[Mesa-dev] [PATCH] intel/fs: Restrict live intervals to the subset possibly reachable from any definition.
Francisco Jerez
currojerez at riseup.net
Sat Sep 9 21:53:59 UTC 2017
Chema Casanova <jmcasanova at igalia.com> writes:
> El 08/09/17 a las 11:06, Alejandro Piñeiro escribió:
>> On 08/09/17 02:50, Francisco Jerez wrote:
>>> Currently the liveness analysis pass would extend a live interval up
>>> to the top of the program when no unconditional and complete
>>> definition of the variable is found that dominates all of its uses.
>>>
>>> This can lead to a serious performance problem in shaders containing
>>> many partial writes, like scalar arithmetic, FP64 and soon FP16
>>> operations.
>> Just tested with the VK_KHR_16bit_storage implementation. Works really
>> fine with the most problematic tests, so we can drop the "i965/fs: Add
>> reuse_16bit_conversions_register optimization" patch (that was already
>> NAKed by both you and Connor).
>>
>> My test was limited to that extension CTS tests, but in case it helps:
>> Tested-by: Alejandro Piñeiro <apinheiro at igalia.com>
>
> I've seen one regression on a full VK-CTS run with this patch over the
> VK_KHR_16bit_storage branch, but I can reproduce the same regression
> applying it on master.
>
> Failing test is:
> dEQP-VK.glsl.return.return_in_dynamic_loop_dynamic_vertex
>
Good finding! I've been digging into this regression some more and it
looks like the inaccuracy of the liveness analysis pass was protecting
us from a long-standing bug of the dataflow analysis code related to
cross-channel variable interference in non-uniformly executed loops.
I had been ranting about exactly this problem for years, but never
bothered to fix it. Of course it had to come back to bite me as if
destiny wanted to teach me some sort of distasteful lesson. :P
I'm amazed that *none* of the GL CTS, dEQP or Piglit test suite managed
to expose this issue, and that the one dEQP-VK test that could have
triggered the problem was passing basically by luck (with a small,
seemingly harmless change to the source code of the test case it would
have exploded regardless of this patch).
Anyway, will come back with a fix, probably next week.
>>> The number of oversize live intervals in such workloads
>>> can cause the compilation time of the shader to explode because of the
>>> worse than quadratic behavior of the register allocator and scheduler
>>> when running out of registers, and it can also cause the running time
>>> of the shader to explode due to the amount of spilling it leads to,
>>> which is orders of magnitude slower than GRF memory.
>>>
>>> This patch fixes it by computing the intersection of our current live
>>> intervals with the subset of the program that can possibly be reached
>>> from any definition of the variable. Extending the storage allocation
>>> of the variable beyond that is pretty useless because its value is
>>> guaranteed to be undefined at a point that cannot be reached from any
>>> definition.
>>>
>>> No significant change in the running time of shader-db (with 5%
>>> statistical significance).
>>>
>>> shader-db results on IVB:
>>>
>>> total cycles in shared programs: 61108780 -> 60932856 (-0.29%)
>>> cycles in affected programs: 16335482 -> 16159558 (-1.08%)
>>> helped: 5121
>>> HURT: 4347
>>>
>>> total spills in shared programs: 1309 -> 1288 (-1.60%)
>>> spills in affected programs: 249 -> 228 (-8.43%)
>>> helped: 3
>>> HURT: 0
>>>
>>> total fills in shared programs: 1652 -> 1597 (-3.33%)
>>> fills in affected programs: 262 -> 207 (-20.99%)
>>> helped: 4
>>> HURT: 0
>>>
>>> LOST: 2
>>> GAINED: 209
>>>
>>> shader-db results on BDW:
>>>
>>> total cycles in shared programs: 67617262 -> 67361220 (-0.38%)
>>> cycles in affected programs: 23397142 -> 23141100 (-1.09%)
>>> helped: 8045
>>> HURT: 6488
>>>
>>> total spills in shared programs: 1456 -> 1252 (-14.01%)
>>> spills in affected programs: 465 -> 261 (-43.87%)
>>> helped: 3
>>> HURT: 0
>>>
>>> total fills in shared programs: 1720 -> 1465 (-14.83%)
>>> fills in affected programs: 471 -> 216 (-54.14%)
>>> helped: 4
>>> HURT: 0
>>>
>>> LOST: 2
>>> GAINED: 162
>>>
>>> shader-db results on SKL:
>>>
>>> total cycles in shared programs: 65436248 -> 65245186 (-0.29%)
>>> cycles in affected programs: 22560936 -> 22369874 (-0.85%)
>>> helped: 8457
>>> HURT: 6247
>>>
>>> total spills in shared programs: 437 -> 437 (0.00%)
>>> spills in affected programs: 0 -> 0
>>> helped: 0
>>> HURT: 0
>>>
>>> total fills in shared programs: 870 -> 854 (-1.84%)
>>> fills in affected programs: 16 -> 0
>>> helped: 1
>>> HURT: 0
>>>
>>> LOST: 0
>>> GAINED: 107
>>> ---
>>> src/intel/compiler/brw_fs_live_variables.cpp | 34 ++++++++++++++++++++++++----
>>> src/intel/compiler/brw_fs_live_variables.h | 12 ++++++++++
>>> 2 files changed, 42 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/src/intel/compiler/brw_fs_live_variables.cpp b/src/intel/compiler/brw_fs_live_variables.cpp
>>> index c449672a519..059f076fa51 100644
>>> --- a/src/intel/compiler/brw_fs_live_variables.cpp
>>> +++ b/src/intel/compiler/brw_fs_live_variables.cpp
>>> @@ -83,9 +83,11 @@ fs_live_variables::setup_one_write(struct block_data *bd, fs_inst *inst,
>>> /* The def[] bitset marks when an initialization in a block completely
>>> * screens off previous updates of that variable (VGRF channel).
