Porting ARM optimization from Android to ios.

Tested APM and iSAC in Android. Bit-exact with original versions.
Changes include removing or changing some GCC derivatives (e.g. .fnstart, .hword), instruction syntax, etc.
Review URL: https://webrtc-codereview.appspot.com/934009

git-svn-id: http://webrtc.googlecode.com/svn/trunk@3124 4adac7df-926f-26a2-2b94-8c16560cd09d

11 files changed, 100 insertions, 158 deletions

diff --git a/webrtc/common_audio/signal_processing/Android.mk b/webrtc/common_audio/signal_processing/Android.mk
index ecbc5ddf4e..aed7e73b5d 100644
--- a/webrtc/common_audio/signal_processing/Android.mk
+++ b/webrtc/common_audio/signal_processing/Android.mk
@@ -60,7 +60,7 @@ LOCAL_C_INCLUDES := \
 
 ifeq ($(ARCH_ARM_HAVE_ARMV7A),true)
 LOCAL_SRC_FILES += \
-    filter_ar_fast_q12_armv7.s
+    filter_ar_fast_q12_armv7.S
 else
 LOCAL_SRC_FILES += \
     filter_ar_fast_q12.c
@@ -68,8 +68,8 @@ endif
 
 ifeq ($(TARGET_ARCH),arm)
 LOCAL_SRC_FILES += \
-    complex_bit_reverse_arm.s \
-    spl_sqrt_floor_arm.s
+    complex_bit_reverse_arm.S \
+    spl_sqrt_floor_arm.S
 else
 LOCAL_SRC_FILES += \
     complex_bit_reverse.c \
@@ -102,10 +102,10 @@ LOCAL_MODULE_CLASS := STATIC_LIBRARIES
 LOCAL_MODULE := libwebrtc_spl_neon
 LOCAL_MODULE_TAGS := optional
 LOCAL_SRC_FILES := \
-    cross_correlation_neon.s \
-    downsample_fast_neon.s \
-    min_max_operations_neon.s \
-    vector_scaling_operations_neon.s
+    cross_correlation_neon.S \
+    downsample_fast_neon.S \
+    min_max_operations_neon.S \
+    vector_scaling_operations_neon.S
 
 # Flags passed to both C and C++ files.
 LOCAL_CFLAGS := \
diff --git a/webrtc/common_audio/signal_processing/complex_bit_reverse_arm.s b/webrtc/common_audio/signal_processing/complex_bit_reverse_arm.S
index 4828077800..e7f8a819bd 100644
--- a/webrtc/common_audio/signal_processing/complex_bit_reverse_arm.s
+++ b/webrtc/common_audio/signal_processing/complex_bit_reverse_arm.S
@@ -12,15 +12,11 @@
 @ for ARMv5 platforms.
 @ Reference C code is in file complex_bit_reverse.c. Bit-exact.
 
-.arch armv5
-
-.global WebRtcSpl_ComplexBitReverse
+#include "webrtc/system_wrappers/interface/asm_defines.h"
 
+GLOBAL_FUNCTION WebRtcSpl_ComplexBitReverse
 .align  2
-
-WebRtcSpl_ComplexBitReverse:
-.fnstart
-
+DEFINE_FUNCTION WebRtcSpl_ComplexBitReverse
   push {r4-r7}
 
   cmp r1, #7
@@ -88,39 +84,36 @@ END:
   pop {r4-r7}
   bx lr
 
-.fnend
-
-
 @ The index tables. Note the values are doubles of the actual indexes for 16-bit
 @ elements, different from the generic C code. It actually provides byte offsets
 @ for the indexes.
 
