GHC.CmmToAsm

Note [Native code generator] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

The native-code generator has machine-independent and machine-dependent modules.

This module (GHC.CmmToAsm) is the top-level machine-independent module. Before entering machine-dependent land, we do some machine-independent optimisations (defined below) on the CmmStmtss. (Which ideally would be folded into CmmOpt ...)

We convert to the machine-specific Instr datatype with cmmCodeGen, assuming an infinite supply of registers. We then use a (mostly) machine-independent register allocator to rejoin reality. Obviously, regAlloc has machine-specific helper functions (see the used register allocator for details).

Finally, we order the basic blocks of the function so as to minimise the number of jumps between blocks, by utilising fallthrough wherever possible.

The machine-dependent bits are generally contained under GHCCmmToAsmArch/* and generally breaks down as follows:

• Regs: Everything about the target platform's machine registers (and immediate operands, and addresses, which tend to intermingle/interact with registers).

• Instr: Includes the Instr datatype plus a miscellany of other things (e.g., targetDoubleSize, smStablePtrTable, ...)

• CodeGen: is where Cmm stuff turns into machine instructions.

• Ppr: pprInstr turns an Instr into text (well, really a SDoc).

The register allocators lives under GHC.CmmToAsm.Reg.*, there is both a Linear and a Graph based register allocator. Both of which have their own notes describing them. They are mostly platform independent but there are some platform specific files encoding architecture details under RegAllocatorArch.hs

• }