Lower byte swap intrinsic.
Clump the negate instruction w/ the bswap instruction as an
"inplace" operation. One difference is that bswap has stricter
requirements the operand type.
BUG= https://code.google.com/p/nativeclient/issues/detail?id=3882
R=stichnot@chromium.org, wala@chromium.org
Review URL: https://codereview.chromium.org/401533002
diff --git a/crosstest/test_bitmanip.cpp b/crosstest/test_bitmanip.cpp
index 2ebe8a4..7106767 100644
--- a/crosstest/test_bitmanip.cpp
+++ b/crosstest/test_bitmanip.cpp
@@ -38,3 +38,8 @@
FOR_ALL_BMI_OP_TYPES(X)
#undef X
+
+#define X(type, builtin_name) \
+ type test_bswap(type a) { return builtin_name(a); }
+BSWAP_TABLE
+#undef X
diff --git a/crosstest/test_bitmanip.def b/crosstest/test_bitmanip.def
index b164ab7..0dac033 100644
--- a/crosstest/test_bitmanip.def
+++ b/crosstest/test_bitmanip.def
@@ -39,4 +39,10 @@
FOR_ALL_BMI_TYPES_INST(X, ctpop)
//#define X(inst, type)
+#define BSWAP_TABLE \
+ /* type, builtin_name */ \
+ X(uint16_t, __builtin_bswap16) \
+ X(uint32_t, __builtin_bswap32) \
+ X(uint64_t, __builtin_bswap64)
+
#endif // TEST_BIT_MANIP_DEF
diff --git a/crosstest/test_bitmanip.h b/crosstest/test_bitmanip.h
index 7c4efdb..d283d7f 100644
--- a/crosstest/test_bitmanip.h
+++ b/crosstest/test_bitmanip.h
@@ -22,3 +22,7 @@
FOR_ALL_BMI_OP_TYPES(X)
#undef X
+
+#define X(type, builtin_name) type test_bswap(type);
+BSWAP_TABLE
+#undef X
diff --git a/crosstest/test_bitmanip_main.cpp b/crosstest/test_bitmanip_main.cpp
index b3ad585..89dc8ba 100644
--- a/crosstest/test_bitmanip_main.cpp
+++ b/crosstest/test_bitmanip_main.cpp
@@ -28,27 +28,31 @@
}
volatile uint64_t Values[] = {
- 0, 1, 0x7e,
- 0x7f, 0x80, 0x81,
- 0xfe, 0xff, 0x7ffe,
- 0x7fff, 0x8000, 0x8001,
- 0xfffe, 0xffff,
- 0x007fffff /*Max subnormal + */,
- 0x00800000 /*Min+ */, 0x7f7fffff /*Max+ */,
- 0x7f800000 /*+Inf*/, 0xff800000 /*-Inf*/,
- 0x7fa00000 /*SNaN*/, 0x7fc00000 /*QNaN*/,
- 0x7ffffffe, 0x7fffffff, 0x80000000,
- 0x80000001, 0xfffffffe, 0xffffffff,
- 0x100000000ll, 0x100000001ll,
- 0x000fffffffffffffll /*Max subnormal + */,
- 0x0010000000000000ll /*Min+ */,
- 0x7fefffffffffffffll /*Max+ */,
- 0x7ff0000000000000ll /*+Inf*/,
- 0xfff0000000000000ll /*-Inf*/,
- 0x7ff0000000000001ll /*SNaN*/,
- 0x7ff8000000000000ll /*QNaN*/,
- 0x7ffffffffffffffell, 0x7fffffffffffffffll, 0x8000000000000000ll,
- 0x8000000000000001ll, 0xfffffffffffffffell, 0xffffffffffffffffll };
+ 0, 1,
+ 0x7e, 0x7f,
+ 0x80, 0x81,
+ 0xfe, 0xff,
+ 0x7ffe, 0x7fff,
+ 0x8000, 0x8001,
+ 0xfffe, 0xffff,
+ 0xc0de, 0xabcd,
+ 0xdcba, 0x007fffff /*Max subnormal + */,
+ 0x00800000 /*Min+ */, 0x7f7fffff /*Max+ */,
+ 0x7f800000 /*+Inf*/, 0xff800000 /*-Inf*/,
+ 0x7fa00000 /*SNaN*/, 0x7fc00000 /*QNaN*/,
+ 0x7ffffffe, 0x7fffffff,
+ 0x80000000, 0x80000001,
+ 0xfffffffe, 0xffffffff,
+ 0x12345678, 0xabcd1234,
+ 0x1234dcba, 0x100000000ll,
+ 0x100000001ll, 0x123456789abcdef1ll,
+ 0x987654321ab1fedcll, 0x000fffffffffffffll /*Max subnormal + */,
+ 0x0010000000000000ll /*Min+ */, 0x7fefffffffffffffll /*Max+ */,
+ 0x7ff0000000000000ll /*+Inf*/, 0xfff0000000000000ll /*-Inf*/,
+ 0x7ff0000000000001ll /*SNaN*/, 0x7ff8000000000000ll /*QNaN*/,
+ 0x7ffffffffffffffell, 0x7fffffffffffffffll,
+ 0x8000000000000000ll, 0x8000000000000001ll,
