[AArch64] Add SME implementation of ScaleRowDown2

Including associated changes for adding a new scale_sme.cc file.

There is no benefit from an SVE version of this kernel for devices with
an SVE vector length of 128-bits, so skip directly to SME instead.  We
do not use the ZA tile here, so this is a purely streaming-SVE (SSVE)
implementation.

Change-Id: I47d149613fbabd8c203605a809811f1a668e8fb7
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/5913883
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: Justin Green <greenjustin@google.com>

BUILD.gn

@@ -263,6 +263,7 @@ if (libyuv_use_sme) {
     sources = [
       "source/rotate_sme.cc",
       "source/row_sme.cc",
+      "source/scale_sme.cc",
     ]
 
     deps = [ ":libyuv_internal" ]

CMakeLists.txt

@@ -130,7 +130,8 @@ int main(void) { return 0; }
       # Enable AArch64 SME kernels.
       add_library(${ly_lib_name}_sme OBJECT
         ${ly_src_dir}/rotate_sme.cc
-        ${ly_src_dir}/row_sme.cc)
+        ${ly_src_dir}/row_sme.cc
+        ${ly_src_dir}/scale_sme.cc)
       target_compile_options(${ly_lib_name}_sme PRIVATE -march=armv9-a+sme)
       list(APPEND ly_lib_parts $<TARGET_OBJECTS:${ly_lib_name}_sme>)
     else()

include/libyuv/scale_row.h

@@ -116,6 +116,12 @@ extern "C" {
 #define HAS_SCALEUVROWUP2_BILINEAR_16_NEON
 #endif
 
+// The following are available on AArch64 SME platforms:
+#if !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) && \
+    defined(__aarch64__)
+#define HAS_SCALEROWDOWN2_SME
+#endif
+
 #if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
 #define HAS_SCALEADDROW_MSA
 #define HAS_SCALEARGBCOLS_MSA
@@ -1397,13 +1403,15 @@ void ScaleUVRowUp2_Bilinear_16_Any_NEON(const uint16_t* src_ptr,
                                         int dst_width);
 
 // ScaleRowDown2Box also used by planar functions
-// NEON downscalers with interpolation.
+// NEON/SME downscalers with interpolation.
-
-// Note - not static due to reuse in convert for 444 to 420.
 void ScaleRowDown2_NEON(const uint8_t* src_ptr,
                         ptrdiff_t src_stride,
                         uint8_t* dst,
                         int dst_width);
+void ScaleRowDown2_SME(const uint8_t* src_ptr,
+                       ptrdiff_t src_stride,
+                       uint8_t* dst,
+                       int dst_width);
 void ScaleRowDown2Linear_NEON(const uint8_t* src_ptr,
                               ptrdiff_t src_stride,
                               uint8_t* dst,

source/scale.cc

@@ -74,6 +74,13 @@ static void ScalePlaneDown2(int src_width,
     }
   }
 #endif
+#if defined(HAS_SCALEROWDOWN2_SME)
+  if (TestCpuFlag(kCpuHasSME)) {
+    if (filtering == kFilterNone) {
+      ScaleRowDown2 = ScaleRowDown2_SME;
+    }
+  }
+#endif
 #if defined(HAS_SCALEROWDOWN2_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3)) {
     ScaleRowDown2 =

source/scale_sme.cc

@@ -0,0 +1,65 @@
+/*
+ *  Copyright 2024 The LibYuv Project Authors. All rights reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS. All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "libyuv/scale_row.h"
+
+#ifdef __cplusplus
+namespace libyuv {
+extern "C" {
+#endif
+
+#if !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) && \
+    defined(__aarch64__)
+
+__arm_locally_streaming void ScaleRowDown2_SME(const uint8_t* src_ptr,
+                                               ptrdiff_t src_stride,
+                                               uint8_t* dst,
+                                               int dst_width) {
+  // Streaming-SVE only, no use of ZA tile.
+  (void)src_stride;
+  int vl;
+  asm volatile(
+      "cntb     %x[vl]                                       \n"
+      "subs     %w[dst_width], %w[dst_width], %w[vl]         \n"
+      "b.lt     2f                                           \n"
+
+      "1:                                                    \n"
+      "ptrue    p0.b                                         \n"
+      "ld2b     {z0.b, z1.b}, p0/z, [%[src_ptr]]             \n"
+      "incb     %[src_ptr], all, mul #2                      \n"
+      "subs     %w[dst_width], %w[dst_width], %w[vl]         \n"
+      "st1b     {z1.b}, p0, [%[dst_ptr]]                     \n"
+      "incb     %[dst_ptr]                                   \n"
+      "b.ge     1b                                           \n"
+
+      "2:                                                    \n"
+      "adds     %w[dst_width], %w[dst_width], %w[vl]         \n"
+      "b.eq     99f                                          \n"
+
+      "whilelt  p0.b, wzr, %w[dst_width]                     \n"
+      "ld2b     {z0.b, z1.b}, p0/z, [%[src_ptr]]             \n"
+      "st1b     {z1.b}, p0, [%[dst_ptr]]                     \n"
+
+      "99:                                                   \n"
+      : [src_ptr] "+r"(src_ptr),      // %[src_ptr]
+        [dst_ptr] "+r"(dst),          // %[dst_ptr]
+        [dst_width] "+r"(dst_width),  // %[dst_width]
+        [vl] "=r"(vl)                 // %[vl]
+      :
+      : "memory", "cc", "z0", "z1", "p0");
+}
+
+#endif  // !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) &&
+        // defined(__aarch64__)
+
+#ifdef __cplusplus
+}  // extern "C"
+}  // namespace libyuv
+#endif