ShuriZma
/
xenia
mirror of https://github.com/xenia-project/xenia.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Adding a bunch of instructions.

This commit is contained in:
Ben Vanik 2013-01-26 21:51:31 -08:00
parent 733fe85d37
commit d9b08f569f
10 changed files with 616 additions and 168 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
TODO.md
View File

@ -1,33 +1,19 @@
## Instructions
```
need cr0:
andix
orx
mullwx
divwux
### XER CA bit (carry)
addx
addix
addic
addzex
subfx
subfex
subficx
Not sure the way I'm doing this is right. addic/subficx/etc set it to the value
of the overflow bit from the LLVM *_with_overflow intrinsic.
```
MISDECODING: andix
rlwinmx
rlwimix
rldiclx
extsbx
slwx
srawix
# can be no-op, or @llvm.prefetch
dcbt
dcbtst
twi # @llvm.debugtrap ?
```
Overflow bits can be set via the intrinsics:

7
include/xenia/cpu/codegen/function_generator.h
View File

@ -66,13 +66,20 @@ public:
llvm::Value* xer_value();
void update_xer_value(llvm::Value* value);
void update_xer_with_overflow(llvm::Value* value);
void update_xer_with_carry(llvm::Value* value);
void update_xer_with_overflow_and_carry(llvm::Value* value);
llvm::Value* lr_value();
void update_lr_value(llvm::Value* value);
llvm::Value* ctr_value();
void update_ctr_value(llvm::Value* value);
llvm::Value* cr_value(uint32_t n);
void update_cr_value(uint32_t n, llvm::Value* value);
void update_cr_with_cond(uint32_t n, llvm::Value* lhs, llvm::Value* rhs,
bool is_signed);
llvm::Value* gpr_value(uint32_t n);
void update_gpr_value(uint32_t n, llvm::Value* value);

11
include/xenia/cpu/ppc/instr.h
View File

@ -166,6 +166,17 @@ typedef struct {
uint32_t : 6;
} M;
// kXEPPCInstrFormatMD
struct {
uint32_t Rc : 1;
uint32_t SH5 : 1;
uint32_t : 3;
uint32_t MB5 : 1;
uint32_t MB : 5;
uint32_t SH : 5;
uint32_t RA : 5;
uint32_t RS : 5;
uint32_t : 6;
} MD;
// kXEPPCInstrFormatMDS
// kXEPPCInstrFormatVA
// kXEPPCInstrFormatVX

5
private/runtest.sh
View File

@ -1,6 +1,11 @@
python xenia-build.py xethunk
python xenia-build.py build
if [ "$?" -ne 0 ]; then
echo "Build failed!"
exit $?
fi
./build/xenia/release/xenia-run \
private/$1 \
--optimize_ir_modules=true \

433
src/cpu/codegen/emit_alu.cc
View File

@ -29,21 +29,35 @@ namespace codegen {
XEEMITTER(addx, 0x7C000214, XO )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// RD <- (RA) + (RB)
if (i.XO.Rc) {
// With cr0 update.
XEINSTRNOTIMPLEMENTED();
return 1;
}
if (i.XO.OE) {
// With XER update.
XEINSTRNOTIMPLEMENTED();
return 1;
// This is a different codepath as we need to use llvm.sadd.with.overflow.
Function* sadd_with_overflow = Intrinsic::getDeclaration(
g.gen_module(), Intrinsic::sadd_with_overflow, b.getInt64Ty());
Value* v = b.CreateCall2(sadd_with_overflow,
g.gpr_value(i.XO.RA), g.gpr_value(i.XO.RB));
g.update_gpr_value(i.XO.RT, b.CreateExtractValue(v, 0));
g.update_xer_with_overflow(b.CreateExtractValue(v, 1));
if (i.XO.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
} else {
// No OE bit setting.
