Unused return parameter pruner #10080

hrkrshnn · 2020-10-20T14:00:12Z

Optimization rule that rewrites functions whose return parameters are not used at callsite.

TODO

Tests in fullsuite
Figure out how to deal with the pop case.
Figure out a good sequence.

hrkrshnn · 2020-10-20T14:02:13Z

libsolidity/interface/OptimiserSettings.h

+			"Tpeiul"                   // Run functional expression inliner
 			"xarulrul"                 // Prune a bit more in SSA
 			"xarrcL"                   // Turn into SSA again and simplify
-			"gvif"                     // Run full inliner
+			"gvPif"                    // Run full inliner


The order needs to be looked at more carefully.

It would be nice if the InlineTiny pass can have its own name. After p (unusedFunctionParameterPruner), only the InlineTiny needs to be run.

hrkrshnn · 2020-10-20T14:02:42Z

libyul/optimiser/Suite.cpp

+		{SSATransform::name,                        'a'},
+		{StructuralSimplifier::name,                't'},
+		{UnusedFunctionParameterPruner::name,       'p'},
+		{UnusedFunctionReturnParameterPruner::name, 'P'},


The only extra line. The rest are indentation changes.

hrkrshnn · 2020-10-20T14:11:09Z

test/libyul/yulOptimizerTests/fullSuite/stack_compressor_msize.yul

@@ -49,7 +49,7 @@
 //         sstore(not(gcd(10, 15)), 1)
 //         sstore(0, 0)
 //         sstore(2, 1)
-//         extcodecopy(1, msize(), 1, 1)
+//         pop(foo_singlereturn_1_78_248(calldataload(0), calldataload(3)))


bad change.

Can you check why this is happening? Does the linking function not fall into the "tiny" category in the full inliner?

hrkrshnn · 2020-10-20T14:12:53Z

test/libyul/yulOptimizerTests/unusedFunctionReturnParameterPruner/pop.yul

+// A test to see if the optimization step can deal with pop
+// Expression splitter converts `pop(f(a))` into


This doesn't work yet. Need to figure out a fix.

The UnusedPruner should remove the pop(z)

UnusedPruner is actually converting it back to pop(f())

{ let a := f() pop(a) function f() -> y { sstore(1, sload(1)) } } // ---- // step: unusedPruner // // { // pop(f()) // function f() -> y // { sstore(1, sload(1)) } // }

chriseth · 2020-10-21T15:22:28Z

libsolidity/interface/OptimiserSettings.h

@@ -41,10 +41,10 @@ struct OptimiserSettings

 			// should have good "compilability" property here.

-			"Tpeul"                    // Run functional expression inliner
+			"Tpeiul"                   // Run functional expression inliner


Why run the full inliner? The comment only mentions the functional expression inliner.

We'll have to run full inliner twice. One after UnusedFunctionParameterPruner and once after this step. The sequence might have to be modified

chriseth · 2020-10-21T15:23:29Z

libyul/optimiser/Suite.cpp

-		{UnusedFunctionParameterPruner::name, 'p'},
-		{UnusedPruner::name,                  'u'},
-		{VarDeclInitializer::name,            'd'},
+		{BlockFlattener::name,                      'f'},


No whitespace-based alignment across lines is a good rule to have in the coding standard :)

libyul/optimiser/UnusedFunctionReturnParameterPruner.h

chriseth · 2020-10-21T15:42:33Z

libyul/optimiser/UnusedFunctionReturnParameterPruner.cpp

+	/// Function name and a boolean mask, where `false` at index `i` indicates that the function
+	/// return-parameter at index `i` in `FunctionDefinition::parameter` is unused at every function
+	/// call-site.
+	map<YulString, vector<bool>> m_callSiteMasks;


Can we again rename the variable to indicate the meaning of the bool directly?

