gh-mirror/mikepaul-LuaJIT
1
0
Fork 0
mirror of https://github.com/LuaJIT/LuaJIT.git synced 2025-02-07 15:14:08 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

DynASM/ARM: Add VFP instructions.

Browse Source
This commit is contained in:
Mike Pall 2012-07-29 12:16:33 +02:00
parent e8af6e9da4
commit 2d58872cb5
2 changed files with 234 additions and 48 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
dynasm/dasm_arm.h
View File

@ -22,7 +22,7 @@ enum {
DASM_ALIGN, DASM_REL_LG, DASM_LABEL_LG, DASM_ALIGN, DASM_REL_LG, DASM_LABEL_LG,
/* The following actions also have an argument. */ /* The following actions also have an argument. */
DASM_REL_PC, DASM_LABEL_PC, DASM_REL_PC, DASM_LABEL_PC,
DASM_IMM, DASM_IMM12, DASM_IMM16, DASM_IMML8, DASM_IMML12, DASM_IMM, DASM_IMM12, DASM_IMM16, DASM_IMML8, DASM_IMML12, DASM_IMMV8,
DASM__MAX DASM__MAX
}; };
@ -250,6 +250,9 @@ void dasm_put(Dst_DECL, int start, ...)
#endif #endif
b[pos++] = n; b[pos++] = n;
break; break;
case DASM_IMMV8:
CK((n & 3) == 0, RANGE_I);
n >>= 2;
case DASM_IMML8: case DASM_IMML8:
case DASM_IMML12: case DASM_IMML12:
CK(n >= 0 ? ((n>>((ins>>5)&31)) == 0) : CK(n >= 0 ? ((n>>((ins>>5)&31)) == 0) :
@ -316,7 +319,7 @@ int dasm_link(Dst_DECL, size_t *szp)
case DASM_REL_LG: case DASM_REL_PC: pos++; break; case DASM_REL_LG: case DASM_REL_PC: pos++; break;
case DASM_LABEL_LG: case DASM_LABEL_PC: b[pos++] += ofs; break; case DASM_LABEL_LG: case DASM_LABEL_PC: b[pos++] += ofs; break;
case DASM_IMM: case DASM_IMM12: case DASM_IMM16: case DASM_IMM: case DASM_IMM12: case DASM_IMM16:
case DASM_IMML8: case DASM_IMML12: pos++; break; case DASM_IMML8: case DASM_IMML12: case DASM_IMMV8: pos++; break;
} }
} }
stop: (void)0; stop: (void)0;
@ -377,9 +380,13 @@ int dasm_encode(Dst_DECL, void *buffer)
} else if ((ins & 0x1000)) { } else if ((ins & 0x1000)) {
CK((n & 3) == 0 && -256 <= n && n <= 256, RANGE_REL); CK((n & 3) == 0 && -256 <= n && n <= 256, RANGE_REL);
goto patchimml8; goto patchimml8;
} else { } else if ((ins & 0x2000) == 0) {
CK((n & 3) == 0 && -4096 <= n && n <= 4096, RANGE_REL); CK((n & 3) == 0 && -4096 <= n && n <= 4096, RANGE_REL);
goto patchimml12; goto patchimml;
} else {
CK((n & 3) == 0 && -1020 <= n && n <= 1020, RANGE_REL);
n >>= 2;
goto patchimml;
} }
break; break;
case DASM_LABEL_LG: case DASM_LABEL_LG:
@ -399,7 +406,7 @@ int dasm_encode(Dst_DECL, void *buffer)
cp[-1] |= n >= 0 ? (0x00800000 | (n & 0x0f) | ((n & 0xf0) << 4)) : cp[-1] |= n >= 0 ? (0x00800000 | (n & 0x0f) | ((n & 0xf0) << 4)) :
((-n & 0x0f) | ((-n & 0xf0) << 4)); ((-n & 0x0f) | ((-n & 0xf0) << 4));
break; break;
case DASM_IMML12: patchimml12: case DASM_IMML12: case DASM_IMMV8: patchimml:
cp[-1] |= n >= 0 ? (0x00800000 | n) : (-n); cp[-1] |= n >= 0 ? (0x00800000 | n) : (-n);
break; break;
default: *cp++ = ins; break; default: *cp++ = ins; break;

265
dynasm/dasm_arm.lua
View File

@ -39,7 +39,7 @@ local wline, werror, wfatal, wwarn
local action_names = { local action_names = {
"STOP", "SECTION", "ESC", "REL_EXT", "STOP", "SECTION", "ESC", "REL_EXT",
"ALIGN", "REL_LG", "LABEL_LG", "ALIGN", "REL_LG", "LABEL_LG",
"REL_PC", "LABEL_PC", "IMM", "IMM12", "IMM16", "IMML8", "IMML12", "REL_PC", "LABEL_PC", "IMM", "IMM12", "IMM16", "IMML8", "IMML12", "IMMV8",
} }
-- Maximum number of section buffer positions for dasm_put(). -- Maximum number of section buffer positions for dasm_put().
