[Mesa-dev] [PATCH 11/11] swr: [rasterizer core] introduce simd16intrin.h
Tim Rowley
timothy.o.rowley at intel.com
Mon Jul 18 17:10:20 UTC 2016
Refactoring to leave existing simd_* intrinsics in "simdintrin.h" unchanged,
adding corresponding simd16_* intrinsics in "simd16intrin.h" on the side,
with emulation, that we can use piecemeal, rather than the all-or-nothing
approach to bring up avx512.
---
src/gallium/drivers/swr/Makefile.sources | 1 +
.../drivers/swr/rasterizer/common/simd16intrin.h | 717 +++++++++++++++++++++
.../drivers/swr/rasterizer/common/simdintrin.h | 16 +-
src/gallium/drivers/swr/rasterizer/core/knobs.h | 23 +-
4 files changed, 751 insertions(+), 6 deletions(-)
create mode 100644 src/gallium/drivers/swr/rasterizer/common/simd16intrin.h
diff --git a/src/gallium/drivers/swr/Makefile.sources b/src/gallium/drivers/swr/Makefile.sources
index 8d97a75..99b3900 100644
--- a/src/gallium/drivers/swr/Makefile.sources
+++ b/src/gallium/drivers/swr/Makefile.sources
@@ -54,6 +54,7 @@ COMMON_CXX_SOURCES := \
rasterizer/common/rdtsc_buckets.h \
rasterizer/common/rdtsc_buckets_shared.h \
rasterizer/common/rdtsc_buckets_shared.h \
+ rasterizer/common/simd16intrin.h \
rasterizer/common/simdintrin.h \
rasterizer/common/swr_assert.cpp \
rasterizer/common/swr_assert.h
diff --git a/src/gallium/drivers/swr/rasterizer/common/simd16intrin.h b/src/gallium/drivers/swr/rasterizer/common/simd16intrin.h
new file mode 100644
index 0000000..3597dbd2
--- /dev/null
+++ b/src/gallium/drivers/swr/rasterizer/common/simd16intrin.h
@@ -0,0 +1,717 @@
+/****************************************************************************
+* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved.
+*
+* Permission is hereby granted, free of charge, to any person obtaining a
+* copy of this software and associated documentation files (the "Software"),
+* to deal in the Software without restriction, including without limitation
+* the rights to use, copy, modify, merge, publish, distribute, sublicense,
+* and/or sell copies of the Software, and to permit persons to whom the
+* Software is furnished to do so, subject to the following conditions:
+*
+* The above copyright notice and this permission notice (including the next
+* paragraph) shall be included in all copies or substantial portions of the
+* Software.
+*
+* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+* IN THE SOFTWARE.
+****************************************************************************/
+
+#ifndef __SWR_SIMD16INTRIN_H__
+#define __SWR_SIMD16INTRIN_H__
+
+#if ENABLE_AVX512_SIMD16
+
+#if KNOB_SIMD16_WIDTH == 16
+
+#if ENABLE_AVX512_EMULATION
+struct simd16scalar
+{
+ __m256 lo;
+ __m256 hi;
+};
+struct simd16scalard
+{
+ __m256d lo;
+ __m256d hi;
+};
+struct simd16scalari
+{
+ __m256i lo;
+ __m256i hi;
+};
+typedef uint16_t simd16mask;
+
+#define _simd16_masklo(mask) ((mask) & 0xFF)
+#define _simd16_maskhi(mask) (((mask) >> 8))
+#define _simd16_setmask(hi, lo) (((hi) << 8) | (lo))
+
+#else
+typedef __m512 simd16scalar;
+typedef __m512d simd16scalard;
+typedef __m512i simd16scalari;
+typedef __mask16 simd16mask;
+#endif//ENABLE_AVX512_EMULATION
+#else
+#error Unsupported vector width
