Working on OpenCL support

Commiting OpenCL implementation does not include Diderot library functions support

Final touches on the cl code

Final touches on the cl code

  Working on command-line option support

  Working on compiler support for command-line options

  remove stuff from sources.cm

  Start to factor common C code support

Added globals support to host side for the OpenCL

  Push input descriptor string through compiler phases.

  Changed syntax of input variables to require literal definitions and to
  allow a description string.

  Don't need a duplicate copy of run-cc.sml in cl-target; use the one in c-target

Added the code for generating OpenCL

  Split input functions into their own file and added Diderot_InputVec2

  Changed the naming convention for Input functions to precision independent

  added some missing cleanup

  Configuration hacking

  Remove debugging output

  Working on support for parallel runtime

  Remove -march=native option from C compiles

  Working on parallel runtime

  Adding general support for command-line options

  Working on command-line options

  Working on command-line option support

  Minor fixes

  Working on multiple-target support

  Added configure checks for OpenCL and made the codegen sources.cm file be a
  generated file

  Starting to work on supporting multiple targets

  Add framework to support command-line control of optimization

  Bug fix: support multidimensional image types

  Lots of work on getting the IL invariants right.

  Split normalization part of HighIL optimization into its own module

  Working on debugging value numbering

  Added ilName to SSA signature

  Working copy of SML/NJ HashCons library, since the version distributed with 110.72 has
  a fatal bug in it.

  Working on value numbering

  Improvements to the replaceOutEdge and added RHS.vars

  Fixed bug in edge editing

  Minor bug fix: remove redundant replaceNode operation

  Create RHS substructure in SSAFn functor

  Fix type errors in ExprFn

  Added check of image range type when loading images during simplify phase.  Also improved
  error message when either range or domain is incorrect.

  Fixed a bug in printing errors when there is no position info

  better distclean

  Adding support for program properties in TreeIL and fix a bug in the handling
  of die in variable analysis.

  Use UnusedVarFn functor

  Bug fix in maintaining census counts

  Added functor to support unused variable elimination

  Added pie as a float literal

  Fixed typo in grammar

  Add -m64 for linking too

  restore -m64 option

  Bug fix: remember to adjust census counts when renaming params to shadow vars

  Fixed bug: matrix subtraction was being translated to matrix addition.

  Added some FIXME notes

  Added some debugging support

  Bug fix: decrement the census counts on state variables that are unused
  outside strand initialization

  Added TreeIL.Var.toString function

  Added dec function to CensusFn

  Overload the infix convolution operation so that it is commutative.

  Changed the name of a function

  More tweaks to get strand state variables right.

  Hook in backtrace monitor

  Fixed potential syntax error in output

  Lots of tweaking of the variable analysis phase and its use in translating
  Low IL to Tree IL.

  IL pretty printers have been moved to top level

  Code reorganization: split the tree-il stuff into its own directory and moved the
  variable analysis into that directory too.  This reorg. is in anticipation of using
  the variable scoping information from VarAnalysis in the translation to TreeIL.

  Fix so that "make build" works from a clean tree

  Incomplete bug fix for parameters in methods

  Finish implementing field scaling, addition, and subtraction

  A lot of changes to better handle variable scoping etc.

  Bug fix: syntax error when attempting to define 1D array of strands

  Added standard method names to BasisNames module

  Added Node.uses and Node.defs to SSAFn functor

  removed debugging output

  Added timing of compiler phases to help isolate performance bugs.

  Added syntax for boolean literals

  debugging timer support

typos

  Include extra libs in link command

  More configuration hacking to better support multiple platforms.

  New module to support timing compiler phases

  Bug fix: generated incorrect code for transforming gradients to world space when the
  result was used more than once.

  Make the -rpath option something that is only used on Linux

  Fixed indentation

  Split code to compile and link C code into its own module.

