Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

softgpu: Use NEON SIMD for alpha blending, etc.

Browse source
This commit is contained in:
Unknown W. Brackets 2023-01-07 16:57:18 -08:00
parent b55dbdab7f
commit 587a322207
2 changed files with 84 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

59
GPU/Software/DrawPixel.cpp
View file

@ -15,6 +15,7 @@
// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
#include "ppsspp_config.h"
#include <mutex>
#include "Common/Common.h"
#include "Common/Data/Convert/ColorConv.h"
@ -416,6 +417,8 @@ static inline Vec3<int> GetSourceFactor(PixelBlendFactor factor, const Vec4<int>
case PixelBlendFactor::SRCALPHA:
#if defined(_M_SSE)
return Vec3<int>(_mm_shuffle_epi32(source.ivec, _MM_SHUFFLE(3, 3, 3, 3)));
#elif PPSSPP_ARCH(ARM64_NEON)
return Vec3<int>(vdupq_laneq_s32(source.ivec, 3));
#else
return Vec3<int>::AssignToAll(source.a());
#endif
@ -423,6 +426,8 @@ static inline Vec3<int> GetSourceFactor(PixelBlendFactor factor, const Vec4<int>
case PixelBlendFactor::INVSRCALPHA:
#if defined(_M_SSE)
return Vec3<int>(_mm_sub_epi32(_mm_set1_epi32(255), _mm_shuffle_epi32(source.ivec, _MM_SHUFFLE(3, 3, 3, 3))));
#elif PPSSPP_ARCH(ARM64_NEON)
return Vec3<int>(vsubq_s32(vdupq_n_s32(255), vdupq_laneq_s32(source.ivec, 3)));
#else
return Vec3<int>::AssignToAll(255 - source.a());
#endif
@ -469,6 +474,8 @@ static inline Vec3<int> GetDestFactor(PixelBlendFactor factor, const Vec4<int> &
case PixelBlendFactor::SRCALPHA:
#if defined(_M_SSE)
return Vec3<int>(_mm_shuffle_epi32(source.ivec, _MM_SHUFFLE(3, 3, 3, 3)));
#elif PPSSPP_ARCH(ARM64_NEON)
return Vec3<int>(vdupq_laneq_s32(source.ivec, 3));
#else
return Vec3<int>::AssignToAll(source.a());
#endif
@ -476,6 +483,8 @@ static inline Vec3<int> GetDestFactor(PixelBlendFactor factor, const Vec4<int> &
case PixelBlendFactor::INVSRCALPHA:
#if defined(_M_SSE)
return Vec3<int>(_mm_sub_epi32(_mm_set1_epi32(255), _mm_shuffle_epi32(source.ivec, _MM_SHUFFLE(3, 3, 3, 3))));
#elif PPSSPP_ARCH(ARM64_NEON)
return Vec3<int>(vsubq_s32(vdupq_n_s32(255), vdupq_laneq_s32(source.ivec, 3)));
#else
return Vec3<int>::AssignToAll(255 - source.a());
#endif
@ -533,6 +542,18 @@ static Vec3<int> AlphaBlendingResult(const PixelFuncID &pixelID, const Vec4<int>
const __m128i d = _mm_mulhi_epi16(drgb, df);
return Vec3<int>(_mm_unpacklo_epi16(_mm_adds_epi16(s, d), _mm_setzero_si128()));
#elif PPSSPP_ARCH(ARM64_NEON)
const int32x4_t half = vdupq_n_s32(1);
const int32x4_t srgb = vaddq_s32(vshlq_n_s32(source.ivec, 1), half);
const int32x4_t sf = vaddq_s32(vshlq_n_s32(srcfactor.ivec, 1), half);
const int32x4_t s = vshrq_n_s32(vmulq_s32(srgb, sf), 10);
const int32x4_t drgb = vaddq_s32(vshlq_n_s32(dst.ivec, 1), half);
const int32x4_t df = vaddq_s32(vshlq_n_s32(dstfactor.ivec, 1), half);
const int32x4_t d = vshrq_n_s32(vmulq_s32(drgb, df), 10);
return Vec3<int>(vaddq_s32(s, d));
#else
static constexpr Vec3<int> half = Vec3<int>::AssignToAll(1);
Vec3<int> lhs = ((source.rgb() * 2 + half) * (srcfactor * 2 + half)) / 1024;
@ -555,6 +576,18 @@ static Vec3<int> AlphaBlendingResult(const PixelFuncID &pixelID, const Vec4<int>
const __m128i d = _mm_mulhi_epi16(drgb, df);
return Vec3<int>(_mm_unpacklo_epi16(_mm_max_epi16(_mm_subs_epi16(s, d), _mm_setzero_si128()), _mm_setzero_si128()));
#elif PPSSPP_ARCH(ARM64_NEON)
const int32x4_t half = vdupq_n_s32(1);
const int32x4_t srgb = vaddq_s32(vshlq_n_s32(source.ivec, 1), half);
const int32x4_t sf = vaddq_s32(vshlq_n_s32(srcfactor.ivec, 1), half);
const int32x4_t s = vshrq_n_s32(vmulq_s32(srgb, sf), 10);
const int32x4_t drgb = vaddq_s32(vshlq_n_s32(dst.ivec, 1), half);
const int32x4_t df = vaddq_s32(vshlq_n_s32(dstfactor.ivec, 1), half);
const int32x4_t d = vshrq_n_s32(vmulq_s32(drgb, df), 10);
return Vec3<int>(vqsubq_s32(s, d));
#else
static constexpr Vec3<int> half = Vec3<int>::AssignToAll(1);
Vec3<int> lhs = ((source.rgb() * 2 + half) * (srcfactor * 2 + half)) / 1024;
@ -577,6 +610,18 @@ static Vec3<int> AlphaBlendingResult(const PixelFuncID &pixelID, const Vec4<int>
const __m128i d = _mm_mulhi_epi16(drgb, df);
return Vec3<int>(_mm_unpacklo_epi16(_mm_max_epi16(_mm_subs_epi16(d, s), _mm_setzero_si128()), _mm_setzero_si128()));
#elif PPSSPP_ARCH(ARM64_NEON)
const int32x4_t half = vdupq_n_s32(1);
const int32x4_t srgb = vaddq_s32(vshlq_n_s32(source.ivec, 1), half);
const int32x4_t sf = vaddq_s32(vshlq_n_s32(srcfactor.ivec, 1), half);
const int32x4_t s = vshrq_n_s32(vmulq_s32(srgb, sf), 10);
const int32x4_t drgb = vaddq_s32(vshlq_n_s32(dst.ivec, 1), half);
const int32x4_t df = vaddq_s32(vshlq_n_s32(dstfactor.ivec, 1), half);
const int32x4_t d = vshrq_n_s32(vmulq_s32(drgb, df), 10);
return Vec3<int>(vqsubq_s32(d, s));
#else
static constexpr Vec3<int> half = Vec3<int>::AssignToAll(1);
Vec3<int> lhs = ((source.rgb() * 2 + half) * (srcfactor * 2 + half)) / 1024;
@ -586,19 +631,31 @@ static Vec3<int> AlphaBlendingResult(const PixelFuncID &pixelID, const Vec4<int>
}
case GE_BLENDMODE_MIN:
#if PPSSPP_ARCH(ARM64_NEON)
return Vec3<int>(vminq_s32(source.ivec, dst.ivec));
#else
return Vec3<int>(std::min(source.r(), dst.r()),
std::min(source.g(), dst.g()),
std::min(source.b(), dst.b()));
#endif
case GE_BLENDMODE_MAX:
#if PPSSPP_ARCH(ARM64_NEON)
return Vec3<int>(vmaxq_s32(source.ivec, dst.ivec));
#else
return Vec3<int>(std::max(source.r(), dst.r()),
std::max(source.g(), dst.g()),
std::max(source.b(), dst.b()));
#endif
case GE_BLENDMODE_ABSDIFF:
#if PPSSPP_ARCH(ARM64_NEON)
return Vec3<int>(vabdq_s32(source.ivec, dst.ivec));
#else
return Vec3<int>(::abs(source.r() - dst.r()),
::abs(source.g() - dst.g()),
::abs(source.b() - dst.b()));
#endif
default:
return source.rgb();
@ -684,7 +741,7 @@ void SOFTRAST_CALL DrawSinglePixel(int x, int y, int z, int fog, Vec4IntArg colo
prim_color += Vec4<int>::AssignToAll(pixelID.cached.ditherMatrix[(y & 3) * 4 + (x & 3)]);
}
#if defined(_M_SSE)
#if defined(_M_SSE) || PPSSPP_ARCH(ARM64_NEON)
new_color = Vec3<int>(prim_color.ivec).ToRGB();
new_color |= stencil << 24;
#else