+#endif//KNOB_SIMD16_WIDTH == 16
+
+OSALIGN(union, KNOB_SIMD16_BYTES) simd16vector
+{
+ simd16scalar v[4];
+ struct
+ {
+ simd16scalar x, y, z, w;
+ };
+
+ simd16scalar& operator[] (const int i) { return v[i]; }
+ const simd16scalar& operator[] (const int i) const { return v[i]; }
+};
+
+#if ENABLE_AVX512_EMULATION
+
+#define SIMD16_EMU_AVX512_0(type, func, intrin) \
+INLINE type func()\
+{\
+ type result;\
+\
+ result.lo = intrin();\
+ result.hi = intrin();\
+\
+ return result;\
+}
+
+#define SIMD16_EMU_AVX512_1(type, func, intrin) \
+INLINE type func(type a)\
+{\
+ type result;\
+\
+ result.lo = intrin(a.lo);\
+ result.hi = intrin(a.hi);\
+\
+ return result;\
+}
+
+#define SIMD16_EMU_AVX512_2(type, func, intrin) \
+INLINE type func(type a, type b)\
+{\
+ type result;\
+\
+ result.lo = intrin(a.lo, b.lo);\
+ result.hi = intrin(a.hi, b.hi);\
+\
+ return result;\
+}
+
+#define SIMD16_EMU_AVX512_3(type, func, intrin) \
+INLINE type func(type a, type b, type c)\
+{\
+ type result;\
+\
+ result.lo = intrin(a.lo, b.lo, c.lo);\
+ result.hi = intrin(a.hi, b.hi, c.hi);\
+\
+ return result;\
+}
+
+SIMD16_EMU_AVX512_0(simd16scalar, _simd16_setzero_ps, _mm256_setzero_ps)
+SIMD16_EMU_AVX512_0(simd16scalari, _simd16_setzero_si, _mm256_setzero_si256)
+
+INLINE simd16scalar _simd16_set1_ps(float a)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_set1_ps(a);
+ result.hi = _mm256_set1_ps(a);
+
+ return result;
+}
+
+INLINE simd16scalari _simd16_set1_epi8(char a)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_set1_epi8(a);
+ result.hi = _mm256_set1_epi8(a);
+
+ return result;
+}
+
+INLINE simd16scalari _simd16_set1_epi32(int a)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_set1_epi32(a);
+ result.hi = _mm256_set1_epi32(a);
+
+ return result;
+}
+
+INLINE simd16scalari _simd16_set_epi32(int e7, int e6, int e5, int e4, int e3, int e2, int e1, int e0)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_set_epi32(e7, e6, e5, e4, e3, e2, e1, e0);
+ result.hi = _mm256_set_epi32(e7, e6, e5, e4, e3, e2, e1, e0);
+
+ return result;
+}
+
+INLINE simd16scalari _simd16_set_epi32(int e15, int e14, int e13, int e12, int e11, int e10, int e9, int e8, int e7, int e6, int e5, int e4, int e3, int e2, int e1, int e0)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_set_epi32(e7, e6, e5, e4, e3, e2, e1, e0);
+ result.hi = _mm256_set_epi32(e15, e14, e13, e12, e11, e10, e9, e8);
+
+ return result;
+}
+
+INLINE simd16scalar _simd16_load_ps(float const *m)
+{
+ simd16scalar result;
+
+ float const *n = reinterpret_cast<float const *>(reinterpret_cast<uint8_t const *>(m) + sizeof(result.lo));
+
+ result.lo = _mm256_load_ps(m);
+ result.hi = _mm256_load_ps(n);
+
+ return result;
+}
+
+INLINE simd16scalar _simd16_loadu_ps(float const *m)
+{
+ simd16scalar result;
+
+ float const *n = reinterpret_cast<float const *>(reinterpret_cast<uint8_t const *>(m) + sizeof(result.lo));
+
+ result.lo = _mm256_loadu_ps(m);
+ result.hi = _mm256_loadu_ps(n);
+
+ return result;
+}
+
+INLINE simd16scalar _simd16_load1_ps(float const *m)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_broadcast_ss(m);
+ result.hi = _mm256_broadcast_ss(m);