@ -405,14 +405,14 @@ local map_op = {
strd_2 = "e00000f0DL", strd_3 = "e00000f0DL", -- v5TE strd_2 = "e00000f0DL", strd_3 = "e00000f0DL", -- v5TE
ldrsh_2 = "e01000f0DL", ldrsh_3 = "e01000f0DL", ldrsh_2 = "e01000f0DL", ldrsh_3 = "e01000f0DL",
ldm_2 = "e8900000nR", ldmia_2 = "e8900000nR", ldmfd_2 = "e8900000nR", ldm_2 = "e8900000oR", ldmia_2 = "e8900000oR", ldmfd_2 = "e8900000oR",
ldmda_2 = "e8100000nR", ldmfa_2 = "e8100000nR", ldmda_2 = "e8100000oR", ldmfa_2 = "e8100000oR",
ldmdb_2 = "e9100000nR", ldmea_2 = "e9100000nR", ldmdb_2 = "e9100000oR", ldmea_2 = "e9100000oR",
ldmib_2 = "e9900000nR", ldmed_2 = "e9900000nR", ldmib_2 = "e9900000oR", ldmed_2 = "e9900000oR",
stm_2 = "e8800000nR", stmia_2 = "e8800000nR", stmfd_2 = "e8800000nR", stm_2 = "e8800000oR", stmia_2 = "e8800000oR", stmfd_2 = "e8800000oR",
stmda_2 = "e8000000nR", stmfa_2 = "e8000000nR", stmda_2 = "e8000000oR", stmfa_2 = "e8000000oR",
stmdb_2 = "e9000000nR", stmea_2 = "e9000000nR", stmdb_2 = "e9000000oR", stmea_2 = "e9000000oR",
stmib_2 = "e9800000nR", stmed_2 = "e9800000nR", stmib_2 = "e9800000oR", stmed_2 = "e9800000oR",
pop_1 = "e8bd0000R", push_1 = "e92d0000R", pop_1 = "e8bd0000R", push_1 = "e92d0000R",
-- Branch instructions. -- Branch instructions.
@ -428,9 +428,89 @@ local map_op = {
svc_1 = "ef000000T", swi_1 = "ef000000T", svc_1 = "ef000000T", swi_1 = "ef000000T",
ud_0 = "e7f001f0", ud_0 = "e7f001f0",
-- NYI: Advanced SIMD and VFP instructions. -- VFP instructions.