  Fixed some makefile issues

  Added -rpath flag to linking phase so that dynamic libraries work on Linux

  Added partial support for mapping tensors from image space back to world space.
  Still need runtime system support for Hessians (see include/Diderot/inline-image.h)

  Adding NaN support to print-as-cl.sml

  Fixed bug in printing of nan

  Add outer loop-to-fixed point

  Fix flag passed to C compiler

  Bug fix: FloatLit.isNeg was not complete

  Fixed subtracting negative literal bug

  Several CLang changes:
    1) fixed bug with printing negative literals (e.g., a - -1)
    2) added support for assignment and shift operators to CLang.
    3) added break and continue.
    4) added printing support for nan

  Trying to make syntax error messages a bit better.

  Minor cleanup of grammar

  fixed spelling in a comment

  Export TreeToC for use by other targets

  Beginnings of OpenCL support

added c2ctmr kernel, an ostensibly C2 copy of the Catmull-Rom kernel

  Change syntax for field probing.

  Fixed field negation

  Added check for variable redefinition in the same scope

  Added FIXME note

  Fix voxel address arithmetic for non-scalar images

  Removed debug output

  Another bug fix for voxel indexing

  Fixed broken voxel address computation in mid-to-low.sml

  Fixed code generation for lerp

  Fixes for working with 1D images

  Added some image-info operations

  Add error check for Select<int,0>

  Catch error when Select is used on scalar

  Working on supporting non-scalar image data

  Working on supporting vector images

  Changes to image-info.sml:
    added support for loading non-scalar image data
    better error handling when Nrrd file does not exist

  Minor name change.

  Adding more 3x3 matrix support

  More bug fixes.

  fixing bugs in refactored code generator

  Working on refactoring

  Change type of Scale operator to work on arbitrary tensor types

  Working on codegen refactoring

  Working on refactoring code generation

  More work on refactoring

  Added more target properties to TARGET signature

  parameterize low-il to tree-il translation by target

  Make the low-to-tree translation be parameterized over the target

  More refactoring work

  Working on refactoring code generation

  Change type of Scale operator to work on arbitrary tensor types

  Working on refactoring code generation.  New file for translating TreeIL expressions
  and statements to CLang

  Removed InvScale from IL.  Instead, we invert the and use Scale in translate-basis.sml

  Added list of all IL basis functions

  Be more consistent about using functions to build CLang terms (instead of using
  the raw constructors)

  Change Norm operator in MidIL and LowIL to work on arbitrary tensor
  types.  Still need codegen support.

  Some debugging and added "^" as power operator.

  Adding support for zeros[shape] and changed I to identity[d]

  Debugged implementation of outer products

  Adding support for outer product

  Added Literal.intLit helper function

  Bug fix: translation of matrix/vector multiplication ot IL ops was broken

  Added codegen support for vector-matrix, matrix-vector, and matrix-matrix
  multiplication

  Added missing support for sqrt

  Working on codegen support for vector-matrix, matrix-vector, and matrix-matrix
  multiplication.

  Generalized dot product to inner product of tensor types.  Added IL support for
  vector-matrix, matrix-vector, and matrix-matrix multiplications.  Code generation
  and runtime support is still missing.

  Use matrix copy when assigning to matrix variables.

  Added matrix subscripting support to code generation

  Fix operator checking code to allow Subscript on higher-order tensors

  Better error message with initially is missing

  Added codegen support for creating matrices

  Add support for strand-local field variables.

  Bug fix: slice operations were breaking eval phase

  Adding support for infix dot and cross-product operators

  Added identity matrix support (still needs codegen support)

  Added lerp as an overloaded function

  Allow for overloaded functions (like lerp)

  We don't need two copies of tent in Kernel

  Starting to add more unicode names

  Bug fix

  Bug fix: handle 0-degree kernels

added a once-differentiable tent kernel called c1tent

  Working on value-numbering

  Working on value-numbering optimization

  Add Trace to codegen

  Better printing of partials

  Added type labels to nodes in the differentiation iterator so that the intermediate
  variables get the right type

  Overhaul in the way that the MidIL and LowIL represent tensor types (especially
  for orders > 1).  We still need to deal with the flattening of tensor-value
  construction

  Added FIXME comment

  Fixed bug in the C code that gets generated for subscripting vectors

  The C-target now compiles and links the generated C code

  Fixed bug in handling of norm

  Have norm map to abs when applied to scalars

  Finish support for normalize

  Adding support for vector normalize

  Fixed type error

  Adding normalization support for (s*F)@x, D(-F), and D(s*F) forms.

