Fix method names of hardware intrinsic APIs #15471
Conversation
| @@ -263,19 +263,19 @@ public static class Sse2 | |||
| /// <summary> | |||
| /// __m128i _mm_cvtps_epi32 (__m128 a) | |||
| /// </summary> | |||
| public static Vector128<int> ConvertToInt(Vector128<float> value) { throw new PlatformNotSupportedException(); } | |||
| public static Vector128<int> ConvertToInt32(Vector128<float> value) { throw new PlatformNotSupportedException(); } | |||
There was a problem hiding this comment.
This should be ConvertToVector128Int32 to match the naming convention the other methods have been given.
| /// <summary> | ||
| /// __m128i _mm_cvtpd_epi32 (__m128d a) | ||
| /// </summary> | ||
| public static Vector128<int> ConvertToInt(Vector128<double> value) { throw new PlatformNotSupportedException(); } | ||
| public static Vector128<int> ConvertToInt32(Vector128<double> value) { throw new PlatformNotSupportedException(); } |
| /// <summary> | ||
| /// __m128 _mm_cvtepi32_ps (__m128i a) | ||
| /// </summary> | ||
| public static Vector128<float> ConvertToFloat(Vector128<int> value) { throw new PlatformNotSupportedException(); } | ||
| public static Vector128<float> ConvertToSingle(Vector128<int> value) { throw new PlatformNotSupportedException(); } |
| /// <summary> | ||
| /// __m128 _mm_cvtpd_ps (__m128d a) | ||
| /// </summary> | ||
| public static Vector128<float> ConvertToFloat(Vector128<double> value) { throw new PlatformNotSupportedException(); } | ||
| public static Vector128<float> ConvertToSingle(Vector128<double> value) { throw new PlatformNotSupportedException(); } |
| @@ -288,11 +288,11 @@ public static class Sse2 | |||
| /// <summary> | |||
| /// __m128i _mm_cvttps_epi32 (__m128 a) | |||
| /// </summary> | |||
| public static Vector128<int> ConvertToIntWithTruncation(Vector128<float> value) { throw new PlatformNotSupportedException(); } | |||
| public static Vector128<int> ConvertToInt32WithTruncation(Vector128<float> value) { throw new PlatformNotSupportedException(); } | |||
There was a problem hiding this comment.
ConvertToVector128Int32WithTruncation
| /// <summary> | ||
| /// __m128i _mm_cvttpd_epi32 (__m128d a) | ||
| /// </summary> | ||
| public static Vector128<int> ConvertToIntWithTruncation(Vector128<double> value) { throw new PlatformNotSupportedException(); } | ||
| public static Vector128<int> ConvertToInt32WithTruncation(Vector128<double> value) { throw new PlatformNotSupportedException(); } |
|
cc @eerhardt |
|
FYI @KrzysztofCwalina @karelz @terrajobst This is changing shape of Intel intrinsics that was approved earlier. It looks like an oversight to me that can just go in without review. |
|
@fiigii Could you please prepare PR with a matching change in CoreFX? |
|
#15341 is adding new public APIs. Adding new APIs does not break anything. It does not need to be done in lock step. Your change is changing existing public APIs, and thus it needs to be done in lock step. |
|
@jkotas Thank you for explaining, will submit the CoreFX PR. |
|
@fiigii, since they are also naming issues. There are a few other inconsistencies that would be nice to fix. We use We use |
Ok, will make consistent to
Will change to |
fiigii
changed the title
fiigii
force-pushed
the
fixapi
branch
2 times, most recently
from
60889e7 to
4360c89
Compare
|
@tannergooding PTAL. I will make the CoreFX PR if this looks good to you. |
|
CoreFX PR dotnet/corefx#25965 |
Avoid language-specific type names in hardware intrinsic APIs. Discussed in #15341 (comment)
@tannergooding @jkotas