util/bufferiszero: Split out host include files

Split out host/bufferiszero.h.inc for x86, aarch64 and generic
in order to avoid an overlong ifdef ladder.

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

4 files changed, 211 insertions, 0 deletions

diff --git a/host/include/aarch64/host/bufferiszero.c.inc b/host/include/aarch64/host/bufferiszero.c.inc
new file mode 100644
index 0000000..947ee7c
--- /dev/null
+++ b/host/include/aarch64/host/bufferiszero.c.inc
@@ -0,0 +1,76 @@
+/*
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ * buffer_is_zero acceleration, aarch64 version.
+ */
+
+#ifdef __ARM_NEON
+#include <arm_neon.h>
+
+/*
+ * Helper for preventing the compiler from reassociating
+ * chains of binary vector operations.
+ */
+#define REASSOC_BARRIER(vec0, vec1) asm("" : "+w"(vec0), "+w"(vec1))
+
+static bool buffer_is_zero_simd(const void *buf, size_t len)
+{
+    uint32x4_t t0, t1, t2, t3;
+
+    /* Align head/tail to 16-byte boundaries.  */
+    const uint32x4_t *p = QEMU_ALIGN_PTR_DOWN(buf + 16, 16);
+    const uint32x4_t *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 16);
+
+    /* Unaligned loads at head/tail.  */
+    t0 = vld1q_u32(buf) | vld1q_u32(buf + len - 16);
+
+    /* Collect a partial block at tail end.  */
+    t1 = e[-7] | e[-6];
+    t2 = e[-5] | e[-4];
+    t3 = e[-3] | e[-2];
+    t0 |= e[-1];
+    REASSOC_BARRIER(t0, t1);
+    REASSOC_BARRIER(t2, t3);
+    t0 |= t1;
+    t2 |= t3;
+    REASSOC_BARRIER(t0, t2);
+    t0 |= t2;
+
+    /*
+     * Loop over complete 128-byte blocks.
+     * With the head and tail removed, e - p >= 14, so the loop
+     * must iterate at least once.
+     */
+    do {
+        /*
+         * Reduce via UMAXV.  Whatever the actual result,
+         * it will only be zero if all input bytes are zero.
+         */
+        if (unlikely(vmaxvq_u32(t0) != 0)) {
+            return false;
+        }
+
+        t0 = p[0] | p[1];
+        t1 = p[2] | p[3];
+        t2 = p[4] | p[5];
+        t3 = p[6] | p[7];
+        REASSOC_BARRIER(t0, t1);
+        REASSOC_BARRIER(t2, t3);
+        t0 |= t1;
+        t2 |= t3;
+        REASSOC_BARRIER(t0, t2);
+        t0 |= t2;
+        p += 8;
+    } while (p < e - 7);
+
+    return vmaxvq_u32(t0) == 0;
+}
+
+static biz_accel_fn const accel_table[] = {
+    buffer_is_zero_int_ge256,
+    buffer_is_zero_simd,
+};
+
+#define best_accel() 1
+#else
+# include "host/include/generic/host/bufferiszero.c.inc"
+#endif
diff --git a/host/include/generic/host/bufferiszero.c.inc b/host/include/generic/host/bufferiszero.c.inc
new file mode 100644
index 0000000..ea0875c
--- /dev/null
+++ b/host/include/generic/host/bufferiszero.c.inc
@@ -0,0 +1,10 @@
+/*
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ * buffer_is_zero acceleration, generic version.
+ */
+
+static biz_accel_fn const accel_table[1] = {
+    buffer_is_zero_int_ge256
+};
+
+#define best_accel() 0
diff --git a/host/include/i386/host/bufferiszero.c.inc b/host/include/i386/host/bufferiszero.c.inc
new file mode 100644
index 0000000..3b9605d
--- /dev/null
+++ b/host/include/i386/host/bufferiszero.c.inc
@@ -0,0 +1,124 @@
+/*
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ * buffer_is_zero acceleration, x86 version.
+ */
+
+#if defined(CONFIG_AVX2_OPT) || defined(__SSE2__)
+#include <immintrin.h>
+
+/* Helper for preventing the compiler from reassociating
+   chains of binary vector operations.  */
+#define SSE_REASSOC_BARRIER(vec0, vec1) asm("" : "+x"(vec0), "+x"(vec1))
+