>>> */
>>> - if (inst->dst.file == VGRF && !inst->is_partial_write()) {
>>> - if (!BITSET_TEST(bd->use, var))
>>> + if (inst->dst.file == VGRF) {
>>> + if (!inst->is_partial_write() && !BITSET_TEST(bd->use, var))
>>> BITSET_SET(bd->def, var);
>>> +
>>> + BITSET_SET(bd->defout, var);
>>> }
>>> }
>>>
>>> @@ -199,6 +201,28 @@ fs_live_variables::compute_live_variables()
>>> }
>>> }
>>> }
>>> +
>>> + /* Propagate defin and defout down the CFG to calculate the union of live
>>> + * variables potentially defined along any possible control flow path.
>>> + */
>>> + do {
>>> + cont = false;
>>> +
>>> + foreach_block (block, cfg) {
>>> + const struct block_data *bd = &block_data[block->num];
>>> +
>>> + foreach_list_typed(bblock_link, child_link, link, &block->children) {
>>> + struct block_data *child_bd = &block_data[child_link->block->num];
>>> +
>>> + for (int i = 0; i < bitset_words; i++) {
>>> + const BITSET_WORD new_def = bd->defout[i] & ~child_bd->defin[i];
>>> + child_bd->defin[i] |= new_def;
>>> + child_bd->defout[i] |= new_def;
>>> + cont |= new_def;
>>> + }
>>> + }
>>> + }
>>> + } while (cont);
>>> }
>>>
>>> /**
>>> @@ -212,12 +236,12 @@ fs_live_variables::compute_start_end()
>>> struct block_data *bd = &block_data[block->num];
>>>
>>> for (int i = 0; i < num_vars; i++) {
>>> - if (BITSET_TEST(bd->livein, i)) {
>>> + if (BITSET_TEST(bd->livein, i) && BITSET_TEST(bd->defin, i)) {
>>> start[i] = MIN2(start[i], block->start_ip);
>>> end[i] = MAX2(end[i], block->start_ip);
>>> }
>>>
>>> - if (BITSET_TEST(bd->liveout, i)) {
>>> + if (BITSET_TEST(bd->liveout, i) && BITSET_TEST(bd->defout, i)) {
>>> start[i] = MIN2(start[i], block->end_ip);
>>> end[i] = MAX2(end[i], block->end_ip);
>>> }
>>> @@ -260,6 +284,8 @@ fs_live_variables::fs_live_variables(fs_visitor *v, const cfg_t *cfg)
>>> block_data[i].use = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
>>> block_data[i].livein = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
>>> block_data[i].liveout = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
>>> + block_data[i].defin = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
>>> + block_data[i].defout = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
>>>
>>> block_data[i].flag_def[0] = 0;
>>> block_data[i].flag_use[0] = 0;
>>> diff --git a/src/intel/compiler/brw_fs_live_variables.h b/src/intel/compiler/brw_fs_live_variables.h
>>> index d2d5898ed1c..9e95e443170 100644
>>> --- a/src/intel/compiler/brw_fs_live_variables.h
>>> +++ b/src/intel/compiler/brw_fs_live_variables.h
>>> @@ -55,6 +55,18 @@ struct block_data {
>>> /** Which defs reach the exit point of the block. */
>>> BITSET_WORD *liveout;
>>>
>>> + /**
>>> + * Variables such that the entry point of the block may be reached from any
>>> + * of their definitions.
>>> + */
>>> + BITSET_WORD *defin;
>>> +
>>> + /**
>>> + * Variables such that the exit point of the block may be reached from any
>>> + * of their definitions.
>>> + */
>>> + BITSET_WORD *defout;
>>> +
>>> BITSET_WORD flag_def[1];
>>> BITSET_WORD flag_use[1];
>>> BITSET_WORD flag_livein[1];
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 212 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170909/c41510c0/attachment-0001.sig>
More information about the mesa-dev
mailing list