diff options
Diffstat (limited to 'src/gallium')
| -rw-r--r-- | src/gallium/drivers/freedreno/Makefile.sources | 2 | ||||
| -rw-r--r-- | src/gallium/drivers/freedreno/freedreno_screen.c | 1 | ||||
| -rw-r--r-- | src/gallium/drivers/freedreno/freedreno_util.h | 1 | ||||
| -rw-r--r-- | src/gallium/drivers/freedreno/ir3/ir3.h | 10 | ||||
| -rw-r--r-- | src/gallium/drivers/freedreno/ir3/ir3_cmdline.c | 3 | ||||
| -rw-r--r-- | src/gallium/drivers/freedreno/ir3/ir3_compiler.c | 34 | ||||
| -rw-r--r-- | src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c | 32 | ||||
| -rw-r--r-- | src/gallium/drivers/freedreno/ir3/ir3_dump.c | 456 | ||||
| -rw-r--r-- | src/gallium/drivers/freedreno/ir3/ir3_print.c | 210 | ||||
| -rw-r--r-- | src/gallium/drivers/freedreno/ir3/ir3_ra.c | 4 |
10 files changed, 228 insertions, 525 deletions
diff --git a/src/gallium/drivers/freedreno/Makefile.sources b/src/gallium/drivers/freedreno/Makefile.sources index a565a9c4e4d..809d1a0f8f5 100644 --- a/src/gallium/drivers/freedreno/Makefile.sources +++ b/src/gallium/drivers/freedreno/Makefile.sources @@ -125,13 +125,13 @@ ir3_SOURCES := \ ir3/ir3_compiler.h \ ir3/ir3_cp.c \ ir3/ir3_depth.c \ - ir3/ir3_dump.c \ ir3/ir3_flatten.c \ ir3/ir3_group.c \ ir3/ir3.h \ ir3/ir3_legalize.c \ ir3/ir3_nir.h \ ir3/ir3_nir_lower_if_else.c \ + ir3/ir3_print.c \ ir3/ir3_ra.c \ ir3/ir3_sched.c \ ir3/ir3_shader.c \ diff --git a/src/gallium/drivers/freedreno/freedreno_screen.c b/src/gallium/drivers/freedreno/freedreno_screen.c index 6a5748c73ca..00b9471095e 100644 --- a/src/gallium/drivers/freedreno/freedreno_screen.c +++ b/src/gallium/drivers/freedreno/freedreno_screen.c @@ -68,7 +68,6 @@ static const struct debug_named_value debug_options[] = { {"fraghalf", FD_DBG_FRAGHALF, "Use half-precision in fragment shader"}, {"nobin", FD_DBG_NOBIN, "Disable hw binning"}, {"optmsgs", FD_DBG_OPTMSGS,"Enable optimizer debug messages"}, - {"optdump", FD_DBG_OPTDUMP,"Dump shader DAG to .dot files"}, {"glsl120", FD_DBG_GLSL120,"Temporary flag to force GLSL 120 (rather than 130) on a3xx+"}, {"nocp", FD_DBG_NOCP, "Disable copy-propagation"}, {"nir", FD_DBG_NIR, "Enable experimental NIR compiler"}, diff --git a/src/gallium/drivers/freedreno/freedreno_util.h b/src/gallium/drivers/freedreno/freedreno_util.h index 2735ae41315..aec09ab6616 100644 --- a/src/gallium/drivers/freedreno/freedreno_util.h +++ b/src/gallium/drivers/freedreno/freedreno_util.h @@ -63,7 +63,6 @@ enum adreno_stencil_op fd_stencil_op(unsigned op); #define FD_DBG_FRAGHALF 0x0080 #define FD_DBG_NOBIN 0x0100 #define FD_DBG_OPTMSGS 0x0400 -#define FD_DBG_OPTDUMP 0x0800 #define FD_DBG_GLSL120 0x1000 #define FD_DBG_NOCP 0x2000 #define FD_DBG_NIR 0x4000 diff --git a/src/gallium/drivers/freedreno/ir3/ir3.h b/src/gallium/drivers/freedreno/ir3/ir3.h index a7fd1814ff5..f37dfab3341 100644 --- a/src/gallium/drivers/freedreno/ir3/ir3.h +++ b/src/gallium/drivers/freedreno/ir3/ir3.h @@ -747,12 +747,8 @@ static inline struct ir3_instruction * __ssa_src_n(struct ir3_instruction *instr /* dump: */ -#include <stdio.h> -void ir3_dump(struct ir3 *shader, const char *name, - struct ir3_block *block /* XXX maybe 'block' ptr should move to ir3? */, - FILE *f); -void ir3_dump_instr_single(struct ir3_instruction *instr); -void ir3_dump_instr_list(struct