fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ome guest function calls can now be resolved and embedded directly in

Browse Source 
the emitted asm as rel32 calls. Disabled by default, enabled via
resolve_rel32_guest_calls
detect whether cpu has fast jrcxz, fast loop/loope/loopne
much more thorough LoadConstantXMM
New cvar elide_e0_check that allows the backend to assume accesses via
the SP or TLS register will not cross into 0xe0 range
Add x64 codegen for Vector shift uint8
If has fast jrcxz use for some traptrue/breaktrue instructions
Use phat nops
Add cvar use_fast_dot_product, which uses a four instruction sequence
for both dot product instructions which ought to be equivalent. disabled
by default.
This commit is contained in:
chss95cs@gmail.com
2022-06-20 15:08:18 -07:00
parent a4ff64c465
commit 549ee28a93
6 changed files with 583 additions and 95 deletions
Download Patch File Download Diff File Expand all files Collapse all files

92
src/xenia/cpu/backend/x64/x64_sequences.cc
View File

@@ -38,6 +38,10 @@
#include "xenia/cpu/hir/hir_builder.h"
#include "xenia/cpu/processor.h"
DEFINE_bool(use_fast_dot_product, false,
"Experimental optimization, much shorter sequence on dot products, treating inf as overflow instead of using mcxsr"
"four insn dotprod",
"CPU");
namespace xe {
namespace cpu {
namespace backend {
@@ -886,7 +890,10 @@ struct COMPARE_EQ_I8
e.cmp(src1, src2);
},
[](X64Emitter& e, const Reg8& src1, int32_t constant) {
e.cmp(src1, constant);
if (constant == 0) {
e.test(src1, src1);
} else
e.cmp(src1, constant);
});
e.sete(i.dest);
}
@@ -900,7 +907,10 @@ struct COMPARE_EQ_I16
e.cmp(src1, src2);
},
[](X64Emitter& e, const Reg16& src1, int32_t constant) {
e.cmp(src1, constant);
if (constant == 0) {
e.test(src1, src1);
} else
e.cmp(src1, constant);
});
e.sete(i.dest);
}
@@ -914,7 +924,10 @@ struct COMPARE_EQ_I32
e.cmp(src1, src2);
},
[](X64Emitter& e, const Reg32& src1, int32_t constant) {
e.cmp(src1, constant);
if (constant == 0) {
e.test(src1, src1);
} else
e.cmp(src1, constant);
});
e.sete(i.dest);
}
@@ -928,7 +941,10 @@ struct COMPARE_EQ_I64
e.cmp(src1, src2);
},
[](X64Emitter& e, const Reg64& src1, int32_t constant) {
e.cmp(src1, constant);
if (constant == 0) {
e.test(src1, src1);
} else
e.cmp(src1, constant);
});
e.sete(i.dest);
}
@@ -1980,6 +1996,8 @@ struct DIV_V128 : Sequence<DIV_V128, I<OPCODE_DIV, V128Op, V128Op, V128Op>> {
assert_true(!i.instr->flags);
EmitAssociativeBinaryXmmOp(e, i,
[](X64Emitter& e, Xmm dest, Xmm src1, Xmm src2) {
// e.vrcpps(e.xmm0, src2);
//e.vmulps(dest, src1, e.xmm0);
e.vdivps(dest, src1, src2);
});
}
@@ -2591,43 +2609,51 @@ EMITTER_OPCODE_TABLE(OPCODE_LOG2, LOG2_F32, LOG2_F64, LOG2_V128);
struct DOT_PRODUCT_V128 {
static void Emit(X64Emitter& e, Xmm dest, Xmm src1, Xmm src2, uint8_t imm) {
// TODO(benvanik): apparently this is very slow
// - find alternative?
Xbyak::Label end;
e.inLocalLabel();
if (cvars::use_fast_dot_product) {
e.vdpps(dest, src1, src2, imm);
e.vandps(e.xmm0, dest, e.GetXmmConstPtr(XMMAbsMaskPS));
e.vcmpgeps(e.xmm0, e.xmm0, e.GetXmmConstPtr(XMMFloatInf));
e.vblendvps(dest, dest, e.GetXmmConstPtr(XMMQNaN), e.xmm0);
// Grab space to put MXCSR.
// TODO(gibbed): stick this in TLS or
// something?
e.sub(e.rsp, 8);
} else {
// TODO(benvanik): apparently this is very slow
// - find alternative?
Xbyak::Label end;
e.inLocalLabel();
// Grab MXCSR and mask off the overflow flag,
// because it's sticky.
e.vstmxcsr(e.dword[e.rsp]);
e.mov(e.eax, e.dword[e.rsp]);
e.and_(e.eax, uint32_t(~8));
e.mov(e.dword[e.rsp], e.eax);
e.vldmxcsr(e.dword[e.rsp]);
// Grab space to put MXCSR.
// TODO(gibbed): stick this in TLS or
// something?
e.sub(e.rsp, 8);
// Hey we can do the dot product now.
e.vdpps(dest, src1, src2, imm);
// Grab MXCSR and mask off the overflow flag,
// because it's sticky.
e.vstmxcsr(e.dword[e.rsp]);
e.mov(e.eax, e.dword[e.rsp]);
e.and_(e.eax, uint32_t(~8));
e.mov(e.dword[e.rsp], e.eax);
e.vldmxcsr(e.dword[e.rsp]);
// Load MXCSR...
e.vstmxcsr(e.dword[e.rsp]);
// Hey we can do the dot product now.
e.vdpps(dest, src1, src2, imm);
// ..free our temporary space and get MXCSR at
// the same time
e.pop(e.rax);
// Load MXCSR...
e.vstmxcsr(e.dword[e.rsp]);
// Did we overflow?
e.test(e.al, 8);
e.jz(end);
// ..free our temporary space and get MXCSR at
// the same time
e.pop(e.rax);
// Infinity? HA! Give NAN.
e.vmovdqa(dest, e.GetXmmConstPtr(XMMQNaN));
// Did we overflow?
e.test(e.al, 8);
e.jz(end);
e.L(end);
e.outLocalLabel();
// Infinity? HA! Give NAN.
e.vmovdqa(dest, e.GetXmmConstPtr(XMMQNaN));
e.L(end);
e.outLocalLabel();
}
}
};