[dpdk-dev] [RFC v2] hash: unify crc32 API header for x86 and ARM

[<pbhagavatula@marvell.com>]

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Merge crc32 hash calculation public API headers for x86 and ARM.
Select the best available CRC32 algorithm when unsupported algorithm
on a given CPU architecture is requested by an application.

Previously, if an application directly includes `rte_crc_arm64.h` without
including `rte_hash_crc.h` it will fail to compile.
Although, `rte_crc_arm64.h` is no longer needed make it a dummy file for
ABI purposes.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 v2 Changes:
 - Don't remove `rte_crc_arm64.h` for ABI purposes.
 - Revert function pointer approach for performance reasons.
 - Select the best available algorithm based on the arch when user passes an
 unsupported crc32 algorithm.

 app/test/test_hash.c            |   6 ++
 lib/librte_hash/meson.build     |   6 +-
 lib/librte_hash/rte_crc_arm64.h | 175 +-------------------------------
 lib/librte_hash/rte_hash_crc.h  | 153 ++++++++++++++++++++--------
 4 files changed, 122 insertions(+), 218 deletions(-)

diff --git a/app/test/test_hash.c b/app/test/test_hash.c
index afa3a1a3c..7bd457dac 100644
--- a/app/test/test_hash.c
+++ b/app/test/test_hash.c
@@ -195,7 +195,13 @@ test_crc32_hash_alg_equiv(void)
 	}

 	/* Resetting to best available algorithm */
+#if defined RTE_ARCH_X86
 	rte_hash_crc_set_alg(CRC32_SSE42_x64);
+#elif defined RTE_ARCH_ARM64
+	rte_hash_crc_set_alg(CRC32_ARM64);
+#else
+	rte_hash_crc_set_alg(CRC32_SW);
+#endif

 	if (i == CRC32_ITERATIONS)
 		return 0;
diff --git a/lib/librte_hash/meson.build b/lib/librte_hash/meson.build
index 6ab46ae9d..8a3cf2f64 100644
--- a/lib/librte_hash/meson.build
+++ b/lib/librte_hash/meson.build
@@ -1,12 +1,14 @@
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright(c) 2017 Intel Corporation

-headers = files('rte_crc_arm64.h',
-	'rte_fbk_hash.h',
+headers = files('rte_fbk_hash.h',
 	'rte_hash_crc.h',
 	'rte_hash.h',
 	'rte_jhash.h',
 	'rte_thash.h')
+if dpdk_conf.has('RTE_ARCH_ARM64')
+	headers += files('rte_crc_arm64.h')
+endif

 sources = files('rte_cuckoo_hash.c', 'rte_fbk_hash.c')
 deps += ['ring']
diff --git a/lib/librte_hash/rte_crc_arm64.h b/lib/librte_hash/rte_crc_arm64.h
index b4628cfc0..adfcafc7d 100644
--- a/lib/librte_hash/rte_crc_arm64.h
+++ b/lib/librte_hash/rte_crc_arm64.h
@@ -5,179 +5,6 @@
 #ifndef _RTE_CRC_ARM64_H_
 #define _RTE_CRC_ARM64_H_