 .align  2
 index_7:  @ Indexes for stages == 7.
-  .hword 4, 256, 8, 128, 12, 384, 16, 64, 20, 320, 24, 192, 28, 448, 36, 288
-  .hword 40, 160, 44, 416, 48, 96, 52, 352, 56, 224, 60, 480, 68, 272, 72, 144
-  .hword 76, 400, 84, 336, 88, 208, 92, 464, 100, 304, 104, 176, 108, 432, 116
-  .hword 368, 120, 240, 124, 496, 132, 264, 140, 392, 148, 328, 152, 200, 156
-  .hword 456, 164, 296, 172, 424, 180, 360, 184, 232, 188, 488, 196, 280, 204
-  .hword 408, 212, 344, 220, 472, 228, 312, 236, 440, 244, 376, 252, 504, 268
-  .hword 388, 276, 324, 284, 452, 300, 420, 308, 356, 316, 484, 332, 404, 348
-  .hword 468, 364, 436, 380, 500, 412, 460, 444, 492
+  .short 4, 256, 8, 128, 12, 384, 16, 64, 20, 320, 24, 192, 28, 448, 36, 288
+  .short 40, 160, 44, 416, 48, 96, 52, 352, 56, 224, 60, 480, 68, 272, 72, 144
+  .short 76, 400, 84, 336, 88, 208, 92, 464, 100, 304, 104, 176, 108, 432, 116
+  .short 368, 120, 240, 124, 496, 132, 264, 140, 392, 148, 328, 152, 200, 156
+  .short 456, 164, 296, 172, 424, 180, 360, 184, 232, 188, 488, 196, 280, 204
+  .short 408, 212, 344, 220, 472, 228, 312, 236, 440, 244, 376, 252, 504, 268
+  .short 388, 276, 324, 284, 452, 300, 420, 308, 356, 316, 484, 332, 404, 348
+  .short 468, 364, 436, 380, 500, 412, 460, 444, 492
 
 index_8:  @ Indexes for stages == 8.
-  .hword 4, 512, 8, 256, 12, 768, 16, 128, 20, 640, 24, 384, 28, 896, 32, 64
-  .hword 36, 576, 40, 320, 44, 832, 48, 192, 52, 704, 56, 448, 60, 960, 68, 544
-  .hword 72, 288, 76, 800, 80, 160, 84, 672, 88, 416, 92, 928, 100, 608, 104
-  .hword 352, 108, 864, 112, 224, 116, 736, 120, 480, 124, 992, 132, 528, 136
-  .hword 272, 140, 784, 148, 656, 152, 400, 156, 912, 164, 592, 168, 336, 172
-  .hword 848, 176, 208, 180, 720, 184, 464, 188, 976, 196, 560, 200, 304, 204
-  .hword 816, 212, 688, 216, 432, 220, 944, 228, 624, 232, 368, 236, 880, 244
-  .hword 752, 248, 496, 252, 1008, 260, 520, 268, 776, 276, 648, 280, 392, 284
-  .hword 904, 292, 584, 296, 328, 300, 840, 308, 712, 312, 456, 316, 968, 324
-  .hword 552, 332, 808, 340, 680, 344, 424, 348, 936, 356, 616, 364, 872, 372
-  .hword 744, 376, 488, 380, 1000, 388, 536, 396, 792, 404, 664, 412, 920, 420
-  .hword 600, 428, 856, 436, 728, 440, 472, 444, 984, 452, 568, 460, 824, 468
-  .hword 696, 476, 952, 484, 632, 492, 888, 500, 760, 508, 1016, 524, 772, 532
-  .hword 644, 540, 900, 548, 580, 556, 836, 564, 708, 572, 964, 588, 804, 596
-  .hword 676, 604, 932, 620, 868, 628, 740, 636, 996, 652, 788, 668, 916, 684
-  .hword 852, 692, 724, 700, 980, 716, 820, 732, 948, 748, 884, 764, 1012, 796
-  .hword 908, 812, 844, 828, 972, 860, 940, 892, 1004, 956, 988
+  .short 4, 512, 8, 256, 12, 768, 16, 128, 20, 640, 24, 384, 28, 896, 32, 64
+  .short 36, 576, 40, 320, 44, 832, 48, 192, 52, 704, 56, 448, 60, 960, 68, 544
+  .short 72, 288, 76, 800, 80, 160, 84, 672, 88, 416, 92, 928, 100, 608, 104
+  .short 352, 108, 864, 112, 224, 116, 736, 120, 480, 124, 992, 132, 528, 136
+  .short 272, 140, 784, 148, 656, 152, 400, 156, 912, 164, 592, 168, 336, 172
+  .short 848, 176, 208, 180, 720, 184, 464, 188, 976, 196, 560, 200, 304, 204
+  .short 816, 212, 688, 216, 432, 220, 944, 228, 624, 232, 368, 236, 880, 244
+  .short 752, 248, 496, 252, 1008, 260, 520, 268, 776, 276, 648, 280, 392, 284
+  .short 904, 292, 584, 296, 328, 300, 840, 308, 712, 312, 456, 316, 968, 324
+  .short 552, 332, 808, 340, 680, 344, 424, 348, 936, 356, 616, 364, 872, 372
+  .short 744, 376, 488, 380, 1000, 388, 536, 396, 792, 404, 664, 412, 920, 420
+  .short 600, 428, 856, 436, 728, 440, 472, 444, 984, 452, 568, 460, 824, 468
+  .short 696, 476, 952, 484, 632, 492, 888, 500, 760, 508, 1016, 524, 772, 532
+  .short 644, 540, 900, 548, 580, 556, 836, 564, 708, 572, 964, 588, 804, 596
+  .short 676, 604, 932, 620, 868, 628, 740, 636, 996, 652, 788, 668, 916, 684
+  .short 852, 692, 724, 700, 980, 716, 820, 732, 948, 748, 884, 764, 1012, 796
+  .short 908, 812, 844, 828, 972, 860, 940, 892, 1004, 956, 988
diff --git a/webrtc/common_audio/signal_processing/cross_correlation_neon.s b/webrtc/common_audio/signal_processing/cross_correlation_neon.S
index a18f67223f..580440c81f 100644
--- a/webrtc/common_audio/signal_processing/cross_correlation_neon.s
+++ b/webrtc/common_audio/signal_processing/cross_correlation_neon.S
@@ -29,24 +29,18 @@
 @ r7: Total iteration of LOOP_DIM_SEQ_RESIDUAL
 @ r8, r9, r10, r11, r12: scratch
 