Value* v = b.CreateAdd(g.gpr_value(i.XO.RA), g.gpr_value(i.XO.RB));
g.update_gpr_value(i.XO.RT, v);
if (i.XO.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
Value* v = b.CreateAdd(g.gpr_value(i.XO.RA), g.gpr_value(i.XO.RB));
g.update_gpr_value(i.XO.RT, v);
return 0;
}
XEEMITTER(addcx, 0X7C000014, XO )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
@ -57,8 +71,10 @@ XEEMITTER(addex, 0x7C000114, XO )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(addi, 0x38000000, D )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// if RA = 0 then RT <- EXTS(SI)
// else RT <- (RA) + EXTS(SI)
// if RA = 0 then
// RT <- EXTS(SI)
// else
// RT <- (RA) + EXTS(SI)
Value* v = b.getInt64(XEEXTS16(i.D.DS));
if (i.D.RA) {
@ -70,8 +86,16 @@ XEEMITTER(addi, 0x38000000, D )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(addic, 0x30000000, D )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// RT <- (RA) + EXTS(SI)
Function* sadd_with_overflow = Intrinsic::getDeclaration(
g.gen_module(), Intrinsic::sadd_with_overflow, b.getInt64Ty());
Value* v = b.CreateCall2(sadd_with_overflow,
g.gpr_value(i.D.RA), b.getInt64(XEEXTS16(i.D.DS)));
g.update_gpr_value(i.D.RT, b.CreateExtractValue(v, 0));
g.update_xer_with_carry(b.CreateExtractValue(v, 1));
return 0;
}
XEEMITTER(addicx, 0x34000000, D )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
@ -80,8 +104,10 @@ XEEMITTER(addicx, 0x34000000, D )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(addis, 0x3C000000, D )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// if RA = 0 then RT <- EXTS(SI) || i16.0
// else RT <- (RA) + EXTS(SI) || i16.0
// if RA = 0 then
// RT <- EXTS(SI) || i16.0
// else
// RT <- (RA) + EXTS(SI) || i16.0
Value* v = b.getInt64(XEEXTS16(i.D.DS) << 16);
if (i.D.RA) {
@ -98,8 +124,29 @@ XEEMITTER(addmex, 0x7C0001D4, XO )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(addzex, 0x7C000194, XO )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// RT <- (RA) + CA
Function* sadd_with_overflow = Intrinsic::getDeclaration(
g.gen_module(), Intrinsic::sadd_with_overflow, b.getInt64Ty());
Value* ca = b.CreateAnd(b.CreateLShr(g.xer_value(), 29), 0x1);
Value* v = b.CreateCall2(sadd_with_overflow,
g.gpr_value(i.XO.RA), ca);
Value* add_value = b.CreateExtractValue(v, 0);
g.update_gpr_value(i.XO.RT, add_value);
if (i.XO.OE) {
// With XER[SO] update too.
g.update_xer_with_overflow_and_carry(b.CreateExtractValue(v, 1));
} else {
// Just CA update.
g.update_xer_with_carry(b.CreateExtractValue(v, 1));
}
if (i.XO.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, add_value, b.getInt64(0), true);
}
return 0;
}
XEEMITTER(divdx, 0x7C0003D2, XO )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
@ -120,24 +167,50 @@ XEEMITTER(divwx, 0x7C0003D6, XO )(FunctionGenerator& g, IRBuilder<>& b, I
XEEMITTER(divwux, 0x7C000396, XO )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// dividend[0:31] <- (RA)[32:63]
// divisor[0:31] <- (RB)[32:63]
// if divisor = 0 then
// if OE = 1 then
// XER[OV] <- 1
// return
// RT[32:63] <- dividend ÷ divisor
// RT[0:31] <- undefined
if (i.XO.Rc) {
// With cr0 update.
XEINSTRNOTIMPLEMENTED();
return 1;
}
if (i.XO.OE) {
// With XER update.
XEINSTRNOTIMPLEMENTED();
return 1;
}
Value* dividend = b.CreateTrunc(g.gpr_value(i.XO.RA), b.getInt32Ty());
Value* divisor = b.CreateTrunc(g.gpr_value(i.XO.RB), b.getInt32Ty());
// Note that we skip the zero handling block and just avoid the divide if
// we are OE=0.
BasicBlock* zero_bb = i.XO.OE ?
BasicBlock::Create(*g.context(), "", g.gen_fn()) : NULL;
BasicBlock* nonzero_bb = BasicBlock::Create(*g.context(), "", g.gen_fn());
BasicBlock* after_bb = BasicBlock::Create(*g.context(), "", g.gen_fn());
b.CreateCondBr(b.CreateICmpEQ(divisor, b.getInt32(0)),
i.XO.OE ? zero_bb : after_bb, nonzero_bb);
if (zero_bb) {
// Divisor was zero - do XER update.
b.SetInsertPoint(zero_bb);
g.update_xer_with_overflow(b.getInt1(1));
b.CreateBr(after_bb);
}
// Divide.
b.SetInsertPoint(nonzero_bb);
Value* v = b.CreateUDiv(dividend, divisor);
v = b.CreateZExt(v, b.getInt64Ty());
g.update_gpr_value(i.XO.RT, v);
// If we are OE=1 we need to clear the overflow bit.
g.update_xer_with_overflow(b.getInt1(0));
if (i.XO.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
b.CreateBr(after_bb);
// Resume.
b.SetInsertPoint(after_bb);
return 0;
}
@ -184,11 +257,6 @@ XEEMITTER(mulli, 0x1C000000, D )(FunctionGenerator& g, IRBuilder<>& b, I
XEEMITTER(mullwx, 0x7C0001D6, XO )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// RT <- (RA)[32:63] × (RB)[32:63]
if (i.XO.Rc) {
// With cr0 update.
XEINSTRNOTIMPLEMENTED();
return 1;
}
if (i.XO.OE) {
// With XER update.
XEINSTRNOTIMPLEMENTED();
@ -199,6 +267,11 @@ XEEMITTER(mullwx, 0x7C0001D6, XO )(FunctionGenerator& g, IRBuilder<>& b, I
b.CreateSExt(g.gpr_value(i.XO.RB), b.getInt64Ty()));
g.update_gpr_value(i.XO.RT, v);
if (i.XO.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
@ -208,8 +281,37 @@ XEEMITTER(negx, 0x7C0000D0, XO )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(subfx, 0x7C000050, XO )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// RT <- ¬(RA) + (RB) + 1
if (i.XO.OE) {
// With XER update.
