X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fintel%2Fcompiler%2Fbrw_compiler.c;h=a5b90e9b3c9e00b4db2c4f8eb7844bb923b5f3a6;hb=8d8a3815ef698348857cd9812eaa0dc76f5b19b5;hp=7ceeb14c70f84a79a1b68ce9aab8f741d1b88d43;hpb=110669c85c3fcfed0a4dff42a36fc3f97c5a559a;p=mesa.git diff --git a/src/intel/compiler/brw_compiler.c b/src/intel/compiler/brw_compiler.c index 7ceeb14c70f..a5b90e9b3c9 100644 --- a/src/intel/compiler/brw_compiler.c +++ b/src/intel/compiler/brw_compiler.c @@ -47,9 +47,13 @@ .vectorize_io = true, \ .use_interpolated_input_intrinsics = true, \ .vertex_id_zero_based = true, \ - .lower_base_vertex = true + .lower_base_vertex = true, \ + .use_scoped_barrier = true, \ + .support_8bit_alu = true, \ + .support_16bit_alu = true #define COMMON_SCALAR_OPTIONS \ + .lower_to_scalar = true, \ .lower_pack_half_2x16 = true, \ .lower_pack_snorm_2x16 = true, \ .lower_pack_snorm_4x8 = true, \ @@ -60,6 +64,9 @@ .lower_unpack_snorm_4x8 = true, \ .lower_unpack_unorm_2x16 = true, \ .lower_unpack_unorm_4x8 = true, \ + .lower_usub_sat64 = true, \ + .lower_hadd64 = true, \ + .lower_bfe_with_two_constants = true, \ .max_unroll_iterations = 32 static const struct nir_shader_compiler_options scalar_nir_options = { @@ -95,16 +102,16 @@ brw_compiler_create(void *mem_ctx, const struct gen_device_info *devinfo) brw_fs_alloc_reg_sets(compiler); brw_vec4_alloc_reg_set(compiler); - brw_init_compaction_tables(devinfo); compiler->precise_trig = env_var_as_boolean("INTEL_PRECISE_TRIG", false); compiler->use_tcs_8_patch = - devinfo->gen >= 9 && (INTEL_DEBUG & DEBUG_TCS_EIGHT_PATCH); + devinfo->gen >= 12 || + (devinfo->gen >= 9 && (INTEL_DEBUG & DEBUG_TCS_EIGHT_PATCH)); if (devinfo->gen >= 10) { /* We don't support vec4 mode on Cannonlake. */ - for (int i = MESA_SHADER_VERTEX; i < MESA_SHADER_STAGES; i++) + for (int i = MESA_SHADER_VERTEX; i < MESA_ALL_SHADER_STAGES; i++) compiler->scalar_stage[i] = true; } else { compiler->scalar_stage[MESA_SHADER_VERTEX] = @@ -137,16 +144,8 @@ brw_compiler_create(void *mem_ctx, const struct gen_device_info *devinfo) nir_lower_dsub | nir_lower_ddiv; - if (!devinfo->has_64bit_types || (INTEL_DEBUG & DEBUG_SOFT64)) { - int64_options |= nir_lower_mov64 | - nir_lower_icmp64 | - nir_lower_iadd64 | - nir_lower_iabs64 | - nir_lower_ineg64 | - nir_lower_logic64 | - nir_lower_minmax64 | - nir_lower_shift64 | - nir_lower_extract64; + if (!devinfo->has_64bit_float || (INTEL_DEBUG & DEBUG_SOFT64)) { + int64_options |= (nir_lower_int64_options)~0; fp64_options |= nir_lower_fp64_full_software; } @@ -158,7 +157,7 @@ brw_compiler_create(void *mem_ctx, const struct gen_device_info *devinfo) int64_options |= nir_lower_imul_2x32_64; /* We want the GLSL compiler to emit code that uses condition codes */ - for (int i = 0; i < MESA_SHADER_STAGES; i++) { + for (int i = 0; i < MESA_ALL_SHADER_STAGES; i++) { compiler->glsl_compiler_options[i].MaxUnrollIterations = 0; compiler->glsl_compiler_options[i].MaxIfDepth = devinfo->gen < 6 ? 