["vadd.f32_3"] = "ee300a00dnm",
["vadd.f64_3"] = "ee300b00Gdnm",
["vsub.f32_3"] = "ee300a40dnm",
["vsub.f64_3"] = "ee300b40Gdnm",
["vmul.f32_3"] = "ee200a00dnm",
["vmul.f64_3"] = "ee200b00Gdnm",
["vnmul.f32_3"] = "ee200a40dnm",
["vnmul.f64_3"] = "ee200b40Gdnm",
["vmla.f32_3"] = "ee000a00dnm",
["vmla.f64_3"] = "ee000b00Gdnm",
["vmls.f32_3"] = "ee000a40dnm",
["vmls.f64_3"] = "ee000b40Gdnm",
["vnmla.f32_3"] = "ee100a40dnm",
["vnmla.f64_3"] = "ee100b40Gdnm",
["vnmls.f32_3"] = "ee100a00dnm",
["vnmls.f64_3"] = "ee100b00Gdnm",
["vdiv.f32_3"] = "ee800a00dnm",
["vdiv.f64_3"] = "ee800b00Gdnm",
-- NYI instructions, since I have no need for them right now: ["vabs.f32_2"] = "eeb00ac0dm",
["vabs.f64_2"] = "eeb00bc0Gdm",
["vneg.f32_2"] = "eeb10a40dm",
["vneg.f64_2"] = "eeb10b40Gdm",
["vsqrt.f32_2"] = "eeb10ac0dm",
["vsqrt.f64_2"] = "eeb10bc0Gdm",
["vcmp.f32_2"] = "eeb40a40dm",
["vcmp.f64_2"] = "eeb40b40Gdm",
["vcmpe.f32_2"] = "eeb40ac0dm",
["vcmpe.f64_2"] = "eeb40bc0Gdm",
["vcmpz.f32_1"] = "eeb50a40d",
["vcmpz.f64_1"] = "eeb50b40Gd",
["vcmpze.f32_1"] = "eeb50ac0d",
["vcmpze.f64_1"] = "eeb50bc0Gd",
vldr_2 = "ed100a00dl|ed100b00Gdl",
vstr_2 = "ed000a00dl|ed000b00Gdl",
vldm_2 = "ec900a00or",
vldmia_2 = "ec900a00or",
vldmdb_2 = "ed100a00or",
vpop_1 = "ecbd0a00r",
vstm_2 = "ec800a00or",
vstmia_2 = "ec800a00or",
vstmdb_2 = "ed000a00or",
vpush_1 = "ed2d0a00r",
["vmov.f32_2"] = "eeb00a40dm|eeb00a00dY", -- #imm is VFPv3 only
["vmov.f64_2"] = "eeb00b40Gdm|eeb00b00GdY", -- #imm is VFPv3 only
vmov_2 = "ee100a10Dn|ee000a10nD",
vmov_3 = "ec500a10DNm|ec400a10mDN|ec500b10GDNm|ec400b10GmDN",
vmrs_0 = "eef1fa10",
vmrs_1 = "eef10a10D",
vmsr_1 = "eee10a10D",
["vcvt.s32.f32_2"] = "eebd0ac0dm",
["vcvt.s32.f64_2"] = "eebd0bc0dGm",
["vcvt.u32.f32_2"] = "eebc0ac0dm",
["vcvt.u32.f64_2"] = "eebc0bc0dGm",
["vcvtr.s32.f32_2"] = "eebd0a40dm",
["vcvtr.s32.f64_2"] = "eebd0b40dGm",
["vcvtr.u32.f32_2"] = "eebc0a40dm",
["vcvtr.u32.f64_2"] = "eebc0b40dGm",
["vcvt.f32.s32_2"] = "eeb80ac0dm",
["vcvt.f64.s32_2"] = "eeb80bc0GdFm",
["vcvt.f32.u32_2"] = "eeb80a40dm",
["vcvt.f64.u32_2"] = "eeb80b40GdFm",
["vcvt.f32.f64_2"] = "eeb70bc0dGm",
["vcvt.f64.f32_2"] = "eeb70ac0GdFm",
-- VFPv4 only:
["vfma.f32_3"] = "eea00a00dnm",
["vfma.f64_3"] = "eea00b00Gdnm",
["vfms.f32_3"] = "eea00a40dnm",
["vfms.f64_3"] = "eea00b40Gdnm",
["vfnma.f32_3"] = "ee900a40dnm",
["vfnma.f64_3"] = "ee900b40Gdnm",
["vfnms.f32_3"] = "ee900a00dnm",
["vfnms.f64_3"] = "ee900b00Gdnm",
-- NYI: Advanced SIMD instructions.
