The current approach is to keep the JITer as simple as possible, and thus as
fast as possible. The generated code quality will suffer from that.
-X86 register allocation:
-========================
+Register Allocation:
+====================
-We can use 8bit or 16bit registers on the x86. If we use that feature we have
-more registers to allocate, which maybe prevents some register spills. We
-currently ignore that ability and always allocate 32 bit registers, because I
-think we would gain very little from that optimisation and it would complicate
-the code.
+With lcc you can assign a fixed register to a tree before register
+allocation. For example this is needed by call, which return the value always
+in EAX on x86. The current implementation works without such system, due to
+special forest generation.
Different Register Sets:
========================
point values, and one for integers. Should we support architectures with more
that two sets? Does someone knows such an architecture?
-Register Allocation:
-====================
+64bit Integer Values:
+=====================
-With lcc you can assign a fixed register to a tree before register
-allocation. For example this is needed by call, which return the value always
-in EAX on x86. The current implementation works without such system (due to
-special forest generation), and I wonder if we really need this feature?
+I can imagine two different implementation. On possibility would be to treat
+long (64bit) values simply like any other value type. This implies that we
+call class methods for ALU operations like add or sub. Sure, this method will
+be be a bit inefficient.
+
+The more performant solution is to allocate two 32bit registers for each 64bit
+value. We add a new non terminal to the monoburg grammar called "lreg". The
+register allocation routines takes care of this non terminal and allocates two
+32 bit registers for them.
Forest generation:
==================
-One idea was to drive the code generation directly from the CIL code, without
-generating an intermediate forest of trees. I think this is not possible,
-because you always have to gather some attributes and attach it to the
-instruction (for example the register allocation info). So I thing generating a
-tree is the right thing and that also works perfectly with monoburg. IMO we
-would not get any benefit from trying to feed monoburg directly with CIL
-instructions.
+Consider the following code:
+
+OPCODE: STACK LOCALS
+LDLOC.0 (5) [5,0]
+LDC.1 (5,1) [5,0]
+STLOC.0 (5) [1,0]
+STLOC.1 () [1,5]
+
+A simple forest generation generates:
-We can also speedup the tree generation by using alloca instead of malloc.
+STLOC.0(LDC.1)
+STLOC.1(LDLOC.0)
+
+Which is wrong, since it stores the wrong value (1 instead of 5). Instead we
+must generate something like:
+
+STLOC.TMP(LDLOC.0)
+STLOC.0(LDC.1)
+STLOC.1(LDLOC.TMP)
+
+Where STLOC.TMP saves the value into a new temporary variable.
+
+We also need a similar solution for basic block boundaries when the stack depth
+is not zero. We can simply save those values to new temporary values. Consider
+the following basic block with one instruction:
+
+LDLOC.1
+This should generate a tree like:
+
+STLOC.TMP(LDLOC.1) Please notice that an intelligent register allocator can
+still allocate registers for those new variables.
DAG handling:
=============
This is what lcc is doing, if I understood 12.8, page 342, 343?
-Value Types:
-============
+Possible Optimisations:
+=======================
+
+Miguel said ORP does some optimisation on IL level, for example moving array
+bounds checking out of loops:
+
+for (i = 0; i < N; i++) { check_range (a, i); a [i] = X; }
+
+id transformed to:
+
+if (in_range (a, 0, N)) { for (i = 0; i < N; i++) a[i] = X; }
+else for (i = 0; i < N; i++) { check_range (a, i); a [i] = X; }
+
+The "else" is only to keep original semantics (exception handling).
-The only CLI instructions which can handle value types are loads and stores,
-either to local variable, to the stack or to array elements. Value types with a
-size smaller than sizeof(int) are handled like any other basic type. For other
-value types we load the base address and emit block copies to store them.
+We need loop detection logic in order to implement this (dominator tree).
+AFAIK CACAO also implements this.
\ No newline at end of file