+ 0xfffffffffffffffell, 0xffffffffffffffffll};
const static size_t NumValues = sizeof(Values) / sizeof(*Values);
@@ -96,6 +100,25 @@
}
}
+template <typename Type>
+void testByteSwap(size_t &TotalTests, size_t &Passes, size_t &Failures) {
+ for (size_t i = 0; i < NumValues; ++i) {
+ Type Value = static_cast<Type>(Values[i]);
+ ++TotalTests;
+ Type ResultSz = test_bswap(Value);
+ Type ResultLlc = Subzero_::test_bswap(Value);
+ if (ResultSz == ResultLlc) {
+ ++Passes;
+ } else {
+ ++Failures;
+ std::cout << "test_bswap" << (CHAR_BIT * sizeof(Type)) << "("
+ << static_cast<uint64_t>(Value)
+ << "): sz=" << static_cast<uint64_t>(ResultSz)
+ << " llc=" << static_cast<uint64_t>(ResultLlc) << "\n";
+ }
+ }
+}
+
int main(int argc, char **argv) {
size_t TotalTests = 0;
size_t Passes = 0;
@@ -103,6 +126,9 @@
testBitManip<uint32_t>(TotalTests, Passes, Failures);
testBitManip<uint64_t>(TotalTests, Passes, Failures);
+ testByteSwap<uint16_t>(TotalTests, Passes, Failures);
+ testByteSwap<uint32_t>(TotalTests, Passes, Failures);
+ testByteSwap<uint64_t>(TotalTests, Passes, Failures);
std::cout << "TotalTests=" << TotalTests << " Passes=" << Passes
<< " Failures=" << Failures << "\n";
diff --git a/src/IceInstX8632.cpp b/src/IceInstX8632.cpp
index baa145f..f1a68da 100644
--- a/src/IceInstX8632.cpp
+++ b/src/IceInstX8632.cpp
@@ -95,11 +95,6 @@
addSource(Source2);
}
-InstX8632Neg::InstX8632Neg(Cfg *Func, Operand *SrcDest)
- : InstX8632(Func, InstX8632::Neg, 1, llvm::dyn_cast<Variable>(SrcDest)) {
- addSource(SrcDest);
-}
-
InstX8632Shld::InstX8632Shld(Cfg *Func, Variable *Dest, Variable *Source1,
Variable *Source2)
: InstX8632(Func, InstX8632::Shld, 3, Dest) {
@@ -440,6 +435,9 @@
}
+// In-place ops
+template <> const char *InstX8632Bswap::Opcode = "bswap";
+template <> const char *InstX8632Neg::Opcode = "neg";
// Unary ops
template <> const char *InstX8632Bsf::Opcode = "bsf";
template <> const char *InstX8632Bsr::Opcode = "bsr";
@@ -473,6 +471,7 @@
template <> const char *InstX8632Divps::Opcode = "divps";
template <> const char *InstX8632Idiv::Opcode = "idiv";
template <> const char *InstX8632Divss::Opcode = "divss";
+template <> const char *InstX8632Rol::Opcode = "rol";
template <> const char *InstX8632Shl::Opcode = "shl";
template <> const char *InstX8632Psll::Opcode = "psll";
template <> const char *InstX8632Shr::Opcode = "shr";
@@ -611,21 +610,6 @@
dumpSources(Func);
}
-void InstX8632Neg::emit(const Cfg *Func) const {
- Ostream &Str = Func->getContext()->getStrEmit();
- assert(getSrcSize() == 1);
- Str << "\tneg\t";
- getSrc(0)->emit(Func);
- Str << "\n";
-}
-
-void InstX8632Neg::dump(const Cfg *Func) const {
- Ostream &Str = Func->getContext()->getStrDump();
- dumpDest(Func);
- Str << " = neg." << getDest()->getType() << " ";
- dumpSources(Func);
-}
-
void InstX8632Shld::emit(const Cfg *Func) const {
Ostream &Str = Func->getContext()->getStrEmit();
assert(getSrcSize() == 3);
diff --git a/src/IceInstX8632.h b/src/IceInstX8632.h
index db60d68..b50199b 100644
--- a/src/IceInstX8632.h
+++ b/src/IceInstX8632.h
@@ -141,6 +141,7 @@
Br,
Bsf,
Bsr,
+ Bswap,
Call,
Cdq,
Cmov,
@@ -188,6 +189,7 @@
Push,
Pxor,
Ret,
+ Rol,
Sar,
Sbb,
Shl,
@@ -352,6 +354,42 @@
virtual ~InstX8632Call() {}
};
+// Instructions of the form x := op(x).