chriseth · 2020-10-21T15:44:36Z

libyul/optimiser/UnusedFunctionReturnParameterPruner.cpp

+			m_callSiteMasks.at(name)[i] =
+				m_callSiteMasks.at(name)[i] || m_references.count(_v.variables[i].name);


Suggested change

m_callSiteMasks.at(name)[i] =

m_callSiteMasks.at(name)[i] || m_references.count(_v.variables[i].name);

if (m_references.count(_v.variables[i].name))

m_callSiteMasks.at(name)[i] = true;

chriseth · 2020-10-21T15:45:20Z

libyul/optimiser/UnusedFunctionReturnParameterPruner.cpp

+	}
+}
+
+/// Returns a map with function name as key and vector of bools such that if the index `i` of the


Can you move the documentation to the declaration?

chriseth · 2020-10-21T15:47:04Z

libyul/optimiser/UnusedFunctionReturnParameterPruner.cpp

+
+/// Returns a map with function name as key and vector of bools such that if the index `i` of the
+/// vector is false, then the function's `i`-th return parameter is unused at *all* callsites.
+map<YulString, vector<bool>> FindFunctionsWithUnusedReturns::functions()


Maybe turn this into a static function and make the constructor private? We don't want to have this run twice if we can avoid it.

libyul/optimiser/UnusedFunctionReturnParameterPruner.cpp

chriseth · 2020-10-21T16:13:32Z

test/libyul/YulOptimizerTest.cpp

@@ -264,6 +265,24 @@ TestCase::TestResult YulOptimizerTest::run(ostream& _stream, string const& _line
 		LiteralRematerialiser::run(*m_context, *m_object->code);
 		UnusedFunctionParameterPruner::run(*m_context, *m_object->code);
 	}
+	else if (m_optimizerStep == "unusedFunctionReturnParameterPruner")


Maybe use this step to test only the unused function return parameter pruner (this can be used to test that it does not modify small enough functions) and only use the "fullSuite" to test that it works efficiently?

chriseth · 2020-10-21T16:16:52Z

test/libyul/yulOptimizerTests/unusedFunctionReturnParameterPruner/all_return_unused.yul

+//     let a, b, c := f_6(sload(0))
+//     function f(d)
+//     {
+//         let x, y, z


Maybe it would be better to declare them separately?

I can do that, however I think it's better to minimize inserting to the front to the vector FunctionDefinition.body.statements. Don't you think a single VariableDeclaration would be better from this perspective?

I would say it is more important that the optimizer can deal nicely with the variables.

Optimization rule that rewrites functions whose return parameters are not used at callsite.

hrkrshnn · 2020-11-20T16:38:12Z

libsolutil/CommonData.h

-		ret += ret;
-	else
-		ret += std::move(_b);
+	assert(&_a != &_b);


In the other version it was not possible to call this operator if T was not copyable.

Is there no solution that works in all cases?

hrkrshnn · 2020-11-20T16:40:28Z

test/libyul/yulOptimizerTests/fullSuite/unusedFunctionReturnParameterPruner_loop.yul

+//         sstore(x, x)
+//         if calldataload(0) { leave }
+//     }
+// }


This test is not working now.

hrkrshnn · 2020-11-20T16:41:06Z

test/libyul/yulOptimizerTests/fullSuite/unusedFunctionReturnParameterPruner_pop.yul

+//     }
+//     function f_17(x) -> y
+//     { f(x) }
+// }


The inliner is not inlining f_17. Not sure why.

hrkrshnn · 2020-11-20T16:45:43Z

test/libyul/yulOptimizerTests/unusedFunctionReturnParameterPruner/pop.yul

+{
+    let a := sload(1)
+    let z := f(a)
+    pop(z)


The step currently removes all pop(identifier) at the beginning. So this test works.

chriseth · 2020-11-23T17:59:43Z

libsolidity/interface/OptimiserSettings.h

@@ -45,6 +45,7 @@ struct OptimiserSettings
 			"xarulrul"                 // Prune a bit more in SSA
 			"xarrcL"                   // Turn into SSA again and simplify
 			"gvif"                     // Run full inliner
+			"xaTPrci"                  // Prune unused return values and inline


Do you need to introduce so many additional steps?

chriseth · 2020-11-23T18:02:15Z

libyul/optimiser/UnusedFunctionReturnParameterPruner.cpp

+ * A helper class that removes expressions of the form `pop(a)` where a is an identifier. Generally,
+ * pop can be the discard function.
+ */
+class PopRemover: public ASTModifier


Why can't we use the UnusedPruner for that?

