add ppc32be, and other fixes

.github/workflows/release.yml

@@ -34,6 +34,8 @@ jobs:
             os: ubuntu-24.04
           - target: linux-arm64
             os: ubuntu-24.04
+          - target: linux-ppc32be
+            os: ubuntu-24.04
           - target: macos-amd64
             os: macos-14
           - target: macos-arm64
@@ -49,6 +51,12 @@ jobs:
           sudo apt-get update
           sudo apt-get install --yes gcc-aarch64-linux-gnu
 
+      - name: Install the Linux PPC32 big-endian cross-compiler
+        if: matrix.target == 'linux-ppc32be'
+        run: |
+          sudo apt-get update
+          sudo apt-get install --yes gcc-powerpc-linux-gnu
+
       - name: Build
         run: make ${{ matrix.target }}
 

Makefile

@@ -3,6 +3,7 @@
 # The assembly kernels are architecture-specific:
 #   src/fossmark.S         AArch64 (ARM64)
 #   src/fossmark_x86_64.S  x86-64  (AMD64)
+#   src/fossmark_ppc32.c   PowerPC 32-bit, including big-endian systems
 # The C driver (src/main.c) is portable across architectures and OSes. A
 # "binary that runs everywhere" is not possible - each OS/arch pair uses a
 # different executable format and instruction set - so output is named per
@@ -42,6 +43,7 @@ DIST      := dist
 DRIVER    := src/main.c
 ASM_ARM64 := src/fossmark.S
 ASM_AMD64 := src/fossmark_x86_64.S
+SRC_PPC32 := src/fossmark_ppc32.c
 
 # ---- host detection: normalise `uname -m` to our arch names ----
 HOST_ARCH := $(shell uname -m)
@@ -50,10 +52,16 @@ ifneq (,$(filter aarch64 arm64,$(HOST_ARCH)))
 	HOST_ASM      := $(ASM_ARM64)
 else ifneq (,$(filter x86_64 amd64,$(HOST_ARCH)))
 	HOST_ARCHNAME := amd64
-	HOST_ASM      := $(ASM_AMD64)
+	HOST_KERNEL   := $(ASM_AMD64)
+else ifneq (,$(filter ppc powerpc ppc32 powerpc32,$(HOST_ARCH)))
+	HOST_ARCHNAME := ppc32be
+	HOST_KERNEL   := $(SRC_PPC32)
 else
 	HOST_ARCHNAME := $(HOST_ARCH)
-	HOST_ASM      := $(ASM_ARM64)
+	$(error unsupported host architecture '$(HOST_ARCH)')
+endif
+ifeq ($(HOST_ARCHNAME),arm64)
+	HOST_KERNEL := $(ASM_ARM64)
 endif
 
 # ---- host OS name for the native binary ----
@@ -81,21 +89,27 @@ else
 endif
 CC_MACOS_ARM64 ?= $(CC)
 CC_MACOS_AMD64 ?= $(CC)
+ifeq ($(HOST_ARCHNAME),ppc32be)
+	CC_PPC32BE ?= $(CC)
+else
+	CC_PPC32BE ?= powerpc-linux-gnu-gcc
+endif
 
 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 macos-arm64 macos-amd64 bench test clean
+.PHONY: all native linux-arm64 linux-amd64 linux-ppc32be macos-arm64 macos-amd64 bench test clean
 
-# `make all` builds both Linux binaries.
+# `make all` builds all Linux binaries.
-all: linux-arm64 linux-amd64
+all: linux-arm64 linux-amd64 linux-ppc32be
 
 # `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-ppc32be: $(DIST)/fossmark-linux-ppc32be
 macos-arm64: $(DIST)/fossmark-macos-arm64
 macos-amd64: $(DIST)/fossmark-macos-amd64
 
@@ -107,6 +121,10 @@ $(DIST)/fossmark-linux-amd64: $(DRIVER) $(ASM_AMD64) | $(DIST)
 	$(CC_AMD64) $(CFLAGS) $(PTHREAD) -o $@ $(DRIVER) $(ASM_AMD64) $(LDLIBS)
 	@echo "built $@"
 
