From ff964916dc5424495a3e884e02c3b2c8f285b78b Mon Sep 17 00:00:00 2001 From: Kenneth Graunke Date: Fri, 3 Nov 2017 14:52:05 -0700 Subject: [PATCH] i965: Use nir_lower_atomics_to_ssbos and delete ABO compiler code. We use the same hardware mechanism for both atomic counters and SSBO atomics, so there's really no benefit to maintaining separate code to handle each case. Instead, we can just use Rob's shiny new NIR pass to convert atomic_uints to SSBOs, and delete piles of code. The ssbo_start section of the binding table becomes a combined ABO and SSBO section, with ABOs first, then SSBOs. Reviewed-by: Jason Ekstrand --- src/intel/compiler/brw_compiler.h | 1 - src/intel/compiler/brw_fs_nir.cpp | 47 ------------------- src/intel/compiler/brw_shader.cpp | 32 ------------- src/intel/compiler/brw_shader.h | 2 - src/intel/compiler/brw_vec4_nir.cpp | 46 ------------------ src/mesa/drivers/dri/i965/brw_link.cpp | 2 + src/mesa/drivers/dri/i965/brw_program.c | 12 ++--- .../drivers/dri/i965/brw_wm_surface_state.c | 5 +- 8 files changed, 8 insertions(+), 139 deletions(-) diff --git a/src/intel/compiler/brw_compiler.h b/src/intel/compiler/brw_compiler.h index df6ee018546..c17eaed946e 100644 --- a/src/intel/compiler/brw_compiler.h +++ b/src/intel/compiler/brw_compiler.h @@ -581,7 +581,6 @@ struct brw_stage_prog_data { uint32_t gather_texture_start; uint32_t ubo_start; uint32_t ssbo_start; - uint32_t abo_start; uint32_t image_start; uint32_t shader_time_start; uint32_t plane_start[3]; diff --git a/src/intel/compiler/brw_fs_nir.cpp b/src/intel/compiler/brw_fs_nir.cpp index ca3707db90f..bed1cd3b492 100644 --- a/src/intel/compiler/brw_fs_nir.cpp +++ b/src/intel/compiler/brw_fs_nir.cpp @@ -3645,53 +3645,6 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr dest = get_nir_dest(instr->dest); switch (instr->intrinsic) { - case nir_intrinsic_atomic_counter_inc: - case nir_intrinsic_atomic_counter_dec: - case nir_intrinsic_atomic_counter_read: - case nir_intrinsic_atomic_counter_add: - case nir_intrinsic_atomic_counter_min: - case nir_intrinsic_atomic_counter_max: - case nir_intrinsic_atomic_counter_and: - case nir_intrinsic_atomic_counter_or: - case nir_intrinsic_atomic_counter_xor: - case nir_intrinsic_atomic_counter_exchange: - case nir_intrinsic_atomic_counter_comp_swap: { - if (stage == MESA_SHADER_FRAGMENT && - instr->intrinsic != nir_intrinsic_atomic_counter_read) - brw_wm_prog_data(prog_data)->has_side_effects = true; - - /* Get some metadata from the image intrinsic. */ - const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; - - /* Get the arguments of the atomic intrinsic. */ - const fs_reg offset = get_nir_src(instr->src[0]); - const unsigned surface = (stage_prog_data->binding_table.abo_start + - instr->const_index[0]); - const fs_reg src0 = (info->num_srcs >= 2 - ? get_nir_src(instr->src[1]) : fs_reg()); - const fs_reg src1 = (info->num_srcs >= 3 - ? get_nir_src(instr->src[2]) : fs_reg()); - fs_reg tmp; - - assert(info->num_srcs <= 3); - - /* Emit a surface read or atomic op. */ - if (instr->intrinsic == nir_intrinsic_atomic_counter_read) { - tmp = emit_untyped_read(bld, brw_imm_ud(surface), offset, 1, 1); - } else { - tmp = emit_untyped_atomic(bld, brw_imm_ud(surface), offset, src0, - src1, 1, 1, - get_atomic_counter_op(instr->intrinsic)); - } - - /* Assign