Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(16)

Issues Search
    My Issues | Starred     Open | Closed | All

Unified Diff: src/arm/assembler-arm.cc

Issue 11428137: ARM: Make use of d16-d31 when available. (Closed) Base URL: https://v8.googlecode.com/svn/branches/bleeding_edge
Patch Set: Address Rodolph's comments Created 8 years ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « src/arm/assembler-arm.h ('k') | src/arm/assembler-arm-inl.h » ('j') | src/arm/code-stubs-arm.h » ('J')
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: src/arm/assembler-arm.cc
diff --git a/src/arm/assembler-arm.cc b/src/arm/assembler-arm.cc
index 47ea0e20666e2f44621e42959b7094067c43ed79..558f48b0fa0586c7d8636bec6c80a569f6da653f 100644
--- a/src/arm/assembler-arm.cc
+++ b/src/arm/assembler-arm.cc
@@ -1680,19 +1680,22 @@ void Assembler::vldr(const DwVfpRegister dst,
int offset,
const Condition cond) {
// Ddst = MEM(Rbase + offset).
- // Instruction details available in ARM DDI 0406A, A8-628.
- // cond(31-28) | 1101(27-24)| U001(23-20) | Rbase(19-16) |
- // Vdst(15-12) | 1011(11-8) | offset
+ // Instruction details available in ARM DDI 0406C.b, A8-924.
+ // cond(31-28) | 1101(27-24)| U(23) | D(22) | 01(21-20) | Rbase(19-16) |
+ // Vd(15-12) | 1011(11-8) | offset
ASSERT(CpuFeatures::IsEnabled(VFP2));
int u = 1;
if (offset < 0) {
offset = -offset;
u = 0;
}
+ ASSERT(offset >= 0);
+ int vd, d;
+ dst.split_code(&vd, &d);
ASSERT(offset >= 0);
if ((offset % 4) == 0 && (offset / 4) < 256) {
- emit(cond | u*B23 | 0xD1*B20 | base.code()*B16 | dst.code()*B12 |
+ emit(cond | 0xD*B24 | u*B23 | d*B22 | B20 | base.code()*B16 | vd*B12 |
0xB*B8 | ((offset / 4) & 255));
} else {
// Larger offsets must be handled by computing the correct address
@@ -1703,7 +1706,7 @@ void Assembler::vldr(const DwVfpRegister dst,
} else {
sub(ip, base, Operand(offset));
}
- emit(cond | 0xD1*B20 | ip.code()*B16 | dst.code()*B12 | 0xB*B8);
+ emit(cond | 0xD*B24 | d*B22 | B20 | ip.code()*B16 | vd*B12 | 0xB*B8);
}
}
@@ -1766,9 +1769,9 @@ void Assembler::vstr(const DwVfpRegister src,
int offset,
const Condition cond) {
// MEM(Rbase + offset) = Dsrc.
- // Instruction details available in ARM DDI 0406A, A8-786.
- // cond(31-28) | 1101(27-24)| U000(23-20) | | Rbase(19-16) |
- // Vsrc(15-12) | 1011(11-8) | (offset/4)
+ // Instruction details available in ARM DDI 0406C.b, A8-1082.
+ // cond(31-28) | 1101(27-24)| U(23) | D(22) | 00(21-20) | Rbase(19-16) |
+ // Vd(15-12) | 1011(11-8) | (offset/4)
ASSERT(CpuFeatures::IsEnabled(VFP2));
int u = 1;
if (offset < 0) {
@@ -1776,9 +1779,12 @@ void Assembler::vstr(const DwVfpRegister src,
u = 0;
}
ASSERT(offset >= 0);
+ int vd, d;
+ src.split_code(&vd, &d);
+
if ((offset % 4) == 0 && (offset / 4) < 256) {
- emit(cond | u*B23 | 0xD0*B20 | base.code()*B16 | src.code()*B12 |
- 0xB*B8 | ((offset / 4) & 255));
+ emit(cond | 0xD*B24 | u*B23 | d*B22 | base.code()*B16 | vd*B12 | 0xB*B8 |
+ ((offset / 4) & 255));
} else {
// Larger offsets must be handled by computing the correct address
// in the ip register.