+$(DIST)/fossmark-linux-ppc32be: $(DRIVER) $(SRC_PPC32) | $(DIST)
+	$(CC_PPC32BE) $(CFLAGS) $(PTHREAD) -o $@ $(DRIVER) $(SRC_PPC32) $(LDLIBS)
+	@echo "built $@"
+
 $(DIST)/fossmark-macos-arm64: $(DRIVER) $(ASM_ARM64) | $(DIST)
 	$(CC_MACOS_ARM64) -arch arm64 $(CFLAGS) $(PTHREAD) -o $@ $(DRIVER) $(ASM_ARM64) $(LDLIBS)
 	@echo "built $@"
@@ -124,6 +142,9 @@ endif
 ifeq ($(OSNAME)-$(HOST_ARCHNAME),linux-amd64)
 NATIVE_HAS_RULE := yes
 endif
+ifeq ($(OSNAME)-$(HOST_ARCHNAME),linux-ppc32be)
+NATIVE_HAS_RULE := yes
+endif
 ifeq ($(OSNAME)-$(HOST_ARCHNAME),macos-arm64)
 NATIVE_HAS_RULE := yes
 endif
@@ -131,8 +152,8 @@ ifeq ($(OSNAME)-$(HOST_ARCHNAME),macos-amd64)
 NATIVE_HAS_RULE := yes
 endif
 ifneq ($(NATIVE_HAS_RULE),yes)
-$(NATIVE_BIN): $(DRIVER) $(HOST_ASM) | $(DIST)
+$(NATIVE_BIN): $(DRIVER) $(HOST_KERNEL) | $(DIST)
-	$(CC) $(CFLAGS) $(PTHREAD) -o $@ $(DRIVER) $(HOST_ASM) $(LDLIBS)
+	$(CC) $(CFLAGS) $(PTHREAD) -o $@ $(DRIVER) $(HOST_KERNEL) $(LDLIBS)
 	@echo "built $@"
 endif
 
@@ -145,7 +166,7 @@ bench: $(NATIVE_BIN)
 
 # Build and run the kernel correctness tests for the host arch.
 test: | $(DIST)
-	$(CC) $(CFLAGS) $(PTHREAD) -o $(DIST)/test_kernels src/test_kernels.c $(HOST_ASM) $(LDLIBS)
+	$(CC) $(CFLAGS) $(PTHREAD) -o $(DIST)/test_kernels src/test_kernels.c $(HOST_KERNEL) $(LDLIBS)
 	./$(DIST)/test_kernels
 
 clean:

README.md

@@ -5,9 +5,10 @@ small C driver. It measures each workload twice: once on a single core and once
 across every available core. The final report includes separate single-core and
 multicore scores.
 
-The repository currently builds an executable named `fossmark` for ARM64 and
+The repository currently builds an executable named `fossmark` for ARM64,
-x86-64. The C driver handles timing, memory, threads, output, and scoring. The
+x86-64, and 32-bit big-endian PowerPC. The C driver handles timing, memory,
-performance-sensitive kernels live in architecture-specific assembly files.
+threads, output, and scoring. Performance-sensitive kernels live in
+architecture-specific backend files.
 
 ## Workloads
 
@@ -48,6 +49,7 @@ Other targets are available for explicit platforms and architectures:
 ```sh
 make linux-arm64
 make linux-amd64
+make linux-ppc32be
 make macos-arm64
 make macos-amd64
 make all
@@ -59,6 +61,7 @@ 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-ppc32be CC_PPC32BE=powerpc-linux-gnu-gcc
 ```
 
 Apple Clang can build either macOS architecture with `-arch`. Windows timing
@@ -77,7 +80,8 @@ The exact filename depends on the host platform and architecture.
 
 Pushing a Git tag runs the GitHub Actions build and correctness tests. If they
 succeed, the workflow creates a GitHub Release named `Release <tag name>` with
-Linux and macOS archives for AMD64 and ARM64, plus a `SHA256SUMS` file.
+Linux archives for AMD64, ARM64, and PPC32 big-endian, macOS archives for AMD64
+and ARM64, and a `SHA256SUMS` file.
 
 ## Scores
 
@@ -115,13 +119,14 @@ normal.
 