// This is a different codepath as we need to use llvm.ssub.with.overflow.
Function* ssub_with_overflow = Intrinsic::getDeclaration(
g.gen_module(), Intrinsic::ssub_with_overflow, b.getInt64Ty());
Value* v = b.CreateCall2(ssub_with_overflow,
g.gpr_value(i.XO.RB), g.gpr_value(i.XO.RA));
g.update_gpr_value(i.XO.RT, b.CreateExtractValue(v, 0));
g.update_xer_with_overflow(b.CreateExtractValue(v, 1));
if (i.XO.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
} else {
// No OE bit setting.
Value* v = b.CreateSub(g.gpr_value(i.XO.RB), g.gpr_value(i.XO.RA));
g.update_gpr_value(i.XO.RT, v);
if (i.XO.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
}
XEEMITTER(subfcx, 0x7C000010, XO )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
@ -218,13 +320,45 @@ XEEMITTER(subfcx, 0x7C000010, XO )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(subficx, 0x20000000, D )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// RT <- ¬(RA) + EXTS(SI) + 1
Function* ssub_with_overflow = Intrinsic::getDeclaration(
g.gen_module(), Intrinsic::ssub_with_overflow, b.getInt64Ty());
Value* v = b.CreateCall2(ssub_with_overflow,
b.getInt64(XEEXTS16(i.D.DS)), g.gpr_value(i.D.RA));
g.update_gpr_value(i.D.RT, b.CreateExtractValue(v, 0));
g.update_xer_with_carry(b.CreateExtractValue(v, 1));
return 0;
}
XEEMITTER(subfex, 0x7C000110, XO )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// RT <- ¬(RA) + (RB) + CA
// TODO(benvanik): possible that the add of rb+ca needs to also check for
// overflow!
Value* ca = b.CreateAnd(b.CreateLShr(g.xer_value(), 29), 0x1);
Function* uadd_with_overflow = Intrinsic::getDeclaration(
g.gen_module(), Intrinsic::uadd_with_overflow, b.getInt64Ty());
Value* v = b.CreateCall2(uadd_with_overflow,
b.CreateNot(g.gpr_value(i.XO.RA)),
b.CreateAdd(g.gpr_value(i.XO.RB), ca));
g.update_gpr_value(i.XO.RT, b.CreateExtractValue(v, 0));
if (i.XO.OE) {
// With XER update.
g.update_xer_with_overflow_and_carry(b.CreateExtractValue(v, 1));
} else {
g.update_xer_with_carry(b.CreateExtractValue(v, 1));
}
if (i.XO.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
XEEMITTER(subfmex, 0x7C0001D0, XO )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
@ -240,29 +374,6 @@ XEEMITTER(subfzex, 0x7C000190, XO )(FunctionGenerator& g, IRBuilder<>& b, I
// Integer compare (A-4)
void XeEmitCompareCore(FunctionGenerator& g, IRBuilder<>& b,
Value* lhs, Value* rhs, uint32_t BF, bool is_signed) {
// bit0 = RA < RB
// bit1 = RA > RB
// bit2 = RA = RB
// bit3 = XER[SO]
// Bits are reversed:
// 0123
// 3210
Value* is_lt = is_signed ?
b.CreateICmpSLT(lhs, rhs) : b.CreateICmpULT(lhs, rhs);
Value* is_gt = is_signed ?
b.CreateICmpSGT(lhs, rhs) : b.CreateICmpUGT(lhs, rhs);
Value* cp = b.CreateSelect(is_gt, b.getInt8(1 << 2), b.getInt8(1 << 1));
Value* c = b.CreateSelect(is_lt, b.getInt8(1 << 3), cp);
// TODO(benvanik): set bit 4 to XER[SO]
// Insert the 4 bits into their location in the CR.
g.update_cr_value(BF, c);
}
XEEMITTER(cmp, 0x7C000000, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// if L = 0 then
// a <- EXTS((RA)[32:63])
@ -291,7 +402,7 @@ XEEMITTER(cmp, 0x7C000000, X )(FunctionGenerator& g, IRBuilder<>& b, I
rhs = b.CreateSExt(rhs, b.getInt64Ty());
}
XeEmitCompareCore(g, b, lhs, rhs, BF, true);
g.update_cr_with_cond(BF, lhs, rhs, true);
return 0;
}
@ -320,7 +431,7 @@ XEEMITTER(cmpi, 0x2C000000, D )(FunctionGenerator& g, IRBuilder<>& b, I
}
Value* rhs = b.getInt64(XEEXTS16(i.D.DS));
XeEmitCompareCore(g, b, lhs, rhs, BF, true);
g.update_cr_with_cond(BF, lhs, rhs, true);
return 0;
}
@ -353,7 +464,7 @@ XEEMITTER(cmpl, 0x7C000040, X )(FunctionGenerator& g, IRBuilder<>& b, I
rhs = b.CreateZExt(rhs, b.getInt64Ty());
}
XeEmitCompareCore(g, b, lhs, rhs, BF, false);
g.update_cr_with_cond(BF, lhs, rhs, false);
return 0;
}
@ -382,7 +493,7 @@ XEEMITTER(cmpli, 0x28000000, D )(FunctionGenerator& g, IRBuilder<>& b, I
}
Value* rhs = b.getInt64(i.D.DS);
XeEmitCompareCore(g, b, lhs, rhs, BF, false);
g.update_cr_with_cond(BF, lhs, rhs, false);
return 0;
}
@ -393,31 +504,29 @@ XEEMITTER(cmpli, 0x28000000, D )(FunctionGenerator& g, IRBuilder<>& b, I
XEEMITTER(andx, 0x7C000038, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// RA <- (RS) & (RB)
if (i.X.Rc) {
// With cr0 update.
