Tachyon-list Décembre 2010

tachyon-list@iro.umontreal.ca

4 participants
22 discussions

Démarrer une nNouvelle discussion

late
by Marc Feeley 22 Déc '10

22 Déc '10

I'll be at my office at 12:45. Marc

1 0

Runtime error when running js2scm
by Marc Feeley 22 Déc '10

22 Déc '10

Maxime, when I run Tachyon using js2scm, I get the error 'the first input must be a pointer' in this function in ir/instructions: var LoadInstr = instrMaker( 'load', function (typeParams, inputVals, branchTargets) { instrMaker.validNumParams(typeParams, 1, 1); instrMaker.validNumInputs(inputVals, 2, 2); assert ( inputVals[0].type.isPtr(), 'the first input must be a pointer' ); instrMaker.validType(inputVals[1], IRType.pint); this.type = typeParams[0]; } ); When running with v8 that error does not happen. Do you have an idea where this might come from? It could be a bug in js2scm, or a bug in Tachyon. Have you seen this problem before? Marc

1 0

Strangeness in constprop.js
by Marc Feeley 22 Déc '10

22 Déc '10

Maxime, can you check this code in ir/constprop.js : var val0 = (v0.type === IRType.box) ? ((v0.value === undefined || v0.value === null) ? 0 : (v0.value << TAG_NUM_BITS_INT)) // FIXME : v0.value; var val1 = (v1.type === IRType.box) ? ((v1.value === undefined || v1.value === null) ? 0 : (v1.value << TAG_NUM_BITS_INT)) // FIXME : v1.value; It seems that v0.value can be undefined or null. In your original code you directly did a v0.value << TAG_NUM_BITS_INT and this seems like a bug to me (v8 surprisingly returns 0 in those cases). Marc

1 0

Bugfixes & IIR Constants
by chevalma＠iro.umontreal.ca 21 Déc '10

21 Déc '10

I spent several hours yesterday and today hunting bugs in the IR that somehow went unnoticed until now. Notably, the constant propagation algorithm would produce infinite loops over programs containing infinite loops (I had to reread the IBM paper on the topic to figure out what was wrong). This is now fixed, and I have improved the implementation a little bit to deal with more cases. I have also decided to change the implementation of IIR constants, which you all seemed puzzled and displeased with. I implemented small primitive functions that will dynamically cast boxed integers into the desired types (eg: pint, i32, u16, i16). The constant propagation algorithm is able to convert those to constants at compile time when the input to the functions is constant. This deals with negative constant without hacks, without the need for constant propagation over the AST, and I also took the opportunity to lighten the syntax. Before the change: iir.cst special form, essentially serves as a macro to generate typed constants. May be confusing as it looks like a function call or an iir instruction, but it is neither. for (var i = iir.cst(IRType.pint, 0); i < STR_TBL_INIT_SIZE; i += iir.cst(IRType.pint, 1)) { ... } Now: call the function "pint" to get the pint value of an integer. This call becomes a pint constant after constant propagation. Similarly, i16, u16 and i32 are defined for those other types. for (var i = pint(0); i < STR_TBL_INIT_SIZE; i += pint(1)) { ... } - Maxime

1 0

JavaScript to JavaScript can run Tachyon!
by Marc Feeley 20 Déc '10

20 Déc '10

I finished the JavaScript to JavaScript compiler and added an option to dynamically log the location of each function that is called, which can be useful for debugging. Below is a transcript of a sample use to run Tachyon. Marc % time parser/js2js -js -debug utility/debug.js utility/system.js utility/iterators.js utility/graph.js utility/arrays.js utility/hashmap.js utility/linkedlist.js utility/strings.js utility/modules.js utility/misc.js utility/xml.js utility/html.js parser/misc.js parser/scanner.js parser/parser.js parser/pp.js parser/ast-passes.js platform/ffi.js platform/memory.js ir/static.js ir/instructions.js ir/iir.js ir/cfg.js ir/functions.js ir/ir.js ir/optpatterns.js ir/constprop.js ir/commelim.js ir/inlining.js ir/lowering.js frontend.js runtime/objlayout.js runtime/context.js runtime/objects.js codegen/asm.js codegen/asm-x86.js codegen/linearscan.js codegen/backend.js codegen/ir-to-asm-x86.js codegen/regalloc-config-x86.js stdlib/errors.js stdlib/math.js stdlib/arrays.js stdlib/strings.js compiler.js init.js main.js real 0m1.708s user 0m1.573s sys 0m0.060s % time d8 main.js.js > trace.log real 0m6.115s user 0m3.847s sys 0m0.476s % tail -25 trace.log "codegen/asm-x86.js"@391.52-391.91: result: 27060 "init.js"@38.1-47.2: "utility/arrays.js"@139.1-154.2: "utility/debug.js"@52.1-56.2: "utility/debug.js"@43.1-49.2: "utility/debug.js"@46.19-46.47: "utility/debug.js"@31.1-37.2: "utility/arrays.js"@169.33-172.2: "utility/arrays.js"@177.31-180.2: "compiler.js"@29.17-33.2: "utility/arrays.js"@161.32-164.2: "utility/arrays.js"@169.33-172.2: "utility/arrays.js"@177.31-180.2: "compiler.js"@29.17-33.2: "utility/arrays.js"@161.32-164.2: "utility/arrays.js"@169.33-172.2: "utility/arrays.js"@177.31-180.2: "compiler.js"@29.17-33.2: "utility/arrays.js"@161.32-164.2: "utility/arrays.js"@169.33-172.2: "utility/arrays.js"@177.31-180.2: "compiler.js"@29.17-33.2: "utility/arrays.js"@161.32-164.2: "utility/arrays.js"@169.33-172.2:

1 0

More bugs and style issues
by Marc Feeley 18 Déc '10

18 Déc '10

I noticed some uses of floating point numbers. Please make sure the code you wrote does not use floats! The code var s = new Array(Math.floor(this.code.length*1.25)); was rewritten to: var s = new Array((this.code.length*5) >> 2); The following have to be rewritten... but what into? while (!(next === Infinity || current.covers(next))) and assert( k <= 9007199254740991 || k >= -9007199254740991, "Internal error: current scheme cannot garantee an exact" + " representation for signed numbers greater than (2^53-1)" + " or lesser than -(2^53-1)"); There are 38 uses of Infinity in the code and 1 use of NaN... these must be removed! I saw many asserts saying "should" when "must" is the correct term. assert(typeof(name) === "string", "'name' argument should be a string"); It happens so often that it would be abstracted: typeString(name, "name"); Too bad there aren't macros in JS... it would then be simply: typeString(name); The following code is suspect: allocator.min = function (a, accessFct) { if (accessFct === undefined) accessFct = function (x) { return x; }; var minIndex = 0; var minValue = 0; var i; for (i=0; i < a.length; ++i) { if (accessFct(a[i]) > minValue) { minIndex = i; minValue = a[i]; } } return { index:minIndex, value:minValue }; }; shouldn't it be: allocator.min = function (a, accessFct) { if (accessFct === undefined) accessFct = function (x) { return x; }; var minIndex = 0; var minValue = 0; var i; for (i=0; i < a.length; ++i) { var x = accessFct(a[i]); if (x < minValue) // NOTE: < instead of > { minIndex = i; minValue = x; // NOTE: x } } return { index:minIndex, value:minValue }; }; Let me know if that is not correct. Has this code been tested? Moreover, it would be simpler to return the index (the value can be gotten by the client using the index). Also, is the accessFct argument really useful? It doesn't seem to be used in the code! Let's not bloat the compiler needlessly! Marc

3 7

Bootstrapping and changes of object representation
by Erick Lavoie 18 Déc '10

18 Déc '10

I am not completely clear on the bootstrap process and would like your opinion on how best to support the following scenario: Let's suppose we boostrapped tachyon and it is currently running independently. Let's say I want to change the object representation of the Context object and/or the javascript objects and end up running in a new tachyon system with the new representation being used. What should I modify with the current system? What is the process to go from the current system to the new one? Is it mandatory to generate a new image and restart anew? Would we want to support dynamically changing the representation used by the compiler while it is running? Erick

2 1

Address representation issue
by Erick Lavoie 18 Déc '10

18 Déc '10

For linking purposes, in codegen/asm.js asm.address, I store an address in two numbers to circumvent the limitation of the Number representation. It allows me to handle 64 bit addresses by doing the proper operations on the "splitted" address. However, once bootstrapped, the integer representation of numbers will only have 30 bits to represent an address value. I am thinking of further splitting the address into four 16 bits integer for the internal representation. Would you guys have a better idea? Erick

2 1

Number Constants in AST
by Maxime Chevalier-Boisvert 17 Déc '10

17 Déc '10

Marc, would it be easy to make it so that negative constants in the AST (eg: -1) show up as negative number constants, instead of showing up as the negation operator applied to a positive constant? It's posing me a minor issue at the IIR level. - Maxime

2 2

Tachyon bugs
by Marc Feeley 17 Déc '10

17 Déc '10

I have started to bootstrap Tachyon with js2scm to find which library functions are required. In the process I have had to resolve a few issues with the Tachyon sources. In particular the js2scm compiler does not support LabelledStatements yet (i.e. <label> : <statement>) and this is used in a few places to control loop exit. I have rewritten such uses when it was easy, but there remains one use in ir/optpatterns.js that I don't dare modify because the control flow and invariants are not obvious. Maxime: could you take a look and rewrite it without "continue <label>;" statements? I've also discovered that the following global variables are referenced but they are not used anywhere. Please fix this! newdest reg rhsValAssg signExtend u2 upperAddr Also, I have encountered this style of code: if (Error.captureStackTrace) to test for the existence of a particular feature (V8 specific?). Please use this instead: if (Error.captureStackTrace !== undefined) which will make implementing "if" easier. In other words, make sure the test of an if yields a boolean. Marc

3 6

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

Tachyon-list Décembre 2010