diff options
author | Francisco Jerez <[email protected]> | 2017-12-12 12:05:04 -0800 |
---|---|---|
committer | Francisco Jerez <[email protected]> | 2018-03-02 11:28:56 -0800 |
commit | c063e88909e630bb4605037eb0fc072f40f8c2a2 (patch) | |
tree | b67886f38f7584466647081a28209d65a772378b /src/intel/compiler | |
parent | e7c9adca5726a8c96de20ae7c5f21a30061db392 (diff) |
intel/fs: Handle surface opcode sample masks via predication.
The main motivation is to enable HDC surface opcodes on ICL which no
longer allows the sample mask to be provided in a message header, but
this is enabled all the way back to IVB when possible because it
decreases the instruction count of some shaders using HDC messages
significantly, e.g. one of the SynMark2 CSDof compute shaders
decreases instruction count by about 40% due to the removal of header
setup boilerplate which in turn makes a number of send message
payloads more easily CSE-able. Shader-db results on SKL:
total instructions in shared programs: 15325319 -> 15314384 (-0.07%)
instructions in affected programs: 311532 -> 300597 (-3.51%)
helped: 491
HURT: 1
Shader-db results on BDW where the optimization needs to be disabled
in some cases due to hardware restrictions:
total instructions in shared programs: 15604794 -> 15598028 (-0.04%)
instructions in affected programs: 220863 -> 214097 (-3.06%)
helped: 351
HURT: 0
The FPS of SynMark2 CSDof improves by 5.09% ±0.36% (n=10) on my SKL
laptop with this change. According to Eero this improves performance
of the same test by 9% on BYT and by 7-8% on BXT J4205 and on SKL GT2
desktop.
Reviewed-by: Kenneth Graunke <[email protected]>
Tested-By: Eero Tamminen <[email protected]>
Diffstat (limited to 'src/intel/compiler')
-rw-r--r-- | src/intel/compiler/brw_fs.cpp | 43 |
1 files changed, 42 insertions, 1 deletions
diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp index c255a3b23b5..b1e1d98f6e6 100644 --- a/src/intel/compiler/brw_fs.cpp +++ b/src/intel/compiler/brw_fs.cpp @@ -4460,6 +4460,8 @@ static void lower_surface_logical_send(const fs_builder &bld, fs_inst *inst, opcode op, const fs_reg &sample_mask) { + const gen_device_info *devinfo = bld.shader->devinfo; + /* Get the logical send arguments. */ const fs_reg &addr = inst->src[0]; const fs_reg &src = inst->src[1]; @@ -4470,7 +4472,20 @@ lower_surface_logical_send(const fs_builder &bld, fs_inst *inst, opcode op, /* Calculate the total number of components of the payload. */ const unsigned addr_sz = inst->components_read(0); const unsigned src_sz = inst->components_read(1); - const unsigned header_sz = (sample_mask.file == BAD_FILE ? 0 : 1); + /* From the BDW PRM Volume 7, page 147: + * + * "For the Data Cache Data Port*, the header must be present for the + * following message types: [...] Typed read/write/atomics" + * + * Earlier generations have a similar wording. Because of this restriction + * we don't attempt to implement sample masks via predication for such + * messages prior to Gen9, since we have to provide a header anyway. On + * Gen11+ the header has been removed so we can only use predication. + */ + const unsigned header_sz = devinfo->gen < 9 && + (op == SHADER_OPCODE_TYPED_SURFACE_READ || + op == SHADER_OPCODE_TYPED_SURFACE_WRITE || + op == SHADER_OPCODE_TYPED_ATOMIC) ? 1 : 0; const unsigned sz = header_sz + addr_sz + src_sz; /* Allocate space for the payload. */ @@ -4490,6 +4505,32 @@ lower_surface_logical_send(const fs_builder &bld, fs_inst *inst, opcode op, bld.LOAD_PAYLOAD(payload, components, sz, header_sz); + /* Predicate the instruction on the sample mask if no header is + * provided. + */ + if (!header_sz && sample_mask.file != BAD_FILE && + sample_mask.file != IMM) { + const fs_builder ubld = bld.group(1, 0).exec_all(); + if (inst->predicate) { + assert(inst->predicate == BRW_PREDICATE_NORMAL); + assert(!inst->predicate_inverse); + assert(inst->flag_subreg < 2); + /* Combine the sample mask with the existing predicate by using a + * vertical predication mode. + */ + inst->predicate = BRW_PREDICATE_ALIGN1_ALLV; + ubld.MOV(retype(brw_flag_subreg(inst->flag_subreg + 2), + sample_mask.type), + sample_mask); + } else { + inst->flag_subreg = 2; + inst->predicate = BRW_PREDICATE_NORMAL; + inst->predicate_inverse = false; + ubld.MOV(retype(brw_flag_subreg(inst->flag_subreg), sample_mask.type), + sample_mask); + } + } + /* Update the original instruction. */ inst->opcode = op; inst->mlen = header_sz + (addr_sz + src_sz) * inst->exec_size / 8; |