[PATCH] crc32: Add crc32 implementation using __builtin_aarch64_crc32b

[Marek Vasut]

ARMv8.0 has optional crc32 instruction for crc32 calculation. The
instruction is mandatory since ARMv8.1. The crc32 calculation is
faster using the dedicated instruction, e.g. 1.4 GHz iMX8MN gives:

  => time crc32 0x50000000 0x2000000
  time: 0.126 seconds # crc32 instruction
  time: 0.213 seconds # software crc32

Add implementation using the compiler builtin wrapper for the crc32
instruction and enable it by default, since we don't support any
platforms which do not implement this instruction.

Signed-off-by: Marek Vasut <marex at denx.de>
Cc: Simon Glass <sjg at chromium.org>
Cc: Tom Rini <trini at konsulko.com>
---
 arch/arm/Kconfig  | 9 +++++++++
 arch/arm/Makefile | 4 ++++
 lib/crc32.c       | 7 +++++++
 3 files changed, 20 insertions(+)

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 2d59562665f..134d3c6563d 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -10,6 +10,15 @@ config ARM64
 	select SYS_CACHE_SHIFT_6
 
 if ARM64
+config ARM64_CRC32
+	bool "Enable support for CRC32 instruction"
+	default y
+	help
+	  ARMv8 implements dedicated crc32 instruction for crc32 calculation.
+	  This is faster than software crc32 calculation. This instruction may
+	  not be present on all ARMv8.0, but is always present on ARMv8.1 and
+	  newer.
+
 config POSITION_INDEPENDENT
 	bool "Generate position-independent pre-relocation code"
 	help
diff --git a/arch/arm/Makefile b/arch/arm/Makefile
index c68e598a675..ce977bf6323 100644
--- a/arch/arm/Makefile
+++ b/arch/arm/Makefile
@@ -18,7 +18,11 @@ arch-$(CONFIG_CPU_V7A)		=$(call cc-option, -march=armv7-a, \
 				 $(call cc-option, -march=armv7))
 arch-$(CONFIG_CPU_V7M)		=-march=armv7-m
 arch-$(CONFIG_CPU_V7R)		=-march=armv7-r
+ifeq ($(CONFIG_ARM64_CRC32),y)
+arch-$(CONFIG_ARM64)		=-march=armv8-a+crc
+else
 arch-$(CONFIG_ARM64)		=-march=armv8-a
+endif
 
 # On Tegra systems we must build SPL for the armv4 core on the device
 # but otherwise we can use the value in CONFIG_SYS_ARM_ARCH
diff --git a/lib/crc32.c b/lib/crc32.c
index f2acc107fe4..a207e3d2b40 100644
--- a/lib/crc32.c
+++ b/lib/crc32.c
@@ -184,6 +184,12 @@ const uint32_t * ZEXPORT get_crc_table()
  */
 uint32_t __efi_runtime crc32_no_comp(uint32_t crc, const Bytef *buf, uInt len)
 {
+#ifdef CONFIG_ARM64_CRC32
+    crc = cpu_to_le32(crc);
+    while (len--)
+        crc = __builtin_aarch64_crc32b(crc, *buf++);
+    return le32_to_cpu(crc);
+#else
     const uint32_t *tab = crc_table;
     const uint32_t *b =(const uint32_t *)buf;
     size_t rem_len;
@@ -221,6 +227,7 @@ uint32_t __efi_runtime crc32_no_comp(uint32_t crc, const Bytef *buf, uInt len)
     }
 
     return le32_to_cpu(crc);
+#endif
 }
 #undef DO_CRC
 
-- 
2.33.0