+template <InstX8632::InstKindX8632 K>
+class InstX8632Inplaceop : public InstX8632 {
+public:
+ static InstX8632Inplaceop *create(Cfg *Func, Operand *SrcDest) {
+ return new (Func->allocate<InstX8632Inplaceop>())
+ InstX8632Inplaceop(Func, SrcDest);
+ }
+ virtual void emit(const Cfg *Func) const {
+ Ostream &Str = Func->getContext()->getStrEmit();
+ assert(getSrcSize() == 1);
+ Str << "\t" << Opcode << "\t";
+ getSrc(0)->emit(Func);
+ Str << "\n";
+ }
+ virtual void dump(const Cfg *Func) const {
+ Ostream &Str = Func->getContext()->getStrDump();
+ dumpDest(Func);
+ Str << " = " << Opcode << "." << getDest()->getType() << " ";
+ dumpSources(Func);
+ }
+ static bool classof(const Inst *Inst) { return isClassof(Inst, K); }
+
+private:
+ InstX8632Inplaceop(Cfg *Func, Operand *SrcDest)
+ : InstX8632(Func, K, 1, llvm::dyn_cast<Variable>(SrcDest)) {
+ addSource(SrcDest);
+ }
+ InstX8632Inplaceop(const InstX8632Inplaceop &) LLVM_DELETED_FUNCTION;
+ InstX8632Inplaceop &
+ operator=(const InstX8632Inplaceop &) LLVM_DELETED_FUNCTION;
+ virtual ~InstX8632Inplaceop() {}
+ static const char *Opcode;
+};
+
+// Instructions of the form x := op(y)
template <InstX8632::InstKindX8632 K>
class InstX8632Unaryop : public InstX8632 {
public:
@@ -506,6 +544,8 @@
static const char *Opcode;
};
+typedef InstX8632Inplaceop<InstX8632::Bswap> InstX8632Bswap;
+typedef InstX8632Inplaceop<InstX8632::Neg> InstX8632Neg;
typedef InstX8632Unaryop<InstX8632::Bsf> InstX8632Bsf;
typedef InstX8632Unaryop<InstX8632::Bsr> InstX8632Bsr;
typedef InstX8632Unaryop<InstX8632::Lea> InstX8632Lea;
@@ -535,6 +575,7 @@
typedef InstX8632Binop<InstX8632::Pmuludq> InstX8632Pmuludq;
typedef InstX8632Binop<InstX8632::Divps> InstX8632Divps;
typedef InstX8632Binop<InstX8632::Divss> InstX8632Divss;
+typedef InstX8632Binop<InstX8632::Rol, true> InstX8632Rol;
typedef InstX8632Binop<InstX8632::Shl, true> InstX8632Shl;
typedef InstX8632Binop<InstX8632::Psll> InstX8632Psll;
typedef InstX8632Binop<InstX8632::Shr, true> InstX8632Shr;
@@ -590,23 +631,6 @@
virtual ~InstX8632Mul() {}
};
-// Neg instruction - Two's complement negation.
-class InstX8632Neg : public InstX8632 {
-public:
- static InstX8632Neg *create(Cfg *Func, Operand *SrcDest) {
- return new (Func->allocate<InstX8632Neg>()) InstX8632Neg(Func, SrcDest);
- }
- virtual void emit(const Cfg *Func) const;
- virtual void dump(const Cfg *Func) const;
- static bool classof(const Inst *Inst) { return isClassof(Inst, Neg); }
-
-private:
- InstX8632Neg(Cfg *Func, Operand *SrcDest);
- InstX8632Neg(const InstX8632Neg &) LLVM_DELETED_FUNCTION;
- InstX8632Neg &operator=(const InstX8632Neg &) LLVM_DELETED_FUNCTION;
- virtual ~InstX8632Neg() {}
-};
-
// Shld instruction - shift across a pair of operands. TODO: Verify
// that the validator accepts the shld instruction.
class InstX8632Shld : public InstX8632 {
diff --git a/src/IceTargetLoweringX8632.cpp b/src/IceTargetLoweringX8632.cpp
index af9ebc5..b38481b 100644
--- a/src/IceTargetLoweringX8632.cpp
+++ b/src/IceTargetLoweringX8632.cpp
@@ -2596,9 +2596,35 @@
_mfence();
return;
}
- case Intrinsics::Bswap:
- Func->setError("Unhandled intrinsic");
+ case Intrinsics::Bswap: {
+ Variable *Dest = Instr->getDest();
+ Operand *Val = Instr->getArg(0);
+ // In 32-bit mode, bswap only works on 32-bit arguments, and the
+ // argument must be a register. Use rotate left for 16-bit bswap.