ir3_instruction *instr); +void ir3_print(struct ir3 *ir); +void ir3_print_instr(struct ir3_instruction *instr); /* flatten if/else: */ int ir3_block_flatten(struct ir3_block *block); @@ -765,7 +761,7 @@ void ir3_block_depth(struct ir3_block *block); /* copy-propagate: */ void ir3_block_cp(struct ir3_block *block); -/* group neightbors and insert mov's to resolve conflicts: */ +/* group neighbors and insert mov's to resolve conflicts: */ void ir3_block_group(struct ir3_block *block); /* scheduling: */ diff --git a/src/gallium/drivers/freedreno/ir3/ir3_cmdline.c b/src/gallium/drivers/freedreno/ir3/ir3_cmdline.c index 0b16cc1eb54..e42afeaeb21 100644 --- a/src/gallium/drivers/freedreno/ir3/ir3_cmdline.c +++ b/src/gallium/drivers/freedreno/ir3/ir3_cmdline.c @@ -30,6 +30,7 @@ #include <fcntl.h> #include <stdint.h> #include <stdlib.h> +#include <stdio.h> #include <err.h> #include "tgsi/tgsi_parse.h" @@ -243,7 +244,7 @@ int main(int argc, char **argv) while (n < argc) { if (!strcmp(argv[n], "--verbose")) { - fd_mesa_debug |= FD_DBG_OPTDUMP | FD_DBG_MSGS | FD_DBG_OPTMSGS; + fd_mesa_debug |= FD_DBG_MSGS | FD_DBG_OPTMSGS; n++; continue; } diff --git a/src/gallium/drivers/freedreno/ir3/ir3_compiler.c b/src/gallium/drivers/freedreno/ir3/ir3_compiler.c index cc049d3fdfd..25af9f91d3f 100644 --- a/src/gallium/drivers/freedreno/ir3/ir3_compiler.c +++ b/src/gallium/drivers/freedreno/ir3/ir3_compiler.c @@ -3515,22 +3515,6 @@ compile_instructions(struct ir3_compile_context *ctx) } } -static void -compile_dump(struct ir3_compile_context *ctx) -{ - const char *name = (ctx->so->type == SHADER_VERTEX) ? "vert" : "frag"; - static unsigned n = 0; - char fname[16]; - FILE *f; - snprintf(fname, sizeof(fname), "%s-%04u.dot", name, n++); - f = fopen(fname, "w"); - if (!f) - return; - ir3_block_depth(ctx->block); - ir3_dump(ctx->ir, name, ctx->block, f); - fclose(f); -} - int ir3_compile_shader(struct ir3_shader_variant *so, const struct tgsi_token *tokens, struct ir3_shader_key key, @@ -3613,20 +3597,15 @@ ir3_compile_shader(struct ir3_shader_variant *so, block->outputs[block->noutputs++] = ctx.kill[i]; } - if (fd_mesa_debug & FD_DBG_OPTDUMP) - compile_dump(&ctx); - ret = ir3_block_flatten(block); if (ret < 0) { DBG("FLATTEN failed!"); goto out; } - if ((ret > 0) && (fd_mesa_debug & FD_DBG_OPTDUMP)) - compile_dump(&ctx); if (fd_mesa_debug & FD_DBG_OPTMSGS) { printf("BEFORE CP:\n"); - ir3_dump_instr_list(block->head); + ir3_print(so->ir); } ir3_block_depth(block); @@ -3641,7 +3620,7 @@ ir3_compile_shader(struct ir3_shader_variant *so, if (fd_mesa_debug & FD_DBG_OPTMSGS) { printf("BEFORE GROUPING:\n"); - ir3_dump_instr_list(block->head); + ir3_print(so->ir); } /* Group left/right neighbors, inserting mov's where needed to @@ -3649,14 +3628,11 @@ ir3_compile_shader(struct ir3_shader_variant *so, */ ir3_block_group(block); - if (fd_mesa_debug & FD_DBG_OPTDUMP) - compile_dump(&ctx); - ir3_block_depth(block); if (fd_mesa_debug & FD_DBG_OPTMSGS) { printf("AFTER DEPTH:\n"); - ir3_dump_instr_list(block->head); + ir3_print(so->ir); } ret = ir3_block_sched(block); @@ -3667,7 +3643,7 @@ ir3_compile_shader(struct ir3_shader_variant *so, if (fd_mesa_debug & FD_DBG_OPTMSGS) { printf("AFTER SCHED:\n"); - ir3_dump_instr_list(block->head); + ir3_print(so->ir); } ret = ir3_block_ra(block, so->type, so->frag_coord, so->frag_face); @@ -3678,7 +3654,7 @@ ir3_compile_shader(struct ir3_shader_variant *so, if (fd_mesa_debug & FD_DBG_OPTMSGS) { printf("AFTER RA:\n"); - ir3_dump_instr_list(block->head); + ir3_print(so->ir); } ir3_block_legalize(block, &so->has_samp, &max_bary); diff --git a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c index 2cf25ea6e0a..8d382e5cf3e 100644 --- a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c +++ b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c @@ -1908,22 +1908,6 @@ fixup_frag_inputs(struct ir3_compile *ctx) block->inputs = inputs; } -static void -compile_dump(struct ir3_compile *ctx) -{ - const char *name = (ctx->so->type == SHADER_VERTEX) ? "vert" : "frag"; - static unsigned n = 0; - char fname[16]; - FILE *f; - snprintf(fname, sizeof(fname), "%s-%04u.dot", name, n++); - f = fopen(fname, "w"); - if (!f) - return; - ir3_block_depth(ctx->block); - ir3_dump(ctx->ir, name, ctx->block, f); - fclose(f); -} - int ir3_compile_shader_nir(struct ir3_shader_variant *so, const struct tgsi_token *tokens, struct ir3_shader_key key) @@ -2008,12 +1992,9 @@ ir3_compile_shader_nir(struct ir3_shader_variant *so, block->outputs[block->noutputs++] = ctx->kill[i]; } - if (fd_mesa_debug & FD_DBG_OPTDUMP) - compile_dump(ctx); - if (fd_mesa_debug & FD_DBG_OPTMSGS) { printf("BEFORE CP:\n"); - ir3_dump_instr_list(block->head); + ir3_print(so->ir); } ir3_block_depth(block); @@ -2022,7 +2003,7 @@ ir3_compile_shader_nir(struct ir3_shader_variant *so, if (fd_mesa_debug & FD_DBG_OPTMSGS) { printf("BEFORE GROUPING:\n"); - ir3_dump_instr_list(block->head); + ir3_print(so->ir); } /* Group left/right neighbors, inserting mov's where needed to @@ -2030,14 +2011,11 @@ ir3_compile_shader_nir(struct ir3_shader_variant *so, */ ir3_block_group(block); - if (fd_mesa_debug & FD_DBG_OPTDUMP) - compile_dump(ctx); - ir3_block_depth(block); if (fd_mesa_debug & FD_DBG_OPTMSGS) { printf("AFTER DEPTH:\n"); - ir3_dump_instr_list(block->head); + ir3_print(so->ir); } ret = ir3_block_sched(block); @@ -2048,7 +2026,7 @@ ir3_compile_shader_nir(struct ir3_shader_variant *so, if (fd_mesa_debug & FD_DBG_OPTMSGS) { printf("AFTER SCHED:\n"); - ir3_dump_instr_list(block->head); + ir3_print(so->ir); } ret = ir3_block_ra(block, so->type, so->frag_coord, so->frag_face); @@ -2059,7 +2037,7 @@ ir3_compile_shader_nir(struct ir3_shader_variant *so, if (fd_mesa_debug & FD_DBG_OPTMSGS) { printf("AFTER RA:\n"); - ir3_dump_instr_list(block->head); + ir3_print(so->ir); } ir3_block_legalize(block, &so->has_samp, &max_bary); diff --git a/src/gallium/drivers/freedreno/ir3/ir3_dump.c b/src/gallium/drivers/freedreno/ir3/ir3_dump.c deleted file mode 100644 index 1614d637b13..00000000000 --- a/src/gallium/drivers/freedreno/ir3/ir3_dump.c +++ /dev/null @@ -1,456 +0,0 @@ -/* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */ - -/* - * Copyright (C) 2014 Rob Clark <[email protected]> - * - * 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 AUTHORS OR COPYRIGHT HOLDERS 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: - * Rob Clark <[email protected]> - */ - -#include <stdarg.h> - -#include "ir3.h" - -#define PTRID(x) ((unsigned long)(x)) - -struct ir3_dump_ctx { - FILE *f; - bool