The more general step would be "remove ExpressionStatements that are movable".

Why can't we use the UnusedPruner for that?

Because of this: #10080 (comment)

It acutally does remove pop(a), but it also removes let a in let a := f(). Since the result of f has to be used somehow, it is turned into pop(f()).

hrkrshnn · 2021-02-03T15:30:30Z

libsolidity/interface/OptimiserSettings.h

@@ -45,6 +45,7 @@ struct OptimiserSettings
 			"xarulrul"                 // Prune a bit more in SSA
 			"xarrcL"                   // Turn into SSA again and simplify
 			"gvif"                     // Run full inliner
+			"xaTPrci"                  // Prune unused return values and inline


@cameel Can you help to determine a good sequence for this, perhaps with the help of yul-phaser? I was expecting the gas cost test test/libsolidity/gasTests/abiv2_optimised.sol to go down. Maybe there is a better sequence.

Well, this sounds like a good opportunity to try out the phaser. Which other input files would you recommend besides test/libsolidity/gasTests/abiv2_optimised.sol?

A metric will be a problem though. Right now I only have a code size metric and even with that weights need some tuning. So I'm not sure if this will be good for finding sequences that specifically remove unused return values.

Maybe test/libsolidity/gasTests/abiv2_optimised.sol is enough to start with.

The code size metric should work, I think.

tl;dr

I tried the phaser today and here's what I found:

The new optimizer step crashes which interferes with the search. The best sequences I found do not include P at all.

It's hard to tell if these sequences are actually any better. I think they're just stealing steps that would be used later anyway.

Phaser is not good at looking for better fragments in the middle of the sequence. I could try to improve it. Adding a --suffix option to be able to test with a full sequence might help.

Running the phaser

solc --ir test/libsolidity/gasTests/abiv2_optimised.sol | grep -v 'IR:' > abiv2_optimised.yul yul-phaser abiv2_optimised.yul \ --show-initial-population \ --min-chromosome-length 1 \ --max-chromosome-length 10 \ --prefix NdhfoDgvulfnTUtnIfxarrscLMcCTUtTOntnfDIulLculVculjjTpeulxarulrulxarrcLgvif --random-population 10 \ --population xaTPrci \ --population P

This creates the initial population out of 10 random optimization sequences plus xaTPrci. Also tells the phaser to pre-optimize each program with NdhfoDgvulfnTUtnIfxarrscLMcCTUtTOntnfDIulLculVculjjTpeulxarulrulxarrcLgvif and only then apply and rate sequences from the population.

Note:

ReasoningBasedSimplifier is very slow. I had to locally remove it from step list in Chromosome::allStepNames() to get reasonable speed. With it, every round takes minutes (even on a RelWithDebInfo build). Without it, it's less than a second per round.

Command-line input validations for some options are incomplete so it can sometimes crash on invalid input. For example make sure you always specify the population and do not use step abbreviations that do not exist. I need to make them more robust.

Phaser output

I did a few runs and selected those that managed to run the longest. Unfortunately there seems to be a bug in this new optimization step and phaser crashes sooner or later (more about it below).

Note that the fitness values are relative to the size of the original program (1000 means no change, 500 means half the size, etc.).