XEINSTRNOTIMPLEMENTED();
return 1;
}
Value* v = b.CreateAnd(g.gpr_value(i.X.RT), g.gpr_value(i.X.RB));
g.update_gpr_value(i.X.RA, v);
if (i.X.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
XEEMITTER(andcx, 0x7C000078, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// RA <- (RS) & ¬(RB)
if (i.X.Rc) {
// With cr0 update.
XEINSTRNOTIMPLEMENTED();
return 1;
}
Value* v = b.CreateXor(g.gpr_value(i.X.RB), -1);
v = b.CreateAnd(g.gpr_value(i.X.RT), v);
g.update_gpr_value(i.X.RA, v);
if (i.X.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
@ -427,8 +536,8 @@ XEEMITTER(andix, 0x70000000, D )(FunctionGenerator& g, IRBuilder<>& b, I
Value* v = b.CreateAnd(g.gpr_value(i.D.RT), (uint64_t)i.D.DS);
g.update_gpr_value(i.D.RA, v);
// TODO(benvanik): update cr0
XEINSTRNOTIMPLEMENTED();
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
return 1;
}
@ -439,8 +548,8 @@ XEEMITTER(andisx, 0x74000000, D )(FunctionGenerator& g, IRBuilder<>& b, I
Value* v = b.CreateAnd(g.gpr_value(i.D.RT), ((uint64_t)i.D.DS) << 16);
g.update_gpr_value(i.D.RA, v);
// TODO(benvanik): update cr0
XEINSTRNOTIMPLEMENTED();
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
return 1;
}
@ -457,11 +566,6 @@ XEEMITTER(cntlzwx, 0x7C000034, X )(FunctionGenerator& g, IRBuilder<>& b, I
// n <- n + 1
// RA <- n - 32
if (i.X.Rc) {
XEINSTRNOTIMPLEMENTED();
return 1;
}
Value* v = g.gpr_value(i.X.RT);
v = b.CreateTrunc(v, b.getInt32Ty());
@ -474,6 +578,11 @@ XEEMITTER(cntlzwx, 0x7C000034, X )(FunctionGenerator& g, IRBuilder<>& b, I
count = b.CreateZExt(count, b.getInt64Ty());
g.update_gpr_value(i.X.RA, count);
if (i.X.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, count, b.getInt64(0), true);
}
return 0;
}
@ -487,16 +596,16 @@ XEEMITTER(extsbx, 0x7C000774, X )(FunctionGenerator& g, IRBuilder<>& b, I
// RA[56:63] <- (RS)[56:63]
// RA[0:55] <- i56.s
if (i.X.Rc) {
XEINSTRNOTIMPLEMENTED();
return 1;
}
Value* v = g.gpr_value(i.X.RT);
v = b.CreateTrunc(v, b.getInt8Ty());
v = b.CreateSExt(v, b.getInt64Ty());
g.update_gpr_value(i.X.RA, v);
if (i.X.Rc) {
// Update cr0.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
@ -518,31 +627,29 @@ XEEMITTER(nandx, 0x7C0003B8, X )(FunctionGenerator& g, IRBuilder<>& b, I
XEEMITTER(norx, 0x7C0000F8, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// RA <- ¬((RS) | (RB))
if (i.X.Rc) {
// With cr0 update.
XEINSTRNOTIMPLEMENTED();
return 1;
}
Value* v = b.CreateOr(g.gpr_value(i.X.RT), g.gpr_value(i.X.RB));
v = b.CreateXor(v, -1);
g.update_gpr_value(i.X.RA, v);
if (i.X.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
XEEMITTER(orx, 0x7C000378, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// RA <- (RS) | (RB)
if (i.X.Rc) {
// With cr0 update.
XEINSTRNOTIMPLEMENTED();
return 1;
}
Value* v = b.CreateOr(g.gpr_value(i.X.RT), g.gpr_value(i.X.RB));
g.update_gpr_value(i.X.RA, v);
if (i.X.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
@ -572,15 +679,14 @@ XEEMITTER(oris, 0x64000000, D )(FunctionGenerator& g, IRBuilder<>& b, I
XEEMITTER(xorx, 0x7C000278, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// RA <- (RS) XOR (RB)
if (i.X.Rc) {
// With cr0 update.