verbose; -}; - -static void dump_instr_name(struct ir3_dump_ctx *ctx, - struct ir3_instruction *instr) -{ - /* for debugging: */ - if (ctx->verbose) { -#ifdef DEBUG - fprintf(ctx->f, "%04u:", instr->serialno); -#endif - fprintf(ctx->f, "%03u: ", instr->depth); - } - - if (instr->flags & IR3_INSTR_SY) - fprintf(ctx->f, "(sy)"); - if (instr->flags & IR3_INSTR_SS) - fprintf(ctx->f, "(ss)"); - - if (is_meta(instr)) { - switch(instr->opc) { - case OPC_META_PHI: - fprintf(ctx->f, "Φ"); - break; - default: - /* shouldn't hit here.. just for debugging: */ - switch (instr->opc) { - case OPC_META_INPUT: fprintf(ctx->f, "_meta:in"); break; - case OPC_META_OUTPUT: fprintf(ctx->f, "_meta:out"); break; - case OPC_META_FO: fprintf(ctx->f, "_meta:fo"); break; - case OPC_META_FI: fprintf(ctx->f, "_meta:fi"); break; - case OPC_META_FLOW: fprintf(ctx->f, "_meta:flow"); break; - - default: fprintf(ctx->f, "_meta:%d", instr->opc); break; - } - break; - } - } else if (instr->category == 1) { - static const char *type[] = { - [TYPE_F16] = "f16", - [TYPE_F32] = "f32", - [TYPE_U16] = "u16", - [TYPE_U32] = "u32", - [TYPE_S16] = "s16", - [TYPE_S32] = "s32", - [TYPE_U8] = "u8", - [TYPE_S8] = "s8", - }; - if (instr->cat1.src_type == instr->cat1.dst_type) - fprintf(ctx->f, "mov"); - else - fprintf(ctx->f, "cov"); - fprintf(ctx->f, ".%s%s", type[instr->cat1.src_type], type[instr->cat1.dst_type]); - } else { - fprintf(ctx->f, "%s", ir3_instr_name(instr)); - if (instr->flags & IR3_INSTR_3D) - fprintf(ctx->f, ".3d"); - if (instr->flags & IR3_INSTR_A) - fprintf(ctx->f, ".a"); - if (instr->flags & IR3_INSTR_O) - fprintf(ctx->f, ".o"); - if (instr->flags & IR3_INSTR_P) - fprintf(ctx->f, ".p"); - if (instr->flags & IR3_INSTR_S) - fprintf(ctx->f, ".s"); - if (instr->flags & IR3_INSTR_S2EN) - fprintf(ctx->f, ".s2en"); - } -} - -static void dump_reg_name(struct ir3_dump_ctx *ctx, - struct ir3_register *reg, bool followssa) -{ - if ((reg->flags & (IR3_REG_FABS | IR3_REG_SABS)) && - (reg->flags & (IR3_REG_FNEG | IR3_REG_SNEG | IR3_REG_BNOT))) - fprintf(ctx->f, "(absneg)"); - else if (reg->flags & (IR3_REG_FNEG | IR3_REG_SNEG | IR3_REG_BNOT)) - fprintf(ctx->f, "(neg)"); - else if (reg->flags & (IR3_REG_FABS | IR3_REG_SABS)) - fprintf(ctx->f, "(abs)"); - - if (reg->flags & IR3_REG_IMMED) { - fprintf(ctx->f, "imm[%f,%d,0x%x]", reg->fim_val, reg->iim_val, reg->iim_val); - } else if (reg->flags & IR3_REG_SSA) { - if (ctx->verbose) { - fprintf(ctx->f, "_"); - if (followssa) { - fprintf(ctx->f, "["); - dump_instr_name(ctx, reg->instr); - fprintf(ctx->f, "]"); - } - } - } else if (reg->flags & IR3_REG_RELATIV) { - if (reg->flags & IR3_REG_HALF) - fprintf(ctx->f, "h"); - if (reg->flags & IR3_REG_CONST) - fprintf(ctx->f, "c<a0.x + %u>", reg->num); - else - fprintf(ctx->f, "\x1b[0;31mr<a0.x + %u>\x1b[0m (%u)", reg->num, reg->size); - } else { - if (reg->flags & IR3_REG_HALF) - fprintf(ctx->f, "h"); - if (reg->flags & IR3_REG_CONST) - fprintf(ctx->f, "c%u.%c", reg_num(reg), "xyzw"[reg_comp(reg)]); - else - fprintf(ctx->f, "\x1b[0;31mr%u.