From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 4F9B7471E7; Sun, 11 Jan 2026 16:18:10 +0100 (CET) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 9D3AE4042C; Sun, 11 Jan 2026 16:18:05 +0100 (CET) Received: from mail-qk1-f169.google.com (mail-qk1-f169.google.com [209.85.222.169]) by mails.dpdk.org (Postfix) with ESMTP id A0A574028F for ; Sun, 11 Jan 2026 16:18:03 +0100 (CET) Received: by mail-qk1-f169.google.com with SMTP id af79cd13be357-8b2f2c5ec36so645320385a.1 for ; Sun, 11 Jan 2026 07:18:03 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20230601; t=1768144683; x=1768749483; darn=dpdk.org; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:from:to:cc:subject:date :message-id:reply-to; bh=yHuot1G8WwDkOcV5c09JhgaxtXJh9o57wX7AL0PYWJI=; b=nQyjJQsDnqOhy1g/fKCvwUFbCsdWxHfN9SNG9PIW9Ey76Cqk5matF9OdxHNlEJx0Lx VD4pSGPNVdqp88zEXLlgbdNF5AcFkECGz7+jqHjs/1dKQV90eDSOEHuNtbUfv4x9KDtR iQJei7XGqgJ3PDluIVvLUArE3JMdgqz/fEAB3WqMoc3t6rZtqYijVGo+69OzlytIFxfd VbQicZyDXuksM53XFOZKiYmVLNr34NFb0wVdvW4psANkzcpCRZIkfqEEPNJ5bwFtg8rE OTtzC0NV6h+kPoairPnGH8uM2EVNc0p1sYpjyiA4zMcNlS2Eq9ZySEnYIvQB0KTLoMgT A3mg== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1768144683; x=1768749483; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-gg:x-gm-message-state:from :to:cc:subject:date:message-id:reply-to; bh=yHuot1G8WwDkOcV5c09JhgaxtXJh9o57wX7AL0PYWJI=; b=gNUy8goNLvTAM0AZyq+pOCGBfEzGwDFOA5GoodQvvSEb9mTzI4uMbOQhLkg9ZCZF92 U/Aqs2Xy1skRY8Xv4TKj0jIRQ7zKjw2jeyKTeSQPLMKGZNyMiBZfhrvOZsc1RpHgMXQA jShM6BORlbRqIY6n5yZxo5RcMLlkilSdnrqBkYvlhm8Dl+/3TOgoac4N05WCU2n1yhnA R3MSw/2ADp45bd3bUYk2R2IrPFEoD2t56yhj8AA5zQOQZZIJapaCQpNLFpqhCzmhyHvo c/YuD3aPKZHcnB51Zj0opEPcvs7bM4XRzVVldb6PRyQD1oHO94dgMDdWmeQ0iBHt723/ BPnw== X-Gm-Message-State: AOJu0YxTJgkN3unW/Erztp2uXvkTLPIac02j69TErpPhKvi42BGPkdtF 328OikFnB9fJYgZrBTpBmlr8r+Hj7n1GyrWDmoFciQmBIbxZBrXUXxJRCsUZLA== X-Gm-Gg: AY/fxX5wBDfznOTJzoTxwVdbXemPDzXxwlPhsf0N8AGYflJWuOYBAji/Wa/Xp9TU4B4 OucchfDeu4FeK6/J0zGaJerSV3XAViSKY3O2can6/63fR9SS+uNHMSK680udZk5KgfjBhqhGJqZ 4btZfDemnuiEmGk9IB6LLDiVEU4UdNDjtAx7CHPZEOHdo+MeeAys83C8QZ0ScUh1XrtgxbwAxH+ YiwW79myWtohZrz2MMGd6XZfGl+WId4sOiLvD5a5ifYrAINz9POojgfXKqGosPbsedisD+JXefv XL7oZjae49knp716aHo3bxaZNoaOVIES7zwAjsQoIEc7+sStyizAkdPdp6sh7GJjRRMebTWjHkP fVi191RKbHeDgXbiK4T8ey1cL3ZuvbRGD6fIl6GmDIqU2OkmZl5vibXm0bb9ZBPMI9py+Tcp1k5 SU6gmvokdp7+STGBCqUvKqG7furyt/a2rkLfz1b+R912Ig8m8nTufSLDxOk78/YkGGRg== X-Google-Smtp-Source: AGHT+IGrnfOJuEtRBZQ3ny7LljpHYCBV9QSoFf8Ue8WWgvQJgPaTD1GiHFWY/2Yc2NPQ3216xf3Kkw== X-Received: by 2002:a05:620a:7001:b0:8b2:eb83:b99 with SMTP id af79cd13be357-8c38936f65cmr2050977185a.15.1768144682701; Sun, 11 Jan 2026 07:18:02 -0800 (PST) Received: from st47p01nt-relayp01.apple.com ([2600:1000:b19c:b9d9:b0a4:2ee1:3986:5e9d]) by smtp.gmail.com with ESMTPSA id 6a1803df08f44-890772346e2sm125909716d6.33.2026.01.11.07.18.02 (version=TLS1_3 cipher=TLS_CHACHA20_POLY1305_SHA256 bits=256/256); Sun, 11 Jan 2026 07:18:02 -0800 (PST) From: scott.k.mitch1@gmail.com