fftr microkernel single data pointer

- always used in-place so use a single data pointer

PiperOrigin-RevId: 469606317

11 files changed, 223 insertions, 312 deletions

diff --git a/bench/cs16-fftr.cc b/bench/cs16-fftr.cc
index b2d63cff4..e6fb60c61 100644
--- a/bench/cs16-fftr.cc
+++ b/bench/cs16-fftr.cc
@@ -33,17 +33,14 @@ void cs16_fftr(
   assert(samples % 2 == 0);
   const size_t sample_size = samples * 2 + 2;
 
-  std::vector<int16_t, AlignedAllocator<int16_t, 64>> input(
-      sample_size + XNN_EXTRA_BYTES / sizeof(int16_t));
-  std::vector<int16_t, AlignedAllocator<int16_t, 64>> output(sample_size);
+  std::vector<int16_t, AlignedAllocator<int16_t, 64>> data(sample_size + XNN_EXTRA_BYTES / sizeof(int16_t));
   std::vector<int16_t, AlignedAllocator<int16_t, 64>> twiddle(samples);
 
-  std::iota(input.begin(), input.end(), 0);
-  std::iota(output.begin(), output.end(), 1);
+  std::iota(data.begin(), data.end(), 0);
   std::iota(twiddle.begin(), twiddle.end(), 2);
 
   for (auto _ : state) {
-    fftr(samples, input.data(), output.data(), twiddle.data());
+    fftr(samples, data.data(), twiddle.data());
   }
 
   const uint64_t cpu_frequency = benchmark::utils::GetCurrentCpuFrequency();
diff --git a/src/cs16-fftr/gen/scalar-x1.c b/src/cs16-fftr/gen/scalar-x1.c
index c0d3c37fe..2131ed918 100644
--- a/src/cs16-fftr/gen/scalar-x1.c
+++ b/src/cs16-fftr/gen/scalar-x1.c
@@ -4,8 +4,8 @@
 //
 // Copyright 2022 Google LLC
 //
-// This source code is licensed under the BSD-style license found il the
-// LICENSE file il the root directory of this source tree.
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
 
 #include <assert.h>
 #include <stddef.h>
@@ -17,43 +17,39 @@
 
 void xnn_cs16_fftr_ukernel__scalar_x1(
     size_t samples,
-    const int16_t* input,
-    int16_t* output,
+    int16_t* data,
     const int16_t* twiddle) {
 
   assert(samples >= 2);
   assert(samples % 2 == 0);
-  assert(input != NULL);
-  assert(output != NULL);
+  assert(data != NULL);
+  assert(data != NULL);
   assert(twiddle != NULL);
 
-  const int16_t* il = input;
-  const int16_t* ir = input + samples * 2;
-  int32_t vdcr = (int32_t) il[0];
-  int32_t vdci = (int32_t) il[1];
-  il += 2;
+  int16_t* dl = data;
+  int16_t* dr = data + samples * 2;
+  int32_t vdcr = (int32_t) dl[0];
+  int32_t vdci = (int32_t) dl[1];
+
   vdcr = math_asr_s32(vdcr * 16383 + 16384, 15);
   vdci = math_asr_s32(vdci * 16383 + 16384, 15);
 
-  int16_t* ol  = output;
-  int16_t* or = output + samples * 2;
-  ol[0] = vdcr + vdci;
-  ol[1] = 0;
-  ol += 2;
-  or[0] = vdcr - vdci;
-  or[1] = 0;
+  dl[0] = vdcr + vdci;
+  dl[1] = 0;
+  dl += 2;
+  dr[0] = vdcr - vdci;
+  dr[1] = 0;
 
   samples >>= 1;
 
 
   if XNN_UNLIKELY(samples != 0) {
     do {
-      int32_t vilr = il[0];
-      int32_t vili = il[1];
-      il += 2;
-      ir -= 2;
-      int32_t virr =  (int32_t) ir[0];
-      int32_t viri = -(int32_t) ir[1];
+      dr -= 2;
+      int32_t vilr = dl[0];
+      int32_t vili = dl[1];
+      int32_t virr =  (int32_t) dr[0];
+      int32_t viri = -(int32_t) dr[1];
       const int32_t vtwr = twiddle[0];
       const int32_t vtwi = twiddle[1];
       twiddle += 2;
@@ -70,13 +66,11 @@ void xnn_cs16_fftr_ukernel__scalar_x1(
       const int32_t twr = math_asr_s32(vacc2r * vtwr - vacc2i * vtwi + 16384, 15);
       const int32_t twi = math_asr_s32(vacc2r * vtwi + vacc2i * vtwr + 16384, 15);
 
-      ol[0] = math_asr_s32(vacc1r + twr, 1);
-      ol[1] = math_asr_s32(vacc1i + twi, 1);
-      ol += 2;
-      or -= 2;
-      or[0] = math_asr_s32(vacc1r - twr, 1);
-      or[1] = math_asr_s32(twi - vacc1i, 1);
-
+      dl[0] = math_asr_s32(vacc1r + twr, 1);
+      dl[1] = math_asr_s32(vacc1i + twi, 1);
+      dr[0] = math_asr_s32(vacc1r - twr, 1);
+      dr[1] = math_asr_s32(twi - vacc1i, 1);
+      dl += 2;
     } while (--samples != 0);
   }
 }
diff --git a/src/cs16-fftr/gen/scalar-x2.c b/src/cs16-fftr/gen/scalar-x2.c
index 2c602f4c7..83dc7c20c 100644
--- a/src/cs16-fftr/gen/scalar-x2.c
+++ b/src/cs16-fftr/gen/scalar-x2.c
@@ -4,8 +4,8 @@
 //
 // Copyright 2022 Google LLC
 //
-// This source code is licensed under the BSD-style license found il the
-// LICENSE file il the root directory of this source tree.
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
 