%c\x1b[0m", reg_num(reg), "xyzw"[reg_comp(reg)]); - } -} - -static void ir3_instr_dump(struct ir3_dump_ctx *ctx, - struct ir3_instruction *instr); -static void ir3_block_dump(struct ir3_dump_ctx *ctx, - struct ir3_block *block, const char *name); - -static void dump_instr(struct ir3_dump_ctx *ctx, - struct ir3_instruction *instr) -{ - /* if we've already visited this instruction, bail now: */ - if (ir3_instr_check_mark(instr)) - return; - - /* some meta-instructions need to be handled specially: */ - if (is_meta(instr)) { - if ((instr->opc == OPC_META_FO) || - (instr->opc == OPC_META_FI)) { - struct ir3_instruction *src; - foreach_ssa_src(src, instr) - dump_instr(ctx, src); - } else if (instr->opc == OPC_META_FLOW) { - struct ir3_register *reg = instr->regs[1]; - ir3_block_dump(ctx, instr->flow.if_block, "if"); - if (instr->flow.else_block) - ir3_block_dump(ctx, instr->flow.else_block, "else"); - if (reg->flags & IR3_REG_SSA) - dump_instr(ctx, reg->instr); - } else if (instr->opc == OPC_META_PHI) { - /* treat like a normal instruction: */ - ir3_instr_dump(ctx, instr); - } - } else { - ir3_instr_dump(ctx, instr); - } -} - -/* arrarraggh! if link is to something outside of the current block, we - * need to defer emitting the link until the end of the block, since the - * edge triggers pre-creation of the node it links to inside the cluster, - * even though it is meant to be outside.. - */ -static struct { - char buf[40960]; - unsigned n; -} edge_buf; - -/* helper to print or defer: */ -static void printdef(struct ir3_dump_ctx *ctx, - bool defer, const char *fmt, ...) -{ - va_list ap; - va_start(ap, fmt); - if (defer) { - unsigned n = edge_buf.n; - n += vsnprintf(&edge_buf.buf[n], sizeof(edge_buf.buf) - n, - fmt, ap); - edge_buf.n = n; - } else { - vfprintf(ctx->f, fmt, ap); - } - va_end(ap); -} - -static void dump_link2(struct ir3_dump_ctx *ctx, - struct ir3_instruction *instr, const char *target, bool defer) -{ - /* some meta-instructions need to be handled specially: */ - if (is_meta(instr)) { - if (instr->opc == OPC_META_INPUT) { - printdef(ctx, defer, "input%lx:<in%u>:w -> %s", - PTRID(instr->inout.block), - instr->regs[0]->num, target); - } else if (instr->opc == OPC_META_FO) { - struct ir3_register *reg = instr->regs[1]; - dump_link2(ctx, reg->instr, target, defer); - printdef(ctx, defer, "[label=\".%c\"]", - "xyzw"[instr->fo.off & 0x3]); - } else if (instr->opc == OPC_META_FI) { - struct ir3_instruction *src; - - foreach_ssa_src_n(src, i, instr) { - dump_link2(ctx, src, target, defer); - printdef(ctx, defer, "[label=\".%c\"]", - "xyzw"[i & 0x3]); - } - } else if (instr->opc == OPC_META_OUTPUT) { - printdef(ctx, defer, "output%lx:<out%u>:w -> %s", - PTRID(instr->inout.block), - instr->regs[0]->num, target); - } else if (instr->opc == OPC_META_PHI) { - /* treat like a normal instruction: */ - printdef(ctx, defer, "instr%lx:<dst0> -> %s", PTRID(instr), target); - } - } else { - printdef(ctx, defer, "instr%lx:<dst0> -> %s", PTRID(instr), target); - } -} - -static void dump_link(struct ir3_dump_ctx *ctx, - struct ir3_instruction *instr, - struct ir3_block *block, const char *target) -{ - bool defer = instr->block != block; - dump_link2(ctx, instr, target, defer); - printdef(ctx, defer, "\n"); -} - -static struct ir3_register *follow_flow(struct ir3_register *reg) -{ - if (reg->flags & IR3_REG_SSA) { - struct ir3_instruction *instr = reg->instr; - /* go with the flow.. */ - if (is_meta(instr) && (instr->opc == OPC_META_FLOW)) - return