-.arch armv7-a
-.fpu neon
+#include "webrtc/system_wrappers/interface/asm_defines.h"
 
+GLOBAL_FUNCTION WebRtcSpl_CrossCorrelationNeon
 .align  2
-.global WebRtcSpl_CrossCorrelationNeon
-
-WebRtcSpl_CrossCorrelationNeon:
-
-.fnstart
-
-.save {r4-r11}
+DEFINE_FUNCTION WebRtcSpl_CrossCorrelationNeon
   push {r4-r11}
 
   @ Put the shift value (-right_shifts) into a Neon register.
   ldrsh r10, [sp, #36]
   rsb r10, r10, #0
   mov r8, r10, asr #31
-  vmov.32 d16, r10, r8
+  vmov d16, r10, r8
 
   @ Initialize loop counters.
   and r7, r3, #7              @ inner_loop_len2 = dim_seq % 8;
@@ -63,7 +57,7 @@ LOOP_DIM_CROSS_CORRELATION:
 
 LOOP_DIM_SEQ:
   vld1.16 {d20, d21}, [r6]!   @ seq1_ptr
-  vld1.16 {d22, d23}, [r5]!   @ seq2_ptr 
+  vld1.16 {d22, d23}, [r5]!   @ seq2_ptr
   subs r8, r8, #1
   vmull.s16 q12, d20, d22
   vmull.s16 q13, d21, d23
@@ -105,9 +99,6 @@ POST_LOOP_DIM_SEQ_RESIDUAL:   @ Sum the results up and do the shift.
   pop {r4-r11}
   bx  lr
 