First run (without P in population)

---------- INITIAL POPULATION ---------- 353 TcMrsppPtu 541 lT 841 ojTDV 1000 f 1000 vre 1000 hihcdOMgL 1001 eNvihpcnDg 1135 tmPgthU 1166 xaTPrci 1167 IdOaLI 1286 rhUgfarDm

---------- ROUND 219 [round: 0.2 s, total: 38.2 s] ---------- 227 vcrsujOmLTu 227 vcrsujOmTLu 227 vcrsujOmTOu 227 vcrsujOmTTu 227 vcrsujOmThu 227 vcrsujOmTsu 231 vcrsujOmLhu 251 vcrsupOmThu 261 vcrsvjOosTu 540 nslnxsP 663 vhrsuVOmTOx 1000 L Segmentation fault (core dumped)

Second run

---------- INITIAL POPULATION ---------- 290 PLppOmhuON 496 gurcdhtmf 541 lof 1000 P 1000 T 1000 Vo 1013 jToiNIeta 1030 fC 1136 DoNogmtp 1149 OTUim 1166 xaTPrci 1197 acnC

---------- ROUND 928 [round: 0.3 s, total: 261.1 s] ---------- 226 crTvmsuOj 226 crNTvmsuOj 226 crOTvmsuOj 226 crTDvmsuOj 226 crTIvmsuOj 226 crTLvmsuOj 249 crOvvsuuj 583 crTvmslOj 998 NstM 1000 oD 1000 hMrv 1030 C Segmentation fault (core dumped)

Optimized programs

If you want to see the optimized programs, you can use this command:

yul-phaser abiv2_optimised.yul \ --mode print-optimised-programs \ --rounds 0 \ --show-initial-population \ --prefix NdhfoDgvulfnTUtnIfxarrscLMcCTUtTOntnfDIulLculVculjjTpeulxarulrulxarrcLgvif \ --population <your sequence> \

Here's the output from some of the sequences for convenience:

empty-sequence.yul

vcrsujOmLTu.yul

xaTPrci.yul

crTvmsuOj.yul

PLppOmhuON.yul

nslnxsP.yul

Analysis

First, your sequence got a score of 1166 which means that it actually makes the optimized program longer (at least for this input file). P alone is 1000. This means that it will be easily outcompeted and P will quickly disappear from the sequence.

If there is a combination of steps where P works better, it might reappear but for that I'd have to tweak algorithm settings to make it converge a bit slower - so that it can try more random things before the single best sequence and its mutations take over the whole population.

I see that the best sequences win by reaching for steps from the "CTUcarrLsTOtfDncarrIulc" // SSA plus simplify part. And they mostly don't contain P. These other steps seem to have much bigger influence than P.

One extra factor to P's disadvantage is the crash - phaser does not crash when P is not in the population so naturally the longest runs are likely not to have P in the winning sequence.

There are some interesting sequences in the output that do contain P, though they are actually just random mutations that happened to be in the population snapshots I grabbed and would disappear in the next round:

nslnxsP is much shorter mostly because it removes unused functions. Apart from that there are just a handful of places where calculations were replaced by a constant.

PLppOmhuON has a lot more removed and simplified but it's hard for me to tell how much of it is due to P.

Overall, I think that finding a good sequence for this with phaser won't be possible without some modifications:

A --suffix option (similar to the current --prefix) would prevent phaser from trying to include steps from the later part of our current sequence.

Maybe some constraints on which steps should be used.

Having a better metric might help too.

Crash

This reproduces it immediately for me:

yul-phaser test/libyul/yulOptimizerTests/fullSuite/aztec.yul --population 'P'

Here's the relevant part of the stack trace:

Program received signal SIGSEGV, Segmentation fault. (...) #0 FindFunctionsWithUnusedReturns::operator() (this=0x7fffffffd470, _v=...) at /usr/include/c++/10.2.0/variant:1554 (...) #17 solidity::yul::ASTWalker::operator() (this=this@entry=0x7fffffffd470, _block=...) at /projects/solidity/libyul/optimiser/ASTWalker.cpp:89 #18 0x000055555568df81 in FindFunctionsWithUnusedReturns::functions (_dialect=..., _references=..., _block=...) at /projects/solidity/libyul/optimiser/UnusedFunctionReturnParameterPruner.cpp:168 #19 0x000055555568e5d8 in solidity::yul::UnusedFunctionReturnParameterPruner::run (_context=..., _ast=...) at /projects/solidity/libyul/optimiser/UnusedFunctionReturnParameterPruner.cpp:187 #20 0x00005555555f7ee2 in solidity::phaser::Program::applyOptimisationSteps (_dialect=..., _nameDispenser=..., _ast=std::unique_ptr<solidity::yul::Block> = {...}, _optimisationSteps=...) at /projects/solidity/tools/yulPhaser/Program.cpp:206 #21 0x00005555555f7fb9 in solidity::phaser::Program::optimise (this=this@entry=0x7fffffffdc80, _optimisationSteps=std::vector of length 1, capacity 1 = {...}) at /usr/include/c++/10.2.0/bits/unique_ptr.h:172 (...)

So it basically happens several levels deep in the AST walker code that runs when FindFunctionsWithUnusedReturns gets created and its operator () called.

That --suffix option should be a pretty simple modification so let me know when you deal with the crash. I'll give it another shot.

hrkrshnn · 2021-04-20T07:48:00Z

Closing this for now. The step doesn't seem to reduce costs that much. I'll potentially come back to this in the future

