NEON versions of DWCONV2D CHW 3x3p1 micro-kernels
PiperOrigin-RevId: 338809743
diff --git a/src/f32-dwconv2d-chw/gen/3x3p1-minmax-neonfma-1x4.c b/src/f32-dwconv2d-chw/gen/3x3p1-minmax-neonfma-1x4.c
index 201e3b6..d09e595 100644
--- a/src/f32-dwconv2d-chw/gen/3x3p1-minmax-neonfma-1x4.c
+++ b/src/f32-dwconv2d-chw/gen/3x3p1-minmax-neonfma-1x4.c
@@ -1,5 +1,5 @@
// Auto-generated file. Do not edit!
-// Template: src/f32-dwconv2d-chw/3x3p1-neonfma.c.in
+// Template: src/f32-dwconv2d-chw/3x3p1-neon.c.in
// Generator: tools/xngen
//
// Copyright 2020 Google LLC
@@ -62,15 +62,15 @@
size_t w = input_width;
for (; w > 4 * sizeof(float); w -= 4 * sizeof(float)) {
- float32x4_t vo0p0 = vdupq_laneq_f32(vw0123, 0);
+ float32x4_t vo0p0 = vdupq_lane_f32(vget_low_f32(vw0123), 0);
const float32x4_t vi0x89AB = vld1q_f32(i0); i0 += 4;
const float32x4_t vi1x89AB = vld1q_f32(i1); i1 += 4;
const float32x4_t vi2x89AB = vld1q_f32(i2); i2 += 4;
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi0x4567, vw0123, 2);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi0x4567, vget_high_f32(vw0123), 0);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi1x4567, vw4567, 1);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi1x4567, vget_low_f32(vw4567), 1);
vo0p0 = vfmaq_lane_f32(vo0p0, vi2x4567, vw89, 0);
@@ -78,11 +78,11 @@
const float32x4_t vi1x3456 = vextq_f32(vi1x0123, vi1x4567, 3);
const float32x4_t vi2x3456 = vextq_f32(vi2x0123, vi2x4567, 3);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi0x3456, vw0123, 1);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi0x3456, vget_low_f32(vw0123), 1);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi1x3456, vw4567, 0);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi1x3456, vget_low_f32(vw4567), 0);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi2x3456, vw4567, 3);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi2x3456, vget_high_f32(vw4567), 1);
vi0x0123 = vi0x4567;
vi1x0123 = vi1x4567;
@@ -92,9 +92,9 @@
const float32x4_t vi1x5678 = vextq_f32(vi1x4567, vi1x89AB, 1);
const float32x4_t vi2x5678 = vextq_f32(vi2x4567, vi2x89AB, 1);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi0x5678, vw0123, 3);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi0x5678, vget_high_f32(vw0123), 1);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi1x5678, vw4567, 2);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi1x5678, vget_high_f32(vw4567), 0);
vo0p0 = vfmaq_lane_f32(vo0p0, vi2x5678, vw89, 1);
@@ -113,15 +113,15 @@
assert(w >= 1 * sizeof(float));
assert(w <= 4 * sizeof(float));
{
- float32x4_t vo0p0 = vdupq_laneq_f32(vw0123, 0);
+ float32x4_t vo0p0 = vdupq_lane_f32(vget_low_f32(vw0123), 0);
vi0x4567 = vreinterpretq_f32_u32(vandq_u32(vmask, vreinterpretq_u32_f32(vi0x4567)));
vi1x4567 = vreinterpretq_f32_u32(vandq_u32(vmask, vreinterpretq_u32_f32(vi1x4567)));
vi2x4567 = vreinterpretq_f32_u32(vandq_u32(vmask, vreinterpretq_u32_f32(vi2x4567)));
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi0x4567, vw0123, 2);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi0x4567, vget_high_f32(vw0123), 0);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi1x4567, vw4567, 1);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi1x4567, vget_low_f32(vw4567), 1);
vo0p0 = vfmaq_lane_f32(vo0p0, vi2x4567, vw89, 0);
@@ -129,20 +129,20 @@
const float32x4_t vi1x3456 = vextq_f32(vi1x0123, vi1x4567, 3);
const float32x4_t vi2x3456 = vextq_f32(vi2x0123, vi2x4567, 3);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi0x3456, vw0123, 1);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi0x3456, vget_low_f32(vw0123), 1);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi1x3456, vw4567, 0);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi1x3456, vget_low_f32(vw4567), 0);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi2x3456, vw4567, 3);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi2x3456, vget_high_f32(vw4567), 1);
const float32x4_t vzero = vmovq_n_f32(0.0f);
const float32x4_t vi0x5678 = vextq_f32(vi0x4567, vzero, 1);
const float32x4_t vi1x5678 = vextq_f32(vi1x4567, vzero, 1);
const float32x4_t vi2x5678 = vextq_f32(vi2x4567, vzero, 1);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi0x5678, vw0123, 3);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi0x5678, vget_high_f32(vw0123), 1);
- vo0p0 = vfmaq_laneq_f32(vo0p0, vi1x5678, vw4567, 2);
+ vo0p0 = vfmaq_lane_f32(vo0p0, vi1x5678, vget_high_f32(vw4567), 0);
vo0p0 = vfmaq_lane_f32(vo0p0, vi2x5678, vw89, 1);