-.fnend
-
-
 @ TODO(kma): Place this piece of reference code into a C code file.
 @ void WebRtcSpl_CrossCorrelationNeon(WebRtc_Word32* cross_correlation,
 @                                     WebRtc_Word16* seq1,
@@ -120,15 +111,15 @@ POST_LOOP_DIM_SEQ_RESIDUAL:   @ Sum the results up and do the shift.
 @   int j = 0;
 @   int inner_loop_len1 = dim_seq >> 3;
 @   int inner_loop_len2 = dim_seq - (inner_loop_len1 << 3);
-@ 
+@
 @   assert(dim_cross_correlation > 0);
 @   assert(dim_seq > 0);
-@ 
+@
 @   for (i = 0; i < dim_cross_correlation; i++) {
 @     int16_t *seq1_ptr = seq1;
 @     int16_t *seq2_ptr = seq2 + (step_seq2 * i);
 @     int64_t sum = 0;
-@ 
+@
 @     for (j = inner_loop_len1; j > 0; j -= 1) {
 @       sum += WEBRTC_SPL_MUL_16_16(*seq1_ptr, *seq2_ptr);
 @       seq1_ptr++;
@@ -155,14 +146,14 @@ POST_LOOP_DIM_SEQ_RESIDUAL:   @ Sum the results up and do the shift.
 @       seq1_ptr++;
 @       seq2_ptr++;
 @     }
-@ 
+@
 @     // Calculate the rest of the samples.
 @     for (j = inner_loop_len2; j > 0; j -= 1) {
 @       sum += WEBRTC_SPL_MUL_16_16(*seq1_ptr, *seq2_ptr);
 @       seq1_ptr++;
 @       seq2_ptr++;
 @     }
-@ 
+@
 @     *cross_correlation++ = (int32_t)(sum >> right_shifts);
 @   }
 @ }
diff --git a/webrtc/common_audio/signal_processing/downsample_fast_neon.s b/webrtc/common_audio/signal_processing/downsample_fast_neon.S
index 13a825d797..4e348ec646 100644
--- a/webrtc/common_audio/signal_processing/downsample_fast_neon.s
+++ b/webrtc/common_audio/signal_processing/downsample_fast_neon.S
@@ -14,17 +14,11 @@
 @
 @ The reference C code is in file downsample_fast.c. Bit-exact.
 
-.arch armv7-a
-.fpu neon
+#include "webrtc/system_wrappers/interface/asm_defines.h"
 
+GLOBAL_FUNCTION WebRtcSpl_DownsampleFastNeon
 .align  2
-.global WebRtcSpl_DownsampleFastNeon
-
-WebRtcSpl_DownsampleFastNeon:
-
-.fnstart
-
-.save {r4-r11}
+DEFINE_FUNCTION WebRtcSpl_DownsampleFastNeon
   push {r4-r11}
 
   cmp r3, #0                                @ data_out_length <= 0?
@@ -168,14 +162,15 @@ LOOP_COEFF_LENGTH_FACTOR4:
   vmlal.s16 q3, d18, d17
   bge LOOP_COEFF_LENGTH_FACTOR4
 
+  add r11, r5, asl #4                       @ r11 -> &data_in[i + factor * 8]
+  add r9, r5, asl #3                        @ Counter i = delay + factor * 8.
+
   @ Shift, saturate, and store the result.
   vqshrn.s32 d0, q2, #12
   vqshrn.s32 d1, q3, #12
+  cmp r9, r3                                @ i < endpos - factor * 7 ?
   vst1.16 {d0, d1}, [r2]!
 
