diff options
Diffstat (limited to 'src/intel')
-rw-r--r-- | src/intel/compiler/brw_eu.h | 2 | ||||
-rw-r--r-- | src/intel/compiler/brw_eu_emit.c | 8 | ||||
-rw-r--r-- | src/intel/compiler/brw_fs_generator.cpp | 9 | ||||
-rw-r--r-- | src/intel/compiler/brw_vec4_generator.cpp | 9 |
4 files changed, 20 insertions, 8 deletions
diff --git a/src/intel/compiler/brw_eu.h b/src/intel/compiler/brw_eu.h index 591d01386fe..96c22ab429a 100644 --- a/src/intel/compiler/brw_eu.h +++ b/src/intel/compiler/brw_eu.h @@ -1148,7 +1148,7 @@ brw_untyped_surface_write(struct brw_codegen *p, unsigned num_channels, bool header_present); -void +unsigned brw_memory_fence(struct brw_codegen *p, struct brw_reg dst, struct brw_reg src, diff --git a/src/intel/compiler/brw_eu_emit.c b/src/intel/compiler/brw_eu_emit.c index 1938dd65f4d..83f7f4a62ca 100644 --- a/src/intel/compiler/brw_eu_emit.c +++ b/src/intel/compiler/brw_eu_emit.c @@ -3145,7 +3145,7 @@ brw_set_memory_fence_message(struct brw_codegen *p, brw_inst_set_binding_table_index(devinfo, insn, bti); } -void +unsigned brw_memory_fence(struct brw_codegen *p, struct brw_reg dst, struct brw_reg src, @@ -3159,6 +3159,8 @@ brw_memory_fence(struct brw_codegen *p, (devinfo->gen == 7 && !devinfo->is_haswell); struct brw_inst *insn; + unsigned fences = 0; + brw_push_insn_state(p); brw_set_default_mask_control(p, BRW_MASK_DISABLE); brw_set_default_exec_size(p, BRW_EXECUTE_1); @@ -3173,6 +3175,7 @@ brw_memory_fence(struct brw_codegen *p, brw_set_src0(p, insn, src); brw_set_memory_fence_message(p, insn, GEN7_SFID_DATAPORT_DATA_CACHE, commit_enable, bti); + fences++; if (devinfo->gen == 7 && !devinfo->is_haswell) { /* IVB does typed surface access through the render cache, so we need to @@ -3184,6 +3187,7 @@ brw_memory_fence(struct brw_codegen *p, brw_set_src0(p, insn, src); brw_set_memory_fence_message(p, insn, GEN6_SFID_DATAPORT_RENDER_CACHE, commit_enable, bti); + fences++; /* Now write the response of the second message into the response of the * first to trigger a pipeline stall -- This way future render and data @@ -3201,6 +3205,8 @@ brw_memory_fence(struct brw_codegen *p, } brw_pop_insn_state(p); + + return fences; } void diff --git a/src/intel/compiler/brw_fs_generator.cpp b/src/intel/compiler/brw_fs_generator.cpp index 8afc075f187..b50f03142b5 100644 --- a/src/intel/compiler/brw_fs_generator.cpp +++ b/src/intel/compiler/brw_fs_generator.cpp @@ -2216,12 +2216,15 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width, generate_shader_time_add(inst, src[0], src[1], src[2]); break; - case SHADER_OPCODE_MEMORY_FENCE: + case SHADER_OPCODE_MEMORY_FENCE: { assert(src[1].file == BRW_IMMEDIATE_VALUE); assert(src[2].file == BRW_IMMEDIATE_VALUE); - brw_memory_fence(p, dst, src[0], BRW_OPCODE_SEND, src[1].ud, src[2].ud); - send_count++; + const unsigned sends = + brw_memory_fence(p, dst, src[0], BRW_OPCODE_SEND, src[1].ud, + src[2].ud); + send_count += sends; break; + } case FS_OPCODE_SCHEDULING_FENCE: if (unlikely(debug_flag)) diff --git a/src/intel/compiler/brw_vec4_generator.cpp b/src/intel/compiler/brw_vec4_generator.cpp index c247c988181..be5eaf43ca0 100644 --- a/src/intel/compiler/brw_vec4_generator.cpp +++ b/src/intel/compiler/brw_vec4_generator.cpp @@ -1910,10 +1910,13 @@ generate_code(struct brw_codegen *p, send_count++; break; - case SHADER_OPCODE_MEMORY_FENCE: - brw_memory_fence(p, dst, src[0], BRW_OPCODE_SEND, false, /* bti */ 0); - send_count++; + case SHADER_OPCODE_MEMORY_FENCE: { + const unsigned sends = + brw_memory_fence(p, dst, src[0], BRW_OPCODE_SEND, false, + /* bti */ 0); + send_count += sends; break; + } case SHADER_OPCODE_FIND_LIVE_CHANNEL: { const struct brw_reg mask = |