@@ -1788,7 +1794,7 @@ void Assembler::vstr(const DwVfpRegister src,
} else {
sub(ip, base, Operand(offset));
}
- emit(cond | 0xD0*B20 | ip.code()*B16 | src.code()*B12 | 0xB*B8);
+ emit(cond | 0xD*B24 | d*B22 | ip.code()*B16 | vd*B12 | 0xB*B8);
}
}
@@ -1850,9 +1856,9 @@ void Assembler::vldm(BlockAddrMode am,
DwVfpRegister first,
DwVfpRegister last,
Condition cond) {
- // Instruction details available in ARM DDI 0406A, A8-626.
+ // Instruction details available in ARM DDI 0406C.b, A8-922.
// cond(31-28) | 110(27-25)| PUDW1(24-20) | Rbase(19-16) |
- // first(15-12) | 1010(11-8) | (count * 2)
+ // first(15-12) | 1011(11-8) | (count * 2)
ASSERT(CpuFeatures::IsEnabled(VFP2));
ASSERT_LE(first.code(), last.code());
ASSERT(am == ia || am == ia_w || am == db_w);
@@ -1872,7 +1878,7 @@ void Assembler::vstm(BlockAddrMode am,
DwVfpRegister first,
DwVfpRegister last,
Condition cond) {
- // Instruction details available in ARM DDI 0406A, A8-784.
+ // Instruction details available in ARM DDI 0406C.b, A8-1080.
// cond(31-28) | 110(27-25)| PUDW0(24-20) | Rbase(19-16) |
// first(15-12) | 1011(11-8) | (count * 2)
ASSERT(CpuFeatures::IsEnabled(VFP2));
@@ -2008,14 +2014,28 @@ void Assembler::vmov(const DwVfpRegister dst,
mov(ip, Operand(lo));
if (scratch.is(no_reg)) {
- // Move the low part of the double into the lower of the corresponsing S
- // registers of D register dst.
- vmov(dst.low(), ip, cond);
-
- // Move the high part of the double into the higher of the corresponsing S
- // registers of D register dst.
- mov(ip, Operand(hi));
- vmov(dst.high(), ip, cond);
+ if (dst.code() < 16) {
+ // Move the low part of the double into the lower of the corresponsing S
+ // registers of D register dst.
+ vmov(dst.low(), ip, cond);
+
+ // Move the high part of the double into the higher of the corresponsing S
+ // registers of D register dst.
+ mov(ip, Operand(hi));
+ vmov(dst.high(), ip, cond);
+ } else {
+ // There are no corresponding S registers for D register dst.
+
+ // FIXME: We need a way to do this! For now, let's use the stack.
+ // Push r4, r5 on the stack.
Rodolph Perfetta 2012/12/05 00:42:10 There is an instruction which lets you move an ARM
hans 2012/12/05 14:12:58 Awesome, I'll just use vmov.32 then. Thanks!
+ push(r4);
+ push(r5);
+ mov(r4, Operand(lo));
+ mov(r5, Operand(hi));
+ vmov(dst, r4, r5, cond);
+ pop(r5);
+ pop(r4);
+ }
} else {
// Move the low and high parts of the double to a D register in one
// instruction.
@@ -2043,10 +2063,16 @@ void Assembler::vmov(const DwVfpRegister dst,
const DwVfpRegister src,
const Condition cond) {
// Dd = Dm
- // Instruction details available in ARM DDI 0406B, A8-642.
+ // Instruction details available in ARM DDI 0406C.b, A8-938.
+ // cond(31-28) | 11101(27-23) | D(22) | 11(21-20) | 0000(19-16) | Vd(15-12) |
+ // 101(11-9) | sz=1(8) | 0(7) | 1(6) | M(5) | 0(4) | Vm(3-0)
ASSERT(CpuFeatures::IsEnabled(VFP2));
- emit(cond | 0xE*B24 | 0xB*B20 |
- dst.code()*B12 | 0x5*B9 | B8 | B6 | src.code());
+ int vd, d;
+ dst.split_code(&vd, &d);
+ int vm, m;
+ src.split_code(&vm, &m);
+ emit(cond | 0x1D*B23 | d*B22 | 0x3*B20 | vd*B12 | 0x5*B9 | B8 | B6 | m*B5 |
+ vm);
}
@@ -2055,13 +2081,15 @@ void Assembler::vmov(const DwVfpRegister dst,
const Register src2,
const Condition cond) {
// Dm = <Rt,Rt2>.