 #include <assert.h>
 #include <stddef.h>
@@ -17,45 +17,41 @@
 
 void xnn_cs16_fftr_ukernel__scalar_x2(
     size_t samples,
-    const int16_t* input,
-    int16_t* output,
+    int16_t* data,
     const int16_t* twiddle) {
 
   assert(samples >= 2);
   assert(samples % 2 == 0);
-  assert(input != NULL);
-  assert(output != NULL);
+  assert(data != NULL);
+  assert(data != NULL);
   assert(twiddle != NULL);
 
-  const int16_t* il = input;
-  const int16_t* ir = input + samples * 2;
-  int32_t vdcr = (int32_t) il[0];
-  int32_t vdci = (int32_t) il[1];
-  il += 2;
+  int16_t* dl = data;
+  int16_t* dr = data + samples * 2;
+  int32_t vdcr = (int32_t) dl[0];
+  int32_t vdci = (int32_t) dl[1];
+
   vdcr = math_asr_s32(vdcr * 16383 + 16384, 15);
   vdci = math_asr_s32(vdci * 16383 + 16384, 15);
 
-  int16_t* ol  = output;
-  int16_t* or = output + samples * 2;
-  ol[0] = vdcr + vdci;
-  ol[1] = 0;
-  ol += 2;
-  or[0] = vdcr - vdci;
-  or[1] = 0;
+  dl[0] = vdcr + vdci;
+  dl[1] = 0;
+  dl += 2;
+  dr[0] = vdcr - vdci;
+  dr[1] = 0;
 
   samples >>= 1;
 
   for (; samples >= 2; samples -= 2) {
-    int32_t vilr0 = il[0];
-    int32_t vili0 = il[1];
-    int32_t vilr1 = il[2];
-    int32_t vili1 = il[3];
-    il += 2 * 2;
-    ir -= 2 * 2;
-    int32_t virr0 =  (int32_t) ir[2];
-    int32_t viri0 = -(int32_t) ir[3];
-    int32_t virr1 =  (int32_t) ir[0];
-    int32_t viri1 = -(int32_t) ir[1];
+    dr -= 2 * 2;
+    int32_t vilr0 = dl[0];
+    int32_t vili0 = dl[1];
+    int32_t vilr1 = dl[2];
+    int32_t vili1 = dl[3];
+    int32_t virr0 =  (int32_t) dr[2];
+    int32_t viri0 = -(int32_t) dr[3];
+    int32_t virr1 =  (int32_t) dr[0];
+    int32_t viri1 = -(int32_t) dr[1];
     const int32_t vtwr0 = twiddle[0];
     const int32_t vtwi0 = twiddle[1];
     const int32_t vtwr1 = twiddle[2];
@@ -84,26 +80,24 @@ void xnn_cs16_fftr_ukernel__scalar_x2(
     const int32_t twi0 = math_asr_s32(vacc2r0 * vtwi0 + vacc2i0 * vtwr0 + 16384, 15);
     const int32_t twi1 = math_asr_s32(vacc2r1 * vtwi1 + vacc2i1 * vtwr1 + 16384, 15);
 
-    ol[0] = math_asr_s32(vacc1r0 + twr0, 1);
-    ol[1] = math_asr_s32(vacc1i0 + twi0, 1);
-    ol[2] = math_asr_s32(vacc1r1 + twr1, 1);
-    ol[3] = math_asr_s32(vacc1i1 + twi1, 1);
-    ol += 2 * 2;
-    or -= 2 * 2;
-    or[2] = math_asr_s32(vacc1r0 - twr0, 1);
-    or[3] = math_asr_s32(twi0 - vacc1i0, 1);
-    or[0] = math_asr_s32(vacc1r1 - twr1, 1);
-    or[1] = math_asr_s32(twi1 - vacc1i1, 1);
+    dl[0] = math_asr_s32(vacc1r0 + twr0, 1);
+    dl[1] = math_asr_s32(vacc1i0 + twi0, 1);
+    dl[2] = math_asr_s32(vacc1r1 + twr1, 1);
+    dl[3] = math_asr_s32(vacc1i1 + twi1, 1);
+    dr[2] = math_asr_s32(vacc1r0 - twr0, 1);
+    dr[3] = math_asr_s32(twi0 - vacc1i0, 1);
+    dr[0] = math_asr_s32(vacc1r1 - twr1, 1);
+    dr[1] = math_asr_s32(twi1 - vacc1i1, 1);
+    dl += 2 * 2;
   }
 
   if XNN_UNLIKELY(samples != 0) {
     do {
-      int32_t vilr = il[0];
-      int32_t vili = il[1];
-      il += 2;
-      ir -= 2;
-      int32_t virr =  (int32_t) ir[0];
-      int32_t viri = -(int32_t) ir[1];
+      dr -= 2;
+      int32_t vilr = dl[0];
+      int32_t vili = dl[1];
+      int32_t virr =  (int32_t) dr[0];
+      int32_t viri = -(int32_t) dr[1];
       const int32_t vtwr = twiddle[0];
       const int32_t vtwi = twiddle[1];
       twiddle += 2;
@@ -120,13 +114,11 @@ void xnn_cs16_fftr_ukernel__scalar_x2(
       const int32_t twr = math_asr_s32(vacc2r * vtwr - vacc2i * vtwi + 16384, 15);
       const int32_t twi = math_asr_s32(vacc2r * vtwi + vacc2i * vtwr + 16384, 15);
 