-  add r11, r5, asl #4                       @ r11 -> &data_in[i + factor * 8]
-  add r9, r5, asl #3                        @ Counter i = delay + factor * 8.
-  cmp r9, r3                                @ i < endpos - factor * 7 ?
   blt LOOP_ENDPOS_FACTOR4
 
 @
@@ -218,5 +213,3 @@ LOOP2_COEFF_LENGTH:
 END:
   pop {r4-r11}
   bx  lr
-
-.fnend
diff --git a/webrtc/common_audio/signal_processing/filter_ar_fast_q12_armv7.s b/webrtc/common_audio/signal_processing/filter_ar_fast_q12_armv7.S
index 5591bb83cd..ff60cc6198 100644
--- a/webrtc/common_audio/signal_processing/filter_ar_fast_q12_armv7.s
+++ b/webrtc/common_audio/signal_processing/filter_ar_fast_q12_armv7.S
@@ -35,16 +35,11 @@
 @ r11: Scratch
 @ r12: &coefficients[j]
 
-.arch armv7-a
+#include "webrtc/system_wrappers/interface/asm_defines.h"
 
+GLOBAL_FUNCTION WebRtcSpl_FilterARFastQ12
 .align  2
-.global WebRtcSpl_FilterARFastQ12
-
-WebRtcSpl_FilterARFastQ12:
-
-.fnstart
-
-.save {r4-r11}
+DEFINE_FUNCTION WebRtcSpl_FilterARFastQ12
   push {r4-r11}
 
   ldrsh r12, [sp, #32]         @ data_length
@@ -155,9 +150,6 @@ END:
   pop {r4-r11}
   bx  lr
 
-.fnend
-
-
 @Reference C code:
 @
 @void WebRtcSpl_FilterARFastQ12(int16_t* data_in,
diff --git a/webrtc/common_audio/signal_processing/include/spl_inl_armv7.h b/webrtc/common_audio/signal_processing/include/spl_inl_armv7.h
index 8461474fb3..c9bcc1c23f 100644
--- a/webrtc/common_audio/signal_processing/include/spl_inl_armv7.h
+++ b/webrtc/common_audio/signal_processing/include/spl_inl_armv7.h
@@ -166,7 +166,7 @@ static __inline int WebRtcSpl_NormW16(WebRtc_Word16 a) {
 static __inline WebRtc_Word16 WebRtcSpl_SatW32ToW16(WebRtc_Word32 value32) {
   WebRtc_Word16 out16 = 0;
 
-  __asm __volatile ("ssat %r0, #16, %r1" : "=r"(out16) : "r"(value32));
+  __asm __volatile ("ssat %0, #16, %1" : "=r"(out16) : "r"(value32));
 
   return out16;
 }
diff --git a/webrtc/common_audio/signal_processing/min_max_operations_neon.s b/webrtc/common_audio/signal_processing/min_max_operations_neon.S
index 85dd2fb9df..c84307f5e4 100644
--- a/webrtc/common_audio/signal_processing/min_max_operations_neon.s
+++ b/webrtc/common_audio/signal_processing/min_max_operations_neon.S
@@ -15,20 +15,18 @@
 @ The reference C code is in file min_max_operations.c. Code here is basically
 @ a loop unrolling by 8 with Neon instructions. Bit-exact.
 
-.arch armv7-a
-.fpu neon
-.global WebRtcSpl_MaxAbsValueW16Neon
-.global WebRtcSpl_MaxAbsValueW32Neon
-.global WebRtcSpl_MaxValueW16Neon
-.global WebRtcSpl_MaxValueW32Neon
-.global WebRtcSpl_MinValueW16Neon
-.global WebRtcSpl_MinValueW32Neon
-.align  2
+#include "webrtc/system_wrappers/interface/asm_defines.h"
 
-@ int16_t WebRtcSpl_MaxAbsValueW16Neon(const int16_t* vector, int length);
-WebRtcSpl_MaxAbsValueW16Neon:
-.fnstart
+GLOBAL_FUNCTION WebRtcSpl_MaxAbsValueW16Neon
+GLOBAL_FUNCTION WebRtcSpl_MaxAbsValueW32Neon
+GLOBAL_FUNCTION WebRtcSpl_MaxValueW16Neon
+GLOBAL_FUNCTION WebRtcSpl_MaxValueW32Neon
+GLOBAL_FUNCTION WebRtcSpl_MinValueW16Neon
+GLOBAL_FUNCTION WebRtcSpl_MinValueW32Neon
 
