iris/kaizen
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Implement rounding as SIMD for now so that the float env is not fucked with by the compiler

Browse Source
This commit is contained in:
SimoneN64
2024-06-25 23:48:02 +02:00
parent 09c45d6750
commit 15838db5b6
2 changed files with 142 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
src/backend/core/registers/cop/cop1instructions.cpp
View File

@@ -3,6 +3,7 @@
#include <core/Interpreter.hpp>
#include <cmath>
#include <cfenv>
#include <utils/FloatingPoint.hpp>
namespace n64 {
template<> auto Cop1::FGR_T<s32>(Cop0Status& status, u32 index) -> s32& {
@@ -431,7 +432,7 @@ void Cop1::ceills(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<float>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s64>(fs)) return;
CHECK_FPE(s64, fd, ceilf32(fs))
CHECK_FPE(s64, fd, Util::roundCeil<s64>(fs))
FGR_D<s64>(regs.cop0.status, FD(instr)) = fd;
}
@@ -439,7 +440,7 @@ void Cop1::ceilld(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<double>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s64>(fs)) return;
CHECK_FPE(s64, fd, ceilf64(fs))
CHECK_FPE(s64, fd, Util::roundCeil<s64>(fs))
FGR_D<s64>(regs.cop0.status, FD(instr)) = fd;
}
@@ -447,7 +448,7 @@ void Cop1::ceilws(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<float>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s32>(fs)) return;
CHECK_FPE_CONV(s32, fd, ceilf32(fs))
CHECK_FPE_CONV(s32, fd, Util::roundCeil<s32>(fs))
FGR_D<s32>(regs.cop0.status, FD(instr)) = fd;
}
@@ -455,7 +456,7 @@ void Cop1::ceilwd(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<double>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s32>(fs)) return;
CHECK_FPE_CONV(s32, fd, ceilf64(fs))
CHECK_FPE_CONV(s32, fd, Util::roundCeil<s32>(fs))
FGR_D<s32>(regs.cop0.status, FD(instr)) = fd;
}
@@ -544,7 +545,7 @@ void Cop1::cvtwd(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<double>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s32>(fs)) return;
CHECK_FPE_CONV(s32, fd, rintf64(fs))
CHECK_FPE_CONV(s32, fd, Util::roundCurrent<s32>(fs))
FGR_D<s32>(regs.cop0.status, FD(instr)) = fd;
}
@@ -552,7 +553,7 @@ void Cop1::cvtws(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<float>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s32>(fs)) return;
CHECK_FPE_CONV(s32, fd, rintf32(fs))
CHECK_FPE_CONV(s32, fd, Util::roundCurrent<s32>(fs))
FGR_D<s32>(regs.cop0.status, FD(instr)) = fd;
}
@@ -560,7 +561,7 @@ void Cop1::cvtls(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<float>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s64>(fs)) return;
CHECK_FPE(s64, fd, rintf32(fs))
CHECK_FPE(s64, fd, Util::roundCurrent<s64>(fs))
FGR_D<s64>(regs.cop0.status, FD(instr)) = fd;
}
@@ -590,7 +591,7 @@ void Cop1::cvtld(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<double>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s64>(fs)) return;
CHECK_FPE(s64, fd, rintf64(fs))
CHECK_FPE(s64, fd, Util::roundCurrent<s64>(fs))
FGR_D<s64>(regs.cop0.status, FD(instr)) = fd;
}
@@ -901,7 +902,7 @@ void Cop1::roundls(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<float>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s64>(fs)) return;
CHECK_FPE(s64, fd, roundevenf(fs))
CHECK_FPE(s64, fd, Util::roundNearest<s64>(fs))
if(fd != fs && SetCauseInexact()) {
regs.cop0.FireException(ExceptionCode::FloatingPointError, 0, regs.oldPC);
return;
@@ -913,7 +914,7 @@ void Cop1::roundld(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<double>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s64>(fs)) return;
CHECK_FPE(s64, fd, roundeven(fs))
CHECK_FPE(s64, fd, Util::roundNearest<s64>(fs))
if(fd != fs && SetCauseInexact()) {