-      ol[0] = math_asr_s32(vacc1r + twr, 1);
-      ol[1] = math_asr_s32(vacc1i + twi, 1);
-      ol += 2;
-      or -= 2;
-      or[0] = math_asr_s32(vacc1r - twr, 1);
-      or[1] = math_asr_s32(twi - vacc1i, 1);
-
+      dl[0] = math_asr_s32(vacc1r + twr, 1);
+      dl[1] = math_asr_s32(vacc1i + twi, 1);
+      dr[0] = math_asr_s32(vacc1r - twr, 1);
+      dr[1] = math_asr_s32(twi - vacc1i, 1);
+      dl += 2;
     } while (--samples != 0);
   }
 }
diff --git a/src/cs16-fftr/gen/scalar-x3.c b/src/cs16-fftr/gen/scalar-x3.c
index 3d492a1ea..24d016848 100644
--- a/src/cs16-fftr/gen/scalar-x3.c
+++ b/src/cs16-fftr/gen/scalar-x3.c
@@ -4,8 +4,8 @@
 //
 // Copyright 2022 Google LLC
 //
-// This source code is licensed under the BSD-style license found il the
-// LICENSE file il the root directory of this source tree.
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
 
 #include <assert.h>
 #include <stddef.h>
@@ -17,49 +17,45 @@
 
 void xnn_cs16_fftr_ukernel__scalar_x3(
     size_t samples,
-    const int16_t* input,
-    int16_t* output,
+    int16_t* data,
     const int16_t* twiddle) {
 
   assert(samples >= 2);
   assert(samples % 2 == 0);
-  assert(input != NULL);
-  assert(output != NULL);
+  assert(data != NULL);
+  assert(data != NULL);
   assert(twiddle != NULL);
 
-  const int16_t* il = input;
-  const int16_t* ir = input + samples * 2;
-  int32_t vdcr = (int32_t) il[0];
-  int32_t vdci = (int32_t) il[1];
-  il += 2;
+  int16_t* dl = data;
+  int16_t* dr = data + samples * 2;
+  int32_t vdcr = (int32_t) dl[0];
+  int32_t vdci = (int32_t) dl[1];
+
   vdcr = math_asr_s32(vdcr * 16383 + 16384, 15);
   vdci = math_asr_s32(vdci * 16383 + 16384, 15);
 
-  int16_t* ol  = output;
-  int16_t* or = output + samples * 2;
-  ol[0] = vdcr + vdci;
-  ol[1] = 0;
-  ol += 2;
-  or[0] = vdcr - vdci;
-  or[1] = 0;
+  dl[0] = vdcr + vdci;
+  dl[1] = 0;
+  dl += 2;
+  dr[0] = vdcr - vdci;
+  dr[1] = 0;
 
   samples >>= 1;
 
   for (; samples >= 3; samples -= 3) {
-    int32_t vilr0 = il[0];
-    int32_t vili0 = il[1];
-    int32_t vilr1 = il[2];
-    int32_t vili1 = il[3];
-    int32_t vilr2 = il[4];
-    int32_t vili2 = il[5];
-    il += 3 * 2;
-    ir -= 3 * 2;
-    int32_t virr0 =  (int32_t) ir[4];
-    int32_t viri0 = -(int32_t) ir[5];
-    int32_t virr1 =  (int32_t) ir[2];
-    int32_t viri1 = -(int32_t) ir[3];
-    int32_t virr2 =  (int32_t) ir[0];
-    int32_t viri2 = -(int32_t) ir[1];
+    dr -= 3 * 2;
+    int32_t vilr0 = dl[0];
+    int32_t vili0 = dl[1];
+    int32_t vilr1 = dl[2];
+    int32_t vili1 = dl[3];
+    int32_t vilr2 = dl[4];
+    int32_t vili2 = dl[5];
+    int32_t virr0 =  (int32_t) dr[4];
+    int32_t viri0 = -(int32_t) dr[5];
+    int32_t virr1 =  (int32_t) dr[2];
+    int32_t viri1 = -(int32_t) dr[3];
+    int32_t virr2 =  (int32_t) dr[0];
+    int32_t viri2 = -(int32_t) dr[1];
     const int32_t vtwr0 = twiddle[0];
     const int32_t vtwi0 = twiddle[1];
     const int32_t vtwr1 = twiddle[2];
@@ -100,30 +96,28 @@ void xnn_cs16_fftr_ukernel__scalar_x3(
     const int32_t twi1 = math_asr_s32(vacc2r1 * vtwi1 + vacc2i1 * vtwr1 + 16384, 15);
     const int32_t twi2 = math_asr_s32(vacc2r2 * vtwi2 + vacc2i2 * vtwr2 + 16384, 15);
 