Click for a table of gas differences

file name	IR-Opti	Legacy-Opti	Legacy
interface_inheritance_conversions.sol	18.709%	0.000%	0.000%
abiEncoderV1/abi_decode_v2_storage.sol	0.037%	0.000%	0.000%
abiEncoderV1/abi_encode_calldata_slice.sol	0.415%	0.000%	0.000%
abiEncoderV1/struct/struct_storage_ptr.sol	0.086%	0.000%	0.000%
abiEncoderV2/abi_encode_calldata_slice.sol	0.415%	0.000%	0.000%
abiEncoderV2/abi_encode_v2.sol	0.062%	0.000%	0.000%
abiEncoderV2/abi_encode_v2_in_function_inherited_in_v1_contract.sol	6.860%	0.000%	0.000%
abiEncoderV2/abi_encode_v2_in_modifier_used_in_v1_contract.sol	9.042%	0.000%	0.000%
abiEncoderV2/storage_array_encoding.sol	0.051%	0.000%	0.000%
abiEncoderV2/calldata_array.sol	0.060%	0.000%	0.000%
abiencodedecode/abi_decode_simple_storage.sol	0.055%	0.000%	0.000%
array/arrays_complex_from_and_to_storage.sol	0.036%	0.000%	0.000%
array/byte_array_storage_layout.sol	0.053%	0.000%	0.000%
array/byte_array_transitional_2.sol	0.011%	0.000%	0.000%
array/bytes_length_member.sol	0.069%	0.000%	0.000%
array/create_memory_array.sol	0.047%	0.000%	0.000%
array/dynamic_array_cleanup.sol	0.012%	0.000%	0.000%
array/fixed_array_cleanup.sol	0.017%	0.000%	0.000%
array/fixed_arrays_as_return_type.sol	12.468%	0.000%	0.000%
array/function_array_cross_calls.sol	2.777%	0.000%	0.000%
array/reusing_memory.sol	1.347%	0.000%	0.000%
array/dynamic_arrays_in_storage.sol	-0.202%	0.000%	0.000%
array/dynamic_multi_array_cleanup.sol	-0.937%	0.000%	0.000%
array/copying/array_copy_calldata_storage.sol	0.012%	0.000%	0.000%
array/copying/array_copy_cleanup_uint128.sol	0.099%	0.000%	0.000%
array/copying/array_copy_cleanup_uint40.sol	0.030%	0.000%	0.000%
array/copying/array_copy_clear_storage.sol	0.044%	0.000%	0.000%
array/copying/array_copy_clear_storage_packed.sol	0.066%	0.000%	0.000%
array/copying/array_copy_different_packing.sol	0.075%	0.000%	0.000%
array/copying/array_copy_storage_storage_different_base.sol	0.030%	0.000%	0.000%
array/copying/array_copy_storage_storage_different_base_nested.sol	0.037%	0.000%	0.000%
array/copying/array_copy_storage_storage_dynamic_dynamic.sol	0.027%	0.000%	0.000%
array/copying/array_copy_storage_storage_static_dynamic.sol	0.072%	0.000%	0.000%
array/copying/array_copy_storage_storage_static_static.sol	0.061%	0.000%	0.000%
array/copying/array_copy_target_leftover.sol	0.032%	0.000%	0.000%
array/copying/array_copy_target_leftover2.sol	0.066%	0.000%	0.000%
array/copying/array_copy_target_simple.sol	0.025%	0.000%	0.000%
array/copying/array_copy_target_simple_2.sol	0.027%	0.000%	0.000%
array/copying/array_storage_multi_items_per_slot.sol	0.075%	0.000%	0.000%
array/copying/arrays_from_and_to_storage.sol	0.059%	0.000%	0.000%
array/copying/bytes_inside_mappings.sol	0.053%	0.000%	0.000%
array/copying/bytes_storage_to_storage.sol	0.019%	0.000%	0.000%
array/copying/calldata_array_dynamic_to_storage.sol	0.067%	0.000%	0.000%
array/copying/copy_byte_array_in_struct_to_storage.sol	0.041%	0.000%	0.000%
array/copying/copy_byte_array_to_storage.sol	0.026%	0.000%	0.000%
array/copying/copy_function_storage_array.sol	0.080%	0.000%	0.000%
array/copying/copy_removes_bytes_data.sol	0.043%	0.000%	0.000%
array/copying/storage_memory_nested_bytes.sol	0.036%	0.000%	0.000%
array/copying/storage_memory_packed_dyn.sol	0.034%	0.000%	0.000%
array/copying/array_copy_including_array.sol	-0.035%	0.000%	0.000%
array/copying/array_copy_nested_array.sol	-0.004%	0.000%	0.000%
array/copying/array_copy_storage_storage_struct.sol	-0.314%	0.000%	0.000%
array/copying/array_copy_storage_to_memory_nested.sol	-0.074%	0.000%	0.000%
array/copying/array_nested_calldata_to_storage.sol	-0.130%	0.000%	0.000%
array/copying/array_nested_memory_to_storage.sol	-0.028%	0.000%	0.000%
array/copying/array_of_struct_calldata_to_storage.sol	-0.001%	0.000%	0.000%
array/copying/array_of_struct_memory_to_storage.sol	-0.001%	0.000%	0.000%