XEINSTRNOTIMPLEMENTED();
return 1;
}
Value* v = b.CreateXor(g.gpr_value(i.X.RT), g.gpr_value(i.X.RB));
g.update_gpr_value(i.X.RA, v);
if (i.X.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
@ -621,6 +727,29 @@ XEEMITTER(rldicx, 0x78000008, MD )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(rldiclx, 0x78000000, MD )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// n <- sh[5] || sh[0:4]
// r <- ROTL64((RS), n)
// b <- mb[5] || mb[0:4]
// m <- MASK(b, 63)
// RA <- r & m
// uint32_t sh = (i.MD.SH5 << 5) | i.MD.SH;
// uint32_t mb = (i.MD.MB5 << 5) | i.MD.MB;
// Value* v = g.gpr_value(i.MD.RS);
// if (sh) {
// v = // rotate by sh
// }
// if (mb) {
// v = // mask b mb->63
// }
// g.update_gpr_value(i.MD.RA, v);
// if (i.MD.Rc) {
// // With cr0 update.
// g.update_cr_with_cond(0, v, b.getInt64(0), true);
// }
XEINSTRNOTIMPLEMENTED();
return 1;
}
@ -636,6 +765,10 @@ XEEMITTER(rldimix, 0x7800000C, MD )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(rlwimix, 0x50000000, M )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
// n <- SH
// r <- ROTL32((RS)[32:63], n)
// m <- MASK(MB+32, ME+32)
// RA <- r&m | (RA)&¬m
XEINSTRNOTIMPLEMENTED();
return 1;
}
@ -646,19 +779,17 @@ XEEMITTER(rlwinmx, 0x54000000, M )(FunctionGenerator& g, IRBuilder<>& b, I
// m <- MASK(MB+32, ME+32)
// RA <- r & m
if (i.M.Rc) {
// With cr0 update.
XEINSTRNOTIMPLEMENTED();
return 1;
}
// The compiler will generate a bunch of these for the special case of
// SH=0, MB=ME
// Which seems to just select a single bit and set cr0 for use with a branch.
// We can detect this and do less work.
if (!i.M.SH && i.M.MB == i.M.ME) {
Value* v = b.CreateAnd(g.gpr_value(i.M.RS), 1 << i.M.MB);
g.update_gpr_value(i.M.RS, v);
g.update_gpr_value(i.M.RA, v);
if (i.M.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
@ -669,6 +800,12 @@ XEEMITTER(rlwinmx, 0x54000000, M )(FunctionGenerator& g, IRBuilder<>& b, I
// v = b.CreateOr(b.CreateShl(v, i.M.SH), b.CreateLShr(v, 32 - i.M.SH));
// v = b.CreateAnd(v, XEMASK(i.M.MB + 32, i.M.ME + 32));
// if (i.M.Rc) {
// // With cr0 update.
// g.update_cr_with_cond(0, v, b.getInt64(0), true);
// }
printf("rlwinmx %d %d %d\n", i.M.SH, i.M.MB, i.M.ME);
XEINSTRNOTIMPLEMENTED();
return 1;
}
@ -687,8 +824,24 @@ XEEMITTER(sldx, 0x7C000036, X )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(slwx, 0x7C000030, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// n <- (RB)[59:63]
// r <- ROTL32((RS)[32:63], n)
// if (RB)[58] = 0 then
// m <- MASK(32, 63-n)
// else
// m <- i64.0
// RA <- r & m
Value* v = b.CreateShl(g.gpr_value(i.X.RT), g.gpr_value(i.X.RB));
v = b.CreateAnd(v, UINT32_MAX);
g.update_gpr_value(i.X.RA, v);
if (i.X.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
XEEMITTER(sradx, 0x7C000634, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
@ -707,8 +860,40 @@ XEEMITTER(srawx, 0x7C000630, X )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(srawix, 0x7C000670, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// n <- SH
// r <- ROTL32((RS)[32:63], 64-n)
// m <- MASK(n+32, 63)
// s <- (RS)[32]
// RA <- r&m | (i64.s)&¬m
// CA <- s & ((r&¬m)[32:63]≠0)
Value* rs64 = g.gpr_value(i.X.RT);
Value* rs32 = b.CreateTrunc(rs64, b.getInt32Ty());
Value* v;
Value* ca;
if (!i.X.RB) {
// No shift, just a fancy sign extend and CA clearer.
v = rs32;
ca = b.getInt64(0);
} else {
v = b.CreateAShr(rs32, i.X.RB);
// CA is set to 1 if the low-order 32 bits of (RS) contain a negative number
// and any 1-bits are shifted out of position 63; otherwise CA is set to 0.
ca = b.CreateAnd(b.CreateICmpSLT(v, b.getInt32(0)),
b.CreateICmpSLT(rs64, b.getInt64(0)));
}
v = b.CreateSExt(v, b.getInt64Ty());
g.update_gpr_value(i.X.RA, v);
g.update_xer_with_carry(ca);
if (i.X.Rc) {
// With cr0 update.
