From 12ee94ffad513f38a973e733e09ca1a1ca558854 Mon Sep 17 00:00:00 2001
From: Hao Sun <hao.sun@arm.com>
Date: Wed, 9 Jun 2021 12:57:33 +0000
Subject: [PATCH] DynASM/arm64: Partially support 'movi' instruction

In the development of PHP JIT/AArch64, `movi d0, #0` is needed to zero
FP registers.

In this patch, we partially support the 64-bit scalar variant of 'movi'
instruction [1].

In this variant, 64-bit immediate value is acceptable with the condition
that each byte is either all zeros or all ones.

Note that the action to handle variables for this 64-bit immediate is
not implemented yet.

Example:
```
  |  movi d0, #0xffff00ffff
  |  movi Rd(reg-ZREG_V0), #0
```

Disassembly:
```
  movi    d0, #0xffff00ffff
  movi    d0, #0x0
```

[1]
https://developer.arm.com/documentation/ddi0602/2020-12/SIMD-FP-Instructions/MOVI--Move-Immediate--vector--?lang=en

Change-Id: I61701fc25b778393b4d1ae1798cd9541f115a351
---
 dynasm/dasm_arm64.lua | 31 +++++++++++++++++++++++++++++++
 1 file changed, 31 insertions(+)

diff --git a/dynasm/dasm_arm64.lua b/dynasm/dasm_arm64.lua
index cb82dc4a..002b2ef5 100644
--- a/dynasm/dasm_arm64.lua
+++ b/dynasm/dasm_arm64.lua
@@ -441,6 +441,34 @@ local function parse_fpimm(imm)
   end
 end
 
+local function parse_imm_movi(n)
+  local lo = tonumber(n % 0x100000000)
+  local hi = tonumber((n-lo) / 0x100000000)
+  local m = 0
+  for j=0,1 do
+    local h = j == 0 and hi or lo
+    for i=3,0,-1 do
+      local b = band(shr(h, i*8), 0xff)
+      if b ~= 0xff and b ~= 0 then return false, _ end
+      m = m*2 + band(b, 1)
+    end
+  end
+  return true, m
+end
+
+local function parse_imm64(imm)
+  imm = match(imm, "^#(.*)$")
+  if not imm then werror("expected immediate operand") end
+  local n = parse_number(imm)
+  if n then
+    local ok, m = parse_imm_movi(n)
+    if ok then return shl(band(m, 0x1f), 5) + shl(band(m, 0xe0), 11) end
+    werror("invalid immediate `"..imm.."'")
+  else
+    werror("NYI movi imm action")
+  end
+end
+
 local function parse_shift(expr)
   local s, s2 = match(expr, "^(%S+)%s*(.*)$")
   s = map_shift[s]
@@ -889,6 +917,7 @@ map_op = {
 
   -- TODO: crc32*, aes*, sha*, pmull
   -- TODO: SIMD instructions.
+  movi_2   = "2f00e400DOd",
 }
 
 for cond,c in pairs(map_cond) do
@@ -975,6 +1004,8 @@ local function parse_template(params, template, nparams, pos)
       op = op + parse_imm(q, 4, 0, 0, false); n = n + 1
     elseif p == "F" then
       op = op + parse_fpimm(q); n = n + 1
+    elseif p == "O" then
+      op = op + parse_imm64(q); n = n + 1
     elseif p == "Z" then
       if q ~= "#0" and q ~= "#0.0" then werror("expected zero immediate") end
       n = n + 1