-- NYI: I have no need for these instructions right now:
-- swp, swpb, strex, ldrex, strexd, ldrexd, strexb, ldrexb, strexh, ldrexh -- swp, swpb, strex, ldrex, strexd, ldrexd, strexb, ldrexb, strexh, ldrexh
-- msr, nopv6, yield, wfe, wfi, sev, dbg, bxj, smc, srs, rfe -- msr, nopv6, yield, wfe, wfi, sev, dbg, bxj, smc, srs, rfe
-- cps, setend, pli, pld, pldw, clrex, dsb, dmb, isb -- cps, setend, pli, pld, pldw, clrex, dsb, dmb, isb
@ -476,6 +556,18 @@ local function parse_gpr_pm(expr)
return parse_gpr(expr2), (pm == "-") return parse_gpr(expr2), (pm == "-")
end end
local function parse_vr(expr, tp)
local t, r = match(expr, "^([sd])([0-9]+)$")
if t == tp then
r = tonumber(r)
if r <= 31 then
if t == "s" then return shr(r, 1), band(r, 1) end
return band(r, 15), shr(r, 4)
end
end
werror("bad register name `"..expr.."'")
end
local function parse_reglist(reglist) local function parse_reglist(reglist)
reglist = match(reglist, "^{%s*([^}]*)}$") reglist = match(reglist, "^{%s*([^}]*)}$")
if not reglist then werror("register list expected") end if not reglist then werror("register list expected") end
@ -490,6 +582,21 @@ local function parse_reglist(reglist)
return rr return rr
end end
local function parse_vrlist(reglist)
local ta, ra, tb, rb = match(reglist,
"^{%s*([sd])([0-9]+)%s*%-%s*([sd])([0-9]+)%s*}$")
ra, rb = tonumber(ra), tonumber(rb)
if ta and ta == tb and ra and rb and ra <= 31 and rb <= 31 and ra <= rb then
local nr = rb+1 - ra
if ta == "s" then
return shl(shr(ra,1),12)+shl(band(ra,1),22) + nr
else
return shl(band(ra,15),12)+shl(shr(ra,4),22) + nr*2 + 0x100
end
end
werror("register list expected")
end
local function parse_imm(imm, bits, shift, scale, signed) local function parse_imm(imm, bits, shift, scale, signed)
imm = match(imm, "^#(.*)$") imm = match(imm, "^#(.*)$")
if not imm then werror("expected immediate operand") end if not imm then werror("expected immediate operand") end
@ -680,81 +787,132 @@ local function parse_load(params, nparams, n, op)
return op return op
end end
local function parse_vload(q)
local reg, imm = match(q, "^%[%s*([^,%s]*)%s*(.*)%]$")
if reg then
local d = shl(parse_gpr(reg), 16)
if imm == "" then return d end
imm = match(imm, "^,%s*#(.*)$")
if imm then
local n = tonumber(imm)
if n then
if n >= -1020 and n <= 1020 and n%4 == 0 then
return d + (n >= 0 and n/4+0x00800000 or -n/4)
end
werror("out of range immediate `"..imm.."'")
else
waction("IMMV8", 32768 + 32*8, imm)
return d
end
end
else
if match(q, "^[<>=%-]") or match(q, "^extern%s+") then
local mode, n, s = parse_label(q, false)
waction("REL_"..mode, n + 0x2800, s, 1)
return 15 * 65536
end
local reg, tailr = match(q, "^([%w_:]+)%s*(.*)$")
if reg and tailr ~= "" then
local d, tp = parse_gpr(reg)
if tp then
waction("IMMV8", 32768 + 32*8, format(tp.ctypefmt, tailr))
return shl(d, 16)
end
end
end
werror("expected address operand")
end
------------------------------------------------------------------------------ ------------------------------------------------------------------------------
-- Handle opcodes defined with template strings. -- Handle opcodes defined with template strings.
map_op[".template__"] = function(params, template, nparams) local function parse_template(params, template, nparams, pos)
if not params then return sub(template, 9) end
local op = tonumber(sub(template, 1, 8), 16) local op = tonumber(sub(template, 1, 8), 16)
local n = 1 local n = 1
local vr = "s"
-- Limit number of section buffer positions used by a single dasm_put().