To: dev@dpdk.org Cc: mb@smartsharesystems.com, stephen@networkplumber.org, Scott Mitchell Subject: [PATCH v2 1/2] eal: RTE_PTR_ADD/SUB char* for compiler optimizations Date: Sun, 11 Jan 2026 10:17:57 -0500 Message-Id: <20260111151758.89015-2-scott.k.mitch1@gmail.com> X-Mailer: git-send-email 2.39.5 (Apple Git-154) In-Reply-To: <20260111151758.89015-1-scott.k.mitch1@gmail.com> References: <20260111151758.89015-1-scott.k.mitch1@gmail.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org From: Scott Mitchell Modify RTE_PTR_ADD and RTE_PTR_SUB to use char* pointer arithmetic instead of uintptr_t casts when operating on pointer types. This enables better compiler optimization, particularly for Clang which can now recognize simple pointer patterns and apply vectorization, loop unrolling, and improved assembly. The implementation uses C11 _Generic to dispatch based on input type: - Integer types (int, uint64_t, etc.) continue using uintptr_t arithmetic for API compatibility - Pointer types use char* arithmetic, with const-qualified pointers using const char* internally to preserve optimization hints Example use case which benefits is __rte_raw_cksum. Performance results from cksum_perf_autotest on Intel Xeon (Cascade Lake, AVX-512) built with Clang 18.1 (TSC cycles/byte): Block size Before After Improvement 100 0.40 0.24 ~40% 1500 0.50 0.06 ~8x 9000 0.49 0.06 ~8x Signed-off-by: Scott Mitchell --- app/test/meson.build | 1 + app/test/test_ptr_add_sub.c | 190 +++++++++++++++++++++++++++++++++++ lib/eal/include/rte_common.h | 76 ++++++++++++++ 3 files changed, 267 insertions(+) create mode 100644 app/test/test_ptr_add_sub.c diff --git a/app/test/meson.build b/app/test/meson.build index efec42a6bf..80a19d65ba 100644 --- a/app/test/meson.build +++ b/app/test/meson.build @@ -152,6 +152,7 @@ source_file_deps = { 'test_power_intel_uncore.c': ['power', 'power_intel_uncore'], 'test_power_kvm_vm.c': ['power', 'power_kvm_vm'], 'test_prefetch.c': [], + 'test_ptr_add_sub.c': [], 'test_ptr_compress.c': ['ptr_compress'], 'test_rand_perf.c': [], 'test_rawdev.c': ['rawdev', 'bus_vdev', 'raw_skeleton'], diff --git a/app/test/test_ptr_add_sub.c b/app/test/test_ptr_add_sub.c new file mode 100644 index 0000000000..ef449ca9af --- /dev/null +++ b/app/test/test_ptr_add_sub.c @@ -0,0 +1,190 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2026 Apple Inc. + */ + +#include "test.h" +#include +#include + +#include + +/* Test all C11 standard integer types */ +static int +test_ptr_add_sub_integer_types(void) +{ + unsigned long long ull = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(ull, 100), 0x1064, + "RTE_PTR_ADD failed for unsigned long long"); + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_SUB(ull, 100), 0x1000 - 100, + "RTE_PTR_SUB failed for unsigned long long"); + + long long ll = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(ll, 100), 0x1064, + "RTE_PTR_ADD failed for long long"); + + unsigned