26
GPU/Software/RasterizerRectangle.cpp
View file

@ -1,5 +1,6 @@
// See comment in header for the purpose of the code in this file.
#include "ppsspp_config.h"
#include <algorithm>
#include <cmath>
@ -25,6 +26,14 @@
#include <emmintrin.h>
#endif
#if PPSSPP_ARCH(ARM_NEON)
#if defined(_MSC_VER) && PPSSPP_ARCH(ARM64)
#include <arm64_neon.h>
#else
#include <arm_neon.h>
#endif
#endif
extern DSStretch g_DarkStalkerStretch;
// For Darkstalkers hack. Ugh.
extern bool currentDialogActive;
@ -57,6 +66,23 @@ static uint32_t StandardAlphaBlend(uint32_t source, uint32_t dst) {
const __m128i blended16 = _mm_adds_epi16(s, d);
return _mm_cvtsi128_si32(_mm_packus_epi16(blended16, blended16));
#elif PPSSPP_ARCH(ARM64_NEON)
uint16x4_t sf = vdup_n_u16((source >> 24) * 2 + 1);
uint16x4_t df = vdup_n_u16((255 - (source >> 24)) * 2 + 1);
// Convert both to 16-bit, double, and add the half before even going to 32 bit.
uint16x8_t sd_c16 = vmovl_u8(vcreate_u8((uint64_t)source | ((uint64_t)dst << 32)));
sd_c16 = vaddq_u16(vshlq_n_u16(sd_c16, 1), vdupq_n_u16(1));
uint16x4_t srgb = vget_low_u16(sd_c16);
uint16x4_t drgb = vget_high_u16(sd_c16);
uint16x4_t s = vshrn_n_u32(vmull_u16(srgb, sf), 10);
uint16x4_t d = vshrn_n_u32(vmull_u16(drgb, df), 10);
uint16x4_t blended = vset_lane_s16(0, vadd_u16(s, d), 3);
uint8x8_t blended8 = vqmovn_u16(vcombine_u16(blended, blended));
return vget_lane_u32(vreinterpret_u32_u8(blended8), 0);
#else
Vec3<int> srcfactor = Vec3<int>::AssignToAll(source >> 24);
Vec3<int> dstfactor = Vec3<int>::AssignToAll(255 - (source >> 24));