 ## Architecture support
 
-The assembly kernels use only baseline instructions for their architecture:
+The kernel backends use only baseline instructions for their architecture:
 
 * `src/fossmark.S` uses ARMv8-A and NEON under AAPCS64.
 * `src/fossmark_x86_64.S` uses baseline x86-64 and SSE2 under the System V ABI.
+* `src/fossmark_ppc32.c` is endian-safe and uses baseline 32-bit PowerPC operations. It avoids AltiVec so it runs on the Wii's PowerPC 750CL-class CPU.
 
-The kernel files contain no system calls or calls into the C library. The same
+The assembly kernel files contain no system calls or calls into the C library.
-ARM64 source can be assembled for Linux, macOS, Windows, and BSD object formats.
+The same ARM64 source can be assembled for Linux, macOS, Windows, and BSD object formats.
 The current x86-64 source supports Linux, macOS, and the BSDs that use the
 System V calling convention.
 
@@ -149,8 +154,8 @@ with a nonzero status if any check fails.
 src/main.c              portable benchmark driver and scoring
 src/fossmark.S          ARM64 kernels
 src/fossmark_x86_64.S   x86-64 kernels
+src/fossmark_ppc32.c    PPC32 big-endian kernels
 src/test_kernels.c      correctness suite
 Makefile                native and cross-build targets
 dist/                   generated binaries
 ```
-

src/fossmark_ppc32.c

@@ -0,0 +1,149 @@
+/*
+ * Portable kernel backend for 32-bit PowerPC.
+ *
+ * Keeping this backend in C lets the compiler implement 64-bit arguments and
+ * returns according to the platform's PPC32 ABI.  All byte-oriented formats
+ * are decoded explicitly, so the code is correct on big-endian systems.
+ */
+#include <math.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+static uint32_t rotl32(uint32_t x, unsigned n)
+{
+	return (x << n) | (x >> (32 - n));
+}
+
+uint64_t fm_int_math(uint64_t iters)
+{
+	uint64_t a = 0x9e3779b97f4a7c15ULL, b = 0xbf58476d1ce4e5b9ULL;
+	uint64_t c = 0x94d049bb133111ebULL, d = 0x2545f4914f6cdd1dULL;
+	uint64_t i;
+	if (!iters) return 0;
+	for (i = 0; i < iters; i++) {
+		a = a * 0xdeadbeefU + b; b = b * 0xdeadbeefU + c;
+		c = c * 0xdeadbeefU + d; d = d * 0xdeadbeefU + a;
+		a ^= c >> 29; b ^= d << 17; c ^= (a >> 31) | (a << 33);
+		d ^= b >> 7; a += c / 0xdeadbeefU; b += d / 0xdeadbeefU;
+	}
+	return a ^ b ^ c ^ d;
+}
+
+uint64_t fm_fp_math(uint64_t iters)
+{
+	double a = 1.5, b = 2.5, c = 3.5, d = .5, out;
+	uint64_t bits, i;
+	if (!iters) return 0;
+	for (i = 0; i < iters; i++) {
+		a = fmin(a * 1.0625 + .0009765625, 2.0);
+		b = fmin(b * 1.0625 + .0009765625, 2.0);
+		c = fmin(c * 1.0625 + .0009765625, 2.0) + sqrt(a);
+		d = fmax(fabs(fmin(d * 1.0625 + .0009765625, 2.0) + sqrt(b)), 1.0);
+		a += 1.0 / (c + 1.0); b += 1.0 / (d + 1.0);
+	}
+	out = a + b + c + d;
+	memcpy(&bits, &out, sizeof bits);
+	return bits;
+}
+
+uint64_t fm_primes(uint64_t limit, uint8_t *sieve)
+{
+	uint64_t i, j, count = 0;
+	if (limit < 2) return 0;
