[BUG] Problem building for AArch64 - " error: cannot convert ‘neon::U16 {aka short unsigned int}’"

[solabc16]

Hello

Building the native library fails when targeting AArch64, with the following error...

../../../src/opts/SkRasterPipeline_opts.h: In function ‘neon::F neon::from_half(neon::U16)’:
../../../src/opts/SkRasterPipeline_opts.h:657:26: error: cannot convert ‘neon::U16 {aka short unsigned int}’ to ‘float16x4_t {aka __vector(4) __fp16}’ for argument ‘1’ to ‘float32x4_t vcvt_f32_f16(float16x4_t)’
     return vcvt_f32_f16(h);

There are references to this elsewhere, for example : https://github.com/OSSystems/meta-browser/blob/master/recipes-browser/chromium/files/aarch64-skia-build-fix.patch

Do you have a preferred or existing resolution?

• Version with issue: 1.68.0
• Last known good version: 1.60.3

Best
-James

[solabc16]

FYI... attached is the patch file I'm currently using against the 1.68 source; which appears to work satisfactorily.

--- ./skiasharp/externals/skia/src/opts/SkRasterPipeline_opts.h-orig	2019-03-26 21:10:09.615000000 +0000
+++ ./skiasharp/externals/skia/src/opts/SkRasterPipeline_opts.h	2019-03-26 21:13:28.961000000 +0000
@@ -653,10 +653,7 @@
 }
 
 SI F from_half(U16 h) {
-#if defined(__aarch64__) && !defined(SK_BUILD_FOR_GOOGLE3)  // Temporary workaround for some Google3 builds.
-    return vcvt_f32_f16(h);
-
-#elif defined(JUMPER_IS_HSW) || defined(JUMPER_IS_AVX512)
+#if defined(JUMPER_IS_HSW) || defined(JUMPER_IS_AVX512)
     return _mm256_cvtph_ps(h);
 
 #else
@@ -673,10 +670,7 @@
 }
 
 SI U16 to_half(F f) {
-#if defined(__aarch64__) && !defined(SK_BUILD_FOR_GOOGLE3)  // Temporary workaround for some Google3 builds.
-    return vcvt_f16_f32(f);
-
-#elif defined(JUMPER_IS_HSW) || defined(JUMPER_IS_AVX512)
+#if defined(JUMPER_IS_HSW) || defined(JUMPER_IS_AVX512)
     return _mm256_cvtps_ph(f, _MM_FROUND_CUR_DIRECTION);
 
 #else

Best
-James
SkiaSharp168-aarch64-20190326.zip

[jclapis]

FYI... attached is the patch file I'm currently using against the 1.68 source; which appears to work satisfactorily.
...
Best
-James

Thanks James! I just ran into this while trying to build SkiaSharp on a Raspberry Pi 4B (which is ARM64) and your patch worked great.

[mattleibow]

Ah, thanks! I believe this may be fixed with the new work in the #986 PR. A lot of the opts bits have been improved. Thanks for the patch for now. I hope to have a new m81 master soon.

[solabc16]

Thanks for the feedback Joe. Matthew, let me know when you have the new master, happy to test.

For anybody else reading, this patch is still good for releases up to and including 1.68.1.1.

[mattleibow]

@solabc16 the master branch is now the new m80.

Let me know if you still need the patch. They did change a whole bunch of code around this. Also, we have switched to clang some time back, so maybe things are better if you use that?

You can also have a look at sending in a PR that builds skia with docker as part of the CI process. For the current x64 ubuntu build, I use this: https://github.com/mono/SkiaSharp/blob/master/scripts/Docker/ubuntu16/amd64/Dockerfile

The process does need a little bit more work, but I hope to get to that soon.