+
+ return result;
+}
+
+INLINE simd16scalari _simd16_load_si(simd16scalari const *m)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_load_si256(&m[0].lo);
+ result.hi = _mm256_load_si256(&m[0].hi);
+
+ return result;
+}
+
+INLINE simd16scalari _simd16_loadu_si(simd16scalari const *m)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_loadu_si256(&m[0].lo);
+ result.hi = _mm256_loadu_si256(&m[0].hi);
+
+ return result;
+}
+
+INLINE simd16scalar _simd16_broadcast_ss(float const *m)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_broadcast_ss(m);
+ result.hi = _mm256_broadcast_ss(m);
+
+ return result;
+}
+
+INLINE simd16scalar _simd16_broadcast_ps(__m128 const *m)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_broadcast_ps(m);
+ result.hi = _mm256_broadcast_ps(m);
+
+ return result;
+}
+
+INLINE void _simd16_store_ps(float *m, simd16scalar a)
+{
+ float *n = reinterpret_cast<float *>(reinterpret_cast<uint8_t *>(m) + sizeof(a.lo));
+
+ _mm256_store_ps(m, a.lo);
+ _mm256_store_ps(n, a.hi);
+}
+
+INLINE void _simd16_maskstore_ps(float *m, simd16scalari mask, simd16scalar a)
+{
+ float *n = reinterpret_cast<float *>(reinterpret_cast<uint8_t *>(m) + sizeof(a.lo));
+
+ _mm256_maskstore_ps(m, mask.lo, a.lo);
+ _mm256_maskstore_ps(n, mask.hi, a.hi);
+}
+
+INLINE void _simd16_store_si(simd16scalari *m, simd16scalari a)
+{
+ _mm256_store_si256(&m[0].lo, a.lo);
+ _mm256_store_si256(&m[0].hi, a.hi);
+}
+
+template <simd16mask mask>
+INLINE simd16scalar _simd16_blend_ps_temp(simd16scalar a, simd16scalar b)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_blend_ps(a.lo, b.lo, _simd16_masklo(mask));
+ result.hi = _mm256_blend_ps(a.hi, b.hi, _simd16_maskhi(mask));
+
+ return result;
+}
+
+#define _simd16_blend_ps(a, b, mask) _simd16_blend_ps_temp<mask>(a, b)
+
+SIMD16_EMU_AVX512_3(simd16scalar, _simd16_blendv_ps, _mm256_blendv_ps)
+
+INLINE simd16scalari _simd16_blendv_epi32(simd16scalari a, simd16scalari b, const simd16scalar mask)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_castps_si256(_mm256_blendv_ps(_mm256_castsi256_ps(a.lo), _mm256_castsi256_ps(b.lo), mask.lo));
+ result.hi = _mm256_castps_si256(_mm256_blendv_ps(_mm256_castsi256_ps(a.hi), _mm256_castsi256_ps(b.hi), mask.hi));
+
+ return result;
+}
+
+INLINE simd16scalari _simd16_blendv_epi32(simd16scalari a, simd16scalari b, const simd16scalari mask)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_castps_si256(_mm256_blendv_ps(_mm256_castsi256_ps(a.lo), _mm256_castsi256_ps(b.lo), _mm256_castsi256_ps(mask.lo)));
+ result.hi = _mm256_castps_si256(_mm256_blendv_ps(_mm256_castsi256_ps(a.hi), _mm256_castsi256_ps(b.hi), _mm256_castsi256_ps(mask.hi)));
+
+ return result;
+}
+
+SIMD16_EMU_AVX512_2(simd16scalar, _simd16_mul_ps, _mm256_mul_ps)
+SIMD16_EMU_AVX512_2(simd16scalar, _simd16_add_ps, _mm256_add_ps)
+SIMD16_EMU_AVX512_2(simd16scalar, _simd16_sub_ps, _mm256_sub_ps)
+SIMD16_EMU_AVX512_1(simd16scalar, _simd16_rsqrt_ps, _mm256_rsqrt_ps)
+SIMD16_EMU_AVX512_2(simd16scalar, _simd16_min_ps, _mm256_min_ps)
+SIMD16_EMU_AVX512_2(simd16scalar, _simd16_max_ps, _mm256_max_ps)
+
+INLINE simd16mask _simd16_movemask_ps(simd16scalar a)
+{