+	memset(sieve, 0, (size_t)limit);
+	sieve[0] = sieve[1] = 1;
+	for (i = 2; i <= (limit - 1) / i; i++)
+		if (!sieve[i]) for (j = i * i; j < limit; j += i) sieve[j] = 1;
+	for (i = 2; i < limit; i++) count += !sieve[i];
+	return count;
+}
+
+uint64_t fm_simd(uint64_t iters, void *memory)
+{
+	uint32_t *v = (uint32_t *)memory;
+	uint32_t a[8]; uint64_t i; unsigned j; uint32_t sum = 0;
+	if (!iters) return 0;
+	memcpy(a, v, sizeof a);
+	for (i = 0; i < iters; i++)
+		for (j = 0; j < 8; j++) a[j] = rotl32(a[j] + a[(j + 1) & 7] * (j + 3), (j + 5) & 31);
+	for (j = 0; j < 8; j++) sum ^= a[j];
+	memcpy(v, a, sizeof a);
+	return sum;
+}
+
+static uint32_t load32_native(const uint8_t *p)
+{
+	uint32_t v; memcpy(&v, p, sizeof v); return v;
+}
+
+uint64_t fm_compress(const uint8_t *src, uint64_t len, uint32_t *ht)
+{
+	uint64_t ip = 0, anchor = 0, out = 0, ref, ml, lit;
+	memset(ht, 0, (size_t)(1U << 16) * sizeof *ht);
+	if (len < 16) return len + 1;
+	while (ip < len - 12) {
+		uint32_t seq = load32_native(src + ip);
+		uint32_t h = (uint32_t)(seq * 2654435761U) >> 16;
+		ref = ht[h]; ht[h] = (uint32_t)ip;
+		if (ref >= ip || ip - ref >= 65536 || load32_native(src + ref) != seq) { ip++; continue; }
+		for (ml = 4; ip + ml < len && src[ip + ml] == src[ref + ml]; ml++) {}
+		lit = ip - anchor; out += lit + 3 + (lit >= 15) + (ml >= 19);
+		ip += ml; anchor = ip;
+	}
+	return out + (len - anchor) + 1;
+}
+
+static uint32_t load32le(const uint8_t *p)
+{
+	return (uint32_t)p[0] | (uint32_t)p[1] << 8 | (uint32_t)p[2] << 16 | (uint32_t)p[3] << 24;
+}
+static void store32le(uint8_t *p, uint32_t v)
+{
+	p[0] = (uint8_t)v; p[1] = (uint8_t)(v >> 8); p[2] = (uint8_t)(v >> 16); p[3] = (uint8_t)(v >> 24);
+}
+#define QR(a,b,c,d) do { a+=b; d=rotl32(d^a,16); c+=d; b=rotl32(b^c,12); a+=b; d=rotl32(d^a,8); c+=d; b=rotl32(b^c,7); } while (0)
+uint64_t fm_chacha20(uint8_t *buf, uint64_t len, const uint8_t key[32], uint64_t passes)
+{
+	static const uint32_t sigma[4] = {0x61707865,0x3320646e,0x79622d32,0x6b206574};
+	uint32_t base[16], x[16], counter = 0, checksum = 0; uint64_t pass, off; int i, r;
+	len &= ~(uint64_t)63; if (!len || !passes) return 0;
+	memcpy(base, sigma, 16); for (i=0;i<8;i++) base[4+i]=load32le(key+4*i);
+	base[13]=base[14]=base[15]=0;
+	for (pass=0;pass<passes;pass++) for (off=0;off<len;off+=64) {
+		base[12]=counter++; memcpy(x,base,sizeof x);
+		for(r=0;r<10;r++) { QR(x[0],x[4],x[8],x[12]); QR(x[1],x[5],x[9],x[13]); QR(x[2],x[6],x[10],x[14]); QR(x[3],x[7],x[11],x[15]); QR(x[0],x[5],x[10],x[15]); QR(x[1],x[6],x[11],x[12]); QR(x[2],x[7],x[8],x[13]); QR(x[3],x[4],x[9],x[14]); }
+		for(i=0;i<16;i++) { uint32_t k=x[i]+base[i]; uint8_t t[4]; store32le(t,k); buf[off+4*i]^=t[0]; buf[off+4*i+1]^=t[1]; buf[off+4*i+2]^=t[2]; buf[off+4*i+3]^=t[3]; checksum^=k; }
+	}
+	return checksum;
+}
+#undef QR
+
+uint64_t fm_physics(double *b, uint64_t n, uint64_t steps)
+{
+	uint64_t s,i,j,bits; double sum=0;