16 : UINT_MAX; @@ -185,11 +184,16 @@ brw_compiler_create(void *mem_ctx, const struct gen_device_info *devinfo) */ nir_options->lower_ffma = devinfo->gen < 6; nir_options->lower_flrp32 = devinfo->gen < 6 || devinfo->gen >= 11; + nir_options->lower_fpow = devinfo->gen >= 12; nir_options->lower_rotate = devinfo->gen < 11; + nir_options->lower_bitfield_reverse = devinfo->gen < 7; nir_options->lower_int64_options = int64_options; nir_options->lower_doubles_options = fp64_options; + + nir_options->unify_interfaces = i < MESA_SHADER_FRAGMENT; + compiler->glsl_compiler_options[i].NirOptions = nir_options; compiler->glsl_compiler_options[i].ClampBlockIndicesToArrayBounds = true; @@ -235,19 +239,14 @@ brw_get_compiler_config_value(const struct brw_compiler *compiler) unsigned brw_prog_data_size(gl_shader_stage stage) { - STATIC_ASSERT(MESA_SHADER_VERTEX == 0); - STATIC_ASSERT(MESA_SHADER_TESS_CTRL == 1); - STATIC_ASSERT(MESA_SHADER_TESS_EVAL == 2); - STATIC_ASSERT(MESA_SHADER_GEOMETRY == 3); - STATIC_ASSERT(MESA_SHADER_FRAGMENT == 4); - STATIC_ASSERT(MESA_SHADER_COMPUTE == 5); static const size_t stage_sizes[] = { - sizeof(struct brw_vs_prog_data), - sizeof(struct brw_tcs_prog_data), - sizeof(struct brw_tes_prog_data), - sizeof(struct brw_gs_prog_data), - sizeof(struct brw_wm_prog_data), - sizeof(struct brw_cs_prog_data), + [MESA_SHADER_VERTEX] = sizeof(struct brw_vs_prog_data), + [MESA_SHADER_TESS_CTRL] = sizeof(struct brw_tcs_prog_data), + [MESA_SHADER_TESS_EVAL] = sizeof(struct brw_tes_prog_data), + [MESA_SHADER_GEOMETRY] = sizeof(struct brw_gs_prog_data), + [MESA_SHADER_FRAGMENT] = sizeof(struct brw_wm_prog_data), + [MESA_SHADER_COMPUTE] = sizeof(struct brw_cs_prog_data), + [MESA_SHADER_KERNEL] = sizeof(struct brw_cs_prog_data), }; assert((int)stage >= 0 && stage < ARRAY_SIZE(stage_sizes)); return stage_sizes[stage]; @@ -257,13 +256,33 @@ unsigned brw_prog_key_size(gl_shader_stage stage) { static const size_t stage_sizes[] = { - sizeof(struct brw_vs_prog_key), - sizeof(struct brw_tcs_prog_key), - sizeof(struct brw_tes_prog_key), - sizeof(struct brw_gs_prog_key), - sizeof(struct brw_wm_prog_key), - sizeof(struct brw_cs_prog_key), + [MESA_SHADER_VERTEX] = sizeof(struct brw_vs_prog_key), + [MESA_SHADER_TESS_CTRL] = sizeof(struct brw_tcs_prog_key), + [MESA_SHADER_TESS_EVAL] = sizeof(struct brw_tes_prog_key), + [MESA_SHADER_GEOMETRY] = sizeof(struct brw_gs_prog_key), + [MESA_SHADER_FRAGMENT] = sizeof(struct brw_wm_prog_key), + [MESA_SHADER_COMPUTE] = sizeof(struct brw_cs_prog_key), + [MESA_SHADER_KERNEL] = sizeof(struct brw_cs_prog_key), }; assert((int)stage >= 0 && stage < ARRAY_SIZE(stage_sizes)); return stage_sizes[stage]; } + +void +brw_write_shader_relocs(const struct gen_device_info *devinfo, + void *program, + const struct brw_stage_prog_data *prog_data, + struct brw_shader_reloc_value *values, + unsigned num_values) +{ + for (unsigned i = 0; i < prog_data->num_relocs; i++) { + assert(prog_data->relocs[i].offset % 8 == 0); + brw_inst *inst = (brw_inst *)(program + prog_data->relocs[i].offset); + for (unsigned j = 0; j < num_values; j++) { + if (prog_data->relocs[i].id == values[j].id) { + brw_update_reloc_imm(devinfo, inst, values[j].value); + break; + } + } + } +}