+ simd16mask mask;
+
+ reinterpret_cast<uint8_t *>(&mask)[0] = _mm256_movemask_ps(a.lo);
+ reinterpret_cast<uint8_t *>(&mask)[1] = _mm256_movemask_ps(a.hi);
+
+ return mask;
+}
+
+INLINE simd16mask _simd16_movemask_pd(simd16scalard a)
+{
+ simd16mask mask;
+
+ reinterpret_cast<uint8_t *>(&mask)[0] = _mm256_movemask_pd(a.lo);
+ reinterpret_cast<uint8_t *>(&mask)[1] = _mm256_movemask_pd(a.hi);
+
+ return mask;
+}
+
+INLINE simd16mask _simd16_movemask_epi8(simd16scalari a)
+{
+ simd16mask mask;
+
+ reinterpret_cast<uint8_t *>(&mask)[0] = _mm256_movemask_epi8(a.lo);
+ reinterpret_cast<uint8_t *>(&mask)[1] = _mm256_movemask_epi8(a.hi);
+
+ return mask;
+}
+
+INLINE simd16scalari _simd16_cvtps_epi32(simd16scalar a)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_cvtps_epi32(a.lo);
+ result.hi = _mm256_cvtps_epi32(a.hi);
+
+ return result;
+}
+
+INLINE simd16scalari _simd16_cvttps_epi32(simd16scalar a)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_cvttps_epi32(a.lo);
+ result.hi = _mm256_cvttps_epi32(a.hi);
+
+ return result;
+}
+
+INLINE simd16scalar _simd16_cvtepi32_ps(simd16scalari a)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_cvtepi32_ps(a.lo);
+ result.hi = _mm256_cvtepi32_ps(a.hi);
+
+ return result;
+}
+
+template <int comp>
+INLINE simd16scalar _simd16_cmp_ps(simd16scalar a, simd16scalar b)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_cmp_ps(a.lo, b.lo, comp);
+ result.hi = _mm256_cmp_ps(a.hi, b.hi, comp);
+
+ return result;
+}
+
+#define _simd16_cmplt_ps(a, b) _simd16_cmp_ps<_CMP_LT_OQ>(a, b)
+#define _simd16_cmpgt_ps(a, b) _simd16_cmp_ps<_CMP_GT_OQ>(a, b)
+#define _simd16_cmpneq_ps(a, b) _simd16_cmp_ps<_CMP_NEQ_OQ>(a, b)
+#define _simd16_cmpeq_ps(a, b) _simd16_cmp_ps<_CMP_EQ_OQ>(a, b)
+#define _simd16_cmpge_ps(a, b) _simd16_cmp_ps<_CMP_GE_OQ>(a, b)
+#define _simd16_cmple_ps(a, b) _simd16_cmp_ps<_CMP_LE_OQ>(a, b)
+
+SIMD16_EMU_AVX512_2(simd16scalar, _simd16_and_ps, _mm256_and_ps)
+SIMD16_EMU_AVX512_2(simd16scalar, _simd16_or_ps, _mm256_or_ps)
+SIMD16_EMU_AVX512_1(simd16scalar, _simd16_rcp_ps, _mm256_rcp_ps)
+SIMD16_EMU_AVX512_2(simd16scalar, _simd16_div_ps, _mm256_div_ps)
+
+INLINE simd16scalar _simd16_castsi_ps(simd16scalari a)
+{
+ return *reinterpret_cast<simd16scalar *>(&a);
+}
+
+INLINE simd16scalari _simd16_castps_si(simd16scalar a)
+{
+ return *reinterpret_cast<simd16scalari *>(&a);
+}
+
+INLINE simd16scalard _simd16_castsi_pd(simd16scalari a)
+{
+ return *reinterpret_cast<simd16scalard *>(&a);
+}
+
+INLINE simd16scalari _simd16_castpd_si(simd16scalard a)
+{
+ return *reinterpret_cast<simd16scalari *>(&a);
+}
+
+INLINE simd16scalar _simd16_castpd_ps(simd16scalard a)
+{
+ return *reinterpret_cast<simd16scalar *>(&a);
+}
+
+INLINE simd16scalard _simd16_castps_pd(simd16scalar a)
+{
+ return *reinterpret_cast<simd16scalard *>(&a);
+}
+
+SIMD16_EMU_AVX512_2(simd16scalar, _simd16_andnot_ps, _mm256_andnot_ps)
+
+template <int mode>
+INLINE simd16scalar _simd16_round_ps_temp(simd16scalar a)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_round_ps(a.lo, mode);
+ result.hi = _mm256_round_ps(a.hi, mode);
+
+ return result;
+}
+
+#define _simd16_cmp_ps(a, mode) _simd16_round_ps_temp<mode>(a)