+ if (Val->getType() == IceType_i64) {
+ Variable *T_Lo = legalizeToVar(loOperand(Val));
+ Variable *T_Hi = legalizeToVar(hiOperand(Val));
+ Variable *DestLo = llvm::cast<Variable>(loOperand(Dest));
+ Variable *DestHi = llvm::cast<Variable>(hiOperand(Dest));
+ _bswap(T_Lo);
+ _bswap(T_Hi);
+ _mov(DestLo, T_Hi);
+ _mov(DestHi, T_Lo);
+ } else if (Val->getType() == IceType_i32) {
+ Variable *T = legalizeToVar(Val);
+ _bswap(T);
+ _mov(Dest, T);
+ } else {
+ assert(Val->getType() == IceType_i16);
+ Val = legalize(Val);
+ Constant *Eight = Ctx->getConstantInt(IceType_i16, 8);
+ Variable *T = NULL;
+ _mov(T, Val);
+ _rol(T, Eight);
+ _mov(Dest, T);
+ }
return;
+ }
case Intrinsics::Ctpop: {
Variable *Dest = Instr->getDest();
Operand *Val = Instr->getArg(0);
diff --git a/src/IceTargetLoweringX8632.h b/src/IceTargetLoweringX8632.h
index fefc7fd..4c0c245 100644
--- a/src/IceTargetLoweringX8632.h
+++ b/src/IceTargetLoweringX8632.h
@@ -196,6 +196,9 @@
void _bsr(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Bsr::create(Func, Dest, Src0));
}
+ void _bswap(Variable *SrcDest) {
+ Context.insert(InstX8632Bswap::create(Func, SrcDest));
+ }
void _cdq(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Cdq::create(Func, Dest, Src0));
}
@@ -342,6 +345,9 @@
void _ret(Variable *Src0 = NULL) {
Context.insert(InstX8632Ret::create(Func, Src0));
}
+ void _rol(Variable *Dest, Operand *Src0) {
+ Context.insert(InstX8632Rol::create(Func, Dest, Src0));
+ }
void _sar(Variable *Dest, Operand *Src0) {
Context.insert(InstX8632Sar::create(Func, Dest, Src0));
}
diff --git a/tests_lit/llvm2ice_tests/nacl-other-intrinsics.ll b/tests_lit/llvm2ice_tests/nacl-other-intrinsics.ll
index cebf923..dbebf11 100644
--- a/tests_lit/llvm2ice_tests/nacl-other-intrinsics.ll
+++ b/tests_lit/llvm2ice_tests/nacl-other-intrinsics.ll
@@ -22,6 +22,9 @@
declare float @llvm.sqrt.f32(float)
declare double @llvm.sqrt.f64(double)
declare void @llvm.trap()
+declare i16 @llvm.bswap.i16(i16)
+declare i32 @llvm.bswap.i32(i32)
+declare i64 @llvm.bswap.i64(i64)
declare i32 @llvm.ctlz.i32(i32, i1)
declare i64 @llvm.ctlz.i64(i64, i1)
declare i32 @llvm.cttz.i32(i32, i1)
@@ -242,6 +245,33 @@
; CHECK-LABEL: test_trap
; CHECK: ud2
+define i32 @test_bswap_16(i32 %x) {
+entry:
+ %x_trunc = trunc i32 %x to i16
+ %r = call i16 @llvm.bswap.i16(i16 %x_trunc)
+ %r_zext = zext i16 %r to i32
+ ret i32 %r_zext
+}
+; CHECK-LABEL: test_bswap_16
+; CHECK: rol {{.*}}, 8
+
+define i32 @test_bswap_32(i32 %x) {
+entry:
+ %r = call i32 @llvm.bswap.i32(i32 %x)
+ ret i32 %r
+}
+; CHECK-LABEL: test_bswap_32
+; CHECK: bswap e{{.*}}
+
+define i64 @test_bswap_64(i64 %x) {
+entry:
+ %r = call i64 @llvm.bswap.i64(i64 %x)
+ ret i64 %r
+}
+; CHECK-LABEL: test_bswap_64
+; CHECK: bswap e{{.*}}
+; CHECK: bswap e{{.*}}
+
define i32 @test_ctlz_32(i32 %x) {
entry:
%r = call i32 @llvm.ctlz.i32(i32 %x, i1 0)