array/copying/array_of_structs_containing_arrays_calldata_to_storage.sol	-0.126%	0.000%	0.000%
array/copying/array_of_structs_containing_arrays_memory_to_storage.sol	-0.035%	0.000%	0.000%
array/copying/memory_dyn_2d_bytes_to_storage.sol	-0.250%	0.000%	0.000%
array/copying/storage_memory_nested.sol	-0.206%	0.000%	0.000%
array/copying/storage_memory_nested_from_pointer.sol	-0.206%	0.000%	0.000%
array/copying/storage_memory_nested_struct.sol	-0.148%	0.000%	0.000%
array/delete/bytes_delete_element.sol	0.013%	0.000%	0.000%
array/delete/delete_storage_array_packed.sol	0.070%	0.000%	0.000%
array/pop/array_pop_uint16_transition.sol	0.013%	0.000%	0.000%
array/pop/array_pop_uint24_transition.sol	0.019%	0.000%	0.000%
array/pop/byte_array_pop_copy_long.sol	0.043%	0.000%	0.000%
array/pop/byte_array_pop_long_storage_empty.sol	0.016%	0.000%	0.000%
array/pop/byte_array_pop_long_storage_empty_garbage_ref.sol	0.024%	0.000%	0.000%
array/pop/byte_array_pop_masking_long.sol	0.044%	0.000%	0.000%
array/pop/array_pop_array_transition.sol	0.000%	0.000%	0.000%
array/push/array_push.sol	0.050%	0.000%	0.000%
array/push/array_push_packed_array.sol	0.063%	0.000%	0.000%
array/push/byte_array_push_transition.sol	0.009%	0.000%	0.000%
array/push/push_no_args_bytes.sol	0.017%	0.000%	0.000%
array/push/array_push_nested_from_calldata.sol	0.004%	0.000%	0.000%
array/push/array_push_struct.sol	-0.017%	0.000%	0.000%
array/push/array_push_struct_from_calldata.sol	-0.015%	0.000%	0.000%
array/push/push_no_args_2d.sol	-0.313%	0.000%	0.000%
array/concat/bytes_concat_different_types.sol	0.000%	0.000%	0.000%
constructor/arrays_in_constructors.sol	2.883%	0.000%	0.000%
constructor/bytes_in_constructors_packer.sol	5.710%	0.000%	0.000%
constructor/no_callvalue_check.sol	0.057%	0.000%	0.000%
functionCall/mapping_array_internal_argument.sol	-0.017%	0.000%	0.000%
functionTypes/store_function.sol	11.057%	0.000%	0.000%
immutable/multi_creation.sol	1.590%	0.000%	0.000%
salted_create/salted_create.sol	0.000%	0.000%	0.000%
salted_create/salted_create_with_value.sol	11.335%	0.000%	0.000%
storage/packed_storage_structs_bytes.sol	0.059%	0.000%	0.000%
structs/struct_containing_bytes_copy_and_delete.sol	0.057%	0.000%	0.000%
structs/struct_copy.sol	0.069%	0.000%	0.000%
structs/struct_copy_via_local.sol	-2.976%	0.000%	0.000%
structs/struct_delete_storage_with_array.sol	0.057%	0.000%	0.000%
structs/struct_delete_storage_with_arrays_small.sol	0.060%	0.000%	0.000%
structs/struct_memory_to_storage_function_ptr.sol	0.065%	0.000%	0.000%
structs/structs.sol	0.055%	0.000%	0.000%
structs/memory_structs_nested_load.sol	0.041%	0.000%	0.000%
structs/struct_delete_storage_nested_small.sol	-0.148%	0.000%	0.000%
structs/calldata/calldata_struct_with_nested_array_to_storage.sol	0.036%	0.000%	0.000%
structs/conversion/recursive_storage_memory.sol	0.065%	0.000%	0.000%
various/staticcall_for_view_and_pure.sol	0.000%	0.000%	0.000%
various/swap_in_storage_overwrite.sol	0.070%	0.000%	0.000%
various/destructuring_assignment.sol	-0.019%	0.000%	0.000%
various/skip_dynamic_types_for_structs.sol	0.076%	0.000%	0.000%
viaYul/array_memory_index_access.sol	0.042%	0.000%	0.000%
viaYul/array_storage_index_boundary_test.sol	0.010%	0.000%	0.000%
viaYul/array_storage_length_access.sol	0.002%	0.000%	0.000%
viaYul/array_storage_push_empty.sol	-1.113%	0.000%	0.000%
viaYul/array_storage_push_empty_length_address.sol	0.002%	0.000%	0.000%
viaYul/array_storage_push_pop.sol	0.001%	0.000%	0.000%
viaYul/array_storage_index_access.sol	-0.249%	0.000%	0.000%
viaYul/array_storage_index_zeroed_test.sol	-5.096%	0.000%	0.000%
externalContracts/deposit_contract.sol	0.650%	0.000%	0.000%
externalContracts/snark.sol	-0.677%	0.000%	0.000%
inheritance/address_overload_resolution.sol	14.062%	0.000%	0.000%
inheritance/inherited_function_calldata_calldata_interface.sol	13.440%	0.000%	0.000%
inheritance/inherited_function_calldata_memory_interface.sol	12.222%	0.000%	0.000%