- // Instruction details available in ARM DDI 0406A, A8-646.
+ // Instruction details available in ARM DDI 0406C.b, A8-948.
// cond(31-28) | 1100(27-24)| 010(23-21) | op=0(20) | Rt2(19-16) |
// Rt(15-12) | 1011(11-8) | 00(7-6) | M(5) | 1(4) | Vm
ASSERT(CpuFeatures::IsEnabled(VFP2));
ASSERT(!src1.is(pc) && !src2.is(pc));
+ int vm, m;
+ dst.split_code(&vm, &m);
emit(cond | 0xC*B24 | B22 | src2.code()*B16 |
- src1.code()*B12 | 0xB*B8 | B4 | dst.code());
+ src1.code()*B12 | 0xB*B8 | m*B5 | B4 | vm);
}
@@ -2070,13 +2098,15 @@ void Assembler::vmov(const Register dst1,
const DwVfpRegister src,
const Condition cond) {
// <Rt,Rt2> = Dm.
- // Instruction details available in ARM DDI 0406A, A8-646.
+ // Instruction details available in ARM DDI 0406C.b, A8-948.
// cond(31-28) | 1100(27-24)| 010(23-21) | op=1(20) | Rt2(19-16) |
// Rt(15-12) | 1011(11-8) | 00(7-6) | M(5) | 1(4) | Vm
ASSERT(CpuFeatures::IsEnabled(VFP2));
ASSERT(!dst1.is(pc) && !dst2.is(pc));
+ int vm, m;
+ src.split_code(&vm, &m);
emit(cond | 0xC*B24 | B22 | B20 | dst2.code()*B16 |
- dst1.code()*B12 | 0xB*B8 | B4 | src.code());
+ dst1.code()*B12 | 0xB*B8 | m*B5 | B4 | vm);
}
@@ -2289,18 +2319,33 @@ void Assembler::vcvt_f32_f64(const SwVfpRegister dst,
void Assembler::vneg(const DwVfpRegister dst,
const DwVfpRegister src,
const Condition cond) {
+ // Instruction details available in ARM DDI 0406C.b, A8-968.
+ // cond(31-28) | 11101(27-23) | D(22) | 11(21-20) | 0001(19-16) | Vd(15-12) |
+ // 101(11-9) | sz=1(8) | 0(7) | 1(6) | M(5) | 0(4) | Vm(3-0)
ASSERT(CpuFeatures::IsEnabled(VFP2));
- emit(cond | 0xE*B24 | 0xB*B20 | B16 | dst.code()*B12 |
- 0x5*B9 | B8 | B6 | src.code());
+ int vd, d;
+ dst.split_code(&vd, &d);
+ int vm, m;
+ src.split_code(&vm, &m);
+
+ emit(cond | 0x1D*B23 | d*B22 | 0x3*B20 | B16 | vd*B12 | 0x5*B9 | B8 | B6 |
+ m*B5 | vm);
}
void Assembler::vabs(const DwVfpRegister dst,
const DwVfpRegister src,
const Condition cond) {
+ // Instruction details available in ARM DDI 0406C.b, A8-524.
+ // cond(31-28) | 11101(27-23) | D(22) | 11(21-20) | 0000(19-16) | Vd(15-12) |
+ // 101(11-9) | sz=1(8) | 1(7) | 1(6) | M(5) | 0(4) | Vm(3-0).
ASSERT(CpuFeatures::IsEnabled(VFP2));
- emit(cond | 0xE*B24 | 0xB*B20 | dst.code()*B12 |
- 0x5*B9 | B8 | 0x3*B6 | src.code());
+ int vd, d;
+ dst.split_code(&vd, &d);
+ int vm, m;
+ src.split_code(&vm, &m);
+ emit(cond | 0x1D*B23 | d*B22 | 0x3*B20 | vd*B12 | 0x5*B9 | B8 | B7 | B6 |
+ m*B5 | vm);
}
@@ -2310,12 +2355,18 @@ void Assembler::vadd(const DwVfpRegister dst,
const Condition cond) {
// Dd = vadd(Dn, Dm) double precision floating point addition.