  Added FIXME comments for flattening tensors

  Check for existence of image file before running dnorm

  Added TODO list to typechecker.sml

  Added subscript for vectors and code generation for Subscript when argument
  is vector.

  Added support for field negation to high-il-opt.

  Catmull-Rom is C1 (not C2)

  Working on implementation of tensor slices

  Added mechanism for supporting "black box" functions (e.g., sin and sqrt).
  Also added atan2, sqrt, and tan functions to Basis

  Working on support for sequences

  Starting work on an OpenCL backend

  Added support for cross product

  Added trace operator for 2x2, 3x3, and 4x4 matrices

  Added evals, evecs, outer, and trace to Basis

  Added T_Array type to AST

  Add missing case

  Added support for vector inputs

  fix handling of killPath

  Bug fix: use consistent axis order

  Bugfix: missing code when exit was stabilize/die

  Bug fix: get printing format right for vectors

  Bug fix: better CFG editing

  Added support for tracking output state variables through ILs and generating
  a print function in the C target.

  Fixed bug when update always ended in stabilize/die.

  Starting to work on partitioning strand states

  Add stabilize method for strands that don't have them

  Started to add field normalization code to high-opt

  Introduce variable decls into SimpleAST representation.  We need these so that
  the result variables that are introduced for conditional expressions can be
  correctly handled by the translation to SSA.

  Fixed voxel index calculation

  Bug fix: flip order of components in coefficient vectors so that they match the
  order of the voxel vectors

  Fix potential type error in generated code

  Bug fix: be more careful about flushing pending code.

  Fix coefficients in probe code

  Add cast to generated code to avoid warning message

  Preparing to support the rest of the build process in the compiler

  Fix file name

  New file

  Use UnixPath module to find dnorm executable

  Switch to compiler-generated initially code.

  Working on generating initially code.  Also changed var and exp types in CTarget to
  be datatypes for better typechecking and documentation.

  Since initially only supports rectangular-shaped iteration, we can lift the range
  computation outside the iterations.  This allows us to determine the dimensions and
  to change the order of loop nesting.

  Fixed a typechecking issue with iteration

  More work on supporting initially in code generation

  Adding for loops to C code generator

  Added ++ and -- as prefix and postfix operators

  Adding for loops to TreeIL to support initially code generation

  Added pretty printing for initially code in IL

  Adding support for initially to IL.  Also factored out more common code from
  high-to-mid and mid-to-low into translate-fn.sml

  Starting to implement support for initially

  Special case for 0 as a floating-point literal

  Fixed bug in computing coefficients for probes

  Fixed bug in construction of CFG.  The order of predecessors was backwards

  Fixed bug in the handling of assignments to literals

  Added missing new lines and function for printing TreeIL statements

  Fixing pretty printing

  Bug fix: the update function was cast to the wrong type

  Added compact flag for printing IL

  Fix bug in order of axes when doing the convolution reduction (should be
  fastest to slowest).

  adding comments

  Fix conversions from float vector to integer vectors

  Count live variables in exit nodes

  Added evaluation support for E_Cons

  Fixed bug in generation of voxel addressing

  Added documentation

  Fix typo in comment

  Fixed bug in code generation for inside test

  Working on missing pieces of code generation

  New file

  Streamlined code for real to int conversions

  Working on code generation

  Working on code generation

  Working on code generation.

  Working on C target and code generation

  Add decls for local variables

  Added additional CM files to CM file list

  Working on code generation

  Working on code generation

  Working on code generation

  Split runtime names and C target types into their own modules.

  Added "allowedInline" function

  Lots of changes to push image info through to the code generator.

  Working on code generation

  Add border parameter to Inside operator and swap argument order to match
  surface language operator.

  Working on code generation

  Added E_Cons support to TreeIL pretty printer

  Fixed issue in low-to-tree conversion when one branch of if ends in die or stabilize

  Debugging code generation

  Adding documentation

  More work on code generation.  Made sure that the census info is initialized for
  all ILs.