-    ol[0] = math_asr_s32(vacc1r0 + twr0, 1);
-    ol[1] = math_asr_s32(vacc1i0 + twi0, 1);
-    ol[2] = math_asr_s32(vacc1r1 + twr1, 1);
-    ol[3] = math_asr_s32(vacc1i1 + twi1, 1);
-    ol[4] = math_asr_s32(vacc1r2 + twr2, 1);
-    ol[5] = math_asr_s32(vacc1i2 + twi2, 1);
-    ol += 3 * 2;
-    or -= 3 * 2;
-    or[4] = math_asr_s32(vacc1r0 - twr0, 1);
-    or[5] = math_asr_s32(twi0 - vacc1i0, 1);
-    or[2] = math_asr_s32(vacc1r1 - twr1, 1);
-    or[3] = math_asr_s32(twi1 - vacc1i1, 1);
-    or[0] = math_asr_s32(vacc1r2 - twr2, 1);
-    or[1] = math_asr_s32(twi2 - vacc1i2, 1);
+    dl[0] = math_asr_s32(vacc1r0 + twr0, 1);
+    dl[1] = math_asr_s32(vacc1i0 + twi0, 1);
+    dl[2] = math_asr_s32(vacc1r1 + twr1, 1);
+    dl[3] = math_asr_s32(vacc1i1 + twi1, 1);
+    dl[4] = math_asr_s32(vacc1r2 + twr2, 1);
+    dl[5] = math_asr_s32(vacc1i2 + twi2, 1);
+    dr[4] = math_asr_s32(vacc1r0 - twr0, 1);
+    dr[5] = math_asr_s32(twi0 - vacc1i0, 1);
+    dr[2] = math_asr_s32(vacc1r1 - twr1, 1);
+    dr[3] = math_asr_s32(twi1 - vacc1i1, 1);
+    dr[0] = math_asr_s32(vacc1r2 - twr2, 1);
+    dr[1] = math_asr_s32(twi2 - vacc1i2, 1);
+    dl += 3 * 2;
   }
 
   if XNN_UNLIKELY(samples != 0) {
     do {
-      int32_t vilr = il[0];
-      int32_t vili = il[1];
-      il += 2;
-      ir -= 2;
-      int32_t virr =  (int32_t) ir[0];
-      int32_t viri = -(int32_t) ir[1];
+      dr -= 2;
+      int32_t vilr = dl[0];
+      int32_t vili = dl[1];
+      int32_t virr =  (int32_t) dr[0];
+      int32_t viri = -(int32_t) dr[1];
       const int32_t vtwr = twiddle[0];
       const int32_t vtwi = twiddle[1];
       twiddle += 2;
@@ -140,13 +134,11 @@ void xnn_cs16_fftr_ukernel__scalar_x3(
       const int32_t twr = math_asr_s32(vacc2r * vtwr - vacc2i * vtwi + 16384, 15);
       const int32_t twi = math_asr_s32(vacc2r * vtwi + vacc2i * vtwr + 16384, 15);
 
-      ol[0] = math_asr_s32(vacc1r + twr, 1);
-      ol[1] = math_asr_s32(vacc1i + twi, 1);
-      ol += 2;
-      or -= 2;
-      or[0] = math_asr_s32(vacc1r - twr, 1);
-      or[1] = math_asr_s32(twi - vacc1i, 1);
-
+      dl[0] = math_asr_s32(vacc1r + twr, 1);
+      dl[1] = math_asr_s32(vacc1i + twi, 1);
+      dr[0] = math_asr_s32(vacc1r - twr, 1);
+      dr[1] = math_asr_s32(twi - vacc1i, 1);
+      dl += 2;
     } while (--samples != 0);
   }
 }
diff --git a/src/cs16-fftr/gen/scalar-x4.c b/src/cs16-fftr/gen/scalar-x4.c
index 409920e86..be23e44c1 100644
--- a/src/cs16-fftr/gen/scalar-x4.c
+++ b/src/cs16-fftr/gen/scalar-x4.c
@@ -4,8 +4,8 @@
 //
 // Copyright 2022 Google LLC
 //
-// This source code is licensed under the BSD-style license found il the
-// LICENSE file il the root directory of this source tree.
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
 
 #include <assert.h>
 #include <stddef.h>
@@ -17,53 +17,49 @@
 
 void xnn_cs16_fftr_ukernel__scalar_x4(
     size_t samples,
-    const int16_t* input,
-    int16_t* output,
+    int16_t* data,
     const int16_t* twiddle) {
 
   assert(samples >= 2);
   assert(samples % 2 == 0);
-  assert(input != NULL);
-  assert(output != NULL);
+  assert(data != NULL);
+  assert(data != NULL);
   assert(twiddle != NULL);
 
-  const int16_t* il = input;
-  const int16_t* ir = input + samples * 2;
-  int32_t vdcr = (int32_t) il[0];
-  int32_t vdci = (int32_t) il[1];
-  il += 2;
+  int16_t* dl = data;
+  int16_t* dr = data + samples * 2;
+  int32_t vdcr = (int32_t) dl[0];
+  int32_t vdci = (int32_t) dl[1];
+
   vdcr = math_asr_s32(vdcr * 16383 + 16384, 15);
   vdci = math_asr_s32(vdci * 16383 + 16384, 15);
 
