Runtime helpers for 128-bit vectors

ocaml/runtime/Makefile

@@ -23,15 +23,15 @@ BYTECODE_C_SOURCES := $(addsuffix .c, \
   interp misc stacks fix_code startup_aux startup_byt freelist major_gc \
   minor_gc memory alloc roots_byt globroots fail_byt signals \
   signals_byt printexc backtrace_byt backtrace compare ints eventlog \
-  floats str array io extern intern hash sys meta parsing gc_ctrl md5 obj \
+  floats simd str array io extern intern hash sys meta parsing gc_ctrl md5 obj \
   lexing callback debugger weak compact finalise custom dynlink \
   afl $(UNIX_OR_WIN32) bigarray main memprof domain \
   skiplist codefrag)
 
 NATIVE_C_SOURCES := $(addsuffix .c, \
   startup_aux startup_nat main fail_nat roots_nat signals \
   signals_nat misc freelist major_gc minor_gc memory alloc compare ints \
-  floats str array io extern intern hash sys parsing gc_ctrl eventlog md5 obj \
+  floats simd str array io extern intern hash sys parsing gc_ctrl eventlog md5 obj \
   lexing $(UNIX_OR_WIN32) printexc callback weak compact finalise custom \
   globroots backtrace_nat backtrace dynlink_nat debugger meta \
   dynlink clambda_checks afl bigarray \

ocaml/runtime/caml/simd.h

@@ -0,0 +1,46 @@
+/**************************************************************************/
+/*                                                                        */
+/*                                 OCaml                                  */
+/*                                                                        */
+/*                      Max Slater, Jane Street                           */
+/*                                                                        */
+/*   Copyright 2023 Jane Street Group LLC                                 */
+/*                                                                        */
+/*   All rights reserved.  This file is distributed under the terms of    */
+/*   the GNU Lesser General Public License version 2.1, with the          */
+/*   special exception on linking described in the file LICENSE.          */
+/*                                                                        */
+/**************************************************************************/
+
+/* SIMD vector instruction support */
+
+#ifndef CAML_SIMD_H
+#define CAML_SIMD_H
+
+#include "mlvalues.h"
+
+#if defined(_M_IX86_FP) || defined(__SSE2__) || defined(__SSE3__) || \
+    defined(__SSSE3__) || defined(__SSE4_1__) || defined(__SSE4_2__)
+#define ARCH_SSE2
+#endif
+
+#if defined(__AVX__) || defined(__AVX2__)
+#define ARCH_AVX
+#endif
+
+#ifdef ARCH_SSE2
+#include <emmintrin.h>
+
+#define Vec128_val(v)  _mm_loadu_ps((const float*)Bp_val(v))
+#define Vec128_vald(v) _mm_loadu_pd((const double*)Bp_val(v))
+#define Vec128_vali(v) _mm_loadu_si128((const __m128i*)Bp_val(v))
+#define Store_vec128_val(v,x)  _mm_storeu_ps((float*)Bp_val(v), x)
+#define Store_vec128_vald(v,x) _mm_storeu_pd((double*)Bp_val(v), x)
+#define Store_vec128_vali(v,x) _mm_storeu_si128((__m128i*)Bp_val(v), x)
+
+CAMLextern value caml_copy_vec128(__m128);
+CAMLextern value caml_copy_vec128i(__m128i);
+CAMLextern value caml_copy_vec128d(__m128d);
+#endif
+
+#endif /* CAML_SIMD_H */

ocaml/runtime/simd.c

@@ -0,0 +1,40 @@
+/**************************************************************************/
+/*                                                                        */
+/*                                 OCaml                                  */
+/*                                                                        */
+/*                      Max Slater, Jane Street                           */
+/*                                                                        */
+/*   Copyright 2023 Jane Street Group LLC                                 */
+/*                                                                        */
+/*   All rights reserved.  This file is distributed under the terms of    */
+/*   the GNU Lesser General Public License version 2.1, with the          */
+/*   special exception on linking described in the file LICENSE.          */
+/*                                                                        */
+/**************************************************************************/
+
+#define CAML_INTERNALS
+
+#include "caml/alloc.h"
+#include "caml/simd.h"
+
+#ifdef ARCH_SSE2
+
+CAMLexport value caml_copy_vec128(__m128 v) {
+    value res = caml_alloc_small(2, Abstract_tag);
+    Store_vec128_val(res, v);
+    return res;
+}
+
+CAMLexport value caml_copy_vec128i(__m128i v) {
+    value res = caml_alloc_small(2, Abstract_tag);
+    Store_vec128_vali(res, v);
+    return res;
+}
+
+CAMLexport value caml_copy_vec128d(__m128d v) {
+    value res = caml_alloc_small(2, Abstract_tag);
+    Store_vec128_vald(res, v);
+    return res;
+}
+
+#endif

tests/simd/stubs.c

@@ -1,9 +1,7 @@
 
-#include <caml/mlvalues.h>
-#include <caml/alloc.h>
 #include <caml/memory.h>
+#include <caml/simd.h>
 #include <smmintrin.h>
-#include <emmintrin.h>
 #include <assert.h>
 
 int64_t vec128_low_int64(__m128i v)
@@ -24,15 +22,12 @@ __m128i vec128_of_int64s(int64_t low, int64_t high)
 CAMLprim value boxed_combine(value v0, value v1)
 {
   CAMLparam2(v0, v1);
-  CAMLlocal1(res);
 
-  __m128i l = _mm_loadu_si128((__m128i*)v0);
-  __m128i r = _mm_loadu_si128((__m128i*)v1);
+  __m128i l = Vec128_vali(v0);
+  __m128i r = Vec128_vali(v1);
   __m128i result = _mm_add_epi64(l, r);
-  res = caml_alloc_small(2, Abstract_tag);
-  _mm_storeu_si128((__m128i*)res, result);
 
-  CAMLreturn(res);
+  CAMLreturn(caml_copy_vec128i(result));
 }
 
 __m128i lots_of_vectors(