From 5d7609715a44d08f29d4b605c4bea2742a194493 Mon Sep 17 00:00:00 2001 From: =?utf8?q?St=C3=A9phane=20Marchesin?= Date: Wed, 29 Jun 2011 16:52:28 -0700 Subject: [PATCH] i915g: Try to do better in the shader compiler. - Copy i915c's support for phases, that should allow us to run a coupe more shaders. - Fix the error messages. - Still try to proceed when we get a shader that's too long. --- src/gallium/drivers/i915/i915_fpc.h | 1 + src/gallium/drivers/i915/i915_fpc_emit.c | 46 +++++++++++++------ src/gallium/drivers/i915/i915_fpc_translate.c | 14 ++++-- 3 files changed, 41 insertions(+), 20 deletions(-) diff --git a/src/gallium/drivers/i915/i915_fpc.h b/src/gallium/drivers/i915/i915_fpc.h index e298e78c6bc..509395cf1f5 100644 --- a/src/gallium/drivers/i915/i915_fpc.h +++ b/src/gallium/drivers/i915/i915_fpc.h @@ -70,6 +70,7 @@ struct i915_fp_compile { uint temp_flag; /**< Tracks temporary regs which are in use */ uint utemp_flag; /**< Tracks TYPE_U temporary regs which are in use */ + uint register_phases[16]; uint nr_tex_indirect; uint nr_tex_insn; uint nr_alu_insn; diff --git a/src/gallium/drivers/i915/i915_fpc_emit.c b/src/gallium/drivers/i915/i915_fpc_emit.c index f92e6a425b5..d28595e0fd3 100644 --- a/src/gallium/drivers/i915/i915_fpc_emit.c +++ b/src/gallium/drivers/i915/i915_fpc_emit.c @@ -67,7 +67,7 @@ i915_get_temp(struct i915_fp_compile *p) { int bit = ffs(~p->temp_flag); if (!bit) { - i915_program_error(p, "i915_get_temp: out of temporaries\n"); + i915_program_error(p, "i915_get_temp: out of temporaries"); return 0; } @@ -92,7 +92,7 @@ i915_get_utemp(struct i915_fp_compile * p) { int bit = ffs(~p->utemp_flag); if (!bit) { - i915_program_error(p, "i915_get_utemp: out of temporaries\n"); + i915_program_error(p, "i915_get_utemp: out of temporaries"); return 0; } @@ -134,7 +134,7 @@ i915_emit_decl(struct i915_fp_compile *p, *(p->decl++) = D2_MBZ; } else - i915_program_error(p, "Out of declarations\n"); + i915_program_error(p, "Out of declarations"); p->nr_decl_insn++; return reg; @@ -197,7 +197,10 @@ i915_emit_arith(struct i915_fp_compile * p, *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2)); } else - i915_program_error(p, "Out of instructions\n"); + i915_program_error(p, "Out of instructions"); + + if (GET_UREG_TYPE(dest) == REG_TYPE_R) + p->register_phases[GET_UREG_NR(dest)] = p->nr_tex_indirect; p->nr_alu_insn++; return dest; @@ -253,21 +256,31 @@ uint i915_emit_texld( struct i915_fp_compile *p, assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST); assert(dest == UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest))); - /* is the sampler coord a texcoord input reg? */ - if (GET_UREG_TYPE(coord) != REG_TYPE_T) { - p->nr_tex_indirect++; - } + /* Output register being oC or oD defines a phase boundary */ + if (GET_UREG_TYPE(dest) == REG_TYPE_OC || + GET_UREG_TYPE(dest) == REG_TYPE_OD) + p->nr_tex_indirect++; + + /* Reading from an r# register whose contents depend on output of the + * current phase defines a phase boundary. + */ + if (GET_UREG_TYPE(coord) == REG_TYPE_R && + p->register_phases[GET_UREG_NR(coord)] == p->nr_tex_indirect) + p->nr_tex_indirect++; if (p->csr< p->program + I915_PROGRAM_SIZE) { *(p->csr++) = (opcode | - T0_DEST( dest ) | - T0_SAMPLER( sampler )); + T0_DEST( dest ) | + T0_SAMPLER( sampler )); *(p->csr++) = T1_ADDRESS_REG( coord ); *(p->csr++) = T2_MBZ; } - else - i915_program_error(p, "Out of instructions\n"); + else + i915_program_error(p, "Out of instructions"); + + if (GET_UREG_TYPE(dest) == REG_TYPE_R) + p->register_phases[GET_UREG_NR(dest)] = p->nr_tex_indirect; p->nr_tex_insn++; } @@ -305,7 +318,7 @@ i915_emit_const1f(struct i915_fp_compile * p, float c0) } } - i915_program_error(p, "i915_emit_const1f: out of constants\n"); + i915_program_error(p, "i915_emit_const1f: out of constants"); return 0; } @@ -343,7 +356,7 @@ i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1) } } - i915_program_error(p, "i915_emit_const2f: out of constants\n"); + i915_program_error(p, "i915_emit_const2f: out of constants"); return 0; } @@ -354,6 +367,9 @@ i915_emit_const4f(struct i915_fp_compile * p, struct i915_fragment_shader *ifs = p->shader; unsigned reg; + // XXX emit swizzle here for 0, 1, -1 and any combination thereof + // we can use swizzle + neg for that + printf("const %f %f %f %f\n",c0,c1,c2,c3); for (reg = 0; reg < I915_MAX_CONSTANT; reg++) { if (ifs->constant_flags[reg] == 0xf && ifs->constants[reg][0] == c0 && @@ -375,7 +391,7 @@ i915_emit_const4f(struct i915_fp_compile * p, } } - i915_program_error(p, "i915_emit_const4f: out of constants\n"); + i915_program_error(p, "i915_emit_const4f: out of constants"); return 0; } diff --git a/src/gallium/drivers/i915/i915_fpc_translate.c b/src/gallium/drivers/i915/i915_fpc_translate.c index 93fe453e9df..0cbd4f2d748 100644 --- a/src/gallium/drivers/i915/i915_fpc_translate.c +++ b/src/gallium/drivers/i915/i915_fpc_translate.c @@ -276,7 +276,7 @@ src_vector(struct i915_fp_compile *p, assert(!source->Register.Absolute); #endif if (source->Register.Absolute) - debug_printf("Unhandler absolute value\n"); + debug_printf("Unhandled absolute value\n"); return src; } @@ -1089,9 +1089,11 @@ i915_translate_instructions(struct i915_fp_compile *p, for (i = parse.FullToken.FullDeclaration.Range.First; i <= parse.FullToken.FullDeclaration.Range.Last; i++) { - assert(i < I915_MAX_TEMPORARY); - /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */ - p->temp_flag |= (1 << i); /* mark temp as used */ + if (i >= I915_MAX_TEMPORARY) + debug_printf("Too many temps (%d)\n",i); + else + /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */ + p->temp_flag |= (1 << i); /* mark temp as used */ } } break; @@ -1163,6 +1165,8 @@ i915_init_compile(struct i915_context *i915, ifs->num_constants = 0; memset(ifs->constant_flags, 0, sizeof(ifs->constant_flags)); + memset(&p->register_phases, 0, sizeof(p->register_phases)); + for (i = 0; i < I915_TEX_UNITS; i++) ifs->generic_mapping[i] = -1; @@ -1198,7 +1202,7 @@ i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p) unsigned long decl_size = (unsigned long) (p->decl - p->declarations); if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT) - i915_program_error(p, "Exceeded max nr indirect texture lookups"); + debug_printf("Exceeded max nr indirect texture lookups\n"); if (p->nr_tex_insn > I915_MAX_TEX_INSN) i915_program_error(p, "Exceeded max TEX instructions"); -- 2.30.2