+	if (!n || !steps) return 0;
+	for(s=0;s<steps;s++) { for(i=0;i<n;i++) { double ax=0,ay=0,az=0; for(j=0;j<n;j++) { double dx=b[8*j]-b[8*i],dy=b[8*j+1]-b[8*i+1],dz=b[8*j+2]-b[8*i+2]; double q=1.0/sqrt(dx*dx+dy*dy+dz*dz+.0625); q=q*q*q*b[8*j+3]; ax+=dx*q; ay+=dy*q; az+=dz*q; } b[8*i+4]+=ax*.0078125; b[8*i+5]+=ay*.0078125; b[8*i+6]+=az*.0078125; } for(i=0;i<n;i++) { b[8*i]+=b[8*i+4]*.0078125; b[8*i+1]+=b[8*i+5]*.0078125; b[8*i+2]+=b[8*i+6]*.0078125; } }
+	for (i = 0; i < n; i++)
+		sum += b[8*i+4] + b[8*i+5] + b[8*i+6];
+	memcpy(&bits, &sum, sizeof bits);
+	return bits;
+}
+
+static void sift(uint32_t *a, uint64_t root, uint64_t end) { for (;;) { uint64_t c=root*2+1; uint32_t t; if(c>=end)return; if(c+1<end&&a[c+1]>a[c])c++; if(a[root]>=a[c])return; t=a[root];a[root]=a[c];a[c]=t;root=c; } }
+uint64_t fm_sort(uint32_t *a, uint64_t n)
+{
+	uint64_t i,end,sum=0; uint32_t t; if(n<2)return n?a[0]:0;
+	for (i = n / 2; i; i--)
+		sift(a, i - 1, n);
+	for (end = n - 1; end; end--) {
+		t = a[0]; a[0] = a[end]; a[end] = t;
+		sift(a, 0, end);
+	}
+	for(i=0;i<n;i++){sum=(sum>>7)|(sum<<57);sum^=a[i];sum+=a[i];} return sum;
+}
+
+uint64_t fm_chase(void **ptrs, uint64_t steps)
+{
+	void **p=ptrs; uint64_t i; if(!steps)return 0; for(i=0;i<steps;i++)p=(void **)*p; return (uint64_t)((uintptr_t)p-(uintptr_t)ptrs);
+}

src/main.c

@@ -45,6 +45,11 @@
 #  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(__powerpc__) && !defined(__powerpc64__)
+#  define FM_ARCH "PowerPC 32-bit big-endian"
+#  define D_INT  "PPC32 integer ALU and software 64-bit arithmetic"
+#  define D_FP   "PowerPC scalar double-precision floating point"
+#  define D_SIMD "PPC32 parallel integer workload"
 #else
 #  define FM_ARCH "unknown"
 #  define D_INT  "64-bit integer ALU"
@@ -603,14 +608,12 @@ static double display_metric(const struct test *t, const struct result *r)
 static void print_header(void)
 {
 	printf("\n");
-	printf("  fossmark 1.0 - multi-core CPU benchmark\n");
+	printf("  fossbench1.0 - multi-core CPU benchmark\n");
 	printf("  ------------------------------------------------------------------\n");
 	printf("  platform: %s/%s\n", FM_OS, FM_ARCH);
-	printf("  cores:    %ld (each test is run once on 1 core, once on all %ld)\n",
+	printf("  cores:    %ld\n",
 	       g_ncores, g_ncores);
 	printf("\n");
-	printf("  RATE/TIME/SCORE below are the all-core (multi-core) pass.\n");
-	printf("\n");
 	printf("  %-24s %12s  %-11s %8s %9s\n",
 	       "TEST", "RATE", "UNIT", "TIME", "SCORE");
 	printf("  --------------------------------------------------------------------------\n");
@@ -700,8 +703,6 @@ int main(int argc, char **argv)
 	       exp(multi_log_sum / weight_sum));
 	printf("  %-24s %44.0f\n", "SINGLECORE SCORE",
 	       exp(single_log_sum / weight_sum));
-	printf("  %-24s (weighted geometric means, reference machine = %.0f)\n",
-	       "", FM_TARGET_SCORE);
 	printf("\n");
 
 	teardown();