// Dd = D:Vd; Dm=M:Vm; Dn=N:Vm.
- // Instruction details available in ARM DDI 0406A, A8-536.
- // cond(31-28) | 11100(27-23)| D=?(22) | 11(21-20) | Vn(19-16) |
- // Vd(15-12) | 101(11-9) | sz(8)=1 | N(7)=0 | 0(6) | M=?(5) | 0(4) | Vm(3-0)
+ // Instruction details available in ARM DDI 0406C.b, A8-830.
+ // cond(31-28) | 11100(27-23)| D(22) | 11(21-20) | Vn(19-16) |
+ // Vd(15-12) | 101(11-9) | sz=1(8) | N(7) | 0(6) | M(5) | 0(4) | Vm(3-0)
ASSERT(CpuFeatures::IsEnabled(VFP2));
- emit(cond | 0xE*B24 | 0x3*B20 | src1.code()*B16 |
- dst.code()*B12 | 0x5*B9 | B8 | src2.code());
+ int vd, d;
+ dst.split_code(&vd, &d);
+ int vn, n;
+ src1.split_code(&vn, &n);
+ int vm, m;
+ src2.split_code(&vm, &m);
+ emit(cond | 0x1C*B23 | d*B22 | 0x3*B20 | vn*B16 | vd*B12 | 0x5*B9 | B8 |
+ n*B7 | m*B5 | vm);
}
@@ -2325,12 +2376,18 @@ void Assembler::vsub(const DwVfpRegister dst,
const Condition cond) {
// Dd = vsub(Dn, Dm) double precision floating point subtraction.
// Dd = D:Vd; Dm=M:Vm; Dn=N:Vm.
- // Instruction details available in ARM DDI 0406A, A8-784.
- // cond(31-28) | 11100(27-23)| D=?(22) | 11(21-20) | Vn(19-16) |
- // Vd(15-12) | 101(11-9) | sz(8)=1 | N(7)=0 | 1(6) | M=?(5) | 0(4) | Vm(3-0)
+ // Instruction details available in ARM DDI 0406C.b, A8-1086.
+ // cond(31-28) | 11100(27-23)| D(22) | 11(21-20) | Vn(19-16) |
+ // Vd(15-12) | 101(11-9) | sz=1(8) | N(7) | 1(6) | M(5) | 0(4) | Vm(3-0)
ASSERT(CpuFeatures::IsEnabled(VFP2));
- emit(cond | 0xE*B24 | 0x3*B20 | src1.code()*B16 |
- dst.code()*B12 | 0x5*B9 | B8 | B6 | src2.code());
+ int vd, d;
+ dst.split_code(&vd, &d);
+ int vn, n;
+ src1.split_code(&vn, &n);
+ int vm, m;
+ src2.split_code(&vm, &m);
+ emit(cond | 0x1C*B23 | d*B22 | 0x3*B20 | vn*B16 | vd*B12 | 0x5*B9 | B8 |
+ n*B7 | B6 | m*B5 | vm);
}
@@ -2340,12 +2397,18 @@ void Assembler::vmul(const DwVfpRegister dst,
const Condition cond) {
// Dd = vmul(Dn, Dm) double precision floating point multiplication.
// Dd = D:Vd; Dm=M:Vm; Dn=N:Vm.
- // Instruction details available in ARM DDI 0406A, A8-784.
- // cond(31-28) | 11100(27-23)| D=?(22) | 10(21-20) | Vn(19-16) |
- // Vd(15-12) | 101(11-9) | sz(8)=1 | N(7)=0 | 0(6) | M=?(5) | 0(4) | Vm(3-0)
+ // Instruction details available in ARM DDI 0406C.b, A8-960.
