Add Pentium 4 Linux support

.github/workflows/release.yml

@@ -34,6 +34,10 @@ jobs:
             os: ubuntu-24.04
             openssl_target: linux-x86_64
             cross_prefix: ''
+          - target: linux-i386
+            os: ubuntu-24.04
+            openssl_target: linux-x86
+            cross_prefix: ''
           - target: linux-arm64
             os: ubuntu-24.04
             openssl_target: linux-aarch64
@@ -61,6 +65,12 @@ jobs:
           sudo apt-get update
           sudo apt-get install --yes gcc-aarch64-linux-gnu
 
+      - name: Install the Linux i386 multilib compiler
+        if: matrix.target == 'linux-i386'
+        run: |
+          sudo apt-get update
+          sudo apt-get install --yes gcc-multilib
+
       - name: Install the Linux PPC32 big-endian cross-compiler
         if: matrix.target == 'linux-ppc32be'
         run: |

Makefile

@@ -48,6 +48,7 @@ DIST      := dist
 DRIVER    := src/main.c
 ASM_ARM64 := src/fossmark.S
 ASM_AMD64 := src/fossmark_x86_64.S
+SRC_I386  := src/fossmark_ppc32.c
 SRC_PPC32 := src/fossmark_ppc32.c
 ASM_PPC32 := src/fossmark_ppc32_ext.S
 SRC_PPC64 := src/fossmark_ppc32.c
@@ -60,6 +61,9 @@ ifneq (,$(filter aarch64 arm64,$(HOST_ARCH)))
 else ifneq (,$(filter x86_64 amd64,$(HOST_ARCH)))
 	HOST_ARCHNAME := amd64
 	HOST_KERNEL   := $(ASM_AMD64)
+else ifneq (,$(filter i386 i486 i586 i686 x86,$(HOST_ARCH)))
+	HOST_ARCHNAME := i386
+	HOST_KERNEL   := $(SRC_I386)
 else ifneq (,$(filter ppc powerpc ppc32 powerpc32,$(HOST_ARCH)))
 	HOST_ARCHNAME := ppc32be
 	HOST_KERNEL   := $(SRC_PPC32) $(ASM_PPC32)
@@ -68,7 +72,10 @@ else ifneq (,$(filter ppc64 powerpc64,$(HOST_ARCH)))
 	HOST_KERNEL   := $(SRC_PPC64)
 else
 	HOST_ARCHNAME := $(HOST_ARCH)
-	$(error unsupported host architecture '$(HOST_ARCH)')
+$(error unsupported host architecture '$(HOST_ARCH)')
+endif
+ifeq ($(HOST_ARCHNAME),i386)
+	CFLAGS += -march=pentium4 -msse2
 endif
 ifeq ($(HOST_ARCHNAME),ppc64be)
 	CFLAGS += -mcpu=970 -maltivec
@@ -100,6 +107,11 @@ ifeq ($(HOST_ARCHNAME),amd64)
 else
 	CC_AMD64 ?= x86_64-linux-gnu-gcc
 endif
+ifeq ($(HOST_ARCHNAME),i386)
+	CC_I386 ?= $(CC)
+else
+	CC_I386 ?= cc
+endif
 CC_MACOS_ARM64 ?= $(CC)
 CC_MACOS_AMD64 ?= $(CC)
 MACOS_AMD64_MIN ?= 10.5
@@ -118,16 +130,17 @@ NATIVE_BIN := $(DIST)/fossmark-$(OSNAME)-$(HOST_ARCHNAME)
 
 # `make` with no target builds the host binary, as before.
 .DEFAULT_GOAL := native
-.PHONY: all native linux-arm64 linux-amd64 linux-ppc32be linux-ppc64be macos-arm64 macos-amd64 bench test clean
+.PHONY: all native linux-arm64 linux-amd64 linux-i386 linux-ppc32be linux-ppc64be macos-arm64 macos-amd64 bench test clean
 