long ul = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(ul, 100), 0x1064, + "RTE_PTR_ADD failed for unsigned long"); + + long l = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(l, 100), 0x1064, + "RTE_PTR_ADD failed for long"); + + unsigned int ui = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(ui, 100), 0x1064, + "RTE_PTR_ADD failed for unsigned int"); + + int i = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(i, 100), 0x1064, + "RTE_PTR_ADD failed for int"); + + unsigned short us = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(us, 100), 0x1064, + "RTE_PTR_ADD failed for unsigned short"); + + short s = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(s, 100), 0x1064, + "RTE_PTR_ADD failed for short"); + + unsigned char uc = 100; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(uc, 50), 150, + "RTE_PTR_ADD failed for unsigned char"); + + signed char sc = 100; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(sc, 50), 150, + "RTE_PTR_ADD failed for signed char"); + + char c = 100; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(c, 50), 150, + "RTE_PTR_ADD failed for char"); + + _Bool b = 1; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(b, 99), 100, + "RTE_PTR_ADD failed for _Bool"); + + bool b2 = true; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(b2, 99), 100, + "RTE_PTR_ADD failed for bool"); + + return 0; +} + +/* Test pointer types including const correctness */ +static int +test_ptr_add_sub_pointer_types(void) +{ + char buffer[256]; + void *result; + + /* Test void* and const void* */ + void *vp = buffer; + result = RTE_PTR_ADD(vp, 100); + TEST_ASSERT_EQUAL(result, (void *)(buffer + 100), + "RTE_PTR_ADD failed for void*"); + result = RTE_PTR_SUB(vp, 50); + TEST_ASSERT_EQUAL(result, (void *)(buffer - 50), + "RTE_PTR_SUB failed for void*"); + + const void *cvp = buffer; + result = RTE_PTR_ADD(cvp, 100); + TEST_ASSERT_EQUAL(result, (void *)(buffer + 100), + "RTE_PTR_ADD failed for const void*"); + result = RTE_PTR_SUB(cvp, 50); + TEST_ASSERT_EQUAL(result, (void *)(buffer - 50), + "RTE_PTR_SUB failed for const void*"); + + /* Test char* and const char* */ + char *cp = buffer; + result = RTE_PTR_ADD(cp, 100); + TEST_ASSERT_EQUAL(result, (void *)(buffer + 100), + "RTE_PTR_ADD failed for char*"); + result = RTE_PTR_SUB(cp, 50); + TEST_ASSERT_EQUAL(result, (void *)(buffer - 50), + "RTE_PTR_SUB failed for char*"); + + const char *ccp = buffer; + result = RTE_PTR_ADD(ccp, 100); + TEST_ASSERT_EQUAL(result, (void *)(buffer + 100), + "RTE_PTR_ADD failed for const char*"); + result = RTE_PTR_SUB(ccp, 50); + TEST_ASSERT_EQUAL(result, (void *)(buffer - 50), + "RTE_PTR_SUB failed for const char*"); + + /* Test uint32_t* and const uint32_t* */ + uint32_t *u32p = (uint32_t *)buffer; + result = RTE_PTR_ADD(u32p, 100); + TEST_ASSERT_EQUAL(result, (void *)(buffer + 100), + "RTE_PTR_ADD failed for uint32_t*"); + result = RTE_PTR_SUB(u32p, 50); + TEST_ASSERT_EQUAL(result, (void *)(buffer - 50), + "RTE_PTR_SUB failed for uint32_t*"); + + const uint32_t *cu32p = (const uint32_t *)buffer; + result = RTE_PTR_ADD(cu32p, 100); + TEST_ASSERT_EQUAL(result, (void *)(buffer + 100), + "RTE_PTR_ADD failed for const uint32_t*"); + result = RTE_PTR_SUB(cu32p, 50); + TEST_ASSERT_EQUAL(result, (void *)(buffer - 50), + "RTE_PTR_SUB failed for const uint32_t*"); + + /* Verify assigning to const pointer works (adding const is safe) */ + const void *result_const = RTE_PTR_ADD(cvp, 100); + TEST_ASSERT_EQUAL(result_const, (const void *)(buffer + 100), + "RTE_PTR_ADD failed when assigning to const void*"); + + return 0; +} + +/* Test that typedefs resolve to native types correctly */ +static int +test_ptr_add_sub_typedefs(void) +{ + uint64_t u64 = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(u64, 100), 0x1064, + "RTE_PTR_ADD failed for uint64_t"); + + uint32_t u32 = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(u32, 100), 0x1064, + "RTE_PTR_ADD failed for uint32_t"); + + uint16_t u16 = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(u16, 100), 0x1064, + "RTE_PTR_ADD failed for uint16_t"); + + uint8_t u8 = 100; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(u8, 50), 150, + "RTE_PTR_ADD failed for uint8_t"); + + uintptr_t uptr = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(uptr, 100), 0x1064, + "RTE_PTR_ADD failed for uintptr_t"); + + size_t sz = 0x1000; + TEST_ASSERT_EQUAL((uintptr_t)RTE_PTR_ADD(sz, 100), 0x1064, + "RTE_PTR_ADD failed for size_t"); + + return 0; +} + +/* Main test function that runs all subtests */ +static int +test_ptr_add_sub(void) +{ + int ret; + + ret = test_ptr_add_sub_integer_types(); + if (ret != 0) + return ret; + + ret = test_ptr_add_sub_pointer_types(); + if (ret != 0) + return ret; + + ret = test_ptr_add_sub_typedefs(); + if (ret != 0) + return ret; + + return 0; +} + +REGISTER_FAST_TEST(ptr_add_sub_autotest, true, true, test_ptr_add_sub); diff --git a/lib/eal/include/rte_common.h b/lib/eal/include/rte_common.h index 9e7d84f929..6f3dd744c1 100644 --- a/lib/eal/include/rte_common.h +++ b/lib/eal/include/rte_common.h @@ -551,12 +551,88 @@ static void __attribute__((destructor(RTE_PRIO(prio)), used)) func(void) /** * add a byte-value offset to a pointer */ +#ifndef RTE_TOOLCHAIN_MSVC +#define RTE_PTR_ADD(ptr, x) \ + (__extension__ ({ \ + /* Diagnostics suppressed for internal macro operations only. \ + * Compiler type-checks all _Generic branches even when unselected, \ + * triggering warnings with no external impact. */ \ + __rte_diagnostic_push \ + __rte_diagnostic_ignored_wcast_qual \ + _Pragma("GCC diagnostic ignored \"-Wconditional-type-mismatch\"") \ + /* Uses uintptr_t arithmetic for integer types (API compatibility), \ + * and char* arithmetic for pointer types (enables optimization). */ \ + __auto_type _ptr_result = _Generic((ptr), \ + unsigned long long: ((void *)((uintptr_t)(ptr) + (x))), \ + long long: ((void *)((uintptr_t)(ptr) + (x))), \ + unsigned long: ((void *)((uintptr_t)(ptr) + (x))), \ + long: ((void *)((uintptr_t)(ptr) + (x))), \ + unsigned int: ((void *)((uintptr_t)(ptr) + (x))), \ + int: ((void *)((uintptr_t)(ptr) + (x))), \ + unsigned short: ((void *)((uintptr_t)(ptr) + (x))), \ + short: ((void *)((uintptr_t)(ptr) + (x))), \ + unsigned char: ((void *)((uintptr_t)(ptr) + (x))), \ + signed char: ((void *)((uintptr_t)(ptr) + (x))), \ + char: ((void *)((uintptr_t)(ptr) + (x))), \ + _Bool: ((void *)((uintptr_t)(ptr) + (x))), \ + /* Ternary with null pointer constant: per C11, if one operand \ + * is a null pointer constant and the other is a pointer, the \ + * result type is qualified per the pointer operand, normalizing \ + * const T* to const void* and T* to void*. */ \ + default: _Generic((1 ? (ptr) : (void *)0), \ + const void *: ((void *)((const char *)(ptr) + (x))), \ + default: ((void *)((char *)(ptr) + (x))) \ + ) \ + ); \ + __rte_diagnostic_pop \ + _ptr_result; \ + })) +#else #define RTE_PTR_ADD(ptr, x) ((void*)((uintptr_t)(ptr) + (x))) +#endif /** * subtract a byte-value offset from a pointer */ +#ifndef RTE_TOOLCHAIN_MSVC +#define RTE_PTR_SUB(ptr, x) \ + (__extension__ ({ \ + /* Diagnostics suppressed for internal macro operations only. \ + * Compiler type-checks all _Generic branches even when unselected, \ + * triggering warnings with no external impact. */ \ + __rte_diagnostic_push \ + __rte_diagnostic_ignored_wcast_qual \ + _Pragma("GCC diagnostic ignored \"-Wconditional-type-mismatch\"") \ + /* Uses uintptr_t arithmetic for integer types (API compatibility), \ + * and char* arithmetic for pointer types (enables optimization). */ \ + __auto_type _ptr_result = _Generic((ptr), \ + unsigned long long: ((void *)((uintptr_t)(ptr) - (x))), \ + long long: ((void *)((uintptr_t)(ptr) - (x))), \ + unsigned long: ((void *)((uintptr_t)(ptr) - (x))), \ + long: ((void *)((uintptr_t)(ptr) - (x))), \ + unsigned int: ((void *)((uintptr_t)(ptr) - (x))), \ + int: ((void *)((uintptr_t)(ptr) - (x))), \ + unsigned short: ((void *)((uintptr_t)(ptr) - (x))), \ + short: ((void *)((uintptr_t)(ptr) - (x))), \ + unsigned char: ((void *)((uintptr_t)(ptr) - (x))), \ + signed char: ((void *)((uintptr_t)(ptr) - (x))), \ + char: ((void *)((uintptr_t)(ptr) - (x))), \ + _Bool: ((void *)((uintptr_t)(ptr) - (x))), \ + /* Ternary with null pointer constant: per C11, if one operand \ + * is a null pointer constant and the other is a pointer, the \ + * result type is qualified per the pointer operand, normalizing \ + * const T* to const void* and T* to void*. */ \ + default: _Generic((1 ? (ptr) : (void *)0), \ + const void *: ((void *)((const char *)(ptr) - (x))), \ + default: ((void *)((char *)(ptr) - (x))) \ + ) \ + ); \ + __rte_diagnostic_pop \ + _ptr_result; \ + })) +#else #define RTE_PTR_SUB(ptr, x) ((void *)((uintptr_t)(ptr) - (x))) +#endif /** * get the difference between two pointer values, i.e. how far apart -- 2.39.5 (Apple Git-154)