g.update_cr_with_cond(0, v, b.getInt64(0), true);
}
return 0;
}
XEEMITTER(srdx, 0x7C000436, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {

151
src/cpu/codegen/emit_control.cc
View File

@ -436,24 +436,159 @@ XEEMITTER(sc, 0x44000002, SC )(FunctionGenerator& g, IRBuilder<>& b, I
// Trap (A-25)
int XeEmitTrap(FunctionGenerator& g, IRBuilder<>& b, InstrData& i,
Value* va, Value* vb, uint32_t TO) {
// if (a < b) & TO[0] then TRAP
// if (a > b) & TO[1] then TRAP
// if (a = b) & TO[2] then TRAP
// if (a <u b) & TO[3] then TRAP
// if (a >u b) & TO[4] then TRAP
// Bits swapped:
// 01234
// 43210
if (!TO) {
return 0;
}
BasicBlock* after_bb = BasicBlock::Create(*g.context(), "", g.gen_fn(),
g.GetNextBasicBlock());
BasicBlock* trap_bb = BasicBlock::Create(*g.context(), "", g.gen_fn(),
after_bb);
// Create the basic blocks (so we can chain).
std::vector<BasicBlock*> bbs;
if (TO & (1 << 4)) {
bbs.push_back(BasicBlock::Create(*g.context(), "", g.gen_fn(), trap_bb));
}
if (TO & (1 << 3)) {
bbs.push_back(BasicBlock::Create(*g.context(), "", g.gen_fn(), trap_bb));
}
if (TO & (1 << 2)) {
bbs.push_back(BasicBlock::Create(*g.context(), "", g.gen_fn(), trap_bb));
}
if (TO & (1 << 1)) {
bbs.push_back(BasicBlock::Create(*g.context(), "", g.gen_fn(), trap_bb));
}
if (TO & (1 << 0)) {
bbs.push_back(BasicBlock::Create(*g.context(), "", g.gen_fn(), trap_bb));
}
bbs.push_back(after_bb);
// Jump to the first bb.
b.CreateBr(bbs.front());
// Setup each basic block.
std::vector<BasicBlock*>::iterator it = bbs.begin();
if (TO & (1 << 4)) {
// a < b
BasicBlock* bb = *(it++);
b.SetInsertPoint(bb);
Value* cmp = b.CreateICmpSLT(va, vb);
b.CreateCondBr(cmp, trap_bb, *it);
}
if (TO & (1 << 3)) {
// a > b
BasicBlock* bb = *(it++);
b.SetInsertPoint(bb);
Value* cmp = b.CreateICmpSGT(va, vb);
b.CreateCondBr(cmp, trap_bb, *it);
}
if (TO & (1 << 2)) {
// a = b
BasicBlock* bb = *(it++);
b.SetInsertPoint(bb);
Value* cmp = b.CreateICmpEQ(va, vb);
b.CreateCondBr(cmp, trap_bb, *it);
}
if (TO & (1 << 1)) {
// a <u b
BasicBlock* bb = *(it++);
b.SetInsertPoint(bb);
Value* cmp = b.CreateICmpULT(va, vb);
b.CreateCondBr(cmp, trap_bb, *it);
}
if (TO & (1 << 0)) {
// a >u b
BasicBlock* bb = *(it++);
b.SetInsertPoint(bb);
Value* cmp = b.CreateICmpUGT(va, vb);
b.CreateCondBr(cmp, trap_bb, *it);
}
// Create trap BB.
b.SetInsertPoint(trap_bb);
g.SpillRegisters();
// TODO(benvanik): use @llvm.debugtrap? could make debugging better
b.CreateCall2(g.gen_module()->getGlobalVariable("XeTrap"),
g.gen_fn()->arg_begin(),
b.getInt32(i.address));
b.CreateBr(after_bb);
// Resume.
