SPU LLVM: rewrite comparison on non-xfloat path of CFLTU, CFLTS

CFLTU on non-xfloat path is accurate as xfloat path now. * Also optimize FCTIW like FCTIWZ (PPU)
RPCS3 · Dec 26, 2019 · 945c435 · 945c435
1 parent e54438d
commit 945c435
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 8 deletions.
diff --git a/rpcs3/Emu/Cell/PPUInterpreter.cpp b/rpcs3/Emu/Cell/PPUInterpreter.cpp
@@ -4832,9 +4832,9 @@ bool ppu_interpreter::FCMPU(ppu_thread& ppu, ppu_opcode_t op)
 
 bool ppu_interpreter::FCTIW(ppu_thread& ppu, ppu_opcode_t op)
 {
-	const f64 b = ppu.fpr[op.frb];
-	const s32 res = b >= f64(INT32_MAX) ? INT32_MAX : _mm_cvtsd_si32(_mm_load_sd(&b));
-	ppu.fpr[op.frd] = std::bit_cast<f64, s64>(res);
+	const auto b = _mm_load_sd(&ppu.fpr[op.frb]);
+	const auto res = _mm_xor_si128(_mm_cvtpd_epi32(b), _mm_castpd_si128(_mm_cmpge_pd(b, _mm_set1_pd(0x80000000))));
+	ppu.fpr[op.frd] = std::bit_cast<f64, s64>(_mm_cvtsi128_si32(res));
 	if (UNLIKELY(op.rc)) fmt::throw_exception("%s: op.rc", __func__); //ppu_cr_set(ppu, 1, ppu.fpscr.fg, ppu.fpscr.fl, ppu.fpscr.fe, ppu.fpscr.fu);
 	return true;
 }

diff --git a/rpcs3/Emu/Cell/PPUTranslator.cpp b/rpcs3/Emu/Cell/PPUTranslator.cpp
@@ -4004,8 +4004,10 @@ void PPUTranslator::FRSP(ppu_opcode_t op)
 void PPUTranslator::FCTIW(ppu_opcode_t op)
 {
 	const auto b = GetFpr(op.frb);
-	SetFpr(op.frd, m_ir->CreateSelect(m_ir->CreateFCmpOGE(b, ConstantFP::get(GetType<f64>(), f64(INT32_MAX))), m_ir->getInt32(INT32_MAX),
-	Call(GetType<s32>(), "llvm.x86.sse2.cvtsd2si", m_ir->CreateInsertElement(GetUndef<f64[2]>(), b, u64{0}))));
+	const auto xormask = m_ir->CreateSExt(m_ir->CreateFCmpOGE(b, ConstantFP::get(GetType<f64>(), std::exp2l(31.))), GetType<s32>());
+
+	// fix result saturation (0x80000000 -> 0x7fffffff)
+	SetFpr(op.frd, m_ir->CreateXor(xormask, Call(GetType<s32>(), "llvm.x86.sse2.cvtsd2si", m_ir->CreateInsertElement(GetUndef<f64[2]>(), b, u64{0}))));
 
 	//SetFPSCR_FR(Call(GetType<bool>(), m_pure_attr, "__fctiw_get_fr", b));
 	//SetFPSCR_FI(Call(GetType<bool>(), m_pure_attr, "__fctiw_get_fi", b));

diff --git a/rpcs3/Emu/Cell/SPURecompiler.cpp b/rpcs3/Emu/Cell/SPURecompiler.cpp
@@ -7509,7 +7509,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
 
 			value_t<s32[4]> r;
 			r.value = m_ir->CreateFPToSI(a.value, get_type<s32[4]>());
-			set_vr(op.rt, r ^ sext<s32[4]>(fcmp_ord(a >= fsplat<f32[4]>(std::exp2(31.f)))));
+			set_vr(op.rt, r ^ sext<s32[4]>(bitcast<s32[4]>(a) > splat<s32[4]>(((31 + 127) << 23) - 1)));
 		}
 	}
 
@@ -7568,7 +7568,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
 			}
 
 			r.value = m_ir->CreateFPToUI(a.value, get_type<s32[4]>());
-			set_vr(op.rt, select(fcmp_uno(a >= fsplat<f64[4]>(std::exp2(32.f))), splat<s32[4]>(-1), r & sext<s32[4]>(fcmp_ord(a >= fsplat<f64[4]>(0.)))));
+			set_vr(op.rt, select(fcmp_ord(a >= fsplat<f64[4]>(std::exp2(32.f))), splat<s32[4]>(-1), r & sext<s32[4]>(fcmp_ord(a >= fsplat<f64[4]>(0.)))));
 		}
 		else
 		{
@@ -7583,7 +7583,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
 
 			value_t<s32[4]> r;
 			r.value = m_ir->CreateFPToUI(a.value, get_type<s32[4]>());
-			set_vr(op.rt, select(fcmp_uno(a >= fsplat<f32[4]>(std::exp2(32.f))), splat<s32[4]>(-1), r & ~(bitcast<s32[4]>(a) >> 31)));
+			set_vr(op.rt, select(bitcast<s32[4]>(a) > splat<s32[4]>(((32 + 127) << 23) - 1), splat<s32[4]>(-1), r & ~(bitcast<s32[4]>(a) >> 31)));
 		}
 	}