regs.cop0.FireException(ExceptionCode::FloatingPointError, 0, regs.oldPC);
return;
@@ -925,7 +926,7 @@ void Cop1::roundws(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<float>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s32>(fs)) return;
CHECK_FPE_CONV(s32, fd, roundevenf(fs))
CHECK_FPE_CONV(s32, fd, Util::roundNearest<s32>(fs))
if(fd != fs && SetCauseInexact()) {
regs.cop0.FireException(ExceptionCode::FloatingPointError, 0, regs.oldPC);
return;
@@ -937,7 +938,7 @@ void Cop1::roundwd(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<double>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s32>(fs)) return;
CHECK_FPE_CONV(s32, fd, roundeven(fs))
CHECK_FPE_CONV(s32, fd, Util::roundNearest<s32>(fs))
if(fd != fs && SetCauseInexact()) {
regs.cop0.FireException(ExceptionCode::FloatingPointError, 0, regs.oldPC);
return;
@@ -949,7 +950,7 @@ void Cop1::floorls(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<float>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s64>(fs)) return;
CHECK_FPE(s64, fd, floorf32(fs))
CHECK_FPE(s64, fd, Util::roundFloor<s64>(fs))
FGR_D<s64>(regs.cop0.status, FD(instr)) = fd;
}
@@ -957,7 +958,7 @@ void Cop1::floorld(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<double>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s64>(fs)) return;
CHECK_FPE(s64, fd, floorf64(fs))
CHECK_FPE(s64, fd, Util::roundFloor<s64>(fs))
FGR_D<s64>(regs.cop0.status, FD(instr)) = fd;
}
@@ -965,7 +966,7 @@ void Cop1::floorws(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<float>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s32>(fs)) return;
CHECK_FPE_CONV(s32, fd, floorf32(fs))
CHECK_FPE_CONV(s32, fd, Util::roundFloor<s32>(fs))
FGR_D<s32>(regs.cop0.status, FD(instr)) = fd;
}
@@ -973,7 +974,7 @@ void Cop1::floorwd(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<double>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s32>(fs)) return;
CHECK_FPE_CONV(s32, fd, floorf64(fs))
CHECK_FPE_CONV(s32, fd, Util::roundFloor<s32>(fs))
FGR_D<s32>(regs.cop0.status, FD(instr)) = fd;
}
@@ -981,7 +982,7 @@ void Cop1::truncws(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<float>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s32>(fs)) return;
CHECK_FPE_CONV(s32, fd, truncf32(fs))
CHECK_FPE_CONV(s32, fd, Util::roundTrunc<s32>(fs))
if((float)fd != fs && SetCauseInexact()) {
regs.cop0.FireException(ExceptionCode::FloatingPointError, 0, regs.oldPC);
return;
@@ -993,7 +994,7 @@ void Cop1::truncwd(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<double>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s32>(fs)) return;
CHECK_FPE_CONV(s32, fd, truncf64(fs))
CHECK_FPE_CONV(s32, fd, Util::roundTrunc<s32>(fs))
if((double)fd != fs && SetCauseInexact()) {
regs.cop0.FireException(ExceptionCode::FloatingPointError, 0, regs.oldPC);
return;
@@ -1005,7 +1006,7 @@ void Cop1::truncls(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<float>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s64>(fs)) return;
CHECK_FPE(s64, fd, truncf32(fs))
CHECK_FPE(s64, fd, Util::roundTrunc<s64>(fs))
if((float)fd != fs && SetCauseInexact()) {
regs.cop0.FireException(ExceptionCode::FloatingPointError, 0, regs.oldPC);
return;
@@ -1017,7 +1018,7 @@ void Cop1::truncld(u32 instr) {
if(!CheckFPUUsable()) return;
auto fs = FGR_S<double>(regs.cop0.status, FS(instr));
if(!CheckCVTArg<s64>(fs)) return;
CHECK_FPE(s64, fd, truncf64(fs))
CHECK_FPE(s64, fd, Util::roundTrunc<s64>(fs))
if((double)fd != fs && SetCauseInexact()) {
regs.cop0.FireException(ExceptionCode::FloatingPointError, 0, regs.oldPC);
return;

121
src/utils/FloatingPoint.hpp Normal file
View File

@@ -0,0 +1,121 @@
//
// Created by simone on 6/25/24.