the result. */ - bld.MOV(retype(dest, BRW_REGISTER_TYPE_UD), tmp); - - /* Mark the surface as used. */ - brw_mark_surface_used(stage_prog_data, surface); - break; - } - case nir_intrinsic_image_load: case nir_intrinsic_image_store: case nir_intrinsic_image_atomic_add: diff --git a/src/intel/compiler/brw_shader.cpp b/src/intel/compiler/brw_shader.cpp index b27a71c3c27..ba61481a0af 100644 --- a/src/intel/compiler/brw_shader.cpp +++ b/src/intel/compiler/brw_shader.cpp @@ -622,38 +622,6 @@ brw_abs_immediate(enum brw_reg_type type, struct brw_reg *reg) return false; } -/** - * Get the appropriate atomic op for an image atomic intrinsic. - */ -unsigned -get_atomic_counter_op(nir_intrinsic_op op) -{ - switch (op) { - case nir_intrinsic_atomic_counter_inc: - return BRW_AOP_INC; - case nir_intrinsic_atomic_counter_dec: - return BRW_AOP_PREDEC; - case nir_intrinsic_atomic_counter_add: - return BRW_AOP_ADD; - case nir_intrinsic_atomic_counter_min: - return BRW_AOP_UMIN; - case nir_intrinsic_atomic_counter_max: - return BRW_AOP_UMAX; - case nir_intrinsic_atomic_counter_and: - return BRW_AOP_AND; - case nir_intrinsic_atomic_counter_or: - return BRW_AOP_OR; - case nir_intrinsic_atomic_counter_xor: - return BRW_AOP_XOR; - case nir_intrinsic_atomic_counter_exchange: - return BRW_AOP_MOV; - case nir_intrinsic_atomic_counter_comp_swap: - return BRW_AOP_CMPWR; - default: - unreachable("Not reachable."); - } -} - backend_shader::backend_shader(const struct brw_compiler *compiler, void *log_data, void *mem_ctx, diff --git a/src/intel/compiler/brw_shader.h b/src/intel/compiler/brw_shader.h index ed434acf5a0..06abdc4d175 100644 --- a/src/intel/compiler/brw_shader.h +++ b/src/intel/compiler/brw_shader.h @@ -286,8 +286,6 @@ struct brw_gs_compile unsigned control_data_header_size_bits; }; -unsigned get_atomic_counter_op(nir_intrinsic_op op); - #ifdef __cplusplus } #endif diff --git a/src/intel/compiler/brw_vec4_nir.cpp b/src/intel/compiler/brw_vec4_nir.cpp index 0a1caa9fad8..c4ea24b8db7 100644 --- a/src/intel/compiler/brw_vec4_nir.cpp +++ b/src/intel/compiler/brw_vec4_nir.cpp @@ -804,52 +804,6 @@ vec4_visitor::nir_emit_intrinsic(nir_intrinsic_instr *instr) break; } - case nir_intrinsic_atomic_counter_inc: - case nir_intrinsic_atomic_counter_dec: - case nir_intrinsic_atomic_counter_read: - case nir_intrinsic_atomic_counter_add: - case nir_intrinsic_atomic_counter_min: - case nir_intrinsic_atomic_counter_max: - case nir_intrinsic_atomic_counter_and: - case nir_intrinsic_atomic_counter_or: - case nir_intrinsic_atomic_counter_xor: - case nir_intrinsic_atomic_counter_exchange: - case nir_intrinsic_atomic_counter_comp_swap: { - unsigned surf_index = prog_data->base.binding_table.abo_start + - (unsigned) instr->const_index[0]; - const vec4_builder bld = - vec4_builder(this).at_end().annotate(current_annotation, base_ir); - - /* Get some metadata from the image intrinsic. */ - const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; - - /* Get the arguments of the atomic intrinsic. */ - src_reg offset = get_nir_src(instr->src[0], nir_type_int32, - instr->num_components); - const src_reg surface = brw_imm_ud(surf_index); - const src_reg src0 = (info->num_srcs >= 2 - ? get_nir_src(instr->src[1]) : src_reg()); - const src_reg src1 = (info->num_srcs >= 3 - ? get_nir_src(instr->src[2]) : src_reg()); - - src_reg