VSTORE_HALFN(3clc)

💅 VSTORE_HALFN (3clc) 💅

okay bestie, we're in the OpenCL Manual today and this page is genuinely one of the most useful little gremlins in the whole vector-data crew, no cap. it's vstore_halfn and she converts your floats to halfs and yeets them into memory. let's go 🦋

---

✨ NAME ✨

vstore_halfn — Convert float vector to half vector and write to a pointer to memory.

here's the whole squad of signatures, every single one, exactly as written, do not @ me:

void vstore_halfn(floatn data, size_t offset, __global half *p);

void vstore_halfn_rte(floatn data, size_t offset, __global half *p);

void vstore_halfn_rtz(floatn data, size_t offset, __global half *p);

void vstore_halfn_rtp(floatn data, size_t offset, __global half *p);

void vstore_halfn_rtn(floatn data, size_t offset, __global half *p);

void vstore_halfn(floatn data, size_t offset, __local half *p);

void vstore_halfn_rte(floatn data, size_t offset, __local half *p);

void vstore_halfn_rtz(floatn data, size_t offset, __local half *p);

void vstore_halfn_rtp(floatn data, size_t offset, __local half *p);

void vstore_halfn_rtn(floatn data, size_t offset, __local half *p);

void vstore_halfn(floatn data, size_t offset, __private half *p);

void vstore_halfn_rte(floatn data, size_t offset, __private half *p);

void vstore_halfn_rtz(floatn data, size_t offset, __private half *p);

void vstore_halfn_rtp(floatn data, size_t offset, __private half *p);

void vstore_halfn_rtn(floatn data, size_t offset, __private half *p);

and if the optional double is supported, you may use doublen instead of floatn. iconic behaviour honestly ✨

---

🌸 DESCRIPTION 🌸

so the tea is: the floatn value in data (or doublen if optional double is supported, we see you) gets converted into a halfn value using the appropriate rounding mode. then that halfn gets written to the address computed as (p + (offset * n)). and — listen, this is the part where people mess up — the address computed as (p + (offset * n)) must be 16-bit aligned. it's giving "read the requirements, bestie." 💀

plain vstore_halfn uses the default rounding mode, and the default rounding mode is round to nearest even. periodt.

---

🎀 NOTES 🎀

Vector Data Load and Store Functions allow you to read and write vector types from a pointer to memory. that's literally their whole job and they ate.

The generic type gentype is used to indicate the built-in data types char, uchar, short, ushort, int, uint, long, ulong, float, or double. so many options, we love a flexible queen 👑

The generic type name gentypen represents n-element vectors of gentype elements. The suffix n is also used in the function names (i.e. vloadn, vstoren, etc.), where n = 2, 3, 4, 8, or 16. those are your only n values, don't get creative.

⚠️ the results of vector data load and store functions are undefined if the address being read from or written to is not correctly aligned. not "probably fine", not "vibes-based", undefined. we do not play with alignment.

The pointer argument p can be a pointer to global(3clc), local(3clc), or private(3clc) memory for store functions. The pointer argument p can be a pointer to __global, __local, constant(3clc) or __private memory for load functions. (loads get one extra address space, good for them 🫶)

vload3 and vload_half3 read x, y, z components from address (p + (offset * 3)) into a 3-component vector. vstore3, and vstore_half3 write x, y, z components from a 3-component vector to address (p + (offset * 3)). threes stay together fr fr.

In addition vloada_half3 reads x, y, z components from address (p + (offset * 4)) into a 3-component vector and vstorea_half3 writes x, y, z components from a 3-component vector to address (p + (offset * 4)). yes the a versions use 4 for three components, that's just the lore, iykyk 🌈

---

📜 SPECIFICATION 📜

OpenCL Specification[1]

---

👀 SEE ALSO 👀

vectorDataLoadandStoreFunctions(3clc) — go read her, she's the mother page, main character energy 💖

---

🧚 AUTHORS 🧚

The Khronos Group

---

💜 COPYRIGHT 💜

Copyright © 2007-2011 The Khronos Group Inc.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and/or associated documentation files (the "Materials"), to deal in the Materials without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Materials, and to permit persons to whom the Materials are furnished to do so, subject to the condition that this copyright notice and permission notice shall be included in all copies or substantial portions of the Materials.

---

🍓 NOTES 🍓

1. OpenCL Specification page 267, section 6.12.7 - Vector Data Load and Store Functions

---

and that's the whole page bestie!! 🎀 brought to you by The Khronos Group, dated 01/14/2021, straight out of the OpenCL Manual (3clc). stay aligned, stay slay. 💅✨🌸