about summary refs log tree commit diff stats
path: root/mu_instructions
diff options
context:
space:
mode:
authorKartik Agaram <vc@akkartik.com>2020-03-07 17:32:39 -0800
committerKartik Agaram <vc@akkartik.com>2020-03-07 17:40:45 -0800
commit3cf03158599472b1f6713192d9fa2b120f9f209b (patch)
tree4982565ff8b289847c1c263f4d9aeb3c2360b98b /mu_instructions
parent9ee4b34e068550462d440ebc522c7e8ad0c0f2e6 (diff)
downloadmu-3cf03158599472b1f6713192d9fa2b120f9f209b.tar.gz
6094 - new 'compute-offset' instruction
If indexing into a type with power-of-2-sized elements we can access them
in one instruction:

  x/reg1: (addr int) <- index A/reg2: (addr array int), idx/reg3: int

This translates to a single instruction because x86 instructions support
an addressing mode with left-shifts.

For non-powers-of-2, however, we need a multiply. To keep things type-safe,
it is performed like this:

  x/reg1: (offset T) <- compute-offset A: (addr array T), idx: int
  y/reg2: (addr T) <- index A, x

An offset is just an int that is guaranteed to be a multiple of size-of(T).
Offsets can only be used in index instructions, and the types will eventually
be required to line up.

In the process, I have to expand Input-size because mu.subx is growing
big.
Diffstat (limited to 'mu_instructions')
-rw-r--r--mu_instructions14
1 files changed, 11 insertions, 3 deletions
diff --git a/mu_instructions b/mu_instructions
index 0b6da8b0..c8be987b 100644
--- a/mu_instructions
+++ b/mu_instructions
@@ -207,13 +207,21 @@ loop-if-addr>= label        {.name="loop-if-addr>=",  .inouts=[label],
 
 Array operations
 
-var/reg <- length var2/reg2: (addr array T)
+var/reg <- length arr/reg2: (addr array T)
                             {.name="length",          .inouts=[reg2], .outputs=[reg1],  .subx-name="8b/copy-from",          .rm32="*" inouts[0],                        .r32=outputs[0]}
 var/reg <- index arr/rega: (addr array T), idx/regi: int
                             {.name="index",           .inouts=[rega, regi], .outputs=[reg], .subx-name="8d/copy-address",   .rm32="*(" inouts[0] "+" inouts[1] "<<2)",  .r32=outputs[0]}
 var/reg <- index arr/rega: (addr array T), n
-compare var, n              {.name="compare",         .inouts=[var, n],                 .subx-name="81 7/subop/compare",    .rm32="*(ebp+" inouts[0].stack-offset ")",                      .imm32=inouts[1]}
-                            {.name="index",           .inouts=[rega, n], .outputs=[reg], .subx-name="8d/copy-address",      .rm32="*(" inouts[0] "+" inouts[1] "<<2)",  .r32=outputs[0]}
+                            {.name="index",           .inouts=[rega, n], .outputs=[reg], .subx-name="8d/copy-address",      .rm32="*(" inouts[0] "+" inouts[1]*size(T) ")",  .r32=outputs[0]}
+
+var/reg: (offset T) <- compute-offset arr: (addr array T), idx/regi: int  # arr can be in reg or mem
+                            {.name="compute-offset",  .inouts=[arr, regi],  .outputs=[reg], .subx-name="69/multiply",       .rm32=inouts[1],                            .r32=outputs[0],    .imm32=sizeof(T)}
+var/reg: (offset T) <- compute-offset arr: (addr array T), idx: int       # arr can be in reg or mem
+                            {.name="compute-offset",  .inouts=[arr, regi],  .outputs=[reg], .subx-name="69/multiply",       .rm32="*(ebp+" inouts[1].stack-offset ")",  .r32=outputs[0],    .imm32=sizeof(T)}
+var: (offset T) <- compute-offset arr: (addr array T), n                  # arr can be in reg or mem
+                            {.name="compute-offset",  .inouts=[var, n],     .outputs=[reg], .subx-name="c7 0/subop/copy",   .rm32=outputs[0],                                               .imm32=sizeof(T)*n}
+var/reg <- index arr/rega: (addr array T), o/rego: offset
+                            {.name="index",           .inouts=[rega, rego], .outputs=[reg], .subx-name="8d/copy-address",   .rm32="*(" inouts[0] "+" inouts[1] "+" "4)", .r32=outputs[0]}
 
 User-defined types