Fix cross-compiling librt by enabling x86_64 optimisations with pragmas (#20815)

Although Python still lacks a supported method to cross-compile
packages, many downstreams do it anyway, so librt should allow this.

The current approach of enabling x86_64 optimisations after checking
`platform.machine()` is broken. When building on x86_64 for another
architecture, the build fails because the optimisations get enabled when
they shouldn't. Conversely, when building on another architecture for
x86_64, the build fails because the optimisations don't get enabled when
they should.

GCC supports enabling an optimisation with a pragma from that line
onwards. Clang requires the optimisation to be pushed on and popped off
the stack. Unfortunately, MSVC does not have an equivalent feature, but
it is unlikely anyone would cross-compile to x86_64 with that.

The remaining logic for MSVC could be simplified, but it looks like
other compilers like Borland are potentially supported?

I considered checking for x86_64 using CCompiler's `preprocess()`
instead, but this seemed awkward.

This has been tested with GCC and Clang, including with AVX512 enabled.

Supersedes mypyc/librt#32.

Author: chewi

mypyc/build_setup.py

@@ -19,18 +19,11 @@
     from distutils import ccompiler
 
 EXTRA_FLAGS_PER_COMPILER_TYPE_PER_PATH_COMPONENT = {
-    "unix": {
-        "base64/arch/ssse3": ["-mssse3"],
-        "base64/arch/sse41": ["-msse4.1"],
-        "base64/arch/sse42": ["-msse4.2"],
-        "base64/arch/avx2": ["-mavx2"],
-        "base64/arch/avx": ["-mavx"],
-    },
     "msvc": {
         "base64/arch/sse42": ["/arch:SSE4.2"],
         "base64/arch/avx2": ["/arch:AVX2"],
         "base64/arch/avx": ["/arch:AVX"],
-    },
+    }
 }
 
 ccompiler.CCompiler.__spawn = ccompiler.CCompiler.spawn  # type: ignore[attr-defined]

mypyc/lib-rt/base64/arch/avx/codec.c

@@ -9,6 +9,11 @@
 #include "../../env.h"
 
 #if HAVE_AVX
+#if defined(__clang__)
+#pragma clang attribute push (__attribute__((target("avx"))), apply_to=function)
+#else
+#pragma GCC target("avx")
+#endif
 #include <immintrin.h>
 
 // Only enable inline assembly on supported compilers and on 64-bit CPUs.
@@ -62,6 +67,9 @@ base64_stream_decode_avx BASE64_DEC_PARAMS
 	#include "../generic/dec_head.c"
 	dec_loop_ssse3(&s, &slen, &o, &olen);
 	#include "../generic/dec_tail.c"
+#if defined(__clang__)
+	#pragma clang attribute pop
+#endif
 #else
 	return base64_dec_stub(state, src, srclen, out, outlen);
 #endif

mypyc/lib-rt/base64/arch/avx2/codec.c

@@ -9,6 +9,11 @@
 #include "../../env.h"
 
 #if HAVE_AVX2
+#if defined(__clang__)
+#pragma clang attribute push (__attribute__((target("avx2"))), apply_to=function)
+#else
+#pragma GCC target("avx2")
+#endif
 #include <immintrin.h>
 
 // Only enable inline assembly on supported compilers and on 64-bit CPUs.
@@ -52,6 +57,9 @@ base64_stream_decode_avx2 BASE64_DEC_PARAMS
 	#include "../generic/dec_head.c"
 	dec_loop_avx2(&s, &slen, &o, &olen);
 	#include "../generic/dec_tail.c"
+#if defined(__clang__)
+	#pragma clang attribute pop
+#endif
 #else
 	return base64_dec_stub(state, src, srclen, out, outlen);
 #endif

mypyc/lib-rt/base64/arch/avx512/codec.c

@@ -9,6 +9,11 @@
 #include "../../env.h"
 
 #if HAVE_AVX512
+#if defined(__clang__)
+#pragma clang attribute push (__attribute__((target("avx512vbmi"))), apply_to=function)
+#else
+#pragma GCC target("avx512vbmi")
+#endif
 #include <immintrin.h>
 
