blob: 44e57eecca6c33f98d24281499ffe3d21a2cd7a3 [file] [log] [blame]
Dan Gohman10e730a2015-06-29 23:51:55 +00001//===-- README.txt - Notes for WebAssembly code gen -----------------------===//
2
3This WebAssembly backend is presently in a very early stage of development.
4The code should build and not break anything else, but don't expect a lot more
5at this point.
6
7For more information on WebAssembly itself, see the design documents:
8 * https://github.com/WebAssembly/design/blob/master/README.md
9
10The following documents contain some information on the planned semantics and
11binary encoding of WebAssembly itself:
Dan Gohman5d3391f2016-10-24 20:35:17 +000012 * https://github.com/WebAssembly/design/blob/master/Semantics.md
Dan Gohman10e730a2015-06-29 23:51:55 +000013 * https://github.com/WebAssembly/design/blob/master/BinaryEncoding.md
14
JF Bastienf05f6fd2015-12-05 19:36:33 +000015The backend is built, tested and archived on the following waterfall:
JF Bastien4383a342016-01-22 04:21:49 +000016 https://wasm-stat.us
JF Bastienf05f6fd2015-12-05 19:36:33 +000017
18The backend's bringup is done using the GCC torture test suite first since it
19doesn't require C library support. Current known failures are in
20known_gcc_test_failures.txt, all other tests should pass. The waterfall will
21turn red if not. Once most of these pass, further testing will use LLVM's own
JF Bastienc2b30482015-12-09 13:29:32 +000022test suite. The tests can be run locally using:
JF Bastien4383a342016-01-22 04:21:49 +000023 https://github.com/WebAssembly/waterfall/blob/master/src/compile_torture_tests.py
JF Bastienf05f6fd2015-12-05 19:36:33 +000024
Dan Gohman10e730a2015-06-29 23:51:55 +000025//===---------------------------------------------------------------------===//
Dan Gohmandfa81d82015-11-20 03:08:27 +000026
Dan Gohmane0405332016-10-03 22:43:53 +000027Br, br_if, and br_table instructions can support having a value on the value
28stack across the jump (sometimes). We should (a) model this, and (b) extend
29the stackifier to utilize it.
Dan Gohmandfa81d82015-11-20 03:08:27 +000030
31//===---------------------------------------------------------------------===//
Dan Gohman753abf82015-12-06 19:29:54 +000032
Dan Gohmane0405332016-10-03 22:43:53 +000033The min/max instructions aren't exactly a<b?a:b because of NaN and negative zero
Dan Gohman753abf82015-12-06 19:29:54 +000034behavior. The ARM target has the same kind of min/max instructions and has
35implemented optimizations for them; we should do similar optimizations for
36WebAssembly.
37
38//===---------------------------------------------------------------------===//
39
40AArch64 runs SeparateConstOffsetFromGEPPass, followed by EarlyCSE and LICM.
41Would these be useful to run for WebAssembly too? Also, it has an option to
42run SimplifyCFG after running the AtomicExpand pass. Would this be useful for
43us too?
44
45//===---------------------------------------------------------------------===//
46
Dan Gohmane0405332016-10-03 22:43:53 +000047Register stackification uses the VALUE_STACK physical register to impose
Dan Gohman753abf82015-12-06 19:29:54 +000048ordering dependencies on instructions with stack operands. This is pessimistic;
49we should consider alternate ways to model stack dependencies.
50
51//===---------------------------------------------------------------------===//
52
53Lots of things could be done in WebAssemblyTargetTransformInfo.cpp. Similarly,
54there are numerous optimization-related hooks that can be overridden in
55WebAssemblyTargetLowering.
56
57//===---------------------------------------------------------------------===//
58
59Instead of the OptimizeReturned pass, which should consider preserving the
60"returned" attribute through to MachineInstrs and extending the StoreResults
61pass to do this optimization on calls too. That would also let the
62WebAssemblyPeephole pass clean up dead defs for such calls, as it does for
63stores.
64
65//===---------------------------------------------------------------------===//
66
Dan Gohman7f86ca12016-01-16 00:20:03 +000067Consider implementing optimizeSelect, optimizeCompareInstr, optimizeCondBranch,
68optimizeLoadInstr, and/or getMachineCombinerPatterns.
69
70//===---------------------------------------------------------------------===//
71
72Find a clean way to fix the problem which leads to the Shrink Wrapping pass
73being run after the WebAssembly PEI pass.
74
75//===---------------------------------------------------------------------===//
Dan Gohmanadf28172016-01-28 01:22:44 +000076
77When setting multiple local variables to the same constant, we currently get
78code like this:
79
80 i32.const $4=, 0
81 i32.const $3=, 0
82
83It could be done with a smaller encoding like this:
84
85 i32.const $push5=, 0
86 tee_local $push6=, $4=, $pop5
87 copy_local $3=, $pop6
88
89//===---------------------------------------------------------------------===//
Dan Gohman49187022016-02-08 03:42:36 +000090
91WebAssembly registers are implicitly initialized to zero. Explicit zeroing is
92therefore often redundant and could be optimized away.
93
94//===---------------------------------------------------------------------===//
Dan Gohman87e368b2016-02-19 19:22:44 +000095
96Small indices may use smaller encodings than large indices.
Dan Gohman3a71ccb2016-02-20 23:11:14 +000097WebAssemblyRegColoring and/or WebAssemblyRegRenumbering should sort registers
98according to their usage frequency to maximize the usage of smaller encodings.
Dan Gohman87e368b2016-02-19 19:22:44 +000099
100//===---------------------------------------------------------------------===//
Dan Gohmane6b81362016-03-04 20:09:57 +0000101
Dan Gohmanddfa1a62016-03-09 04:17:36 +0000102Many cases of irreducible control flow could be transformed more optimally
103than via the transform in WebAssemblyFixIrreducibleControlFlow.cpp.
104
105It may also be worthwhile to do transforms before register coloring,
106particularly when duplicating code, to allow register coloring to be aware of
107the duplication.
108
109//===---------------------------------------------------------------------===//
Dan Gohman6627e5f2016-05-16 18:51:03 +0000110
111WebAssemblyRegStackify could use AliasAnalysis to reorder loads and stores more
112aggressively.
113
114//===---------------------------------------------------------------------===//
115
116WebAssemblyRegStackify is currently a greedy algorithm. This means that, for
117example, a binary operator will stackify with its user before its operands.
118However, if moving the binary operator to its user moves it to a place where
119its operands can't be moved to, it would be better to leave it in place, or
120perhaps move it up, so that it can stackify its operands. A binary operator
121has two operands and one result, so in such cases there could be a net win by
122prefering the operands.
123
124//===---------------------------------------------------------------------===//
Dan Gohmane045f672016-05-18 20:19:02 +0000125
126Instruction ordering has a significant influence on register stackification and
127coloring. Consider experimenting with the MachineScheduler (enable via
128enableMachineScheduler) and determine if it can be configured to schedule
129instructions advantageously for this purpose.
130
131//===---------------------------------------------------------------------===//
Dan Gohmane0405332016-10-03 22:43:53 +0000132
133WebAssembly is now officially a stack machine, rather than an AST, and this
134comes with additional opportunities for WebAssemblyRegStackify. Specifically,
135the stack doesn't need to be empty after an instruction with no return values.
136WebAssemblyRegStackify could be extended, or possibly rewritten, to take
137advantage of the new opportunities.
138
139//===---------------------------------------------------------------------===//