instr->regs[1]; - } - return reg; -} - -static void ir3_instr_dump(struct ir3_dump_ctx *ctx, - struct ir3_instruction *instr) -{ - struct ir3_register *src; - - fprintf(ctx->f, "instr%lx [shape=record,style=filled,fillcolor=lightgrey,label=\"{", - PTRID(instr)); - dump_instr_name(ctx, instr); - - /* destination register: */ - fprintf(ctx->f, "|<dst0>"); - - /* source register(s): */ - foreach_src_n(src, i, instr) { - struct ir3_register *reg = follow_flow(src); - - fprintf(ctx->f, "|"); - - if (reg->flags & IR3_REG_SSA) - fprintf(ctx->f, "<src%u> ", i); - - dump_reg_name(ctx, reg, true); - } - - fprintf(ctx->f, "}\"];\n"); - - /* and recursively dump dependent instructions: */ - foreach_src_n(src, i, instr) { - struct ir3_register *reg = follow_flow(src); - char target[32]; /* link target */ - - if (!(reg->flags & IR3_REG_SSA)) - continue; - - snprintf(target, sizeof(target), "instr%lx:<src%u>", - PTRID(instr), i); - - dump_instr(ctx, reg->instr); - dump_link(ctx, reg->instr, instr->block, target); - } -} - -static void ir3_block_dump(struct ir3_dump_ctx *ctx, - struct ir3_block *block, const char *name) -{ - unsigned i, n; - - n = edge_buf.n; - - fprintf(ctx->f, "subgraph cluster%lx {\n", PTRID(block)); - fprintf(ctx->f, "label=\"%s\";\n", name); - - /* draw inputs: */ - fprintf(ctx->f, "input%lx [shape=record,label=\"inputs", PTRID(block)); - for (i = 0; i < block->ninputs; i++) - if (block->inputs[i]) - fprintf(ctx->f, "|<in%u> i%u.%c", i, (i >> 2), "xyzw"[i & 0x3]); - fprintf(ctx->f, "\"];\n"); - - /* draw instruction graph: */ - for (i = 0; i < block->noutputs; i++) - if (block->outputs[i]) - dump_instr(ctx, block->outputs[i]); - - /* draw outputs: */ - fprintf(ctx->f, "output%lx [shape=record,label=\"outputs", PTRID(block)); - for (i = 0; i < block->noutputs; i++) - fprintf(ctx->f, "|<out%u> o%u.%c", i, (i >> 2), "xyzw"[i & 0x3]); - fprintf(ctx->f, "\"];\n"); - - /* and links to outputs: */ - for (i = 0; i < block->noutputs; i++) { - char target[32]; /* link target */ - - /* NOTE: there could be outputs that are never assigned, - * so skip them - */ - if (!block->outputs[i]) - continue; - - snprintf(target, sizeof(target), "output%lx:<out%u>:e", - PTRID(block), i); - - dump_link(ctx, block->outputs[i], block, target); - } - - fprintf(ctx->f, "}\n"); - - /* and links to inputs: */ - if (block->parent) { - for (i = 0; i < block->ninputs; i++) { - char target[32]; /* link target */ - - if (!block->inputs[i]) - continue; - - dump_instr(ctx, block->inputs[i]); - - snprintf(target, sizeof(target), "input%lx:<in%u>:e", - PTRID(block), i); - - dump_link(ctx, block->inputs[i], block, target); - } - } - - /* dump deferred edges: */ - if (edge_buf.n > n) { - fprintf(ctx->f, "%*s", edge_buf.n - n, &edge_buf.buf[n]); - edge_buf.n = n; - } -} - -void ir3_dump(struct ir3 *shader, const char *name, - struct ir3_block *block /* XXX maybe 'block' ptr should move to ir3? */, - FILE *f) -{ - struct ir3_dump_ctx ctx = { - .f = f, - }; - ir3_clear_mark(shader); - fprintf(ctx.f, "digraph G {\n"); - fprintf(ctx.f, "rankdir=RL;\n"); - fprintf(ctx.f, "nodesep=0.25;\n"); - fprintf(ctx.f, "ranksep=1.5;\n"); - ir3_block_dump(&ctx, block, name); - fprintf(ctx.f, "}\n"); -} - -/* - * For Debugging: - */ - -void -ir3_dump_instr_single(struct