+ // cond(31-28) | 11100(27-23)| D(22) | 10(21-20) | Vn(19-16) |
+ // Vd(15-12) | 101(11-9) | sz=1(8) | N(7) | 0(6) | M(5) | 0(4) | Vm(3-0)
ASSERT(CpuFeatures::IsEnabled(VFP2));
- emit(cond | 0xE*B24 | 0x2*B20 | src1.code()*B16 |
- dst.code()*B12 | 0x5*B9 | B8 | src2.code());
+ int vd, d;
+ dst.split_code(&vd, &d);
+ int vn, n;
+ src1.split_code(&vn, &n);
+ int vm, m;
+ src2.split_code(&vm, &m);
+ emit(cond | 0x1C*B23 | d*B22 | 0x2*B20 | vn*B16 | vd*B12 | 0x5*B9 | B8 |
+ n*B7 | m*B5 | vm);
}
@@ -2353,13 +2416,17 @@ void Assembler::vmla(const DwVfpRegister dst,
const DwVfpRegister src1,
const DwVfpRegister src2,
const Condition cond) {
- // Instruction details available in ARM DDI 0406C.b, A8-892.
- // cond(31-28) | 11100(27-23) | D=?(22) | 00(21-20) | Vn(19-16) |
- // Vd(15-12) | 101(11-9) | sz(8)=1 | N=?(7) | op(6)=0 | M=?(5) | 0(4) |
- // Vm(3-0)
- unsigned x = (cond | 0x1C*B23 | src1.code()*B16 |
- dst.code()*B12 | 0x5*B9 | B8 | src2.code());
- emit(x);
+ // Instruction details available in ARM DDI 0406C.b, A8-932.
+ // cond(31-28) | 11100(27-23) | D(22) | 00(21-20) | Vn(19-16) |
+ // Vd(15-12) | 101(11-9) | sz=1(8) | N(7) | op=0(6) | M(5) | 0(4) | Vm(3-0)
+ int vd, d;
+ dst.split_code(&vd, &d);
+ int vn, n;
+ src1.split_code(&vn, &n);
+ int vm, m;
+ src2.split_code(&vm, &m);
+ emit(cond | 0x1C*B23 | d*B22 | vn*B16 | vd*B12 | 0x5*B9 | B8 | n*B7 | m*B5 |
+ vm);
}
@@ -2369,12 +2436,18 @@ void Assembler::vdiv(const DwVfpRegister dst,
const Condition cond) {
// Dd = vdiv(Dn, Dm) double precision floating point division.
// Dd = D:Vd; Dm=M:Vm; Dn=N:Vm.
- // Instruction details available in ARM DDI 0406A, A8-584.
- // cond(31-28) | 11101(27-23)| D=?(22) | 00(21-20) | Vn(19-16) |
- // Vd(15-12) | 101(11-9) | sz(8)=1 | N(7)=? | 0(6) | M=?(5) | 0(4) | Vm(3-0)
+ // Instruction details available in ARM DDI 0406C.b, A8-882.
+ // cond(31-28) | 11101(27-23)| D(22) | 00(21-20) | Vn(19-16) |
+ // Vd(15-12) | 101(11-9) | sz=1(8) | N(7) | 0(6) | M(5) | 0(4) | Vm(3-0)
ASSERT(CpuFeatures::IsEnabled(VFP2));
- emit(cond | 0xE*B24 | B23 | src1.code()*B16 |
- dst.code()*B12 | 0x5*B9 | B8 | src2.code());
+ int vd, d;
+ dst.split_code(&vd, &d);
+ int vn, n;
+ src1.split_code(&vn, &n);
+ int vm, m;
+ src2.split_code(&vm, &m);
+ emit(cond | 0x1D*B23 | d*B22 | vn*B16 | vd*B12 | 0x5*B9 | B8 | n*B7 | m*B5 |
+ vm);
}
@@ -2382,26 +2455,31 @@ void Assembler::vcmp(const DwVfpRegister src1,
const DwVfpRegister src2,
const Condition cond) {
// vcmp(Dd, Dm) double precision floating point comparison.
- // Instruction details available in ARM DDI 0406A, A8-570.