+
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_mul_epi32, _mm256_mul_epi32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_mullo_epi32, _mm256_mullo_epi32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_sub_epi32, _mm256_sub_epi32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_sub_epi64, _mm256_sub_epi64)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_min_epi32, _mm256_min_epi32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_max_epi32, _mm256_max_epi32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_min_epu32, _mm256_min_epu32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_max_epu32, _mm256_max_epu32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_add_epi32, _mm256_add_epi32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_and_si, _mm256_and_si256)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_andnot_si, _mm256_andnot_si256)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_or_si, _mm256_or_si256)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_xor_si, _mm256_xor_si256)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_cmpeq_epi32, _mm256_cmpeq_epi32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_cmpgt_epi32, _mm256_cmpgt_epi32)
+
+INLINE int _simd16_testz_ps(simd16scalar a, simd16scalar b)
+{
+ int lo = _mm256_testz_ps(a.lo, b.lo);
+ int hi = _mm256_testz_ps(a.hi, b.hi);
+
+ return lo & hi;
+}
+
+#define _simd16_cmplt_epi32(a, b) _simd16_cmpgt_epi32(b, a)
+
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_unpacklo_epi32, _mm256_unpacklo_epi32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_unpackhi_epi32, _mm256_unpackhi_epi32)
+
+template <int imm8>
+INLINE simd16scalari _simd16_slli_epi32_temp(simd16scalari a)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_slli_epi32(a.lo, imm8);
+ result.hi = _mm256_slli_epi32(a.hi, imm8);
+
+ return result;
+}
+
+#define _simd16_slli_epi32(a, imm8) _simd16_slli_epi32_temp<imm8>(a)
+
+template <int imm8>
+INLINE simd16scalari _simd16_srai_epi32_temp(simd16scalari a)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_srai_epi32(a.lo, imm8);
+ result.hi = _mm256_srai_epi32(a.hi, imm8);
+
+ return result;
+}
+
+#define _simd16_srai_epi32(a, imm8) _simd16_srai_epi32_temp<imm8>(a)
+
+template <int imm8>
+INLINE simd16scalari _simd16_srli_epi32_temp(simd16scalari a)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_srli_epi32(a.lo, imm8);
+ result.hi = _mm256_srli_epi32(a.hi, imm8);
+
+ return result;
+}
+
+#define _simd16_srli_epi32(a, imm8) _simd16_srli_epi32_temp<imm8>(a)
+
+SIMD16_EMU_AVX512_3(simd16scalar, _simd16_fmadd_ps, _mm256_fmadd_ps)
+SIMD16_EMU_AVX512_3(simd16scalar, _simd16_fmsub_ps, _mm256_fmsub_ps)
+
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_shuffle_epi8, _mm256_shuffle_epi8)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_adds_epu8, _mm256_adds_epu8)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_subs_epu8, _mm256_subs_epu8)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_add_epi8, _mm256_add_epi8)
+
+template <int imm8>
+INLINE simd16scalar _simd16_i32gather_ps_temp(float const *m, simd16scalari a)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_i32gather_ps(m, a.lo, imm8);
