|
| 1 | +:data-uri: |
| 2 | +:icons: font |
| 3 | +include::../config/attribs.txt[] |
| 4 | +:source-highlighter: coderay |
| 5 | + |
| 6 | += cl_img_bitwise_ops |
| 7 | + |
| 8 | +== Name Strings |
| 9 | + |
| 10 | +`cl_img_bitwise_ops` |
| 11 | + |
| 12 | +== Contact |
| 13 | + |
| 14 | +Imagination Technologies Developer Forum: + |
| 15 | +https://forums.imgtec.com/ |
| 16 | + |
| 17 | +Tomasz Platek, Imagination Technologies (Tomasz.Platek 'at' imgtec.com) |
| 18 | + |
| 19 | +== Contributors |
| 20 | + |
| 21 | +CY Cheng, Imagination Technologies. + |
| 22 | +Tomasz Platek, Imagination Technologies. |
| 23 | + |
| 24 | +== Notice |
| 25 | + |
| 26 | +Copyright (c) 2024 Imagination Technologies Ltd. All Rights Reserved. |
| 27 | + |
| 28 | +== Status |
| 29 | + |
| 30 | +Final Draft |
| 31 | + |
| 32 | +== Version |
| 33 | + |
| 34 | +Built On: {docdate} + |
| 35 | +Version: 1.0.0 |
| 36 | + |
| 37 | +== Dependencies |
| 38 | + |
| 39 | +This extension is written against the OpenCL C Specification Version V3.0.16. |
| 40 | + |
| 41 | +== Overview |
| 42 | + |
| 43 | +This extension adds built-in functions that expose the bitwise operations of Imagination GPU IP that are not accessible by standard OpenCL C functions. |
| 44 | + |
| 45 | +== New OpenCL C Feature Names |
| 46 | + |
| 47 | +[source,c] |
| 48 | +---- |
| 49 | +__opencl_img_bit_interleave |
| 50 | +---- |
| 51 | + |
| 52 | +== New OpenCL C Functions |
| 53 | + |
| 54 | +Performs the bit interleave operation: |
| 55 | + |
| 56 | +[source,c] |
| 57 | +---- |
| 58 | +gentype img_bit_interleave(gentype a, gentype b); |
| 59 | +---- |
| 60 | + |
| 61 | +== Modifications to the OpenCL C Specification |
| 62 | + |
| 63 | +(Add to Table 16 - Built-in Scalar and Vector Argument Common Functions in Section 6.15.4 - Common Functions) :: |
| 64 | ++ |
| 65 | +-- |
| 66 | +[cols="1,2",options="header"] |
| 67 | +|==== |
| 68 | +| Function | Description |
| 69 | +| gentype *img_bit_interleave*(gentype a, gentype b) |
| 70 | + a| `img_bit_interleave` interleaves the first `n` bits from two sources where `n` is half of the size of gentype in bits. |
| 71 | + |
| 72 | +For `a` and `b`, where a0 and b0 are the least significant bits: |
| 73 | +[source] |
| 74 | +---- |
| 75 | +a = a(N-1)\|a(N-2)\|a(N-3)\|...\|a3\|a2\|a1\|a0 |
| 76 | +b = b(N-1)\|b(N-2)\|b(N-3)\|...\|b3\|b2\|b1\|b0 |
| 77 | +---- |
| 78 | + |
| 79 | +the output is: |
| 80 | +[source] |
| 81 | +---- |
| 82 | +res = b(N/2-1)\|a(N/2-1)\|b(N/2-2)\|a(N/2-2)\|b(N/2-3)\|a(N/2-3)\|...\|b3\|a3\|b2\|a2\|b1\|a1\|b0\|a0 |
| 83 | +---- |
| 84 | +so the sizes of `a`,`b`, and `res` are equal. |
| 85 | + |
| 86 | +Requires that the `__opencl_img_bit_interleave` feature macro is defined. |
| 87 | +|==== |
| 88 | +-- |
| 89 | + |
| 90 | +== Coding Sample |
| 91 | + |
| 92 | +This coding sample shows how to use the *img_bit_interleave* function: |
| 93 | +[source] |
| 94 | +---- |
| 95 | +int4 a = (int4) ( 0x00000000, 0x00000000, 0x0000FFFF, 0xFFFFFFFF); |
| 96 | +int4 b = (int4) ( 0xFFFFFFFF, 0x0000FFFF, 0x00000000, 0x00000000); |
| 97 | +
|
| 98 | +int4 res = img_bit_interleave(a,b); |
| 99 | +
|
| 100 | +printf("res = [ 0x%x 0x%x 0x%x 0x%x]\n", res.s0, res.s1, res.s2, res.s3); |
| 101 | +---- |
| 102 | + |
| 103 | +Executing a work-item of this kernel gives the following result: |
| 104 | +[source] |
| 105 | +---- |
| 106 | +res = [ 0xaaaaaaaa 0xaaaaaaaa 0x55555555 0x55555555] |
| 107 | +---- |
| 108 | + |
| 109 | +== Version History |
| 110 | + |
| 111 | +[cols="5,15,15,70"] |
| 112 | +[grid="rows"] |
| 113 | +[options="header"] |
| 114 | +|==== |
| 115 | +| Version | Date | Author | Changes |
| 116 | +| 1.0.0 | 2024-06-19 | Tomasz Platek | *Initial revision* |
| 117 | +|==== |
| 118 | + |
0 commit comments