-- A single opcode needs a maximum of 3 positions.
if secpos+3 > maxsecpos then wflush() end
local pos = wpos()
-- Process each character. -- Process each character.
for p in gmatch(sub(template, 9), ".") do for p in gmatch(sub(template, 9), ".") do
local q = params[n]
if p == "D" then if p == "D" then
op = op + shl(parse_gpr(params[n]), 12); n = n + 1 op = op + shl(parse_gpr(q), 12); n = n + 1
elseif p == "N" then elseif p == "N" then
op = op + shl(parse_gpr(params[n]), 16); n = n + 1 op = op + shl(parse_gpr(q), 16); n = n + 1
elseif p == "S" then elseif p == "S" then
op = op + shl(parse_gpr(params[n]), 8); n = n + 1 op = op + shl(parse_gpr(q), 8); n = n + 1
elseif p == "M" then elseif p == "M" then
op = op + parse_gpr(params[n]); n = n + 1 op = op + parse_gpr(q); n = n + 1
elseif p == "d" then
local r,h = parse_vr(q, vr); op = op+shl(r,12)+shl(h,22); n = n + 1
elseif p == "n" then
local r,h = parse_vr(q, vr); op = op+shl(r,16)+shl(h,7); n = n + 1
elseif p == "m" then
local r,h = parse_vr(q, vr); op = op+r+shl(h,5); n = n + 1
elseif p == "P" then elseif p == "P" then
local imm = match(params[n], "^#(.*)$") local imm = match(q, "^#(.*)$")
if imm then if imm then
op = op + parse_imm12(imm) + 0x02000000 op = op + parse_imm12(imm) + 0x02000000
else else
op = op + parse_gpr(params[n]) op = op + parse_gpr(q)
end end
n = n + 1 n = n + 1
elseif p == "p" then elseif p == "p" then
op = op + parse_shift(params[n], true); n = n + 1 op = op + parse_shift(q, true); n = n + 1
elseif p == "L" then elseif p == "L" then
op = parse_load(params, nparams, n, op) op = parse_load(params, nparams, n, op)
elseif p == "l" then
op = op + parse_vload(q)
elseif p == "B" then elseif p == "B" then
local mode, n, s = parse_label(params[n], false) local mode, n, s = parse_label(q, false)
waction("REL_"..mode, n, s, 1) waction("REL_"..mode, n, s, 1)
elseif p == "C" then -- blx gpr vs. blx label. elseif p == "C" then -- blx gpr vs. blx label.
local p = params[n] if match(q, "^([%w_]+):(r1?[0-9])$") or match(q, "^r(1?[0-9])$") then
if match(p, "^([%w_]+):(r1?[0-9])$") or match(p, "^r(1?[0-9])$") then op = op + parse_gpr(q)
op = op + parse_gpr(p)
else else
if op < 0xe0000000 then werror("unconditional instruction") end if op < 0xe0000000 then werror("unconditional instruction") end
local mode, n, s = parse_label(p, false) local mode, n, s = parse_label(q, false)
waction("REL_"..mode, n, s, 1) waction("REL_"..mode, n, s, 1)
op = 0xfa000000 op = 0xfa000000
end end
elseif p == "n" then elseif p == "F" then
local r, wb = match(params[n], "^([^!]*)(!?)$") vr = "s"
elseif p == "G" then
vr = "d"
elseif p == "o" then
local r, wb = match(q, "^([^!]*)(!?)$")
op = op + shl(parse_gpr(r), 16) + (wb == "!" and 0x00200000 or 0) op = op + shl(parse_gpr(r), 16) + (wb == "!" and 0x00200000 or 0)
n = n + 1 n = n + 1
elseif p == "R" then elseif p == "R" then
op = op + parse_reglist(params[n]); n = n + 1 op = op + parse_reglist(q); n = n + 1
elseif p == "r" then
op = op + parse_vrlist(q); n = n + 1
elseif p == "W" then elseif p == "W" then
op = op + parse_imm16(params[n]); n = n + 1 op = op + parse_imm16(q); n = n + 1