+/* Note that these vectorized functions may assume len >= 256.  */
+
+static bool __attribute__((target("sse2")))
+buffer_zero_sse2(const void *buf, size_t len)
+{
+    /* Unaligned loads at head/tail.  */
+    __m128i v = *(__m128i_u *)(buf);
+    __m128i w = *(__m128i_u *)(buf + len - 16);
+    /* Align head/tail to 16-byte boundaries.  */
+    const __m128i *p = QEMU_ALIGN_PTR_DOWN(buf + 16, 16);
+    const __m128i *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 16);
+    __m128i zero = { 0 };
+
+    /* Collect a partial block at tail end.  */
+    v |= e[-1]; w |= e[-2];
+    SSE_REASSOC_BARRIER(v, w);
+    v |= e[-3]; w |= e[-4];
+    SSE_REASSOC_BARRIER(v, w);
+    v |= e[-5]; w |= e[-6];
+    SSE_REASSOC_BARRIER(v, w);
+    v |= e[-7]; v |= w;
+
+    /*
+     * Loop over complete 128-byte blocks.
+     * With the head and tail removed, e - p >= 14, so the loop
+     * must iterate at least once.
+     */
+    do {
+        v = _mm_cmpeq_epi8(v, zero);
+        if (unlikely(_mm_movemask_epi8(v) != 0xFFFF)) {
+            return false;
+        }
+        v = p[0]; w = p[1];
+        SSE_REASSOC_BARRIER(v, w);
+        v |= p[2]; w |= p[3];
+        SSE_REASSOC_BARRIER(v, w);
+        v |= p[4]; w |= p[5];
+        SSE_REASSOC_BARRIER(v, w);
+        v |= p[6]; w |= p[7];
+        SSE_REASSOC_BARRIER(v, w);
+        v |= w;
+        p += 8;
+    } while (p < e - 7);
+
+    return _mm_movemask_epi8(_mm_cmpeq_epi8(v, zero)) == 0xFFFF;
+}
+
+#ifdef CONFIG_AVX2_OPT
+static bool __attribute__((target("avx2")))
+buffer_zero_avx2(const void *buf, size_t len)
+{
+    /* Unaligned loads at head/tail.  */
+    __m256i v = *(__m256i_u *)(buf);
+    __m256i w = *(__m256i_u *)(buf + len - 32);
+    /* Align head/tail to 32-byte boundaries.  */
+    const __m256i *p = QEMU_ALIGN_PTR_DOWN(buf + 32, 32);
+    const __m256i *e = QEMU_ALIGN_PTR_DOWN(buf + len - 1, 32);
+    __m256i zero = { 0 };
+
+    /* Collect a partial block at tail end.  */
+    v |= e[-1]; w |= e[-2];
+    SSE_REASSOC_BARRIER(v, w);
+    v |= e[-3]; w |= e[-4];
+    SSE_REASSOC_BARRIER(v, w);
+    v |= e[-5]; w |= e[-6];
+    SSE_REASSOC_BARRIER(v, w);
+    v |= e[-7]; v |= w;
+
+    /* Loop over complete 256-byte blocks.  */
+    for (; p < e - 7; p += 8) {
+        /* PTEST is not profitable here.  */
+        v = _mm256_cmpeq_epi8(v, zero);
+        if (unlikely(_mm256_movemask_epi8(v) != 0xFFFFFFFF)) {
+            return false;
+        }
+        v = p[0]; w = p[1];
+        SSE_REASSOC_BARRIER(v, w);
+        v |= p[2]; w |= p[3];
+        SSE_REASSOC_BARRIER(v, w);
+        v |= p[4]; w |= p[5];
+        SSE_REASSOC_BARRIER(v, w);
+        v |= p[6]; w |= p[7];
+        SSE_REASSOC_BARRIER(v, w);
+        v |= w;
+    }
+
+    return _mm256_movemask_epi8(_mm256_cmpeq_epi8(v, zero)) == 0xFFFFFFFF;
+}
+#endif /* CONFIG_AVX2_OPT */
+
+static biz_accel_fn const accel_table[] = {
+    buffer_is_zero_int_ge256,
+    buffer_zero_sse2,
+#ifdef CONFIG_AVX2_OPT
+    buffer_zero_avx2,
+#endif
+};
+
+static unsigned best_accel(void)
+{
+#ifdef CONFIG_AVX2_OPT
+    unsigned info = cpuinfo_init();
+    if (info & CPUINFO_AVX2) {
+        return 2;
+    }
+#endif
+    return 1;
+}
+
+#else
+# include "host/include/generic/host/bufferiszero.c.inc"
+#endif
diff --git a/host/include/x86_64/host/bufferiszero.c.inc b/host/include/x86_64/host/bufferiszero.c.inc
new file mode 100644
index 0000000..1d3f1fd
--- /dev/null
+++ b/host/include/x86_64/host/bufferiszero.c.inc
@@ -0,0 +1 @@
+#include "host/include/i386/host/bufferiszero.c.inc"