-/**
- * @file
- *
- * RTE CRC arm64 Hash
- */
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-#include <stdint.h>
-#include <rte_cpuflags.h>
-#include <rte_branch_prediction.h>
-#include <rte_common.h>
-
-static inline uint32_t
-crc32c_arm64_u8(uint8_t data, uint32_t init_val)
-{
-	__asm__ volatile(
-			"crc32cb %w[crc], %w[crc], %w[value]"
-			: [crc] "+r" (init_val)
-			: [value] "r" (data));
-	return init_val;
-}
-
-static inline uint32_t
-crc32c_arm64_u16(uint16_t data, uint32_t init_val)
-{
-	__asm__ volatile(
-			"crc32ch %w[crc], %w[crc], %w[value]"
-			: [crc] "+r" (init_val)
-			: [value] "r" (data));
-	return init_val;
-}
-
-static inline uint32_t
-crc32c_arm64_u32(uint32_t data, uint32_t init_val)
-{
-	__asm__ volatile(
-			"crc32cw %w[crc], %w[crc], %w[value]"
-			: [crc] "+r" (init_val)
-			: [value] "r" (data));
-	return init_val;
-}
-
-static inline uint32_t
-crc32c_arm64_u64(uint64_t data, uint32_t init_val)
-{
-	__asm__ volatile(
-			"crc32cx %w[crc], %w[crc], %x[value]"
-			: [crc] "+r" (init_val)
-			: [value] "r" (data));
-	return init_val;
-}
-
-/**
- * Allow or disallow use of arm64 SIMD instrinsics for CRC32 hash
- * calculation.
- *
- * @param alg
- *   An OR of following flags:
- *   - (CRC32_SW) Don't use arm64 crc intrinsics
- *   - (CRC32_ARM64) Use ARMv8 CRC intrinsic if available
- *
- */
-static inline void
-rte_hash_crc_set_alg(uint8_t alg)
-{
-	switch (alg) {
-	case CRC32_ARM64:
-		if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_CRC32))
-			alg = CRC32_SW;
-		/* fall-through */
-	case CRC32_SW:
-		crc32_alg = alg;
-		/* fall-through */
-	default:
-		break;
-	}
-}
-
-/* Setting the best available algorithm */
-RTE_INIT(rte_hash_crc_init_alg)
-{
-	rte_hash_crc_set_alg(CRC32_ARM64);
-}
-
-/**
- * Use single crc32 instruction to perform a hash on a 1 byte value.
- * Fall back to software crc32 implementation in case arm64 crc intrinsics is
- * not supported
- *
- * @param data
- *   Data to perform hash on.
- * @param init_val
- *   Value to initialise hash generator.
- * @return
- *   32bit calculated hash value.
- */
-static inline uint32_t
-rte_hash_crc_1byte(uint8_t data, uint32_t init_val)
-{
-	if (likely(crc32_alg & CRC32_ARM64))
-		return crc32c_arm64_u8(data, init_val);
-
-	return crc32c_1byte(data, init_val);
-}
-
-/**
- * Use single crc32 instruction to perform a hash on a 2 bytes value.
- * Fall back to software crc32 implementation in case arm64 crc intrinsics is
- * not supported
- *
- * @param data
- *   Data to perform hash on.
- * @param init_val
- *   Value to initialise hash generator.
- * @return
- *   32bit calculated hash value.
- */
-static inline uint32_t
-rte_hash_crc_2byte(uint16_t data, uint32_t init_val)
-{
-	if (likely(crc32_alg & CRC32_ARM64))
-		return crc32c_arm64_u16(data, init_val);
-
-	return crc32c_2bytes(data, init_val);
-}
-
-/**
- * Use single crc32 instruction to perform a hash on a 4 byte value.
- * Fall back to software crc32 implementation in case arm64 crc intrinsics is
- * not supported
- *
- * @param data
- *   Data to perform hash on.
- * @param init_val
- *   Value to initialise hash generator.
- * @return
- *   32bit calculated hash value.
- */
-static inline uint32_t
-rte_hash_crc_4byte(uint32_t data, uint32_t init_val)
-{
-	if (likely(crc32_alg & CRC32_ARM64))
-		return crc32c_arm64_u32(data, init_val);
-
-	return crc32c_1word(data, init_val);
-}
-
-/**
- * Use single crc32 instruction to perform a hash on a 8 byte value.
- * Fall back to software crc32 implementation in case arm64 crc intrinsics is
- * not supported
- *
- * @param data
- *   Data to perform hash on.
- * @param init_val
- *   Value to initialise hash generator.
- * @return
- *   32bit calculated hash value.
- */
-static inline uint32_t
-rte_hash_crc_8byte(uint64_t data, uint32_t init_val)
-{
-	if (likely(crc32_alg == CRC32_ARM64))
-		return crc32c_arm64_u64(data, init_val);
-
-	return crc32c_2words(data, init_val);
-}
-
-#ifdef __cplusplus
-}
-#endif
+#include "rte_hash_crc.h"

 #endif /* _RTE_CRC_ARM64_H_ */
diff --git a/lib/librte_hash/rte_hash_crc.h b/lib/librte_hash/rte_hash_crc.h
index cf28031b3..eaba70c12 100644
--- a/lib/librte_hash/rte_hash_crc.h
+++ b/lib/librte_hash/rte_hash_crc.h
@@ -16,10 +16,12 @@ extern "C" {
 #endif

 #include <stdint.h>
-#include <rte_config.h>
-#include <rte_cpuflags.h>
+
 #include <rte_branch_prediction.h>
 #include <rte_common.h>
+#include <rte_config.h>
+#include <rte_cpuflags.h>
+#include <rte_log.h>