-  int16_t* ol  = output;
-  int16_t* or = output + samples * 2;
-  ol[0] = vdcr + vdci;
-  ol[1] = 0;
-  ol += 2;
-  or[0] = vdcr - vdci;
-  or[1] = 0;
+  dl[0] = vdcr + vdci;
+  dl[1] = 0;
+  dl += 2;
+  dr[0] = vdcr - vdci;
+  dr[1] = 0;
 
   samples >>= 1;
 
   for (; samples >= 4; samples -= 4) {
-    int32_t vilr0 = il[0];
-    int32_t vili0 = il[1];
-    int32_t vilr1 = il[2];
-    int32_t vili1 = il[3];
-    int32_t vilr2 = il[4];
-    int32_t vili2 = il[5];
-    int32_t vilr3 = il[6];
-    int32_t vili3 = il[7];
-    il += 4 * 2;
-    ir -= 4 * 2;
-    int32_t virr0 =  (int32_t) ir[6];
-    int32_t viri0 = -(int32_t) ir[7];
-    int32_t virr1 =  (int32_t) ir[4];
-    int32_t viri1 = -(int32_t) ir[5];
-    int32_t virr2 =  (int32_t) ir[2];
-    int32_t viri2 = -(int32_t) ir[3];
-    int32_t virr3 =  (int32_t) ir[0];
-    int32_t viri3 = -(int32_t) ir[1];
+    dr -= 4 * 2;
+    int32_t vilr0 = dl[0];
+    int32_t vili0 = dl[1];
+    int32_t vilr1 = dl[2];
+    int32_t vili1 = dl[3];
+    int32_t vilr2 = dl[4];
+    int32_t vili2 = dl[5];
+    int32_t vilr3 = dl[6];
+    int32_t vili3 = dl[7];
+    int32_t virr0 =  (int32_t) dr[6];
+    int32_t viri0 = -(int32_t) dr[7];
+    int32_t virr1 =  (int32_t) dr[4];
+    int32_t viri1 = -(int32_t) dr[5];
+    int32_t virr2 =  (int32_t) dr[2];
+    int32_t viri2 = -(int32_t) dr[3];
+    int32_t virr3 =  (int32_t) dr[0];
+    int32_t viri3 = -(int32_t) dr[1];
     const int32_t vtwr0 = twiddle[0];
     const int32_t vtwi0 = twiddle[1];
     const int32_t vtwr1 = twiddle[2];
@@ -116,34 +112,32 @@ void xnn_cs16_fftr_ukernel__scalar_x4(
     const int32_t twi2 = math_asr_s32(vacc2r2 * vtwi2 + vacc2i2 * vtwr2 + 16384, 15);
     const int32_t twi3 = math_asr_s32(vacc2r3 * vtwi3 + vacc2i3 * vtwr3 + 16384, 15);
 
-    ol[0] = math_asr_s32(vacc1r0 + twr0, 1);
-    ol[1] = math_asr_s32(vacc1i0 + twi0, 1);
-    ol[2] = math_asr_s32(vacc1r1 + twr1, 1);
-    ol[3] = math_asr_s32(vacc1i1 + twi1, 1);
-    ol[4] = math_asr_s32(vacc1r2 + twr2, 1);
-    ol[5] = math_asr_s32(vacc1i2 + twi2, 1);
-    ol[6] = math_asr_s32(vacc1r3 + twr3, 1);
-    ol[7] = math_asr_s32(vacc1i3 + twi3, 1);
-    ol += 4 * 2;
-    or -= 4 * 2;
-    or[6] = math_asr_s32(vacc1r0 - twr0, 1);
-    or[7] = math_asr_s32(twi0 - vacc1i0, 1);
-    or[4] = math_asr_s32(vacc1r1 - twr1, 1);
-    or[5] = math_asr_s32(twi1 - vacc1i1, 1);
-    or[2] = math_asr_s32(vacc1r2 - twr2, 1);
-    or[3] = math_asr_s32(twi2 - vacc1i2, 1);
-    or[0] = math_asr_s32(vacc1r3 - twr3, 1);
-    or[1] = math_asr_s32(twi3 - vacc1i3, 1);
+    dl[0] = math_asr_s32(vacc1r0 + twr0, 1);
+    dl[1] = math_asr_s32(vacc1i0 + twi0, 1);
+    dl[2] = math_asr_s32(vacc1r1 + twr1, 1);
+    dl[3] = math_asr_s32(vacc1i1 + twi1, 1);
+    dl[4] = math_asr_s32(vacc1r2 + twr2, 1);
+    dl[5] = math_asr_s32(vacc1i2 + twi2, 1);
+    dl[6] = math_asr_s32(vacc1r3 + twr3, 1);
+    dl[7] = math_asr_s32(vacc1i3 + twi3, 1);
+    dr[6] = math_asr_s32(vacc1r0 - twr0, 1);
+    dr[7] = math_asr_s32(twi0 - vacc1i0, 1);
+    dr[4] = math_asr_s32(vacc1r1 - twr1, 1);
+    dr[5] = math_asr_s32(twi1 - vacc1i1, 1);
+    dr[2] = math_asr_s32(vacc1r2 - twr2, 1);
+    dr[3] = math_asr_s32(twi2 - vacc1i2, 1);
+    dr[0] = math_asr_s32(vacc1r3 - twr3, 1);
+    dr[1] = math_asr_s32(twi3 - vacc1i3, 1);
+    dl += 4 * 2;
   }
 
