) { /* obvious ones like KIL, CALL, etc. not included */ case NV_OP_PHI: case NV_OP_MOV: case NV_OP_LINTERP: case NV_OP_PINTERP: case NV_OP_LDA: return FALSE; default: break; } if (opcode >= NV_OP_RCP && opcode <= NV_OP_PREEX2) return FALSE; return TRUE; } int nv_nvi_refcount(struct nv_instruction *nvi) { int i, rc; rc = nvi->flags_def ? nvi->flags_def->refc : 0; for (i = 0; i < 4; ++i) { if (!nvi->def[i]) return rc; rc += nvi->def[i]->refc; } return rc; } int nvcg_replace_value(struct nv_pc *pc, struct nv_value *old_val, struct nv_value *new_val) { int i, n; if (old_val == new_val) return old_val->refc; for (i = 0, n = 0; i < pc->num_refs; ++i) { if (pc->refs[i]->value == old_val) { ++n; nv_reference(pc, &pc->refs[i], new_val); } } return n; } struct nv_value * nvcg_find_constant(struct nv_ref *ref) { struct nv_value *src; if (!ref) return NULL; src = ref->value; while (src->insn && src->insn->opcode == NV_OP_MOV) { assert(!src->insn->src[0]->mod); src = src->insn->src[0]->value; } if ((src->reg.file == NV_FILE_IMM) || (src->insn && src->insn->opcode == NV_OP_LDA && src->insn->src[0]->value->reg.file >= NV_FILE_MEM_C(0) && src->insn->src[0]->value->reg.file <= NV_FILE_MEM_C(15))) return src; return NULL; } struct nv_value * nvcg_find_immediate(struct nv_ref *ref) { struct nv_value *src = nvcg_find_constant(ref); return (src && src->reg.file == NV_FILE_IMM) ? src : NULL; } static void nv_pc_free_refs(struct nv_pc *pc) { int i; for (i = 0; i < pc->num_refs; i += 64) FREE(pc->refs[i]); FREE(pc->refs); } static const char * edge_name(ubyte type) { switch (type) { case CFG_EDGE_FORWARD: return "forward"; case CFG_EDGE_BACK: return "back"; case CFG_EDGE_LOOP_ENTER: return "loop"; case CFG_EDGE_LOOP_LEAVE: return "break"; case CFG_EDGE_FAKE: return "fake"; default: return "?"; } } void nv_pc_pass_in_order(struct nv_basic_block *root, nv_pc_pass_func f, void *priv) { struct nv_basic_block *bb[64], *bbb[16], *b; int j, p, pp; bb[0] = root; p = 1; pp = 0; while (p > 0) { b = bb[--p]; b->priv = 0; for (j = 1; j >= 0; --j) { if (!b->out[j]) continue; switch (b->out_kind[j]) { case CFG_EDGE_BACK: continue; case CFG_EDGE_FORWARD: case CFG_EDGE_FAKE: if (++b->out[j]->priv == b->out[j]->num_in) bb[p++] = b->out[j]; break; case CFG_EDGE_LOOP_ENTER: bb[p++] = b->out[j]; break; case CFG_EDGE_LOOP_LEAVE: bbb[pp++] = b->out[j]; break; default: assert(0); break; } } f(priv, b); if (!p) { p = pp; for (; pp > 0; --pp) bb[pp - 1] = bbb[pp - 1]; } } } static void nv_do_print_function(void *priv, struct nv_basic_block *b) { struct nv_instruction *i = b->phi; debug_printf("=== BB %i ", b->id); if (b->out[0]) debug_printf("[%s -> %i] ", edge_name(b->out_kind[0]), b->out[0]->id); if (b->out[1]) debug_printf("[%s -> %i] ", edge_name(b->out_kind[1]), b->out[1]->id); debug_printf("===\n"); i = b->phi; if (!i) i = b->entry; for (; i; i = i->next) nv_print_instruction(i); } void nv_print_function(struct nv_basic_block *root) { if (root->subroutine) debug_printf("SUBROUTINE %i\n", root->subroutine); else debug_printf("MAIN\n"); nv_pc_pass_in_order(root, nv_do_print_function, root); } void nv_print_program(struct nv_pc *pc) { int i; for (i = 0; i < pc->num_subroutines + 1; ++i) if (pc->root[i]) nv_print_function(pc->root[i]); } #ifdef NV50PC_DEBUG static void nv_do_print_cfgraph(struct nv_pc *pc, FILE *f, struct nv_basic_block *b) { int i; b->pass_seq = pc->pass_seq; fprintf(f, "\t%i [shape=box]\n", b->id); for (i = 0; i < 2; ++i) { if (!b->out[i]) continue; switch (b->out_kind[i]) { case CFG_EDGE_FORWARD: fprintf(f, "\t%i -> %i;\n", b->id, b->out[i]->id); break; case CFG_EDGE_LOOP_ENTER: fprintf(f, "\t%i -> %i [color=green];\n", b->id, b->out[i]->id); break; case CFG_EDGE_LOOP_LEAVE: fprintf(f, "\t%i -> %i [color=red];\n", b->id, b->out[i]->id); break; case CFG_EDGE_BACK: fprintf(f, "\t%i -> %i;\n", b->id, b->out[i]->id); continue; case CFG_EDGE_FAKE: fprintf(f, "\t%i -> %i [style=dotted];\n", b->id, b->out[i]->id); break; default: assert(0); break; } if (b->out[i]->pass_seq < pc->pass_seq) nv_do_print_cfgraph(pc, f, b->out[i]); } } /* Print the control flow graph of subroutine @subr (0 == MAIN) to a file. */ static void nv_print_cfgraph(struct nv_pc *pc, const char *filepath, int subr) { FILE *f; f = fopen(filepath, "a"); if (!f) return; fprintf(f, "digraph G {\n"); ++pc->pass_seq; nv_do_print_cfgraph(pc, f, pc->root[subr]); fprintf(f, "}\n"); fclose(f); } #endif static INLINE void nvcg_show_bincode(struct nv_pc *pc) { unsigned i; for (i = 0; i < pc->bin_size / 4; ++i) { debug_printf("0x%08x ", pc->emit[i]); if ((i % 16) == 15) debug_printf("\n"); } debug_printf("\n"); } static int nv50_emit_program(struct nv_pc *pc) { uint32_t *code = pc->emit; int n; NV50_DBGMSG("emitting program: size = %u\n", pc->bin_size); for (n = 0; n < pc->num_blocks; ++n) { struct nv_instruction *i; struct nv_basic_block *b = pc->bb_list[n]; for (i = b->entry; i; i = i->next) { nv50_emit_instruction(pc, i); pc->bin_pos += 1 + (pc->emit[0] & 1); pc->emit += 1 + (pc->emit[0] & 1); } } assert(pc->emit == &code[pc->bin_size / 4]); /* XXX: we can do better than this ... */ if (!(pc->emit[-2] & 1) || (pc->emit[-2] & 2) || (pc->emit[-1] & 3)) { pc->emit[0] = 0xf0000001; pc->emit[1] = 0xe0000000; pc->bin_size += 8; } pc->emit = code; code[pc->bin_size / 4 - 1] |= 1; #ifdef NV50PC_DEBUG nvcg_show_bincode(pc); #endif return 0; } int nv50_generate_code(struct nv50_translation_info *ti) { struct nv_pc *pc; int ret; int i; pc = CALLOC_STRUCT(nv_pc); if (!pc) return 1; pc->root = CALLOC(ti->subr_nr + 1, sizeof(pc->root[0])); if (!pc->root) { FREE(pc); return 1; } pc->num_subroutines = ti->subr_nr; ret = nv50_tgsi_to_nc(pc, ti); if (ret) goto out; #ifdef NV50PC_DEBUG nv_print_program(pc); #endif pc->opt_reload_elim = ti->store_to_memory ? FALSE : TRUE; /* optimization */ ret = nv_pc_exec_pass0(pc); if (ret) goto out; #ifdef NV50PC_DEBUG nv_print_program(pc); #endif /* register allocation */ ret = nv_pc_exec_pass1(pc); if (ret) goto out; #ifdef NV50PC_DEBUG nv_print_program(pc); nv_print_cfgraph(pc, "nv50_shader_cfgraph.dot", 0); #endif /* prepare for emission */ ret = nv_pc_exec_pass2(pc); if (ret) goto out; pc->emit = CALLOC(pc->bin_size / 4 + 2, 4); if (!pc->emit) { ret = 3; goto out; } ret = nv50_emit_program(pc); if (ret) goto out; ti->p->code_size = pc->bin_size; ti->p->code = pc->emit; ti->p->immd_size = pc->immd_count * 4; ti->p->immd = pc->immd_buf; /* highest 16 bit reg to num of 32 bit regs */ ti->p->max_gpr = (pc->max_reg[NV_FILE_GPR] >> 1) + 1; ti->p->fixups = pc->fixups; ti->p->num_fixups = pc->num_fixups; NV50_DBGMSG("SHADER TRANSLATION - %s\n", ret ? "failure" : "success"); out: nv_pc_free_refs(pc); for (i = 0; i < pc->num_blocks; ++i) FREE(pc->bb_list[i]); if (pc->root) FREE(pc->root); if (ret) { /* on success, these will be referenced by nv50_program */ if (pc->emit) FREE(pc->emit); if (pc->immd_buf) FREE(pc->immd_buf); if (pc->fixups) FREE(pc->fixups); } FREE(pc); return ret; } static void nvbb_insert_phi(struct nv_basic_block *b, struct nv_instruction *i) { if (!b->phi) { i->prev = NULL; b->phi = i; i->next = b->entry; if (b->entry) { assert(!b->entry->prev && b->exit); b->entry->prev = i; } else { b->entry = i; b->exit = i; } } else { assert(b->entry); if (b->entry->opcode == NV_OP_PHI) { /* insert after entry */ assert(b->entry == b->exit); b->entry->next = i; i->prev = b->entry; b->entry = i; b->exit = i; } else { /* insert before entry */ assert(b->entry->prev && b->exit); i->next = b->entry; i->prev = b->entry->prev; b->entry->prev = i; i->prev->next = i; } } } void nvbb_insert_tail(struct nv_basic_block *b, struct nv_instruction *i) { if (i->opcode == NV_OP_PHI) { nvbb_insert_phi(b, i); } else { i->prev = b->exit; if (b->exit) b->exit->next = i; b->exit = i; if (!b->entry) b->entry = i; else if (i->prev && i->prev->opcode == NV_OP_PHI) b->entry = i; } i->bb = b; b->num_instructions++; } void nvi_insert_after(struct nv_instruction *at, struct nv_instruction *ni) { if (!at->next) { nvbb_insert_tail(at->bb, ni); return; } ni->next = at->next; ni->prev = at; ni->next->prev = ni; ni->prev->next = ni; } void nv_nvi_delete(struct nv_instruction *nvi) { struct nv_basic_block *b = nvi->bb; int j; /* debug_printf("REM: "); nv_print_instruction(nvi); */ for (j = 0; j < 5; ++j) nv_reference(NULL, &nvi->src[j], NULL); nv_reference(NULL, &nvi->flags_src, NULL); if (nvi->next) nvi->next->prev = nvi->prev; else { assert(nvi == b->exit); b->exit = nvi->prev; } if (nvi->prev) nvi->prev->next = nvi->next; if (nvi == b->entry) { /* PHIs don't get hooked to b->entry */ b->entry = nvi->next; assert(!nvi->prev || nvi->prev->opcode == NV_OP_PHI); } if (nvi == b->phi) { if (nvi->opcode != NV_OP_PHI) NV50_DBGMSG("NOTE: b->phi points to non-PHI instruction\n"); assert(!nvi->prev); if (!nvi->next || nvi->next->opcode != NV_OP_PHI) b->phi = NULL; else b->phi = nvi->next; } } void nv_nvi_permute(struct nv_instruction *i1, struct nv_instruction *i2) { struct nv_basic_block *b = i1->bb; assert(i1->opcode != NV_OP_PHI && i2->opcode != NV_OP_PHI); assert(i1->next == i2); if (b->exit == i2) b->exit = i1; if (b->entry == i1) b->entry = i2; i2->prev = i1->prev; i1->next = i2->next; i2->next = i1; i1->prev = i2; if (i2->prev) i2->prev->next = i2; if (i1->next) i1->next->prev = i1; } void nvbb_attach_block(struct nv_basic_block *parent, struct nv_basic_block *b, ubyte edge_kind) { assert(b->num_in < 8); if (parent->out[0]) { assert(!parent->out[1]); parent->out[1] = b; parent->out_kind[1] = edge_kind; } else { parent->out[0] = b; parent->out_kind[0] = edge_kind; } b->in[b->num_in] = parent; b->in_kind[b->num_in++] = edge_kind; } /* NOTE: all BRKs are treated as conditional, so there are 2 outgoing BBs */ boolean nvbb_dominated_by(struct nv_basic_block *b, struct nv_basic_block *d) { int j; if (b == d) return TRUE; for (j = 0; j < b->num_in; ++j) if ((b->in_kind[j] != CFG_EDGE_BACK) && !nvbb_dominated_by(b->in[j], d)) return FALSE; return j ? TRUE : FALSE; } /* check if @bf (future) can be reached from @bp (past), stop at @bt */ boolean nvbb_reachable_by(struct nv_basic_block *bf, struct nv_basic_block *bp, struct nv_basic_block *bt) { struct nv_basic_block *q[NV_PC_MAX_BASIC_BLOCKS], *b; int i, p, n; p = 0; n = 1; q[0] = bp; while (p < n) { b = q[p++]; if (b == bf) break; if (b == bt) continue; assert(n <= (1024 - 2)); for (i = 0; i < 2; ++i) { if (b->out[i] && !IS_WALL_EDGE(b->out_kind[i]) && !b->out[i]->priv) { q[n] = b->out[i]; q[n++]->priv = 1; } } } for (--n; n >= 0; --n) q[n]->priv = 0; return (b == bf); } static struct nv_basic_block * nvbb_find_dom_frontier(struct nv_basic_block *b, struct nv_basic_block *df) { struct nv_basic_block *out; int i; if (!nvbb_dominated_by(df, b)) { for (i = 0; i < df->num_in; ++i) { if (df->in_kind[i] == CFG_EDGE_BACK) continue; if (nvbb_dominated_by(df->in[i], b)) return df; } } for (i = 0; i < 2 && df->out[i]; ++i) { if (df->out_kind[i] == CFG_EDGE_BACK) continue; if ((out = nvbb_find_dom_frontier(b, df->out[i]))) return out; } return NULL; } struct nv_basic_block * nvbb_dom_frontier(struct nv_basic_block *b) { struct nv_basic_block *df; int i; for (i = 0; i < 2 && b->out[i]; ++i) if ((df = nvbb_find_dom_frontier(b, b->out[i]))) return df; return NULL; }