b.SetInsertPoint(after_bb);
return 0;
}
XEEMITTER(td, 0x7C000088, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// a <- (RA)
// b <- (RB)
// if (a < b) & TO[0] then TRAP
// if (a > b) & TO[1] then TRAP
// if (a = b) & TO[2] then TRAP
// if (a <u b) & TO[3] then TRAP
// if (a >u b) & TO[4] then TRAP
return XeEmitTrap(g, b, i,
g.gpr_value(i.X.RA),
g.gpr_value(i.X.RB),
i.X.RT);
}
XEEMITTER(tdi, 0x08000000, D )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// a <- (RA)
// if (a < EXTS(SI)) & TO[0] then TRAP
// if (a > EXTS(SI)) & TO[1] then TRAP
// if (a = EXTS(SI)) & TO[2] then TRAP
// if (a <u EXTS(SI)) & TO[3] then TRAP
// if (a >u EXTS(SI)) & TO[4] then TRAP
return XeEmitTrap(g, b, i,
g.gpr_value(i.D.RA),
b.getInt64(XEEXTS16(i.D.DS)),
i.D.RT);
}
XEEMITTER(tw, 0x7C000008, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// a <- EXTS((RA)[32:63])
// b <- EXTS((RB)[32:63])
// if (a < b) & TO[0] then TRAP
// if (a > b) & TO[1] then TRAP
// if (a = b) & TO[2] then TRAP
// if (a <u b) & TO[3] then TRAP
// if (a >u b) & TO[4] then TRAP
return XeEmitTrap(g, b, i,
b.CreateSExt(b.CreateTrunc(g.gpr_value(i.X.RA),
b.getInt32Ty()),
b.getInt64Ty()),
b.CreateSExt(b.CreateTrunc(g.gpr_value(i.X.RB),
b.getInt32Ty()),
b.getInt64Ty()),
i.X.RT);
}
XEEMITTER(twi, 0x0C000000, D )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// a <- EXTS((RA)[32:63])
// if (a < EXTS(SI)) & TO[0] then TRAP
// if (a > EXTS(SI)) & TO[1] then TRAP
// if (a = EXTS(SI)) & TO[2] then TRAP
// if (a <u EXTS(SI)) & TO[3] then TRAP
// if (a >u EXTS(SI)) & TO[4] then TRAP
return XeEmitTrap(g, b, i,
b.CreateSExt(b.CreateTrunc(g.gpr_value(i.D.RA),
b.getInt32Ty()),
b.getInt64Ty()),
b.getInt64(XEEXTS16(i.D.DS)),
i.D.RT);
}

10
src/cpu/codegen/emit_memory.cc
View File

@ -658,13 +658,15 @@ XEEMITTER(dcbst, 0x7C00006C, X )(FunctionGenerator& g, IRBuilder<>& b, I
}
XEEMITTER(dcbt, 0x7C00022C, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// No-op for now.
// TODO(benvanik): use @llvm.prefetch
return 0;
}
XEEMITTER(dcbtst, 0x7C0001EC, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {
XEINSTRNOTIMPLEMENTED();
return 1;
// No-op for now.
// TODO(benvanik): use @llvm.prefetch
return 0;
}
XEEMITTER(dcbz, 0x7C0007EC, X )(FunctionGenerator& g, IRBuilder<>& b, InstrData& i) {

123
src/cpu/codegen/function_generator.cc
View File

@ -489,7 +489,7 @@ void FunctionGenerator::FillRegisters() {
}
// Note that we skip zero.
for (uint32_t n = 1; n < XECOUNT(locals_.gpr); n++) {
for (size_t n = 0; n < XECOUNT(locals_.gpr); n++) {
if (locals_.gpr[n]) {
b.CreateStore(LoadStateValue(
offsetof(xe_ppc_state_t, r) + 8 * n,
@ -550,7 +550,7 @@ void FunctionGenerator::SpillRegisters() {
}
// Note that we skip zero.
for (uint32_t n = 1; n < XECOUNT(locals_.gpr); n++) {
for (uint32_t n = 0; n < XECOUNT(locals_.gpr); n++) {
Value* v = locals_.gpr[n];
if (v) {
StoreStateValue(
@ -591,6 +591,62 @@ void FunctionGenerator::update_xer_value(Value* value) {
b.CreateStore(value, locals_.xer);
}
void FunctionGenerator::update_xer_with_overflow(Value* value) {
IRBuilder<>& b = *builder_;
setup_xer();
// Expects a i1 indicating overflow.
// Trust the caller that if it's larger than that it's already truncated.
if (!value->getType()->isIntegerTy(64)) {
value = b.CreateZExt(value, b.getInt64Ty());
}
Value* xer = xer_value();
xer = b.CreateAnd(xer, 0xFFFFFFFFBFFFFFFF); // clear bit 30
xer = b.CreateOr(xer, b.CreateShl(value, 31));
xer = b.CreateOr(xer, b.CreateShl(value, 30));
b.CreateStore(xer, locals_.xer);
}
void FunctionGenerator::update_xer_with_carry(Value* value) {
IRBuilder<>& b = *builder_;
setup_xer();
// Expects a i1 indicating carry.
// Trust the caller that if it's larger than that it's already truncated.
if (!value->getType()->isIntegerTy(64)) {
value = b.CreateZExt(value, b.getInt64Ty());
}
Value* xer = xer_value();
xer = b.CreateAnd(xer, 0xFFFFFFFFDFFFFFFF); // clear bit 29
xer = b.CreateOr(xer, b.CreateShl(value, 29));
b.CreateStore(xer, locals_.xer);
}
void FunctionGenerator::update_xer_with_overflow_and_carry(Value* value) {
IRBuilder<>& b = *builder_;
setup_xer();
// Expects a i1 indicating overflow.
// Trust the caller that if it's larger than that it's already truncated.
if (!value->getType()->isIntegerTy(64)) {
value = b.CreateZExt(value, b.getInt64Ty());
}
// This is effectively an update_xer_with_overflow followed by an
// update_xer_with_carry, but since the logic is largely the same share it.