   if XNN_UNLIKELY(samples != 0) {
     do {
-      int32_t vilr = il[0];
-      int32_t vili = il[1];
-      il += 2;
-      ir -= 2;
-      int32_t virr =  (int32_t) ir[0];
-      int32_t viri = -(int32_t) ir[1];
+      dr -= 2;
+      int32_t vilr = dl[0];
+      int32_t vili = dl[1];
+      int32_t virr =  (int32_t) dr[0];
+      int32_t viri = -(int32_t) dr[1];
       const int32_t vtwr = twiddle[0];
       const int32_t vtwi = twiddle[1];
       twiddle += 2;
@@ -160,13 +154,11 @@ void xnn_cs16_fftr_ukernel__scalar_x4(
       const int32_t twr = math_asr_s32(vacc2r * vtwr - vacc2i * vtwi + 16384, 15);
       const int32_t twi = math_asr_s32(vacc2r * vtwi + vacc2i * vtwr + 16384, 15);
 
-      ol[0] = math_asr_s32(vacc1r + twr, 1);
-      ol[1] = math_asr_s32(vacc1i + twi, 1);
-      ol += 2;
-      or -= 2;
-      or[0] = math_asr_s32(vacc1r - twr, 1);
-      or[1] = math_asr_s32(twi - vacc1i, 1);
-
+      dl[0] = math_asr_s32(vacc1r + twr, 1);
+      dl[1] = math_asr_s32(vacc1i + twi, 1);
+      dr[0] = math_asr_s32(vacc1r - twr, 1);
+      dr[1] = math_asr_s32(twi - vacc1i, 1);
+      dl += 2;
     } while (--samples != 0);
   }
 }
diff --git a/src/cs16-fftr/scalar.c.in b/src/cs16-fftr/scalar.c.in
index 1b592288d..8af6c6c88 100644
--- a/src/cs16-fftr/scalar.c.in
+++ b/src/cs16-fftr/scalar.c.in
@@ -1,7 +1,7 @@
 // Copyright 2022 Google LLC
 //
-// This source code is licensed under the BSD-style license found il the
-// LICENSE file il the root directory of this source tree.
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
 
 $assert SAMPLE_TILE >= 1
 #include <assert.h>
@@ -14,44 +14,40 @@ $assert SAMPLE_TILE >= 1
 
 void xnn_cs16_fftr_ukernel__scalar_x${SAMPLE_TILE}(
     size_t samples,
-    const int16_t* input,
-    int16_t* output,
+    int16_t* data,
     const int16_t* twiddle) {
 
   assert(samples >= 2);
   assert(samples % 2 == 0);
-  assert(input != NULL);
-  assert(output != NULL);
+  assert(data != NULL);
+  assert(data != NULL);
   assert(twiddle != NULL);
 
-  const int16_t* il = input;
-  const int16_t* ir = input + samples * 2;
-  int32_t vdcr = (int32_t) il[0];
-  int32_t vdci = (int32_t) il[1];
-  il += 2;
+  int16_t* dl = data;
+  int16_t* dr = data + samples * 2;
+  int32_t vdcr = (int32_t) dl[0];
+  int32_t vdci = (int32_t) dl[1];
+
   vdcr = math_asr_s32(vdcr * 16383 + 16384, 15);
   vdci = math_asr_s32(vdci * 16383 + 16384, 15);
 
-  int16_t* ol  = output;
-  int16_t* or = output + samples * 2;
-  ol[0] = vdcr + vdci;
-  ol[1] = 0;
-  ol += 2;
-  or[0] = vdcr - vdci;
-  or[1] = 0;
+  dl[0] = vdcr + vdci;
+  dl[1] = 0;
+  dl += 2;
+  dr[0] = vdcr - vdci;
+  dr[1] = 0;
 
   samples >>= 1;
 
   $if SAMPLE_TILE > 1:
     for (; samples >= ${SAMPLE_TILE}; samples -= ${SAMPLE_TILE}) {
+      dr -= ${SAMPLE_TILE} * 2;
       $for C in range(SAMPLE_TILE):
-        int32_t vilr${C} = il[${C * 2 + 0}];
-        int32_t vili${C} = il[${C * 2 + 1}];
-      il += ${SAMPLE_TILE} * 2;
-      ir -= ${SAMPLE_TILE} * 2;
+        int32_t vilr${C} = dl[${C * 2 + 0}];
+        int32_t vili${C} = dl[${C * 2 + 1}];
       $for C in range(SAMPLE_TILE):
-        int32_t virr${C} =  (int32_t) ir[${(SAMPLE_TILE - 1 - C) * 2 + 0}];
-        int32_t viri${C} = -(int32_t) ir[${(SAMPLE_TILE - 1 - C) * 2 + 1}];
+        int32_t virr${C} =  (int32_t) dr[${(SAMPLE_TILE - 1 - C) * 2 + 0}];
+        int32_t viri${C} = -(int32_t) dr[${(SAMPLE_TILE - 1 - C) * 2 + 1}];
       $for C in range(SAMPLE_TILE):
         const int32_t vtwr${C} = twiddle[${C * 2 + 0}];
         const int32_t vtwi${C} = twiddle[${C * 2 + 1}];
@@ -76,23 +72,21 @@ void xnn_cs16_fftr_ukernel__scalar_x${SAMPLE_TILE}(
         const int32_t twi${C} = math_asr_s32(vacc2r${C} * vtwi${C} + vacc2i${C} * vtwr${C} + 16384, 15);
 