ir3_instruction *instr) -{ - struct ir3_dump_ctx ctx = { - .f = stdout, - .verbose = true, - }; - unsigned i; - - dump_instr_name(&ctx, instr); - for (i = 0; i < instr->regs_count; i++) { - struct ir3_register *reg = instr->regs[i]; - printf(i ? ", " : " "); - dump_reg_name(&ctx, reg, !!i); - } - - if (instr->address) { - fprintf(ctx.f, ", address=_"); - fprintf(ctx.f, "["); - dump_instr_name(&ctx, instr->address); - fprintf(ctx.f, "]"); - } - - if (instr->fanin) { - fprintf(ctx.f, ", fanin=_"); - fprintf(ctx.f, "["); - dump_instr_name(&ctx, instr->fanin); - fprintf(ctx.f, "]"); - } - - if (is_meta(instr)) { - if (instr->opc == OPC_META_FO) { - printf(", off=%d", instr->fo.off); - } else if ((instr->opc == OPC_META_FI) && instr->fi.aid) { - printf(", aid=%d", instr->fi.aid); - } - } - - printf("\n"); -} - -void -ir3_dump_instr_list(struct ir3_instruction *instr) -{ - struct ir3_block *block = instr->block; - unsigned n = 0; - - while (instr) { - ir3_dump_instr_single(instr); - if (!is_meta(instr)) - n++; - instr = instr->next; - } - printf("%u instructions\n", n); - - for (n = 0; n < block->noutputs; n++) { - if (!block->outputs[n]) - continue; - printf("out%d: ", n); - ir3_dump_instr_single(block->outputs[n]); - } -} diff --git a/src/gallium/drivers/freedreno/ir3/ir3_print.c b/src/gallium/drivers/freedreno/ir3/ir3_print.c new file mode 100644 index 00000000000..a5c5d3c8efa --- /dev/null +++ b/src/gallium/drivers/freedreno/ir3/ir3_print.c @@ -0,0 +1,210 @@ +/* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */ + +/* + * Copyright (C) 2014 Rob Clark <[email protected]> + * + * 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 AUTHORS OR COPYRIGHT HOLDERS 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: + * Rob Clark <[email protected]> + */ + +#include <stdarg.h> +#include <stdio.h> + +#include "ir3.h" + +#define PTRID(x) ((unsigned long)(x)) + +static void print_instr_name(struct ir3_instruction *instr) +{ +#ifdef DEBUG + printf("%04u:", instr->serialno); +#endif + printf("%03u: ", instr->depth); + + if (instr->flags & IR3_INSTR_SY) + printf("(sy)"); + if (instr->flags & IR3_INSTR_SS) + printf("(ss)"); + + if (is_meta(instr)) { + switch(instr->opc) { + case OPC_META_PHI: + printf("Φ"); + break; + default: + /* shouldn't hit here.. just for debugging: */ + switch (instr->opc) { + case OPC_META_INPUT: printf("_meta:in"); break; + case OPC_META_OUTPUT: printf("_meta:out"); break; + case OPC_META_FO: printf("_meta:fo"); break; + case OPC_META_FI: printf("_meta:fi"); break; + case OPC_META_FLOW: printf("_meta:flow"); break; + + default: printf("_meta:%d", instr->opc); break; + } + break; + } + } else if (instr->category == 1) { + static const char *type[] = { + [TYPE_F16] = "f16", + [TYPE_F32] = "f32", + [TYPE_U16] = "u16", + [TYPE_U32] = "u32", + [TYPE_S16] = "s16", + [TYPE_S32] = "s32", + [TYPE_U8] = "u8", + [TYPE_S8] = "s8", + }; + if (instr->cat1.src_type == instr->cat1.dst_type) + printf("mov"); + else + printf("cov"); + printf(".