notes

compwocont.md

@@ -0,0 +1,21 @@
+# Summarise Contribution & Motivation
+
+- Both CPS and ANF inhibit some optimisations
+- Aim to get benefits of both
+- Extend direct-style lambda calculus by adding join points
+- How to infer join points
+- Join points can be recursive which gives new optimisation
+- Show that approach works in GHC
+
+# Methodology
+
+- Define new intermediate language Fj
+- Perform single pass to find let bindings that are join points
+- These are let bound functions that will never be captured in closure
+- This is simpler than contification because they only look for tail calls
+
+# Critical Assessment
+
+- Achieves big reduction in allocations for some programs
+- Other programs lose out on potential optimisations
+- Transformations can be applied broadly, even to call by value languages

energysched.md

@@ -0,0 +1,27 @@
+# Summarise Contribution & Motivation
+
+- Energy efficiency is a primary design goal for all systems
+- Power management not exposed to end user
+- This makes it difficult to design general purpose runtime to schedule work between the CPU and GPU
+- Black box approach is used where power model is computed once for each processor
+- This is done because users don't have access to DVFS (dynamic voltage and frequency scaling)
+- Distributes work across CPU and GPU in such a way to optimise for power metric
+
+# Methodology
+
+- Automatically characterising behaviour is non trivial.
+- Measure cpu cores, gpu cores, ring interconnect and LLC = package power
+- Probed processor power use under various workloads to model power consuptions
+- Used microbenchmarks to get power characterisations
+- Application is profiled too to determine memory or compute bound
+- Use a mathematical algorithm to produce gpu offload ratio to minimise a given metric
+- To implement, they use Concord framework, heterogeneous C++ but algorithm is not tied
+- Work stealing from CPU
+- Ran a number of different benchmarks to evaluate performance
+
+# Critical Assessment
+
+- Does well by using black-box approach
+- Other solutions relied on static analysis or existing knowledge of processor
+- Their solution is run at user level and doesn't require controlling of DVFS
+- Limited to applications that benefit from offloading work to GPU

funcmemo.md

@@ -0,0 +1,19 @@
+# Summarise Contribution & Motivation
+
+- Build on existing work of function memoization
+- Extend the scope by doing it at compile time allowing user defined functions instead of dynamic linked
+- Memoization is just lookup in table of function results
+- Software solution and hardware for increased gains
+
+# Methodology
+
+- Identify functions which can be memoized
+- Functions that can be are replaced with memoization wrapper which looks up in table, or reverts to default storing value in table
+- Global variables are considered as extra arguments
+
+# Critical Assessment
+
+- Achieves better speedup than load-time optimisation in previous work
+- Causes memory overhead
+- Inlining can increase code size
+- Hardware solution is heavyweight and even with powergating has area implications

manmem.md

@@ -0,0 +1,27 @@
+# Summarise Contribution & Motivation
+
+- There are lots of safe programming languages
+- People don't use them
+- Garbage collection is the source of inefficiency in unsafe languages
+- Safe manual memory management is a potential solution to this.
+- Simply add a delete operator to free memory, and an exception if that memory is then dereferenced
+
+# Methodology
+
+- Programming model changes
+- Replace GC heap with manually managed heap allocated from new keyword
+- New delete operator
+- Guarantee memory safety with new exception
+- Does not impact on the compiler or programmer too much, no restriction on aliasing
+- Delete semantics are intentionally weak for performance reasons, but maintain safety from use-after-free bugs
+- Uses 64 bit hardware to assign each object new virtual addresses _without_ reusing one until safe to do so
+- The processors MMU will then detect violations as objects are unmapped from the applications address space
+- Other operations allocate objects on new virtual pages as virtual operations are only allowed on pages
+- Included an allocator in .NET toolchain
+
+# Critical Assessment
+
+- Good solution that builds on other works shortcomings
+- Still places burden on programmer unlikely to make changes
+- Using hardware to detect violations is good idea to keep overhead low
+- Somewhat non-deterministic, but this doesn't matter as the original program also was and they address this(extensive testing + debug option)

webasm.md

@@ -0,0 +1,29 @@
+# Summarise Contribution & Motivation
+
+- Web is becoming mature and now requires more complex applications to run on it
+- Javascript is currently only built in language
+- Javascript is not well equipped to deal with these applications or be a compile target
+- WebAssembly is a low level bytecode for the web
+- Aims to provide safe, low overhead execution
+- Better solution than plugins for safety
+- Better than asm.js for consistent performance
+
+# Methodology
+
+- Defines modules for each binary and therefore allows imports
+- Defines functions, not first class and not nested, call stack not exposed
+- Instructions based in a stack machine, for compactness
+- Only defines 4 types, integers and IEEE floating points
+- Has global and local variables
+- Memories defined by modules with little endianness, disjoint from code space and stack so programs can only mess up their own environment
+- Does not offer simple jumps, has structured control flow, gives single pass validation/compilation/SSA
+- Can do JIT and validation in single pass
+- Transmitted over wire in binary form, streaming compilation possible due to layout
+
+# Critical Assessment
+
+- Improves on native client by being available on all browsers as it is not compact and still requires knowledge of underlying system
+- Traps are not handled but need JS intervention
+- Still missing features for higher level languages
+- A compile target for the web with safety by design should prevent many exploits
+-