+ result.hi = _mm256_i32gather_ps(m, a.hi, imm8);
+
+ return result;
+}
+
+#define _simd16_i32gather_ps(m, a, imm8) _simd16_i32gather_ps_temp<imm8>(m, a)
+
+SIMD16_EMU_AVX512_1(simd16scalari, _simd16_abs_epi32, _mm256_abs_epi32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_cmpeq_epi64, _mm256_cmpeq_epi64)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_cmpgt_epi64, _mm256_cmpgt_epi64)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_cmpeq_epi16, _mm256_cmpeq_epi16)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_cmpgt_epi16, _mm256_cmpgt_epi16)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_cmpeq_epi8, _mm256_cmpeq_epi8)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_cmpgt_epi8, _mm256_cmpgt_epi8)
+
+INLINE simd16scalar _simd16_permute_ps(simd16scalar a, simd16scalari b)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_permutevar8x32_ps(a.lo, b.lo);
+ result.hi = _mm256_permutevar8x32_ps(a.hi, b.hi);
+
+ return result;
+}
+
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_permute_epi32, _mm256_permutevar8x32_epi32)
+
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_srlv_epi32, _mm256_srlv_epi32)
+SIMD16_EMU_AVX512_2(simd16scalari, _simd16_sllv_epi32, _mm256_sllv_epi32)
+
+template <int imm8>
+INLINE simd16scalar _simd16_shuffle_ps_temp(simd16scalar a, simd16scalar b)
+{
+ simd16scalar result;
+
+ result.lo = _mm256_shuffle_ps(a.lo, b.lo, imm8);
+ result.hi = _mm256_shuffle_ps(a.hi, b.hi, imm8);
+
+ return result;
+}
+
+#define _simd16_shuffle_ps(a, b, imm8) _simd16_shuffle_ps_temp<imm8>(a, b)
+
+template <int imm8>
+INLINE simd16scalari _simd16_permute_128_temp(simd16scalari a, simd16scalari b)
+{
+ simd16scalari result;
+
+ result.lo = _mm256_permute2x128_si256(a.lo, b.lo, imm8);
+ result.hi = _mm256_permute2x128_si256(a.hi, b.hi, imm8);
+
+ return result;
+}
+
+#define _simd16_permute_128(a, b, imm8) _simd16_permute_128_temp<imm8>(a, b)
+
+// convert bitmask to vector mask
+INLINE simd16scalar vMask16(int32_t mask)
+{
+ simd16scalari temp = _simd16_set1_epi32(mask);
+
+ simd16scalari bits = _simd16_set_epi32(0x8000, 0x4000, 0x2000, 0x1000, 0x0800, 0x0400, 0x0200, 0x0100, 0x0080, 0x0040, 0x0020, 0x0010, 0x0008, 0x0004, 0x0002, 0x0001);
+
+ simd16scalari result = _simd16_cmplt_epi32(_simd16_setzero_si(), _simd16_and_si(temp, bits));
+
+ return _simd16_castsi_ps(result);
+}
+
+#else
+
+INLINE __m512 _m512_broadcast_ss(void const *m)
+{
+ return _mm512_extload_ps(m, _MM_UPCONV_PS_NONE, _MM_BROADCAST_1X16, 0);
+}
+
+INLINE __m512 _m512_broadcast_ps(void const *m)
+{
+ return _mm512_extload_ps(m, _MM_UPCONV_PS_NONE, _MM_BROADCAST_4X16, 0);
+}
+
+INLINE __m512 _m512_blend_ps(__m512 a, __m512 b, const int mask)
+{
+ const __mask16 mask16 = _mm512_int2mask(mask);
+
+ return _mm512_mask_blend_ps(mask16, a, b);
+}
+
+INLINE __m512 _m512_blendv_ps(__m512 a, __m512 b, __m512 mask)
+{
+ const __mask16 mask16 = _mm512_cmpeq_ps_mask(mask, _mm512_setzero_ps());
+
+ return _mm512_mask_blend_ps(mask16, a, b);
+}
+
+INLINE int _m512_movemask_ps(__m512 a)
+{
+ __m512 mask = _mm512_set1_epi32(0x80000000);
+
+ __m512 temp = _mm512_and_epi32(a, mask);