elseif p == "v" then elseif p == "v" then
op = op + parse_imm(params[n], 5, 7, 0, false); n = n + 1 op = op + parse_imm(q, 5, 7, 0, false); n = n + 1
elseif p == "w" then elseif p == "w" then
local imm = match(params[n], "^#(.*)$") local imm = match(q, "^#(.*)$")
if imm then if imm then
op = op + parse_imm(params[n], 5, 7, 0, false); n = n + 1 op = op + parse_imm(q, 5, 7, 0, false); n = n + 1
else else
op = op + shl(parse_gpr(params[n]), 8) + 16 op = op + shl(parse_gpr(q), 8) + 16
end end
elseif p == "X" then elseif p == "X" then
op = op + parse_imm(params[n], 5, 16, 0, false); n = n + 1 op = op + parse_imm(q, 5, 16, 0, false); n = n + 1
elseif p == "Y" then
local imm = tonumber(match(q, "^#(.*)$")); n = n + 1
if not imm or shr(imm, 8) ~= 0 then
werror("bad immediate operand")
end
op = op + shl(band(imm, 0xf0), 12) + band(imm, 0x0f)
elseif p == "K" then elseif p == "K" then
local imm = tonumber(match(params[n], "^#(.*)$")); n = n + 1 local imm = tonumber(match(q, "^#(.*)$")); n = n + 1
if not imm or shr(imm, 16) ~= 0 then if not imm or shr(imm, 16) ~= 0 then
werror("bad immediate operand") werror("bad immediate operand")
end end
op = op + shl(band(imm, 0xfff0), 4) + band(imm, 0x000f) op = op + shl(band(imm, 0xfff0), 4) + band(imm, 0x000f)
elseif p == "T" then elseif p == "T" then
op = op + parse_imm(params[n], 24, 0, 0, false); n = n + 1 op = op + parse_imm(q, 24, 0, 0, false); n = n + 1
elseif p == "s" then elseif p == "s" then
-- Ignored. -- Ignored.
else else
@ -764,6 +922,27 @@ map_op[".template__"] = function(params, template, nparams)
wputpos(pos, op) wputpos(pos, op)
end end
map_op[".template__"] = function(params, template, nparams)
if not params then return sub(template, 9) end
-- Limit number of section buffer positions used by a single dasm_put().
-- A single opcode needs a maximum of 3 positions.
if secpos+3 > maxsecpos then wflush() end
local pos = wpos()
local apos, spos = #actargs, secpos
local ok, err
for t in gmatch(template, "[^|]+") do
ok, err = pcall(parse_template, params, t, nparams, pos)
if ok then return end
secpos = spos
actargs[apos+1] = nil
actargs[apos+2] = nil
actargs[apos+3] = nil
end
error(err, 0)
end
------------------------------------------------------------------------------ ------------------------------------------------------------------------------
-- Pseudo-opcode to mark the position where the action list is to be emitted. -- Pseudo-opcode to mark the position where the action list is to be emitted.
@ -923,10 +1102,10 @@ function _M.mergemaps(map_coreop, map_def)
setmetatable(map_op, { __index = function(t, k) setmetatable(map_op, { __index = function(t, k)
local v = map_coreop[k] local v = map_coreop[k]
if v then return v end if v then return v end
local cc = sub(k, -4, -3) local k1, cc, k2 = match(k, "^(.-)(..)([._].*)$")
local cv = map_cond[cc] local cv = map_cond[cc]
if cv then if cv then
local v = rawget(t, sub(k, 1, -5)..sub(k, -2)) local v = rawget(t, k1..k2)
if type(v) == "string" then return format("%x%s", cv, sub(v, 2)) end if type(v) == "string" then return format("%x%s", cv, sub(v, 2)) end
end end
end }) end })