 /* Lookup tables for software implementation of CRC32C */
 static const uint32_t crc32c_tables[8][256] = {{
@@ -322,7 +324,7 @@ crc32c_2bytes(uint16_t data, uint32_t init_val)
 }

 static inline uint32_t
-crc32c_1word(uint32_t data, uint32_t init_val)
+crc32c_4bytes(uint32_t data, uint32_t init_val)
 {
 	uint32_t crc, term1, term2;
 	crc = init_val;
@@ -336,7 +338,7 @@ crc32c_1word(uint32_t data, uint32_t init_val)
 }

 static inline uint32_t
-crc32c_2words(uint64_t data, uint32_t init_val)
+crc32c_8bytes(uint64_t data, uint32_t init_val)
 {
 	uint32_t crc, term1, term2;
 	union {
@@ -358,6 +360,48 @@ crc32c_2words(uint64_t data, uint32_t init_val)
 	return crc;
 }

+#if defined(RTE_ARCH_ARM64) && defined(RTE_MACHINE_CPUFLAG_CRC32)
+static inline uint32_t
+crc32c_arm64_u8(uint8_t data, uint32_t init_val)
+{
+	__asm__ volatile(
+			"crc32cb %w[crc], %w[crc], %w[value]"
+			: [crc] "+r" (init_val)
+			: [value] "r" (data));
+	return init_val;
+}
+
+static inline uint32_t
+crc32c_arm64_u16(uint16_t data, uint32_t init_val)
+{
+	__asm__ volatile(
+			"crc32ch %w[crc], %w[crc], %w[value]"
+			: [crc] "+r" (init_val)
+			: [value] "r" (data));
+	return init_val;
+}
+
+static inline uint32_t
+crc32c_arm64_u32(uint32_t data, uint32_t init_val)
+{
+	__asm__ volatile(
+			"crc32cw %w[crc], %w[crc], %w[value]"
+			: [crc] "+r" (init_val)
+			: [value] "r" (data));
+	return init_val;
+}
+
+static inline uint32_t
+crc32c_arm64_u64(uint64_t data, uint32_t init_val)
+{
+	__asm__ volatile(
+			"crc32cx %w[crc], %w[crc], %x[value]"
+			: [crc] "+r" (init_val)
+			: [value] "r" (data));
+	return init_val;
+}
+#endif
+
 #if defined(RTE_ARCH_X86)
 static inline uint32_t
 crc32c_sse42_u8(uint8_t data, uint32_t init_val)
@@ -424,42 +468,69 @@ crc32c_sse42_u64(uint64_t data, uint64_t init_val)

 static uint8_t crc32_alg = CRC32_SW;

-#if defined(RTE_ARCH_ARM64) && defined(RTE_MACHINE_CPUFLAG_CRC32)
-#include "rte_crc_arm64.h"
-#else
-
 /**
- * Allow or disallow use of SSE4.2 instrinsics for CRC32 hash
+ * Allow or disallow use of SSE4.2/ARMv8 instrinsics for CRC32 hash
  * calculation.
  *
  * @param alg
  *   An OR of following flags:
- *   - (CRC32_SW) Don't use SSE4.2 intrinsics
+ *   - (CRC32_SW) Don't use SSE4.2 intrinsics (default non-[x86/ARMv8])
  *   - (CRC32_SSE42) Use SSE4.2 intrinsics if available
- *   - (CRC32_SSE42_x64) Use 64-bit SSE4.2 intrinsic if available (default)
- *
+ *   - (CRC32_SSE42_x64) Use 64-bit SSE4.2 intrinsic if available (default x86)
+ *   - (CRC32_ARM64) Use ARMv8 CRC intrinsic if available
  */
 static inline void
 rte_hash_crc_set_alg(uint8_t alg)
 {
-#if defined(RTE_ARCH_X86)
-	if (alg == CRC32_SSE42_x64 &&
-			!rte_cpu_get_flag_enabled(RTE_CPUFLAG_EM64T))
-		alg = CRC32_SSE42;
+	switch (alg) {
+	case CRC32_SSE42_x64:
+	case CRC32_SSE42:
+#if defined RTE_ARCH_X86
+		if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_EM64T))
+			crc32_alg = CRC32_SSE42;
+		else
+			crc32_alg = alg;
+#endif
+#if defined RTE_ARCH_ARM64
+		RTE_LOG(WARNING, HASH,
+			"Incorrect CRC32 algorithm requested setting best"
+			"available algorithm on the architecture\n");
+		rte_hash_crc_set_alg(CRC32_ARM64);
+#endif
+		break;
+	case CRC32_ARM64:
+#if defined RTE_ARCH_ARM64
+		if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_CRC32))
+			crc32_alg = CRC32_ARM64;
 #endif
-	crc32_alg = alg;
+#if defined RTE_ARCH_X86
+		RTE_LOG(WARNING, HASH,
+			"Incorrect CRC32 algorithm requested setting best"
+			"available algorithm on the architecture\n");
+		rte_hash_crc_set_alg(CRC32_SSE42_x64);
+#endif
+		break;
+	case CRC32_SW:
+	default:
+		crc32_alg = CRC32_SW;
+	break;
+	}
 }