 #include "../avx2/dec_reshuffle.c"
@@ -38,6 +43,9 @@ base64_stream_decode_avx512 BASE64_DEC_PARAMS
 	#include "../generic/dec_head.c"
 	dec_loop_avx2(&s, &slen, &o, &olen);
 	#include "../generic/dec_tail.c"
+#if defined(__clang__)
+	#pragma clang attribute pop
+#endif
 #else
 	return base64_dec_stub(state, src, srclen, out, outlen);
 #endif

mypyc/lib-rt/base64/arch/sse41/codec.c

@@ -9,6 +9,11 @@
 #include "../../env.h"
 
 #if HAVE_SSE41
+#if defined(__clang__)
+#pragma clang attribute push (__attribute__((target("sse4.1"))), apply_to=function)
+#else
+#pragma GCC target("sse4.1")
+#endif
 #include <smmintrin.h>
 
 // Only enable inline assembly on supported compilers and on 64-bit CPUs.
@@ -52,6 +57,9 @@ base64_stream_decode_sse41 BASE64_DEC_PARAMS
 	#include "../generic/dec_head.c"
 	dec_loop_ssse3(&s, &slen, &o, &olen);
 	#include "../generic/dec_tail.c"
+#if defined(__clang__)
+	#pragma clang attribute pop
+#endif
 #else
 	return base64_dec_stub(state, src, srclen, out, outlen);
 #endif

mypyc/lib-rt/base64/arch/sse42/codec.c

@@ -9,6 +9,11 @@
 #include "../../env.h"
 
 #if HAVE_SSE42
+#if defined(__clang__)
+#pragma clang attribute push (__attribute__((target("sse4.2"))), apply_to=function)
+#else
+#pragma GCC target("sse4.2")
+#endif
 #include <nmmintrin.h>
 
 // Only enable inline assembly on supported compilers and on 64-bit CPUs.
@@ -52,6 +57,9 @@ base64_stream_decode_sse42 BASE64_DEC_PARAMS
 	#include "../generic/dec_head.c"
 	dec_loop_ssse3(&s, &slen, &o, &olen);
 	#include "../generic/dec_tail.c"
+#if defined(__clang__)
+	#pragma clang attribute pop
+#endif
 #else
 	return base64_dec_stub(state, src, srclen, out, outlen);
 #endif

mypyc/lib-rt/base64/arch/ssse3/codec.c

@@ -9,6 +9,11 @@
 #include "../../env.h"
 
 #if HAVE_SSSE3
+#if defined(__clang__)
+#pragma clang attribute push (__attribute__((target("ssse3"))), apply_to=function)
+#else
+#pragma GCC target("ssse3")
+#endif
 #include <tmmintrin.h>
 
 // Only enable inline assembly on supported compilers and on 64-bit CPUs.
@@ -54,6 +59,9 @@ base64_stream_decode_ssse3 BASE64_DEC_PARAMS
 	#include "../generic/dec_head.c"
 	dec_loop_ssse3(&s, &slen, &o, &olen);
 	#include "../generic/dec_tail.c"
+#if defined(__clang__)
+	#pragma clang attribute pop
+#endif
 #else
 	return base64_dec_stub(state, src, srclen, out, outlen);
 #endif

mypyc/lib-rt/build_setup.py

@@ -19,18 +19,11 @@
     from distutils import ccompiler
 
 EXTRA_FLAGS_PER_COMPILER_TYPE_PER_PATH_COMPONENT = {
-    "unix": {
-        "base64/arch/ssse3": ["-mssse3"],
-        "base64/arch/sse41": ["-msse4.1"],
-        "base64/arch/sse42": ["-msse4.2"],
-        "base64/arch/avx2": ["-mavx2"],
-        "base64/arch/avx": ["-mavx"],
-    },
     "msvc": {
         "base64/arch/sse42": ["/arch:SSE4.2"],
         "base64/arch/avx2": ["/arch:AVX2"],
         "base64/arch/avx": ["/arch:AVX"],
-    },
+    }
 }
 
 ccompiler.CCompiler.__spawn = ccompiler.CCompiler.spawn  # type: ignore[attr-defined]