//
#pragma once
#include <common.hpp>
#include <cmath>
#include <immintrin.h>
namespace Util {
template <typename T>
static inline T roundCeil(float f) {
#ifdef SIMD_SUPPORT
__m128 t = _mm_set_ss(f);
t = _mm_round_ss(t, t, _MM_FROUND_TO_POS_INF);
return _mm_cvtss_f32(t);
#else
return ceilf(f);
#endif
}
template <typename T>
static inline T roundCeil(double f) {
#ifdef SIMD_SUPPORT
__m128d t = _mm_set_sd(f);
t = _mm_round_sd(t, t, _MM_FROUND_TO_POS_INF);
return _mm_cvtsd_f64(t);
#else
return ceil(f);
#endif
}
template <typename T>
static inline T roundNearest(float f) {
#ifdef SIMD_SUPPORT
__m128 t = _mm_set_ss(f);
t = _mm_round_ss(t, t, _MM_FROUND_TO_NEAREST_INT);
return _mm_cvtss_f32(t);
#else
return roundf(f);
#endif
}
template <typename T>
static inline T roundNearest(double f) {
#ifdef SIMD_SUPPORT
__m128d t = _mm_set_sd(f);
t = _mm_round_sd(t, t, _MM_FROUND_TO_NEAREST_INT);
return _mm_cvtsd_f64(t);
#else
return round(f);
#endif
}
template<typename T>
static inline T roundCurrent(float f) {
#ifdef SIMD_SUPPORT
auto t = _mm_set_ss(f);
t = _mm_round_ss(t, t, _MM_FROUND_CUR_DIRECTION);
return _mm_cvtss_f32(t);
#else
return rint(f);
#endif
}
template<typename T>
static inline T roundCurrent(double f) {
#ifdef SIMD_SUPPORT
auto t = _mm_set_sd(f);
t = _mm_round_sd(t, t, _MM_FROUND_CUR_DIRECTION);
return _mm_cvtsd_f64(t);
#else
return rint(f);
#endif
}
template <typename T>
static inline T roundFloor(float f) {
#ifdef SIMD_SUPPORT
__m128 t = _mm_set_ss(f);
t = _mm_round_ss(t, t, _MM_FROUND_TO_NEG_INF);
return _mm_cvtss_f32(t);
#else
return floor(f);
#endif
}
template <typename T>
static inline T roundFloor(double f) {
#ifdef SIMD_SUPPORT
__m128d t = _mm_set_sd(f);
t = _mm_round_sd(t, t, _MM_FROUND_TO_NEG_INF);
return _mm_cvtsd_f64(t);
#else
return floor(f);
#endif
}
template <typename T>
static inline T roundTrunc(float f) {
#ifdef SIMD_SUPPORT
__m128 t = _mm_set_ss(f);
t = _mm_round_ss(t, t, _MM_FROUND_TO_ZERO);
return _mm_cvtss_f32(t);
#else
return trunc(f);
#endif
}
template <typename T>
static inline T roundTrunc(double f) {
#ifdef SIMD_SUPPORT
__m128d t = _mm_set_sd(f);
t = _mm_round_sd(t, t, _MM_FROUND_TO_ZERO);
return _mm_cvtsd_f64(t);
#else
return trunc(f);
#endif
}
}