  Working on code generation

  Debugging code generation

  Remove old IL types.sml file

  Working on code generation: added printing for TreeIL

  Working on code generation

  Working on code generation

  Working on code generation

  Working on code generation

  Working on code generation

  Working on code generation

  More work on code generation

  Working on code generation

  Add codegen hook into driver

  Working on code generation

  Added Controls module

  Adding some info about x86 intrinsics

  Working on code generation

  Working on code generation

  Working on code gen

  Working on code generation

  Working on code generation

  Many changes to IL operators for fields and images.

  Working on new field operations

  Working on code generation

  Working on code generation

  Working on code generation 

  Working no mid to low conversion

  Debugging new IL

  Debugging new IL

  Debugging new IL

  Debugging new IL

  Add better IL pretty printing

  Update pretty printer for new IL

  working on porting to new IL

  Working on porting to new IL

  Made cfg type immutable

  Change "actor" to "strand"

  Working on porting to new IL

  Working on porting to new IR

  Working on port to new IR

  Work around for new IL

  Working on porting to new CFG IR

  Working on porting to new IL

  Porting code to new IL

  Fixing type errors

  Added splice implementation

  Fixing type errors

  Working on pure CFG representation

  Creating a branch to explore a pure CFG IL

  Changed typing rules for tensor construction.  The convention is now slow-to-fast
  for tensor dimensions.

  Added additional operations on fields (F+F, F-F, sF, Fs, F/s)

  Fixed typos

  Fix typos

  Working on mid-to-low translation

  Working on MidIL to LowIL translation

  Working on MidIL to LowIL translation

  Added fromDigits function

  Added explode function

  Working on mid-to-low translation

  Working on Mid to Low translation

  Working on LowIL

  Working on low-il

  Working on codegen

  Working on codegen

  Working on coed generation

  Minor edit

  Working on code generator API

  Change name of MidIL Transform operator to PosToImgSpace.  Also added GradToWorldSpace operator to MidIL.

  Clarify description of simplify

  Upated URL in header to diderot-language.cs.uchicago.edu

  Beginnings of code generation infrastructure

  fix comment

  Starting to work on mid-to-low translation

  update descriptions of passes

  Slightly better error reporting

  Bug fix: realTy == VecTy 1 (not 0)

  Filter out dimensions of size one in tensor types

  Starting to work on the LowIL representation

  Bug fix: get type of Inside right

  Bug fix: wrong signature for EvalKernel

  Bug fix: get type of Select right

  Adding MidIL checking

  Better error messages

  Bug fix: add missing code to set CFG edges in destination

  Added checking for HighIL representation

  Bug fix: get the result type of condtional expressions right when simplifying

  bug fix for conditional expressions

  Working on checking

  More work on checking

  Working on IL checking

  Rename file

  Working on checking code

  Working on HighIL checking code

  Added more type info to HighIL operators to aid checking.

  Simplify basis by removing cruft that is not being used.

  Working on checking for HighIL

  Started to work on checking code for HighIL

  Minor edits

  reorganize operators a bit

  fixed typo

  Slice support in HighIL

  Working on slice support

  Added types to pretty printing

  bug fixes

  Fix inexhaustive matches by adding E_Slice placeholders

  Added types to IL variables

  new files

  Moving IL types out of the Op structures into their own per-IL structures.

  Move MidIL types into their own module

  Copy some of the translation boilerplate from the high-to-mid translation
  into the mid-to-low translation

  normalize tensor shapes in prune functions (e.g., tensor[1] ==> tensor[]).

  Fixed bug in grammar

  Working on adding slice notation and fixed precedence of differentiation

  Added conditional expression syntax

  Debugging probe.sml.