Value* xer = xer_value();
xer = b.CreateAnd(xer, 0xFFFFFFFF9FFFFFFF); // clear bit 30 & 29
xer = b.CreateOr(xer, b.CreateShl(value, 31));
xer = b.CreateOr(xer, b.CreateShl(value, 30));
xer = b.CreateOr(xer, b.CreateShl(value, 29));
b.CreateStore(xer, locals_.xer);
}
void FunctionGenerator::setup_lr() {
IRBuilder<>& b = *builder_;
@ -679,10 +735,50 @@ void FunctionGenerator::update_cr_value(uint32_t n, Value* value) {
setup_cr(n);
value = b.CreateTrunc(value, b.getInt8Ty());
// Truncate to 8 bits if needed.
// TODO(benvanik): also widen?
if (!value->getType()->isIntegerTy(8)) {
value = b.CreateTrunc(value, b.getInt8Ty());
}
b.CreateStore(value, locals_.cr[n]);
}
void FunctionGenerator::update_cr_with_cond(
uint32_t n, Value* lhs, Value* rhs, bool is_signed) {
IRBuilder<>& b = *builder_;
// bit0 = RA < RB
// bit1 = RA > RB
// bit2 = RA = RB
// bit3 = XER[SO]
// Bits are reversed:
// 0123
// 3210
// TODO(benvanik): inline this using the x86 cmp instruction - this prevents
// the need for a lot of the compares and ensures we lower to the best
// possible x86.
// Value* cmp = InlineAsm::get(
// FunctionType::get(),
// "cmp $0, $1 \n"
// "mov from compare registers \n",
// "r,r", ??
// true);
Value* is_lt = is_signed ?
b.CreateICmpSLT(lhs, rhs) : b.CreateICmpULT(lhs, rhs);
Value* is_gt = is_signed ?
b.CreateICmpSGT(lhs, rhs) : b.CreateICmpUGT(lhs, rhs);
Value* cp = b.CreateSelect(is_gt, b.getInt8(1 << 2), b.getInt8(1 << 1));
Value* c = b.CreateSelect(is_lt, b.getInt8(1 << 3), cp);
// TODO(benvanik): set bit 4 to XER[SO]
// Insert the 4 bits into their location in the CR.
update_cr_value(n, c);
}
void FunctionGenerator::setup_gpr(uint32_t n) {
IRBuilder<>& b = *builder_;
@ -699,10 +795,13 @@ Value* FunctionGenerator::gpr_value(uint32_t n) {
IRBuilder<>& b = *builder_;
XEASSERT(n >= 0 && n < 32);
if (n == 0) {
// Always force zero to a constant - this should help LLVM.
return b.getInt64(0);
}
// Actually r0 is writable, even though nobody should ever do that.
// Perhaps we can check usage and enable this if safe?
// if (n == 0) {
// // Always force zero to a constant - this should help LLVM.
// return b.getInt64(0);
// }
setup_gpr(n);
@ -713,10 +812,12 @@ void FunctionGenerator::update_gpr_value(uint32_t n, Value* value) {
IRBuilder<>& b = *builder_;
XEASSERT(n >= 0 && n < 32);
if (n == 0) {
// Ignore writes to zero.
return;
}
// See above - r0 can be written.
// if (n == 0) {
// // Ignore writes to zero.
// return;
// }
setup_gpr(n);

1
src/cpu/codegen/module_generator.cc
View File

@ -250,6 +250,7 @@ void ModuleGenerator::BuildFunction(CodegenFunction* cgf) {
void ModuleGenerator::OptimizeFunction(Module* m, Function* fn) {
FunctionPassManager pm(m);
//fn->dump();
if (FLAGS_optimize_ir_functions) {
PassManagerBuilder pmb;
pmb.OptLevel = 3;

15
src/cpu/exec_module.cc
View File

@ -202,6 +202,11 @@ XECLEANUP:
return result_code;
}
void XeTrap(xe_ppc_state_t* state, uint32_t cia) {
printf("TRAP");
XEASSERTALWAYS();
}
void XeIndirectBranch(xe_ppc_state_t* state, uint64_t target, uint64_t br_ia) {
printf("INDIRECT BRANCH %.8X -> %.8X\n", (uint32_t)br_ia, (uint32_t)target);
XEASSERTALWAYS();
@ -255,6 +260,16 @@ int ExecModule::InjectGlobals() {
int8PtrTy));
// Control methods:
std::vector<Type*> trapArgs;
trapArgs.push_back(int8PtrTy);
trapArgs.push_back(Type::getInt32Ty(context));
FunctionType* trapTy = FunctionType::get(
Type::getVoidTy(context), trapArgs, false);
gv = new GlobalVariable(*gen_module_, trapTy, true,
GlobalVariable::ExternalLinkage, 0,
"XeTrap");
engine_->addGlobalMapping(gv, (void*)&XeTrap);
std::vector<Type*> indirectBranchArgs;
indirectBranchArgs.push_back(int8PtrTy);
indirectBranchArgs.push_back(Type::getInt64Ty(context));