 # `make all` builds all Linux binaries.
-all: linux-arm64 linux-amd64 linux-ppc32be linux-ppc64be
+all: linux-arm64 linux-amd64 linux-i386 linux-ppc32be linux-ppc64be
 
 # `make native` (and bare `make`) build for whatever host you are on.
 native: $(NATIVE_BIN)
 
 linux-arm64: $(DIST)/fossmark-linux-arm64
 linux-amd64: $(DIST)/fossmark-linux-amd64
+linux-i386: $(DIST)/fossmark-linux-i386
 linux-ppc32be: $(DIST)/fossmark-linux-ppc32be
 linux-ppc64be: $(DIST)/fossmark-linux-ppc64be
 macos-arm64: $(DIST)/fossmark-macos-arm64
@@ -141,6 +154,10 @@ $(DIST)/fossmark-linux-amd64: $(DRIVER) $(ASM_AMD64) | $(DIST)
 	$(CC_AMD64) $(CFLAGS) $(TLS_CFLAGS) $(PTHREAD) $(LDFLAGS) -o $@ $(DRIVER) $(ASM_AMD64) $(LDLIBS)
 	@echo "built $@"
 
+$(DIST)/fossmark-linux-i386: $(DRIVER) $(SRC_I386) | $(DIST)
+	$(CC_I386) -m32 -march=pentium4 -msse2 $(CFLAGS) $(TLS_CFLAGS) $(PTHREAD) $(LDFLAGS) -o $@ $(DRIVER) $(SRC_I386) $(LDLIBS)
+	@echo "built $@"
+
 $(DIST)/fossmark-linux-ppc32be: $(DRIVER) $(SRC_PPC32) $(ASM_PPC32) | $(DIST)
 	$(CC_PPC32BE) $(CFLAGS) $(TLS_CFLAGS) $(PTHREAD) $(LDFLAGS) -o $@ $(DRIVER) $(SRC_PPC32) $(ASM_PPC32) $(LDLIBS)
 	@echo "built $@"
@@ -166,6 +183,9 @@ endif
 ifeq ($(OSNAME)-$(HOST_ARCHNAME),linux-amd64)
 NATIVE_HAS_RULE := yes
 endif
+ifeq ($(OSNAME)-$(HOST_ARCHNAME),linux-i386)
+NATIVE_HAS_RULE := yes
+endif
 ifeq ($(OSNAME)-$(HOST_ARCHNAME),linux-ppc32be)
 NATIVE_HAS_RULE := yes
 endif

README.md

@@ -6,7 +6,7 @@ across every available core. The final report includes separate single-core and
 multicore scores.
 
 The repository currently builds an executable named `fossmark` for ARM64,
-x86-64, and 32- or 64-bit big-endian PowerPC. The C driver handles timing, memory,
+x86 (Pentium 4 or newer), x86-64, and 32- or 64-bit big-endian PowerPC. The C driver handles timing, memory,
 threads, output, and scoring. Performance-sensitive kernels live in
 architecture-specific backend files.
 