+.align  2
+@ int16_t WebRtcSpl_MaxAbsValueW16Neon(const int16_t* vector, int length);
+DEFINE_FUNCTION WebRtcSpl_MaxAbsValueW16Neon
   mov r2, #-1                 @ Initialize the return value.
   cmp r0, #0
   beq END_MAX_ABS_VALUE_W16
@@ -50,8 +48,8 @@ LOOP_UNROLLED_BY_8_MAX_ABS_VALUE_W16:
 
   @ Find the maximum value in the Neon registers and move it to r2.
   vmax.u16 d24, d25
-  vpmax.u16 d24, d24
-  vpmax.u16 d24, d24
+  vpmax.u16 d24, d24, d24
+  vpmax.u16 d24, d24, d24
   adds r1, #8
   vmov.u16 r2, d24[0]
   beq END_MAX_ABS_VALUE_W16
@@ -71,12 +69,10 @@ END_MAX_ABS_VALUE_W16:
   mov r0, r2
   bx  lr
 
-.fnend
 
-@ int32_t WebRtcSpl_MaxAbsValueW32Neon(const int32_t* vector, int length);
-WebRtcSpl_MaxAbsValueW32Neon:
-.fnstart
 
+@ int32_t WebRtcSpl_MaxAbsValueW32Neon(const int32_t* vector, int length);
+DEFINE_FUNCTION WebRtcSpl_MaxAbsValueW32Neon
   cmp r0, #0
   moveq r0, #-1
   beq EXIT                    @ Return -1 for a NULL pointer.
@@ -103,7 +99,7 @@ LOOP_UNROLLED_BY_8_MAX_ABS_VALUE_W32:
   @ Find the maximum value in the Neon registers and move it to r2.
   vmax.u32 q12, q11
   vmax.u32 d24, d25
-  vpmax.u32 d24, d24
+  vpmax.u32 d24, d24, d24
   adds r1, #8
   vmov.u32 r2, d24[0]
   beq END_MAX_ABS_VALUE_W32
@@ -125,12 +121,8 @@ END_MAX_ABS_VALUE_W32:
 EXIT:
   bx  lr
 
-.fnend
-
 @ int16_t WebRtcSpl_MaxValueW16Neon(const int16_t* vector, int length);
-WebRtcSpl_MaxValueW16Neon:
-.fnstart
-
+DEFINE_FUNCTION WebRtcSpl_MaxValueW16Neon
   mov r2, #0x8000             @ Initialize the return value.
   cmp r0, #0
   beq END_MAX_VALUE_W16
@@ -151,8 +143,8 @@ LOOP_UNROLLED_BY_8_MAX_VALUE_W16:
 
   @ Find the maximum value in the Neon registers and move it to r2.
   vmax.s16 d24, d25
-  vpmax.s16 d24, d24
-  vpmax.s16 d24, d24
+  vpmax.s16 d24, d24, d24
+  vpmax.s16 d24, d24, d24
   adds r1, #8
   vmov.u16 r2, d24[0]
   beq END_MAX_VALUE_W16
@@ -168,12 +160,8 @@ END_MAX_VALUE_W16:
   mov r0, r2
   bx  lr
 
-.fnend
-
 @ int32_t WebRtcSpl_MaxValueW32Neon(const int32_t* vector, int length);
-WebRtcSpl_MaxValueW32Neon:
-.fnstart
-
+DEFINE_FUNCTION WebRtcSpl_MaxValueW32Neon
   mov r2, #0x80000000         @ Initialize the return value.
   cmp r0, #0
   beq END_MAX_VALUE_W32
@@ -196,8 +184,8 @@ LOOP_UNROLLED_BY_8_MAX_VALUE_W32:
 
