Re: [PATCH v4 1/2] arm: lib: xor-neon: remove unnecessary GCC < 4.6 warning

[Nick Desaulniers]

On Tue, Jan 19, 2021 at 1:35 PM Arnd Bergmann <arnd@xxxxxxxxxx> wrote:
>
> On Tue, Jan 19, 2021 at 10:18 PM 'Nick Desaulniers' via Clang Built
> Linux <clang-built-linux@xxxxxxxxxxxxxxxx> wrote:
> >
> > On Tue, Jan 19, 2021 at 5:17 AM Adrian Ratiu <adrian.ratiu@xxxxxxxxxxxxx> wrote:
> > > diff --git a/arch/arm/lib/xor-neon.c b/arch/arm/lib/xor-neon.c
> > > index b99dd8e1c93f..f9f3601cc2d1 100644
> > > --- a/arch/arm/lib/xor-neon.c
> > > +++ b/arch/arm/lib/xor-neon.c
> > > @@ -14,20 +14,22 @@ MODULE_LICENSE("GPL");
> > >  #error You should compile this file with '-march=armv7-a -mfloat-abi=softfp -mfpu=neon'
> > >  #endif
> > >
> > > +/*
> > > + * TODO: Even though -ftree-vectorize is enabled by default in Clang, the
> > > + * compiler does not produce vectorized code due to its cost model.
> > > + * See: https://github.com/ClangBuiltLinux/linux/issues/503
> > > + */
> > > +#ifdef CONFIG_CC_IS_CLANG
> > > +#warning Clang does not vectorize code in this file.
> > > +#endif
> >
> > Arnd, remind me again why it's a bug that the compiler's cost model
> > says it's faster to not produce a vectorized version of these loops?
> > I stand by my previous comment: https://bugs.llvm.org/show_bug.cgi?id=40976#c8
>
> The point is that without vectorizing the code, there is no point in building
> both the default xor code and a "neon" version that has to save/restore
> the neon registers but doesn't actually use them.
>
>           Arnd

Thoughts?  Also, Nathan brings up my previous point about restrict.
This would benefit both GCC and Clang as they would not produce 2
"versions" of the loop; one vectorized if the std::distance between x
& y >= 8, one scalar otherwise.  But the callers would have to ensure
no overlap otherwise UB.

diff --git a/include/asm-generic/xor.h b/include/asm-generic/xor.h
index b62a2a56a4d4..7db16adc7d89 100644
--- a/include/asm-generic/xor.h
+++ b/include/asm-generic/xor.h
@@ -7,12 +7,21 @@

 #include <linux/prefetch.h>

+/* Overrule LLVM's cost model in order to always produce a vectorized loop
+ * version.
+ */
+#if defined(__clang__) && defined(CONFIG_ARM)
+#define FORCE_VECTORIZE _Pragma("clang loop vectorize(enable)")
+#else
+#define CLANG_FORCE_VECTORIZE
+#endif
+
 static void
 xor_8regs_2(unsigned long bytes, unsigned long *p1, unsigned long *p2)
 {
        long lines = bytes / (sizeof (long)) / 8;

-       do {
+       FORCE_VECTORIZE do {
                p1[0] ^= p2[0];
                p1[1] ^= p2[1];
                p1[2] ^= p2[2];
@@ -32,7 +41,7 @@ xor_8regs_3(unsigned long bytes, unsigned long *p1,
unsigned long *p2,
 {
        long lines = bytes / (sizeof (long)) / 8;

-       do {
+       FORCE_VECTORIZE do {
                p1[0] ^= p2[0] ^ p3[0];
                p1[1] ^= p2[1] ^ p3[1];
                p1[2] ^= p2[2] ^ p3[2];
@@ -53,7 +62,7 @@ xor_8regs_4(unsigned long bytes, unsigned long *p1,
unsigned long *p2,
 {
        long lines = bytes / (sizeof (long)) / 8;

-       do {
+       FORCE_VECTORIZE do {
                p1[0] ^= p2[0] ^ p3[0] ^ p4[0];
                p1[1] ^= p2[1] ^ p3[1] ^ p4[1];
                p1[2] ^= p2[2] ^ p3[2] ^ p4[2];
@@ -75,7 +84,7 @@ xor_8regs_5(unsigned long bytes, unsigned long *p1,
unsigned long *p2,
 {
        long lines = bytes / (sizeof (long)) / 8;

-       do {
+       FORCE_VECTORIZE do {
                p1[0] ^= p2[0] ^ p3[0] ^ p4[0] ^ p5[0];
                p1[1] ^= p2[1] ^ p3[1] ^ p4[1] ^ p5[1];
                p1[2] ^= p2[2] ^ p3[2] ^ p4[2] ^ p5[2];
-- 
Thanks,
~Nick Desaulniers