@@ -50,6 +50,7 @@ Other targets are available for explicit platforms and architectures:
 ```sh
 make linux-arm64
 make linux-amd64
+make linux-i386          # Pentium 4 / SSE2 baseline
 make linux-ppc32be
 make linux-ppc64be       # PowerPC 970 / iMac G5
 make macos-arm64
@@ -63,6 +64,7 @@ toolchain. Override the target compiler when its name differs from the default:
 ```sh
 make linux-arm64 CC_ARM64=aarch64-linux-gnu-gcc
 make linux-amd64 CC_AMD64=x86_64-linux-gnu-gcc
+make linux-i386 CC_I386=gcc
 make linux-ppc32be CC_PPC32BE=powerpc-linux-gnu-gcc
 make linux-ppc64be CC_PPC64BE=powerpc64-linux-gnu-gcc
 ```

src/fossmark_ppc32.c

@@ -11,6 +11,10 @@
 #include <stdio.h>
 #include <string.h>
 
+#if defined(__i386__) || defined(__x86_64__) || defined(_M_IX86) || defined(_M_X64)
+#include <emmintrin.h>
+#endif
+
 #if defined(__linux__)
 #include <sys/auxv.h>
 #endif
@@ -64,7 +68,7 @@ uint64_t fm_primes(uint64_t limit, uint8_t *sieve)
 	return count;
 }
 
-#if !defined(__powerpc64__)
+#if defined(__powerpc__) && !defined(__powerpc64__)
 static uint64_t fm_simd_scalar(uint64_t iters, void *memory)
 {
 	uint32_t *v = (uint32_t *)memory;
@@ -109,7 +113,7 @@ uint64_t fm_simd(uint64_t iters, void *memory)
 		sum ^= v[j];
 	return sum;
 }
-#else
+#elif defined(__powerpc__)
 /* These are kept in fossmark_ppc32_ext.S so this translation unit, and thus
  * the executable's default code path, only requires baseline PPC32. */
 extern void fm_simd_ps_kernel(uint64_t iters, void *memory);
@@ -192,6 +196,31 @@ uint64_t fm_simd(uint64_t iters, void *memory)
 		sum ^= v[j];
 	return sum;
 }
+#else
+uint64_t fm_simd(uint64_t iters, void *memory)
+{
+	__m128i a, b;
+	uint32_t *v = (uint32_t *)memory;
+	uint32_t sum = 0;
+	uint64_t i;
+	unsigned j;
+
+	if (!iters)
+		return 0;
+	a = _mm_loadu_si128((const __m128i *)v);
+	b = _mm_loadu_si128((const __m128i *)(v + 4));
+	for (i = 0; i < iters; i++) {
+		a = _mm_add_epi32(a, b);
+		b = _mm_xor_si128(b, a);
+		a = _mm_add_epi32(a, b);
+		b = _mm_xor_si128(b, a);
+	}
+	_mm_storeu_si128((__m128i *)v, a);
+	_mm_storeu_si128((__m128i *)(v + 4), b);
+	for (j = 0; j < 8; j++)
+		sum ^= v[j];
+	return sum;
+}
 #endif
 
 static uint32_t load32_native(const uint8_t *p)

src/main.c

@@ -42,7 +42,7 @@
 #ifndef FM_API_BASE_URL
 #  define FM_API_BASE_URL "https://fossbench.net"
 #endif
-#define FM_VERSION "0.1.3-hotfix3"
+#define FM_VERSION "0.1.3-hotfix4"
 
 /* ---------- platform identification (for the banner only) ---------- */
 
@@ -66,6 +66,11 @@
 #  define D_INT  "64-bit ALU: imul, mul, div, bitops"
 #  define D_FP   "double: mulsd/addsd, divsd, sqrtsd"
 #  define D_SIMD "SSE2: 128-bit integer + float"
+#elif defined(__i386__) || defined(_M_IX86)
+#  define FM_ARCH "x86 32-bit"
+#  define D_INT  "Pentium 4 integer ALU and software 64-bit arithmetic"
+#  define D_FP   "x87 scalar double-precision floating point"
+#  define D_SIMD "SSE2: 128-bit integer vectors"
 #elif defined(__powerpc64__)
 #  define FM_ARCH "PowerPC 64-bit big-endian"
 #  define D_INT  "64-bit PowerPC integer ALU"
@@ -145,6 +150,9 @@ extern uint64_t fm_chase(void **ptrs, uint64_t steps);
 #if defined(__powerpc__) && !defined(__powerpc64__)
 #  define CHASE_NODES	(1u << 19)		/* 2 MiB with 32-bit pointers */
 #  define CHASE_DETAIL	"dependent-load pointer chase, 2 MiB"
+#elif UINTPTR_MAX == UINT32_MAX
+#  define CHASE_NODES	(1u << 21)		/* 8 MiB with 32-bit pointers */
+#  define CHASE_DETAIL	"dependent-load pointer chase, 8 MiB"
 #else
 #  define CHASE_NODES	(1u << 21)		/* 16 MiB cycle, > any L2 */
 #  define CHASE_DETAIL	"dependent-load pointer chase, 16 MiB"