%s%s", type[instr->cat1.src_type], type[instr->cat1.dst_type]); + } else { + printf("%s", ir3_instr_name(instr)); + if (instr->flags & IR3_INSTR_3D) + printf(".3d"); + if (instr->flags & IR3_INSTR_A) + printf(".a"); + if (instr->flags & IR3_INSTR_O) + printf(".o"); + if (instr->flags & IR3_INSTR_P) + printf(".p"); + if (instr->flags & IR3_INSTR_S) + printf(".s"); + if (instr->flags & IR3_INSTR_S2EN) + printf(".s2en"); + } +} + +static void print_reg_name(struct ir3_register *reg, bool followssa) +{ + if ((reg->flags & (IR3_REG_FABS | IR3_REG_SABS)) && + (reg->flags & (IR3_REG_FNEG | IR3_REG_SNEG | IR3_REG_BNOT))) + printf("(absneg)"); + else if (reg->flags & (IR3_REG_FNEG | IR3_REG_SNEG | IR3_REG_BNOT)) + printf("(neg)"); + else if (reg->flags & (IR3_REG_FABS | IR3_REG_SABS)) + printf("(abs)"); + + if (reg->flags & IR3_REG_IMMED) { + printf("imm[%f,%d,0x%x]", reg->fim_val, reg->iim_val, reg->iim_val); + } else if (reg->flags & IR3_REG_SSA) { + printf("_"); + if (followssa) { + printf("["); + print_instr_name(reg->instr); + printf("]"); + } + } else if (reg->flags & IR3_REG_RELATIV) { + if (reg->flags & IR3_REG_HALF) + printf("h"); + if (reg->flags & IR3_REG_CONST) + printf("c<a0.x + %u>", reg->num); + else + printf("\x1b[0;31mr<a0.x + %u>\x1b[0m (%u)", reg->num, reg->size); + } else { + if (reg->flags & IR3_REG_HALF) + printf("h"); + if (reg->flags & IR3_REG_CONST) + printf("c%u.%c", reg_num(reg), "xyzw"[reg_comp(reg)]); + else + printf("\x1b[0;31mr%u.%c\x1b[0m", reg_num(reg), "xyzw"[reg_comp(reg)]); + } +} + +static void +tab(int lvl) +{ + for (int i = 0; i < lvl; i++) + printf("\t"); +} + +static void +print_instr(struct ir3_instruction *instr, int lvl) +{ + unsigned i; + + tab(lvl); + + print_instr_name(instr); + for (i = 0; i < instr->regs_count; i++) { + struct ir3_register *reg = instr->regs[i]; + printf(i ? ", " : " "); + print_reg_name(reg, !!i); + } + + if (instr->address) { + printf(", address=_"); + printf("["); + print_instr_name(instr->address); + printf("]"); + } + + if (instr->fanin) { + printf(", fanin=_"); + printf("["); + print_instr_name(instr->fanin); + printf("]"); + } + + if (is_meta(instr)) { + if (instr->opc == OPC_META_FO) { + printf(", off=%d", instr->fo.off); + } else if ((instr->opc == OPC_META_FI) && instr->fi.aid) { + printf(", aid=%d", instr->fi.aid); + } + } + + printf("\n"); +} + +void ir3_print_instr(struct ir3_instruction *instr) +{ + print_instr(instr, 0); +} + +static void +print_block(struct ir3_block *block, int lvl) +{ + struct ir3_instruction *instr; + tab(lvl); printf("block {\n"); + for (instr = block->head; instr; instr = instr->next) { + print_instr(instr, lvl+1); + } + tab(lvl); printf("}\n"); +} + +void +ir3_print(struct ir3 *ir) +{ + struct ir3_block *block = ir->block; + + print_block(block, 0); + + for (unsigned i = 0; i < block->noutputs; i++) { + if (!block->outputs[i]) + continue; + printf("out%d: ", i); + print_instr(block->outputs[i], 0); + } +} diff --git a/src/gallium/drivers/freedreno/ir3/ir3_ra.c b/src/gallium/drivers/freedreno/ir3/ir3_ra.c index a4235a77a15..501352515b5 100644 --- a/src/gallium/drivers/freedreno/ir3/ir3_ra.c +++ b/src/gallium/drivers/freedreno/ir3/ir3_ra.c @@ -78,14 +78,14 @@ struct ir3_ra_ctx { #define ra_dump_list(msg, n) do { \ if (ra_debug) { \ debug_printf("-- " msg); \ - ir3_dump_instr_list(n); \ + ir3_print(n->block->shader); \ } \ } while (0) #define ra_dump_instr(msg, n) do { \ if (ra_debug) { \ debug_printf(">> " msg); \ - ir3_dump_instr_single(n); \ + ir3_print_instr(n); \ } \ } while (0) |