- // cond(31-28) | 11101 (27-23)| D=?(22) | 11 (21-20) | 0100 (19-16) |
- // Vd(15-12) | 101(11-9) | sz(8)=1 | E(7)=0 | 1(6) | M(5)=? | 0(4) | Vm(3-0)
+ // Instruction details available in ARM DDI 0406C.b, A8-864.
+ // cond(31-28) | 11101(27-23)| D(22) | 11(21-20) | 0100(19-16) |
+ // Vd(15-12) | 101(11-9) | sz=1(8) | E=0(7) | 1(6) | M(5) | 0(4) | Vm(3-0)
ASSERT(CpuFeatures::IsEnabled(VFP2));
- emit(cond | 0xE*B24 |B23 | 0x3*B20 | B18 |
- src1.code()*B12 | 0x5*B9 | B8 | B6 | src2.code());
+ int vd, d;
+ src1.split_code(&vd, &d);
+ int vm, m;
+ src2.split_code(&vm, &m);
+ emit(cond | 0x1D*B23 | d*B22 | 0x3*B20 | 0x4*B16 | vd*B12 | 0x5*B9 | B8 | B6 |
+ m*B5 | vm);
}
void Assembler::vcmp(const DwVfpRegister src1,
const double src2,
const Condition cond) {
- // vcmp(Dd, Dm) double precision floating point comparison.
- // Instruction details available in ARM DDI 0406A, A8-570.
- // cond(31-28) | 11101 (27-23)| D=?(22) | 11 (21-20) | 0101 (19-16) |
- // Vd(15-12) | 101(11-9) | sz(8)=1 | E(7)=0 | 1(6) | M(5)=? | 0(4) | 0000(3-0)
+ // vcmp(Dd, #0.0) double precision floating point comparison.
+ // Instruction details available in ARM DDI 0406C.b, A8-864.
+ // cond(31-28) | 11101(27-23)| D(22) | 11(21-20) | 0101(19-16) |
+ // Vd(15-12) | 101(11-9) | sz=1(8) | E=0(7) | 1(6) | 0(5) | 0(4) | 0000(3-0)
ASSERT(CpuFeatures::IsEnabled(VFP2));
ASSERT(src2 == 0.0);
- emit(cond | 0xE*B24 |B23 | 0x3*B20 | B18 | B16 |
- src1.code()*B12 | 0x5*B9 | B8 | B6);
+ int vd, d;
+ src1.split_code(&vd, &d);
+ emit(cond | 0x1D*B23 | d*B22 | 0x3*B20 | 0x5*B16 | vd*B12 | 0x5*B9 | B8 | B6);
}
@@ -2428,11 +2506,16 @@ void Assembler::vmrs(Register dst, Condition cond) {
void Assembler::vsqrt(const DwVfpRegister dst,
const DwVfpRegister src,
const Condition cond) {
- // cond(31-28) | 11101 (27-23)| D=?(22) | 11 (21-20) | 0001 (19-16) |
- // Vd(15-12) | 101(11-9) | sz(8)=1 | 11 (7-6) | M(5)=? | 0(4) | Vm(3-0)
+ // Instruction details available in ARM DDI 0406C.b, A8-1058.
+ // cond(31-28) | 11101(27-23)| D(22) | 11(21-20) | 0001(19-16) |
+ // Vd(15-12) | 101(11-9) | sz=1(8) | 11(7-6) | M(5) | 0(4) | Vm(3-0)
ASSERT(CpuFeatures::IsEnabled(VFP2));
- emit(cond | 0xE*B24 | B23 | 0x3*B20 | B16 |
- dst.code()*B12 | 0x5*B9 | B8 | 3*B6 | src.code());
+ int vd, d;
+ dst.split_code(&vd, &d);
+ int vm, m;
+ src.split_code(&vm, &m);
+ emit(cond | 0x1D*B23 | d*B22 | 0x3*B20 | B16 | vd*B12 | 0x5*B9 | B8 | 0x3*B6 |
+ m*B5 | vm);
}
« no previous file with comments | « src/arm/assembler-arm.h ('k') | src/arm/assembler-arm-inl.h » ('j') | src/arm/code-stubs-arm.h » ('J')

Powered by Google App Engine
This is Rietveld 408576698