+
+ const __mask16 mask16 = _mm512_cmpeq_epu32_mask(temp, mask);
+
+ return _mm512mask2int(mask16);
+}
+
+INLINE int _m512_movemask_pd(__m512 a)
+{
+ __m512 mask = _mm512_set1_epi64(0x8000000000000000);
+
+ __m512 temp = _mm512_and_epi64(a, mask);
+
+ const __mask16 mask16 = _mm512_cmpeq_epu64_mask(temp, mask);
+
+ return _mm512mask2int(mask16);
+}
+
+INLINE __m512 _m512_cmp_ps(__m512 a, __m512 b, __m512 comp)
+{
+ const __mask16 mask16 = _mm512_cmpeq_ps_mask(a, b, comp);
+
+ return _mm512_mask_blend_epi32(mask16, _mm512_setzero_epi32(), _mm512_set1_epi32(0xFFFFFFFF));
+}
+
+INLINE __m512 _mm512_cmplt_epi32(__m512 a, __m512 b)
+{
+ const __mask16 mask16 = _mm512_cmplt_epi32_mask(a, b);
+
+ return _mm512_mask_blend_epi32(mask16, _mm512_setzero_epi32(), _mm512_set1_epi32(0xFFFFFFFF));
+}
+
+INLINE __m512 _mm512_cmpgt_epi32(__m512 a, __m512 b)
+{
+ const __mask16 mask16 = _mm512_cmpgt_epi32_mask(a, b);
+
+ return _mm512_mask_blend_epi32(mask16, _mm512_setzero_epi32(), _mm512_set1_epi32(0xFFFFFFFF));
+}
+
+#define _simd16_load_ps _mm512_load_ps
+#define _simd16_load1_ps _mm256_broadcast_ss
+#define _simd16_loadu_ps _mm512_loadu_ps
+#define _simd16_setzero_ps _mm512_setzero_ps
+#define _simd16_set1_ps _mm512_set1_ps
+#define _simd16_blend_ps _mm512_blend_ps
+#define _simd16_blendv_ps _mm512_blendv_ps
+#define _simd16_store_ps _mm512_store_ps
+#define _simd16_mul_ps _mm512_mul_ps
+#define _simd16_add_ps _mm512_add_ps
+#define _simd16_sub_ps _mm512_sub_ps
+#define _simd16_rsqrt_ps _mm512_rsqrt28_ps
+#define _simd16_min_ps _mm512_min_ps
+#define _simd16_max_ps _mm512_max_ps
+#define _simd16_movemask_ps _mm512_movemask_ps
+#define _simd16_cvtps_epi32 _mm512_cvtps_epi32
+#define _simd16_cvttps_epi32 _mm512_cvttps_epi32
+#define _simd16_cvtepi32_ps _mm512_cvtepi32_ps
+#define _simd16_cmplt_ps(a, b) _mm512_cmp_ps(a, b, _CMP_LT_OQ)
+#define _simd16_cmpgt_ps(a, b) _mm512_cmp_ps(a, b, _CMP_GT_OQ)
+#define _simd16_cmpneq_ps(a, b) _mm512_cmp_ps(a, b, _CMP_NEQ_OQ)
+#define _simd16_cmpeq_ps(a, b) _mm512_cmp_ps(a, b, _CMP_EQ_OQ)
+#define _simd16_cmpge_ps(a, b) _mm512_cmp_ps(a, b, _CMP_GE_OQ)
+#define _simd16_cmple_ps(a, b) _mm512_cmp_ps(a, b, _CMP_LE_OQ)
+#define _simd16_cmp_ps(a, b, comp) _mm512_cmp_ps(a, b, comp)
+#define _simd16_and_ps _mm512_and_ps
+#define _simd16_or_ps _mm512_or_ps
+#define _simd16_rcp_ps _mm512_rcp28_ps
+#define _simd16_div_ps _mm512_div_ps
+#define _simd16_castsi_ps _mm512_castsi512_ps
+#define _simd16_andnot_ps _mm512_andnot_ps
+#define _simd16_round_ps _mm512_round_ps
+#define _simd16_castpd_ps _mm512_castpd_ps
+#define _simd16_broadcast_ps _m512_broadcast_ps
+#define _simd16_movemask_pd _mm512_movemask_pd
+#define _simd16_castsi_pd _mm512_castsi512_pd
+
+#define _simd16_mul_epi32 _mm512_mul_epi32
+#define _simd16_mullo_epi32 _mm512_mullo_epi32
+#define _simd16_sub_epi32 _mm512_sub_epi32
+#define _simd16_sub_epi64 _mm512_sub_epi64
+#define _simd16_min_epi32 _mm512_min_epi32
+#define _simd16_max_epi32 _mm512_max_epi32
+#define _simd16_min_epu32 _mm512_min_epu32
+#define _simd16_max_epu32 _mm512_max_epu32
+#define _simd16_add_epi32 _mm512_add_epi32
+#define _simd16_and_si _mm512_and_si512