       $for C in range(SAMPLE_TILE):
-        ol[${C * 2 + 0}] = math_asr_s32(vacc1r${C} + twr${C}, 1);
-        ol[${C * 2 + 1}] = math_asr_s32(vacc1i${C} + twi${C}, 1);
-      ol += ${SAMPLE_TILE} * 2;
-      or -= ${SAMPLE_TILE} * 2;
+        dl[${C * 2 + 0}] = math_asr_s32(vacc1r${C} + twr${C}, 1);
+        dl[${C * 2 + 1}] = math_asr_s32(vacc1i${C} + twi${C}, 1);
       $for C in range(SAMPLE_TILE):
-        or[${(SAMPLE_TILE - 1 - C) * 2 + 0}] = math_asr_s32(vacc1r${C} - twr${C}, 1);
-        or[${(SAMPLE_TILE - 1 - C) * 2 + 1}] = math_asr_s32(twi${C} - vacc1i${C}, 1);
+        dr[${(SAMPLE_TILE - 1 - C) * 2 + 0}] = math_asr_s32(vacc1r${C} - twr${C}, 1);
+        dr[${(SAMPLE_TILE - 1 - C) * 2 + 1}] = math_asr_s32(twi${C} - vacc1i${C}, 1);
+      dl += ${SAMPLE_TILE} * 2;
     }
 
   if XNN_UNLIKELY(samples != 0) {
     do {
-      int32_t vilr = il[0];
-      int32_t vili = il[1];
-      il += 2;
-      ir -= 2;
-      int32_t virr =  (int32_t) ir[0];
-      int32_t viri = -(int32_t) ir[1];
+      dr -= 2;
+      int32_t vilr = dl[0];
+      int32_t vili = dl[1];
+      int32_t virr =  (int32_t) dr[0];
+      int32_t viri = -(int32_t) dr[1];
       const int32_t vtwr = twiddle[0];
       const int32_t vtwi = twiddle[1];
       twiddle += 2;
@@ -109,13 +103,11 @@ void xnn_cs16_fftr_ukernel__scalar_x${SAMPLE_TILE}(
       const int32_t twr = math_asr_s32(vacc2r * vtwr - vacc2i * vtwi + 16384, 15);
       const int32_t twi = math_asr_s32(vacc2r * vtwi + vacc2i * vtwr + 16384, 15);
 
-      ol[0] = math_asr_s32(vacc1r + twr, 1);
-      ol[1] = math_asr_s32(vacc1i + twi, 1);
-      ol += 2;
-      or -= 2;
-      or[0] = math_asr_s32(vacc1r - twr, 1);
-      or[1] = math_asr_s32(twi - vacc1i, 1);
-
+      dl[0] = math_asr_s32(vacc1r + twr, 1);
+      dl[1] = math_asr_s32(vacc1i + twi, 1);
+      dr[0] = math_asr_s32(vacc1r - twr, 1);
+      dr[1] = math_asr_s32(twi - vacc1i, 1);
+      dl += 2;
     } while (--samples != 0);
   }
 }
diff --git a/src/xnnpack/fft.h b/src/xnnpack/fft.h
index 9aa0afd79..374241136 100644
--- a/src/xnnpack/fft.h
+++ b/src/xnnpack/fft.h
@@ -31,8 +31,7 @@ DECLARE_CS16_BFLY4_UKERNEL_FUNCTION(xnn_cs16_bfly4m1_ukernel__scalar)
 #define DECLARE_CS16_FFTR_UKERNEL_FUNCTION(fn_name) \
   XNN_INTERNAL void fn_name(                        \
     size_t samples,                                 \
-    const int16_t* input,                           \
-    int16_t* output,                                \
+    int16_t* data,                                  \
     const int16_t* twiddle);
 
 DECLARE_CS16_FFTR_UKERNEL_FUNCTION(xnn_cs16_fftr_ukernel__scalar_x1)
diff --git a/src/xnnpack/microfnptr.h b/src/xnnpack/microfnptr.h
index 93525b6fe..a522c0fe9 100644
--- a/src/xnnpack/microfnptr.h
+++ b/src/xnnpack/microfnptr.h
@@ -1687,8 +1687,7 @@ typedef void (*xnn_cs16_bfly4_ukernel_function)(
 
 typedef void (*xnn_cs16_fftr_ukernel_function)(
     size_t samples,
-    const int16_t* input,
-    int16_t* output,
+    int16_t* data,
     const int16_t* twiddle);
 
 
diff --git a/test/cs16-fftr.cc b/test/cs16-fftr.cc
index 223e4914f..6e761bfbd 100644
--- a/test/cs16-fftr.cc
+++ b/test/cs16-fftr.cc
@@ -23,12 +23,6 @@ TEST(CS16_FFTR__SCALAR_X1, samples_eq_256) {
     .Test(xnn_cs16_fftr_ukernel__scalar_x1);
 }
 