tmp; - - dest = get_nir_dest(instr->dest); - - if (instr->intrinsic == nir_intrinsic_atomic_counter_read) { - tmp = emit_untyped_read(bld, surface, offset, 1, 1); - } else { - tmp = emit_untyped_atomic(bld, surface, offset, - src0, src1, - 1, 1, - get_atomic_counter_op(instr->intrinsic)); - } - - bld.MOV(retype(dest, tmp.type), tmp); - brw_mark_surface_used(stage_prog_data, surf_index); - break; - } - case nir_intrinsic_load_ubo: { nir_const_value *const_block_index = nir_src_as_const_value(instr->src[0]); src_reg surf_index; diff --git a/src/mesa/drivers/dri/i965/brw_link.cpp b/src/mesa/drivers/dri/i965/brw_link.cpp index 24545d52ecb..d18521e792d 100644 --- a/src/mesa/drivers/dri/i965/brw_link.cpp +++ b/src/mesa/drivers/dri/i965/brw_link.cpp @@ -299,6 +299,8 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg) NIR_PASS_V(prog->nir, nir_lower_samplers, shProg); NIR_PASS_V(prog->nir, nir_lower_atomics, shProg); + NIR_PASS_V(prog->nir, nir_lower_atomics_to_ssbo, + prog->nir->info.num_abos); if (brw->ctx.Cache) { struct blob writer; diff --git a/src/mesa/drivers/dri/i965/brw_program.c b/src/mesa/drivers/dri/i965/brw_program.c index 5ecfb9f5b11..755d4973cc0 100644 --- a/src/mesa/drivers/dri/i965/brw_program.c +++ b/src/mesa/drivers/dri/i965/brw_program.c @@ -732,10 +732,11 @@ brw_assign_common_binding_table_offsets(const struct gen_device_info *devinfo, stage_prog_data->binding_table.ubo_start = 0xd0d0d0d0; } - if (prog->info.num_ssbos) { + if (prog->info.num_ssbos || prog->info.num_abos) { + assert(prog->info.num_abos <= BRW_MAX_ABO); assert(prog->info.num_ssbos <= BRW_MAX_SSBO); stage_prog_data->binding_table.ssbo_start = next_binding_table_offset; - next_binding_table_offset += prog->info.num_ssbos; + next_binding_table_offset += prog->info.num_abos + prog->info.num_ssbos; } else { stage_prog_data->binding_table.ssbo_start = 0xd0d0d0d0; } @@ -759,13 +760,6 @@ brw_assign_common_binding_table_offsets(const struct gen_device_info *devinfo, stage_prog_data->binding_table.gather_texture_start = 0xd0d0d0d0; } - if (prog->info.num_abos) { - stage_prog_data->binding_table.abo_start = next_binding_table_offset; - next_binding_table_offset += prog->info.num_abos; - } else { - stage_prog_data->binding_table.abo_start = 0xd0d0d0d0; - } - if (prog->info.num_images) { stage_prog_data->binding_table.image_start = next_binding_table_offset; next_binding_table_offset += prog->info.num_images; diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c index bcd82d03e7c..a4804401fc5 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c @@ -1284,7 +1284,8 @@ brw_upload_ubo_surfaces(struct brw_context *brw, struct gl_program *prog, } uint32_t *ssbo_surf_offsets = - &stage_state->surf_offset[prog_data->binding_table.ssbo_start]; + &stage_state->surf_offset[prog_data->binding_table.ssbo_start] + + prog->info.num_abos; for (int i = 0; i < prog->info.num_ssbos; i++) { struct gl_buffer_binding *binding = @@ -1351,7 +1352,7 @@ brw_upload_abo_surfaces(struct brw_context *brw, { struct gl_context *ctx = &brw->ctx; uint32_t *surf_offsets = - &stage_state->surf_offset[prog_data->binding_table.abo_start]; + &stage_state->surf_offset[prog_data->binding_table.ssbo_start]; if (prog->info.num_abos) { for (unsigned i = 0; i < prog->info.num_abos; i++) { -- 2.30.2