  Added fld2d' test case

  Bug fixes to HigiIL optimization

  Added parse-tree support for conditional expressions (still need parser support)

  Started to work on IL checker

  Working on optimization and translation

  Added HighIL to MidIL translation phase

  Working on translation

  Working on high to mid translation

  removed Transform from high IL

  Working on translation functor

  working on translation

  Adding new file for translation support

  Working on high to mid translation

  Move test-kernel stuff to fields directory

  Added hash-cons representation of IL expressions

  Debugging generation of probe code

  Debugging probe.sml

  Working on test driver for probe code

  Added assignToString function

  working on HighIL to MidIL translation

  Fixed header comment

deleting a random file

adding the translate directory for mid-to-low il

Adding the files for the low-il specs

  Adding comments

  New files

  Implementing the census functor

  Working on debugging HighIL optimization

  Working on HighIL optimization

  Put optimization passes in IL directories

  Added some comments

  Working on high to mid translation

  Tweaking MidIL definition

  Fixed bugs in mid-il and IL generator

  Working on HighIL to MidIL translation

  Added compiler README file

  Adding placeholder directories

  Working on probe code generation

  Working on HighIL optimizer

  comments

  Working on compiler

  tweaking mid IL

  new files

  added convolve function

  Added Stats module

  More general representation of image samples

  Added newProp and newFlag functions for nodes, stmts, and vars

  Made the IL generator more general to handle MidIL features

  Working on MidIL representation

  New directory

  Make the "lift" phase be more agressive about evaluating

  merge DIFF and CONV constructors

  tweaking printing of SSA IL

  Moving high-IL stuff to its own directory

  Some work on SSA checking code

  Change kernel representation to not rely on symmetry

  Bug fix

  Added BackwardDFAFn functor

  Added ForwardDFAFn functor

  New version of IL with translation to HighIL and pretty printing.

  Adding more Unicode operators

  Added infix convolve syntax

 Output to log file

  Support evaluation for more functions

  Added smart constructors

  Bug fix: get the order of instantiated meta variables right

  Bug fix: add missing case to pruneShape

  Working on lifting and translation

  Working on image-info input

  Use the log file for debugging output

  Added Log structure

  Hook in evaluation of statics

  Bug fixes

  Working on static evalution

  Working on static evaluation

  Working on evaluation of static globals

  Added Inputs module to handle static input variables

  Added type field to E_Input in Simple AST.

  Made typechecking more strict.

  Working on lifting

  Working on translation to HighIL

  Added E_Input and E_Field constructors to Simple AST

  Reorganize stuff to do with images, kernels, and fields into its own directory.

  Reorganizing source

  Working on translateion

  Added Var substructure to SSAFn

  Starting to add utility code for SSA representations

  Working on translation to HighIL

  Update

  Adding operations for fields

  Working on translation to IL

  Added Tbl structure

  Working on translation to IL

  Minor work on translation

  Working on translation to HighIL

  Working on translation to IL

  Working on IL representation

  Add generator for IL operator structures.

  Renamed internal variable input to optInput and added input variable.

  Working on translation

  Working on translation to HighIL

  Added toType, toDiff, toShape, and toDim functions

  Added file

  Working on translation to HighIL

  Simplified AST code now uses default input values when no input is given.

  Switch the translate code to take simple AST as its input type

  Added pretty printing for simplified AST

  Adding simplify pass to driver

  Cleanup and simplification of AST representation

  New file: working on pretty printing for Simple AST

  Adding simple-ast representation

  Restrict polymorphism to operators/functions.

  Split function and variable namespaces and add notion of restricted operations
  that cannot appear inside actor or initialization code.

  Working on translation to HighIL

  Added comment

  New file

  Adding equality tests and hashing support

  Added output state variables and stabilize methods

  Added kernel printing code

  Debugging kernel support using test driver

  Added min and max on reals.

  Added RawTypes module

  Starting to write a test driver for the kernel code

  Made kernel type abstract and added derivative cache.

  Added toReal function

  Working on IR

  Added variable sets and maps

  Working on IL representation

  Merging image-info into IL directory

  merge image-info directory into the IL directory

  Added normalize function

  Working on IL

  Parameterize the SSA representation over the operators

  Fixed syntax error

  New file

  Adding configuration support

  Fix syntax errors

  Added compare to Rational

  Switched kernels to use rational coefficients and added definitions of common
  kernels.

  Fix bugs

  Working on rational numbers for the kernels

  Restore Gordon's edits

  Moving compiler sources into src/compiler