-TEST(CS16_FFTR__SCALAR_X1, inplace) {
-  FftrMicrokernelTester()
-    .samples(256)
-    .inplace(true)
-    .Test(xnn_cs16_fftr_ukernel__scalar_x1);
-}
 
 TEST(CS16_FFTR__SCALAR_X2, samples_eq_256) {
   FftrMicrokernelTester()
@@ -36,12 +30,6 @@ TEST(CS16_FFTR__SCALAR_X2, samples_eq_256) {
     .Test(xnn_cs16_fftr_ukernel__scalar_x2);
 }
 
-TEST(CS16_FFTR__SCALAR_X2, inplace) {
-  FftrMicrokernelTester()
-    .samples(256)
-    .inplace(true)
-    .Test(xnn_cs16_fftr_ukernel__scalar_x2);
-}
 
 TEST(CS16_FFTR__SCALAR_X3, samples_eq_256) {
   FftrMicrokernelTester()
@@ -49,22 +37,9 @@ TEST(CS16_FFTR__SCALAR_X3, samples_eq_256) {
     .Test(xnn_cs16_fftr_ukernel__scalar_x3);
 }
 
-TEST(CS16_FFTR__SCALAR_X3, inplace) {
-  FftrMicrokernelTester()
-    .samples(256)
-    .inplace(true)
-    .Test(xnn_cs16_fftr_ukernel__scalar_x3);
-}
 
 TEST(CS16_FFTR__SCALAR_X4, samples_eq_256) {
   FftrMicrokernelTester()
     .samples(256)
     .Test(xnn_cs16_fftr_ukernel__scalar_x4);
 }
-
-TEST(CS16_FFTR__SCALAR_X4, inplace) {
-  FftrMicrokernelTester()
-    .samples(256)
-    .inplace(true)
-    .Test(xnn_cs16_fftr_ukernel__scalar_x4);
-}
\ No newline at end of file
diff --git a/test/fftr-microkernel-tester.h b/test/fftr-microkernel-tester.h
index 6f4fd256d..55dfab769 100644
--- a/test/fftr-microkernel-tester.h
+++ b/test/fftr-microkernel-tester.h
@@ -96,15 +96,6 @@ class FftrMicrokernelTester {
     return this->samples_;
   }
 
-  inline FftrMicrokernelTester& inplace(bool inplace) {
-    this->inplace_ = inplace;
-    return *this;
-  }
-
-  inline bool inplace() const {
-    return this->inplace_;
-  }
-
   inline FftrMicrokernelTester& iterations(size_t iterations) {
     this->iterations_ = iterations;
     return *this;
@@ -120,23 +111,20 @@ class FftrMicrokernelTester {
     auto i16rng = std::bind(std::uniform_int_distribution<int16_t>(), std::ref(rng));
     const size_t sample_size = samples() * 2 + 2;
 
-    std::vector<int16_t> x(sample_size + XNN_EXTRA_BYTES / sizeof(int16_t));
-    std::vector<int16_t> twiddle(samples() + XNN_EXTRA_BYTES / sizeof(int16_t));
-    std::vector<int16_t> y(sample_size + (inplace() ? XNN_EXTRA_BYTES / sizeof(int16_t) : 0));
+    std::vector<int16_t> twiddle(samples());
+    std::vector<int16_t> y(sample_size);
     std::vector<int16_t> y_ref(sample_size);
-    const int16_t* x_data = inplace() ? y.data() : x.data();
 
     for (size_t iteration = 0; iteration < iterations(); iteration++) {
-      std::generate(x.begin(), x.end(), std::ref(i16rng));
       std::generate(twiddle.begin(), twiddle.end(), std::ref(i16rng));
       std::generate(y.begin(), y.end(), std::ref(i16rng));
-      std::generate(y_ref.begin(), y_ref.end(), std::ref(i16rng));
+      std::copy(y.begin(), y.end(), y_ref.begin());
 
       // Compute reference results.
-      xnn_cs16_fftr_reference(samples(), x_data, y_ref.data(), twiddle.data());
+      xnn_cs16_fftr_reference(samples(), y_ref.data(), y_ref.data(), twiddle.data());
 
       // Call optimized micro-kernel.
-      fftr(samples(), x_data, y.data(), twiddle.data());
+      fftr(samples(), y.data(), twiddle.data());
 
       // Verify results.
       for (size_t n = 0; n < sample_size; n++) {
@@ -148,6 +136,5 @@ class FftrMicrokernelTester {
 
  private:
   size_t samples_{256};
-  bool inplace_{false};
   size_t iterations_{15};
 };
diff --git a/tools/generate-fftr-test.py b/tools/generate-fftr-test.py
index 53de31293..fa84f602b 100755
--- a/tools/generate-fftr-test.py
+++ b/tools/generate-fftr-test.py
@@ -44,14 +44,6 @@ TEST(${TEST_NAME}, samples_eq_256) {
     .Test(${", ".join(TEST_ARGS)});
 }
 
-TEST(${TEST_NAME}, inplace) {
-  $if ISA_CHECK:
-    ${ISA_CHECK};
-  FftrMicrokernelTester()
-    .samples(256)
-    .inplace(true)
-    .Test(${", ".join(TEST_ARGS)});
-}
 """