#pragma once
#include "utils.h"
+#include "common/simdintrin.h"
//////////////////////////////////////////////////////////////////////////
/// PackTraits - Helpers for packing / unpacking same pixel sizes
static simdscalar pack(simdscalar &in) = delete;
#if ENABLE_AVX512_SIMD16
static simd16scalar loadSOA_16(const uint8_t *pSrc) = delete;
- static void SIMDAPI storeSOA(uint8_t *pDst, simd16scalar src) = delete;
+ static void SIMDCALL storeSOA(uint8_t *pDst, simd16scalar src) = delete;
static simd16scalar unpack(simd16scalar &in) = delete;
static simd16scalar pack(simd16scalar &in) = delete;
#endif
static simdscalar pack(simdscalar &in) { return _simd_setzero_ps(); }
#if ENABLE_AVX512_SIMD16
static simd16scalar loadSOA_16(const uint8_t *pSrc) { return _simd16_setzero_ps(); }
- static void SIMDAPI storeSOA(uint8_t *pDst, simd16scalar src) { return; }
+ static void SIMDCALL storeSOA(uint8_t *pDst, simd16scalar src) { return; }
static simd16scalar unpack(simd16scalar &in) { return _simd16_setzero_ps(); }
static simd16scalar pack(simd16scalar &in) { return _simd16_setzero_ps(); }
#endif
__m256i result = _mm256_castsi128_si256(resLo);
result = _mm256_insertf128_si256(result, resHi, 1);
- return _mm256_castsi256_ps(result);
+ return simdscalar{ _mm256_castsi256_ps(result) };
#else
return _mm256_castsi256_ps(_mm256_cvtepu8_epi32(_mm_castps_si128(_mm256_castps256_ps128(in))));
#endif
return result;
}
- static void SIMDAPI storeSOA(uint8_t *pDst, simd16scalar src)
+ static void SIMDCALL storeSOA(uint8_t *pDst, simd16scalar src)
{
// store simd16 bytes
_mm_store_ps(reinterpret_cast<float *>(pDst), _mm256_castps256_ps128(_simd16_extract_ps(src, 0)));
static simd16scalar unpack(simd16scalar &in)
{
- simd16scalari result = _simd16_cvtepu8_epi32(_mm_castps_si128(_mm256_castps256_ps128(_simd16_extract_ps(in, 0))));
+ simd4scalari tmp = _mm_castps_si128(_mm256_castps256_ps128(_simd16_extract_ps(in, 0)));
+ simd16scalari result = _simd16_cvtepu8_epi32(tmp);
return _simd16_castsi_ps(result);
}
return result;
}
- static void SIMDAPI storeSOA(uint8_t *pDst, simd16scalar src)
+ static void SIMDCALL storeSOA(uint8_t *pDst, simd16scalar src)
{
// store simd16 bytes
_mm_store_ps(reinterpret_cast<float *>(pDst), _mm256_castps256_ps128(_simd16_extract_ps(src, 0)));
static simd16scalar unpack(simd16scalar &in)
{
- simd16scalari result = _simd16_cvtepu8_epi32(_mm_castps_si128(_mm256_castps256_ps128(_simd16_extract_ps(in, 0))));
+ simd4scalari tmp = _mm_castps_si128(_mm256_castps256_ps128(_simd16_extract_ps(in, 0)));
+ simd16scalari result = _simd16_cvtepu8_epi32(tmp);
return _simd16_castsi_ps(result);
}
return result;
}
- static void SIMDAPI storeSOA(uint8_t *pDst, simd16scalar src)
+ static void SIMDCALL storeSOA(uint8_t *pDst, simd16scalar src)
{
_simd_store_ps(reinterpret_cast<float *>(pDst), _simd16_extract_ps(src, 0));
}
return result;
}
- static void SIMDAPI storeSOA(uint8_t *pDst, simd16scalar src)
+ static void SIMDCALL storeSOA(uint8_t *pDst, simd16scalar src)
{
_simd_store_ps(reinterpret_cast<float *>(pDst), _simd16_extract_ps(src, 0));
}
return _simd16_load_ps(reinterpret_cast<const float *>(pSrc));
}
- static void SIMDAPI storeSOA(uint8_t *pDst, simd16scalar src)
+ static void SIMDCALL storeSOA(uint8_t *pDst, simd16scalar src)
{
_simd16_store_ps(reinterpret_cast<float *>(pDst), src);
}
#if ENABLE_AVX512_SIMD16
template< unsigned expnum, unsigned expden, unsigned coeffnum, unsigned coeffden >
-inline static simd16scalar SIMDAPI fastpow(simd16scalar value)
+inline static simd16scalar SIMDCALL fastpow(simd16scalar value)
{
static const float factor1 = exp2(127.0f * expden / expnum - 127.0f)
* powf(1.0f * coeffnum / coeffden, 1.0f * expden / expnum);
return result;
}
-inline static simd16scalar SIMDAPI pow512_4(simd16scalar arg)
+inline static simd16scalar SIMDCALL pow512_4(simd16scalar arg)
{
// 5/12 is too small, so compute the 4th root of 20/12 instead.
// 20/12 = 5/3 = 1 + 2/3 = 2 - 1/3. 2/3 is a suitable argument for fastpow.
return xavg;
}
-inline static simd16scalar SIMDAPI powf_wrapper(const simd16scalar base, float exp)
+inline static simd16scalar SIMDCALL powf_wrapper(const simd16scalar base, float exp)
{
const float *f = reinterpret_cast<const float *>(&base);
};
//////////////////////////////////////////////////////////////////////////
-/// Format1 - Bitfield for single component formats.
