ShuriZma
/
xenia-canary
mirror of https://github.com/xenia-canary/xenia-canary.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Removing comments.

This commit is contained in:
Ben Vanik 2013-12-07 06:29:58 -08:00
parent 3b268f07ef
commit e5d867a92f
1 changed files with 1 additions and 70 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

71
src/alloy/compiler/passes.h
View File

@ -48,6 +48,7 @@
// //
// Run order: // Run order:
// ContextPromotion // ContextPromotion
// Simplification
// ConstantPropagation // ConstantPropagation
// TypePropagation // TypePropagation
// ByteSwapElimination // ByteSwapElimination
@ -55,42 +56,6 @@
// DeadStoreElimination // DeadStoreElimination
// DeadCodeElimination // DeadCodeElimination
// //
// - ContextPromotion
// Like mem2reg, but because context memory is unaliasable it's easier to
// check and convert LoadContext/StoreContext into value operations.
// Example of load->value promotion:
// v0 = load_context +100
// store_context +200, v0
// v1 = load_context +100 <-- replace with v1 = v0
// store_context +200, v1
//
// It'd be possible in this stage to also remove redundant context stores:
// Example of dead store elimination:
// store_context +100, v0 <-- removed due to following store
// store_context +100, v1
// This is more generally done by DSE, however if it could be done here
// instead as it may be faster (at least on the block-level).
//
// - ConstantPropagation
// Once ContextPromotion has run there will likely be a whole slew of
// constants that can be pushed through the function.
// Example:
// store_context +100, 1000
// v0 = load_context +100
// v1 = add v0, v0
// store_context +200, v1
// after PromoteContext:
// store_context +100, 1000
// v0 = 1000
// v1 = add v0, v0
// store_context +200, v1
// after PropagateConstants:
// store_context +100, 1000
// v0 = 1000
// v1 = add 1000, 1000
// store_context +200, 2000
// A DCE run after this should clean up any of the values no longer needed.
//
// - TypePropagation // - TypePropagation
// There are many extensions/truncations in generated code right now due to // There are many extensions/truncations in generated code right now due to
// various load/stores of varying widths. Being able to find and short- // various load/stores of varying widths. Being able to find and short-
@ -143,19 +108,6 @@
// ... (DCE takes care of this) ... // ... (DCE takes care of this) ...
// store v87.i64, v21.i32 // store v87.i64, v21.i32
// //
// - Simplification
// Run over the instructions and rename assigned variables:
// v1 = v0
// v2 = v1
// v3 = add v0, v2
// becomes:
// v1 = v0 (will be removed by DCE)
// v2 = v0 (will be removed by DCE)
// v3 = add v0, v0
// This could be run several times, as it could make other passes faster
// to compute (for example, ConstantPropagation). DCE will take care of
// the useless assigns.
//
// - DeadStoreElimination // - DeadStoreElimination
// Generic DSE pass, removing all redundant stores. ContextPromotion may be // Generic DSE pass, removing all redundant stores. ContextPromotion may be
// able to take care of most of these, as the input assembly is generally // able to take care of most of these, as the input assembly is generally
@ -180,26 +132,5 @@
// store_context +302, v5 // store_context +302, v5
// branch_true v5, ... // branch_true v5, ...
// //
// - DeadCodeElimination
// ContextPromotion/DSE will likely leave around a lot of dead statements.
// Code generated for comparison/testing produces many unused statements and
// with proper use analysis it should be possible to remove most of them:
// After context promotion/simplification:
// v33.i8 = compare_ult v31.i32, 0
// v34.i8 = compare_ugt v31.i32, 0
// v35.i8 = compare_eq v31.i32, 0
// store_context +300, v33.i8
// store_context +301, v34.i8
// store_context +302, v35.i8
// branch_true v35.i8, loc_8201A484
// After DSE:
// v33.i8 = compare_ult v31.i32, 0
// v34.i8 = compare_ugt v31.i32, 0
// v35.i8 = compare_eq v31.i32, 0
// branch_true v35.i8, loc_8201A484
// After DCE:
// v35.i8 = compare_eq v31.i32, 0
// branch_true v35.i8, loc_8201A484
//
#endif // ALLOY_COMPILER_PASSES_H_ #endif // ALLOY_COMPILER_PASSES_H_