   @ Find the maximum value in the Neon registers and move it to r2.
   vmax.s32 q12, q11
-  vpmax.s32 d24, d25
-  vpmax.s32 d24, d24
+  vpmax.s32 d24, d24, d25
+  vpmax.s32 d24, d24, d24
   adds r1, #8
   vmov.s32 r2, d24[0]
   beq END_MAX_VALUE_W32
@@ -213,12 +201,8 @@ END_MAX_VALUE_W32:
   mov r0, r2
   bx  lr
 
-.fnend
-
 @ int16_t WebRtcSpl_MinValueW16Neon(const int16_t* vector, int length);
-WebRtcSpl_MinValueW16Neon:
-.fnstart
-
+DEFINE_FUNCTION WebRtcSpl_MinValueW16Neon
   movw r2, #0x7FFF            @ Initialize the return value.
   cmp r0, #0
   beq END_MIN_VALUE_W16
@@ -239,8 +223,8 @@ LOOP_UNROLLED_BY_8_MIN_VALUE_W16:
 
   @ Find the maximum value in the Neon registers and move it to r2.
   vmin.s16 d24, d25
-  vpmin.s16 d24, d24
-  vpmin.s16 d24, d24
+  vpmin.s16 d24, d24, d24
+  vpmin.s16 d24, d24, d24
   adds r1, #8
   vmov.s16 r2, d24[0]
   sxth  r2, r2
@@ -257,12 +241,8 @@ END_MIN_VALUE_W16:
   mov r0, r2
   bx  lr
 
-.fnend
-
 @ int32_t WebRtcSpl_MinValueW32Neon(const int32_t* vector, int length);
-WebRtcSpl_MinValueW32Neon:
-.fnstart
-
+DEFINE_FUNCTION WebRtcSpl_MinValueW32Neon
   mov r2, #0x7FFFFFFF         @ Initialize the return value.
   cmp r0, #0
   beq END_MIN_VALUE_W32
@@ -285,8 +265,8 @@ LOOP_UNROLLED_BY_8_MIN_VALUE_W32:
 
   @ Find the maximum value in the Neon registers and move it to r2.
   vmin.s32 q12, q11
-  vpmin.s32 d24, d25
-  vpmin.s32 d24, d24
+  vpmin.s32 d24, d24, d25
+  vpmin.s32 d24, d24, d24
   adds r1, #8
   vmov.s32 r2, d24[0]
   beq END_MIN_VALUE_W32
@@ -301,5 +281,3 @@ LOOP_MIN_VALUE_W32:
 END_MIN_VALUE_W32:
   mov r0, r2
   bx  lr
-
-.fnend
diff --git a/webrtc/common_audio/signal_processing/resample_by_2.c b/webrtc/common_audio/signal_processing/resample_by_2.c
index c1d8b37844..e6692e8a79 100644
--- a/webrtc/common_audio/signal_processing/resample_by_2.c
+++ b/webrtc/common_audio/signal_processing/resample_by_2.c
@@ -31,8 +31,8 @@ static __inline WebRtc_Word32 MUL_ACCUM_1(WebRtc_Word32 tbl_value,
                                           WebRtc_Word32 diff,
                                           WebRtc_Word32 state) {
   WebRtc_Word32 result;
-  __asm__("smlawb %r0, %r1, %r2, %r3": "=r"(result): "r"(diff),
-                                       "r"(tbl_value), "r"(state));
+  __asm __volatile ("smlawb %0, %1, %2, %3": "=r"(result): "r"(diff),
+                                   "r"(tbl_value), "r"(state));
   return result;
 }
 
