Code

Intrinsicpr#53

Code

Open

Brugarolas wants to merge 23 commits intoBrugarolas:v3.0-simpleBrugarolas/luajit-experimental:v3.0-simplefrom fsfod:intrinsicprfsfod/LuaJIT:intrinsicprCopy head branch name to clipboard

+6,242-97

Commits

Commits on Dec 8, 2019

• Added support for separate mcode areas

  fsfod

  committed
  235ab3b

  Copy full SHA for 235ab3b

• Initial support for intrinsics on x86/x64 interpreter only

  fsfod

  committed
  c84b176

  Copy full SHA for c84b176

• Template intrinsics user machine code

  fsfod

  committed
  e1a1721

  Copy full SHA for e1a1721

• Extended op_emit to support 2 byte vex opcodes and optionally expand to 3 byte form if needed

  fsfod

  committed
  c0797d3

  Copy full SHA for c0797d3

• Added support for ymm registers in intrinsics

  fsfod

  committed
  f21526d

  Copy full SHA for f21526d

• Added support for casting vectors to a pointer when using lj_cconv_ct_tv by using a special cast flag(CCF_INTRINS_ARG) for intrinsic vector arguments

  fsfod

  committed
  e6fecee

  Copy full SHA for e6fecee

• Implement support for opcodes with dynamic registers

  fsfod

  committed
  de4c0b6

  Copy full SHA for de4c0b6

• Extended emit_op to support 4 byte opcodes based on checking a new flag OP4B in rr

  fsfod

  committed
  7c697b0

  Copy full SHA for 7c697b0

• Added support for 4 byte opcode intrinsics

  fsfod

  committed
  57ff675

  Copy full SHA for 57ff675

• Added JIT support for intrinsics. Support for vector registers is NYI.

  fsfod

  committed
  f7331e9

  Copy full SHA for f7331e9

• Treat IR_INTRN as potential load with respect to DSE

  fsfod

  committed
  275d0dc

  Copy full SHA for 275d0dc

• Added a flag(s) for opcodes with non memory store side effects and enabled DCE of intrinsics

  Show description for a374e90
  fsfod

  committed
  a374e90

  Copy full SHA for a374e90

• Added a JIT flag for AVX1 support

  fsfod

  committed
  239f8ad

  Copy full SHA for 239f8ad

• Added VEX opcode support for intrinsics

  fsfod

  committed
  befcdc6

  Copy full SHA for befcdc6

• CSE support for intrinsics only enabled for single value returning intrinsics that have no side effects and are not forced indirect ModRM which could be a load or store

  fsfod

  committed
  748091c

  Copy full SHA for 748091c

• Fix store opcodes with dynamic destructive out register not being correctly allocated an input register

  fsfod

  committed
  4eaf7f6

  Copy full SHA for 4eaf7f6

• Wip Intrinsic documentation

  fsfod

  committed
  96f1f83

  Copy full SHA for 96f1f83

• Don't allow stitching to defeat our black listing of the loop in jit_tester

  fsfod

  committed
  cb3c483

  Copy full SHA for cb3c483

• Fixed LJ_GC64 builds breaking intrinsic interpreter wrappers in various ways.

  Show description for ede73f2
  fsfod

  committed
  ede73f2

  Copy full SHA for ede73f2

• Fix Intrinsics crashing in the JIT for LJ_GC64 because they were using RID_DISPATCH

  Show description for a81dcf6
  fsfod

  committed
  a81dcf6

  Copy full SHA for a81dcf6

• Fix wrong function name in some assert_cdef calls for intrinsic CSE tests causing random test failures

  fsfod

  committed
  7a82653

  Copy full SHA for 7a82653

• Improve the runtests shell script and add support with testing with different builds of LuaJIT

  fsfod

  committed
  1e9a1bb

  Copy full SHA for 1e9a1bb

• Change intrinsic REX opcode mode X to be silently ignored in 32 builds to allow pointer based intrinsics to work in both 64 bit and 32 bit with the same definion.

  fsfod

  committed
  26808ed

  Copy full SHA for 26808ed