c6f5c89669
In addition to float<->half functions to convert one number (#127708), add float_to_half_array and half_to_float_array functions: - On x64, this uses SSE2 4-wide implementation to do the conversion (2x faster half->float, 4x faster float->half compared to scalar), - There's also an AVX2 codepath that uses CPU hardware F16C instructions (8-wide), to be used when/if blender codebase will start to be built for AVX2 (today it is not yet). - On arm64, this uses NEON VCVT instructions to do the conversion. Use these functions in Vulkan buffer/texture conversion code. Time taken to convert float->half texture while viewing EXR file in image space (22M numbers to convert): 39.7ms -> 10.1ms (would be 6.9ms if building for AVX2) Pull Request: https://projects.blender.org/blender/blender/pulls/127838
46 lines
1.1 KiB
C++
46 lines
1.1 KiB
C++
/* SPDX-FileCopyrightText: 2024 Blender Authors
|
|
*
|
|
* SPDX-License-Identifier: GPL-2.0-or-later */
|
|
|
|
#pragma once
|
|
|
|
/** \file
|
|
* \ingroup bli
|
|
*/
|
|
|
|
#include <cstddef>
|
|
#include <cstdint>
|
|
|
|
namespace blender::math {
|
|
|
|
/**
|
|
* Float (FP32) <-> Half (FP16) conversion functions.
|
|
*
|
|
* Behavior matches hardware (x64 F16C, ARM NEON FCVT),
|
|
* including handling of denormals, infinities, NaNs, rounding
|
|
* is to nearest even, etc. When NaNs are produced, the exact
|
|
* bit pattern might not match hardware, but it will still be a NaN.
|
|
*
|
|
* When compiling for ARM NEON (e.g. Apple Silicon),
|
|
* hardware VCVT instructions are used.
|
|
*
|
|
* For anything involving more than a handful of numbers,
|
|
* prefer #float_to_half_array and #half_to_float_array for
|
|
* performance.
|
|
*/
|
|
|
|
/**
|
|
* Converts float (FP32) number to half-precision (FP16).
|
|
*/
|
|
uint16_t float_to_half(float v);
|
|
|
|
/**
|
|
* Converts half-precision (FP16) number to float (FP32).
|
|
*/
|
|
float half_to_float(uint16_t v);
|
|
|
|
void float_to_half_array(const float *src, uint16_t *dst, size_t length);
|
|
void half_to_float_array(const uint16_t *src, float *dst, size_t length);
|
|
|
|
} // namespace blender::math
|