@@ -47,8 +47,8 @@ static __inline WebRtc_Word32 MUL_ACCUM_2(WebRtc_Word32 tbl_value,
                                           WebRtc_Word32 diff,
                                           WebRtc_Word32 state) {
   WebRtc_Word32 result;
-  __asm__("smmla %r0, %r1, %r2, %r3": "=r"(result): "r"(diff << 1),
-                                      "r"(tbl_value), "r"(state));
+  __asm __volatile ("smmla %0, %1, %2, %3": "=r"(result): "r"(diff << 1),
+                                  "r"(tbl_value), "r"(state));
   return result;
 }
 
diff --git a/webrtc/common_audio/signal_processing/signal_processing.gypi b/webrtc/common_audio/signal_processing/signal_processing.gypi
index b09c767bed..91592ea7c4 100644
--- a/webrtc/common_audio/signal_processing/signal_processing.gypi
+++ b/webrtc/common_audio/signal_processing/signal_processing.gypi
@@ -65,8 +65,8 @@
       'conditions': [
         ['target_arch=="arm"', {
           'sources': [
-            'complex_bit_reverse_arm.s',
-            'spl_sqrt_floor_arm.s',
+            'complex_bit_reverse_arm.S',
+            'spl_sqrt_floor_arm.S',
           ],
           'sources!': [
             'complex_bit_reverse.c',
@@ -76,7 +76,7 @@
             ['armv7==1', {
               'dependencies': ['signal_processing_neon',],
               'sources': [
-                'filter_ar_fast_q12_armv7.s',
+                'filter_ar_fast_q12_armv7.S',
               ],
               'sources!': [
                 'filter_ar_fast_q12.c',
@@ -112,10 +112,10 @@
           'type': '<(library)',
           'includes': ['../../build/arm_neon.gypi',],
           'sources': [
-            'cross_correlation_neon.s',
-            'downsample_fast_neon.s',
-            'min_max_operations_neon.s',
-            'vector_scaling_operations_neon.s',
+            'cross_correlation_neon.S',
+            'downsample_fast_neon.S',
+            'min_max_operations_neon.S',
+            'vector_scaling_operations_neon.S',
           ],
         },
       ],
diff --git a/webrtc/common_audio/signal_processing/spl_sqrt_floor_arm.s b/webrtc/common_audio/signal_processing/spl_sqrt_floor_arm.S
index a2c5b7d0d2..c49ef1f3bc 100644
--- a/webrtc/common_audio/signal_processing/spl_sqrt_floor_arm.s
+++ b/webrtc/common_audio/signal_processing/spl_sqrt_floor_arm.S
@@ -8,10 +8,11 @@
 @ Output:             r0 = INT (SQRT (r0)), precision is 16 bits
 @ Registers touched:  r1, r2
 
-.global WebRtcSpl_SqrtFloor
+#include "webrtc/system_wrappers/interface/asm_defines.h"
 
+GLOBAL_FUNCTION WebRtcSpl_SqrtFloor
 .align  2
-WebRtcSpl_SqrtFloor:
+DEFINE_FUNCTION WebRtcSpl_SqrtFloor
   mov    r1, #3 << 30
   mov    r2, #1 << 30
 
diff --git a/webrtc/common_audio/signal_processing/vector_scaling_operations_neon.s b/webrtc/common_audio/signal_processing/vector_scaling_operations_neon.S
index 562425bf05..07db741b00 100644
--- a/webrtc/common_audio/signal_processing/vector_scaling_operations_neon.s
+++ b/webrtc/common_audio/signal_processing/vector_scaling_operations_neon.S
@@ -13,15 +13,11 @@
 @ optimized for ARM Neon platform. Output is bit-exact with the reference
 @ C code in vector_scaling_operations.c.
 
-.arch armv7-a
-.fpu neon
+#include "webrtc/system_wrappers/interface/asm_defines.h"
 
+GLOBAL_FUNCTION WebRtcSpl_ScaleAndAddVectorsWithRoundNeon
 .align  2
-.global WebRtcSpl_ScaleAndAddVectorsWithRoundNeon
-
-WebRtcSpl_ScaleAndAddVectorsWithRoundNeon:
-.fnstart
-
+DEFINE_FUNCTION WebRtcSpl_ScaleAndAddVectorsWithRoundNeon
   push {r4-r9}
 
   ldr r4, [sp, #32]           @ length
@@ -84,5 +80,3 @@ LOOP_NO_UNROLLING:
 END:
   pop {r4-r9}
   bx  lr
-
-.fnend