chriseth · 2021-04-20T08:57:48Z

Do you think it may be more useful in combination with #11246 ?

hrkrshnn · 2021-04-22T08:02:54Z

Do you think it may be more useful in combination with #11246 ?

I don't think it should help. But I will rebase and check the costs again once it's merged.

hrkrshnn commented Oct 20, 2020

View reviewed changes

hrkrshnn force-pushed the unused-return-parameter-pruner branch 2 times, most recently from add9b96 to 7755b71 Compare October 20, 2020 14:05

ethereum deleted a comment from stackenbotten Oct 20, 2020

hrkrshnn force-pushed the unused-return-parameter-pruner branch from 7755b71 to 03eeb5e Compare October 20, 2020 14:08

ethereum deleted a comment from stackenbotten Oct 20, 2020

hrkrshnn commented Oct 20, 2020

View reviewed changes

hrkrshnn marked this pull request as draft October 20, 2020 14:13

hrkrshnn mentioned this pull request Oct 20, 2020

Function unused parameter part 2 #9662

Closed

hrkrshnn requested a review from chriseth October 20, 2020 14:56

chriseth reviewed Oct 21, 2020

View reviewed changes

libyul/optimiser/UnusedFunctionReturnParameterPruner.h Outdated Show resolved Hide resolved

chriseth reviewed Oct 21, 2020

View reviewed changes

libyul/optimiser/UnusedFunctionReturnParameterPruner.cpp Outdated Show resolved Hide resolved

chriseth reviewed Oct 21, 2020

View reviewed changes

hrkrshnn force-pushed the unused-return-parameter-pruner branch 2 times, most recently from 918339e to e32984f Compare November 20, 2020 16:28

Implemented UnusedFunctionReturnParameterPruner

4553b8c

Optimization rule that rewrites functions whose return parameters are not used at callsite.

hrkrshnn force-pushed the unused-return-parameter-pruner branch from e32984f to 15eeddd Compare November 20, 2020 16:36

hrkrshnn commented Nov 20, 2020

View reviewed changes

Tests for UnusedReturnParameterPruner

03cbc5f

hrkrshnn force-pushed the unused-return-parameter-pruner branch from 15eeddd to 03cbc5f Compare November 20, 2020 16:47

chriseth reviewed Nov 23, 2020

View reviewed changes

hrkrshnn commented Feb 3, 2021

View reviewed changes

cameel mentioned this pull request Feb 12, 2021

[yul] Long run time of reasoning based simplifier against tests with nested expressions #10954

Closed

hrkrshnn closed this Apr 20, 2021

		// A test to see if the optimization step can deal with pop
		// Expression splitter converts `pop(f(a))` into

		m_callSiteMasks.at(name)[i] =
		m_callSiteMasks.at(name)[i] \|\| m_references.count(_v.variables[i].name);

Unused return parameter pruner #10080

Unused return parameter pruner #10080

Conversation

hrkrshnn commented Oct 20, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

hrkrshnn Nov 20, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cameel Feb 5, 2021 • edited Loading

Choose a reason for hiding this comment

tl;dr

Running the phaser

Phaser output

First run (without P in population)

Second run

Optimized programs

Analysis

Crash

Choose a reason for hiding this comment

hrkrshnn commented Apr 20, 2021

chriseth commented Apr 20, 2021

hrkrshnn commented Apr 22, 2021

hrkrshnn commented Oct 20, 2020 •

edited

Loading

hrkrshnn Nov 20, 2020 •

edited

Loading

cameel Feb 5, 2021 •

edited

Loading

First run (without `P` in population)