+#define _simd16_andnot_si _mm512_andnot_si512
+#define _simd16_cmpeq_epi32 _mm512_cmpeq_epi32
+#define _simd16_cmplt_epi32(a,b) _mm256_cmpgt_epi32(b,a)
+#define _simd16_cmpgt_epi32(a,b) _mm256_cmpgt_epi32(a,b)
+#define _simd16_or_si _mm512_or_si512
+#define _simd16_castps_si _mm512_castps_si512
+
+#endif//ENABLE_AVX512_EMULATION
+
+#endif//ENABLE_AVX512_SIMD16
+
+#endif//__SWR_SIMD16INTRIN_H_
diff --git a/src/gallium/drivers/swr/rasterizer/common/simdintrin.h b/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
index 6dcd7c3..4a91e95 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
+++ b/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
@@ -43,19 +43,20 @@ struct simdscalar
__m256 lo;
__m256 hi;
};
-struct simdscalari
-{
- __m256i lo;
- __m256i hi;
-};
struct simdscalard
{
__m256d lo;
__m256d hi;
};
+struct simdscalari
+{
+ __m256i lo;
+ __m256i hi;
+};
typedef uint16_t simdmask;
#else
typedef __m512 simdscalar;
+typedef __m512d simdscalard;
typedef __m512i simdscalari;
typedef __mask16 simdmask;
#endif
@@ -487,6 +488,7 @@ simdscalari _simd_permute_epi32(simdscalari a, simdscalari index)
}
#endif
+#define _simd_permute_128 _mm256_permute2f128_si256
#define _simd_shuffleps_epi32(vA, vB, imm) _mm256_castps_si256(_mm256_shuffle_ps(_mm256_castsi256_ps(vA), _mm256_castsi256_ps(vB), imm))
#define _simd_shuffle_ps _mm256_shuffle_ps
#define _simd_set1_epi32 _mm256_set1_epi32
@@ -1700,4 +1702,8 @@ UINT pext_u32(UINT a, UINT mask)
#endif
}
+#if ENABLE_AVX512_SIMD16
+#include "simd16intrin.h"
+#endif//ENABLE_AVX512_SIMD16
+
#endif//__SWR_SIMDINTRIN_H__
diff --git a/src/gallium/drivers/swr/rasterizer/core/knobs.h b/src/gallium/drivers/swr/rasterizer/core/knobs.h
index 738e47d..61eb3b4 100644
--- a/src/gallium/drivers/swr/rasterizer/core/knobs.h
+++ b/src/gallium/drivers/swr/rasterizer/core/knobs.h
@@ -59,7 +59,8 @@
#define KNOB_SIMD_WIDTH 8
#define KNOB_SIMD_BYTES 32
#elif (KNOB_ARCH == KNOB_ARCH_AVX512)
-#if ENABLE_AVX512_SIMD16
+#if 0
+// not ready to enable this globally, enabled on the side (below)
#define KNOB_ARCH_ISA AVX512F
#define KNOB_ARCH_STR "AVX512"
#define KNOB_SIMD_WIDTH 16
@@ -74,6 +75,11 @@
#error "Unknown architecture"
#endif
+#if ENABLE_AVX512_SIMD16
+#define KNOB_SIMD16_WIDTH 16
+#define KNOB_SIMD16_BYTES 64
+#endif
+
#define MAX_KNOB_ARCH_STR_LEN sizeof("AVX512_PLUS_PADDING")
///////////////////////////////////////////////////////////////////////////////
@@ -133,6 +139,12 @@
#error "incompatible width/tile dimensions"
#endif
+#if ENABLE_AVX512_SIMD16
+#if KNOB_SIMD16_WIDTH==16 && KNOB_TILE_X_DIM < 4
+#error "incompatible width/tile dimensions"
+#endif
+#endif
+
#if KNOB_SIMD_WIDTH == 8
#define SIMD_TILE_X_DIM 4
#define SIMD_TILE_Y_DIM 2
@@ -143,6 +155,15 @@
#error "Invalid simd width"
#endif
+#if ENABLE_AVX512_SIMD16
+#if KNOB_SIMD16_WIDTH == 16
+#define SIMD16_TILE_X_DIM 4
+#define SIMD16_TILE_Y_DIM 4
+#else
+#error "Invalid simd width"
+#endif
+#endif
+
///////////////////////////////////////////////////////////////////////////////
// Optimization knobs
///////////////////////////////////////////////////////////////////////////////
--
1.9.1
More information about the mesa-dev
mailing list