+/// FormatIntType - Calculate base integer type for pixel components based
+/// on total number of bits. Components can be smaller
+/// that this type, but the entire pixel must not be
+/// any smaller than this type.
//////////////////////////////////////////////////////////////////////////
-template<uint32_t x>
-struct Format1
+template <uint32_t bits, bool bits8 = bits <= 8, bool bits16 = bits <= 16>
+struct FormatIntType
{
- union
- {
- uint32_t r : x;
-
- ///@ The following are here to provide full template needed in Formats.
- uint32_t g : x;
- uint32_t b : x;
- uint32_t a : x;
- };
+ typedef uint32_t TYPE;
};
-//////////////////////////////////////////////////////////////////////////
-/// Format1 - Bitfield for single component formats - 8 bit specialization
-//////////////////////////////////////////////////////////////////////////
-template<>
-struct Format1<8>
+template <uint32_t bits>
+struct FormatIntType<bits, true, true>
{
- union
- {
- uint8_t r;
-
- ///@ The following are here to provide full template needed in Formats.
- uint8_t g;
- uint8_t b;
- uint8_t a;
- };
+ typedef uint8_t TYPE;
};
-//////////////////////////////////////////////////////////////////////////
-/// Format1 - Bitfield for single component formats - 16 bit specialization
-//////////////////////////////////////////////////////////////////////////
-template<>
-struct Format1<16>
+template <uint32_t bits>
+struct FormatIntType<bits, false, true>
{
- union
- {
- uint16_t r;
-
- ///@ The following are here to provide full template needed in Formats.
- uint16_t g;
- uint16_t b;
- uint16_t a;
- };
+ typedef uint16_t TYPE;
};
//////////////////////////////////////////////////////////////////////////
-/// Format2 - Bitfield for 2 component formats.
+/// Format1 - Bitfield for single component formats.
//////////////////////////////////////////////////////////////////////////
-template<uint32_t x, uint32_t y>
-union Format2
+template<uint32_t x>
+union Format1
{
+ typedef typename FormatIntType<x>::TYPE TYPE;
struct
{
- uint32_t r : x;
- uint32_t g : y;
+ TYPE r : x;
};
+
+ ///@ The following are here to provide full template needed in Formats.
struct
{
- ///@ The following are here to provide full template needed in Formats.
- uint32_t b : x;
- uint32_t a : y;
+ TYPE g : x;
+ };
+ struct
+ {
+ TYPE b : x;
+ };
+ struct
+ {
+ TYPE a : x;
};
};
//////////////////////////////////////////////////////////////////////////
-/// Format2 - Bitfield for 2 component formats - 16 bit specialization
+/// Format2 - Bitfield for 2 component formats.
//////////////////////////////////////////////////////////////////////////
-template<>
-union Format2<8,8>
+template<uint32_t x, uint32_t y>
+union Format2
{
+ typedef typename FormatIntType<x + y>::TYPE TYPE;
+
struct
{
- uint16_t r : 8;
- uint16_t g : 8;
+ TYPE r : x;
+ TYPE g : y;
};
struct
{
///@ The following are here to provide full template needed in Formats.
- uint16_t b : 8;
- uint16_t a : 8;
+ TYPE b : x;
+ TYPE a : y;
};
};
template<uint32_t x, uint32_t y, uint32_t z>
union Format3
{
- struct
- {
- uint32_t r : x;
- uint32_t g : y;
- uint32_t b : z;
- };
- uint32_t a; ///@note This is here to provide full template needed in Formats.
-};
+ typedef typename FormatIntType<x + y + z>::TYPE TYPE;
-//////////////////////////////////////////////////////////////////////////
-/// Format3 - Bitfield for 3 component formats - 16 bit specialization
-//////////////////////////////////////////////////////////////////////////
-template<>
-union Format3<5,6,5>
-{
struct
{
- uint16_t r : 5;
- uint16_t g : 6;
- uint16_t b : 5;
+ TYPE r : x;
+ TYPE g : y;
+ TYPE b : z;
};
- uint16_t a; ///@note This is here to provide full template needed in Formats.
+ TYPE a; ///@note This is here to provide full template needed in Formats.
};
//////////////////////////////////////////////////////////////////////////
template<uint32_t x, uint32_t y, uint32_t z, uint32_t w>
struct Format4
{
- uint32_t r : x;
- uint32_t g : y;
- uint32_t b : z;
- uint32_t a : w;
-};
-
-//////////////////////////////////////////////////////////////////////////
-/// Format4 - Bitfield for 4 component formats - 16 bit specialization
-//////////////////////////////////////////////////////////////////////////
-template<>
-struct Format4<5,5,5,1>
-{
- uint16_t r : 5;
- uint16_t g : 5;
- uint16_t b : 5;
- uint16_t a : 1;
-};
+ typedef typename FormatIntType<x + y + z + w>::TYPE TYPE;
-//////////////////////////////////////////////////////////////////////////
-/// Format4 - Bitfield for 4 component formats - 16 bit specialization
-//////////////////////////////////////////////////////////////////////////
-template<>
-struct Format4<4,4,4,4>
-{
- uint16_t r : 4;
- uint16_t g : 4;
- uint16_t b : 4;
- uint16_t a : 4;
+ TYPE r : x;
+ TYPE g : y;
+ TYPE b : z;
+ TYPE a : w;
};
//////////////////////////////////////////////////////////////////////////
return TypeTraits<X, NumBitsX>::loadSOA_16(pSrc);
}
- INLINE static void SIMDAPI storeSOA(uint32_t comp, uint8_t *pDst, simd16scalar src)
+ INLINE static void SIMDCALL storeSOA(uint32_t comp, uint8_t *pDst, simd16scalar src)
{
switch (comp)
{
projects / mesa.git / blobdiff