 /* Setting the best available algorithm */
 RTE_INIT(rte_hash_crc_init_alg)
 {
+#if defined RTE_ARCH_X86
 	rte_hash_crc_set_alg(CRC32_SSE42_x64);
+#elif defined RTE_ARCH_ARM64
+	rte_hash_crc_set_alg(CRC32_ARM64);
+#else
+	rte_hash_crc_set_alg(CRC32_SW);
+#endif
 }

 /**
- * Use single crc32 instruction to perform a hash on a byte value.
- * Fall back to software crc32 implementation in case SSE4.2 is
- * not supported
+ * Calculate crc32 hash value of 1bytes.
  *
  * @param data
  *   Data to perform hash on.
@@ -474,15 +545,15 @@ rte_hash_crc_1byte(uint8_t data, uint32_t init_val)
 #if defined RTE_ARCH_X86
 	if (likely(crc32_alg & CRC32_SSE42))
 		return crc32c_sse42_u8(data, init_val);
+#elif defined RTE_ARCH_ARM64
+	if (likely(crc32_alg & CRC32_ARM64))
+		return crc32c_arm64_u8(data, init_val);
 #endif
-
 	return crc32c_1byte(data, init_val);
 }

 /**
- * Use single crc32 instruction to perform a hash on a 2 bytes value.
- * Fall back to software crc32 implementation in case SSE4.2 is
- * not supported
+ * Calculate crc32 hash value of 2bytes.
  *
  * @param data
  *   Data to perform hash on.
@@ -497,15 +568,15 @@ rte_hash_crc_2byte(uint16_t data, uint32_t init_val)
 #if defined RTE_ARCH_X86
 	if (likely(crc32_alg & CRC32_SSE42))
 		return crc32c_sse42_u16(data, init_val);
+#elif defined RTE_ARCH_ARM64
+	if (likely(crc32_alg & CRC32_ARM64))
+		return crc32c_arm64_u16(data, init_val);
 #endif
-
 	return crc32c_2bytes(data, init_val);
 }

 /**
- * Use single crc32 instruction to perform a hash on a 4 byte value.
- * Fall back to software crc32 implementation in case SSE4.2 is
- * not supported
+ * Calculate crc32 hash value of 4bytes.
  *
  * @param data
  *   Data to perform hash on.
@@ -520,15 +591,15 @@ rte_hash_crc_4byte(uint32_t data, uint32_t init_val)
 #if defined RTE_ARCH_X86
 	if (likely(crc32_alg & CRC32_SSE42))
 		return crc32c_sse42_u32(data, init_val);
+#elif defined RTE_ARCH_ARM64
+	if (likely(crc32_alg & CRC32_ARM64))
+		return crc32c_arm64_u32(data, init_val);
 #endif
-
-	return crc32c_1word(data, init_val);
+	return crc32c_4bytes(data, init_val);
 }

 /**
- * Use single crc32 instruction to perform a hash on a 8 byte value.
- * Fall back to software crc32 implementation in case SSE4.2 is
- * not supported
+ * Calculate crc32 hash value of 8bytes.
  *
  * @param data
  *   Data to perform hash on.
@@ -540,21 +611,19 @@ rte_hash_crc_4byte(uint32_t data, uint32_t init_val)
 static inline uint32_t
 rte_hash_crc_8byte(uint64_t data, uint32_t init_val)
 {
-#ifdef RTE_ARCH_X86_64
-	if (likely(crc32_alg == CRC32_SSE42_x64))
+#if defined RTE_ARCH_X86_64
+	if (likely(crc32_alg & CRC32_SSE42_x64))
 		return crc32c_sse42_u64(data, init_val);
-#endif
-
-#if defined RTE_ARCH_X86
+#elif defined RTE_ARCH_X86
 	if (likely(crc32_alg & CRC32_SSE42))
 		return crc32c_sse42_u64_mimic(data, init_val);
+#elif defined RTE_ARCH_ARM64
+	if (likely(crc32_alg & CRC32_ARM64))
+		return crc32c_arm64_u64(data, init_val);
 #endif
-
-	return crc32c_2words(data, init_val);
+	return crc32c_8bytes(data, init_val);
 }

-#endif
-
 /**
  * Calculate CRC32 hash on user-supplied byte array.
  *
--
2.17.1