diff --git a/README.chromium b/README.chromium
index 1a852d7f8..ac18acfe4 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,6 +1,6 @@
 Name: libyuv
 URL: https://chromium.googlesource.com/libyuv/libyuv/
-Version: 1890
+Version: 1891
 License: BSD
 License File: LICENSE
 Shipped: yes
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index 761bce145..535a939ac 100644
--- a/include/libyuv/version.h
+++ b/include/libyuv/version.h
@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1890
+#define LIBYUV_VERSION 1891
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_
diff --git a/source/row_gcc.cc b/source/row_gcc.cc
index 782382d0c..69babb453 100644
--- a/source/row_gcc.cc
+++ b/source/row_gcc.cc
@@ -2626,7 +2626,7 @@ void OMITFP I444ToARGBRow_SSSE3(const uint8_t* y_buf,
                                 uint8_t* dst_argb,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
@@ -2686,7 +2686,7 @@ void OMITFP I422ToRGB24Row_SSSE3(const uint8_t* y_buf,
                                  uint8_t* dst_rgb24,
                                  const struct YuvConstants* yuvconstants,
                                  int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "movdqa      %[kShuffleMaskARGBToRGB24_0],%%xmm5 \n"
       "movdqa      %[kShuffleMaskARGBToRGB24],%%xmm6 \n"
@@ -2722,7 +2722,7 @@ void OMITFP I444ToRGB24Row_SSSE3(const uint8_t* y_buf,
                                  uint8_t* dst_rgb24,
                                  const struct YuvConstants* yuvconstants,
                                  int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "movdqa      %[kShuffleMaskARGBToRGB24_0],%%xmm5 \n"
       "movdqa      %[kShuffleMaskARGBToRGB24],%%xmm6 \n"
@@ -2758,7 +2758,7 @@ void OMITFP I422ToARGBRow_SSSE3(const uint8_t* y_buf,
                                 uint8_t* dst_argb,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
@@ -2787,7 +2787,7 @@ void OMITFP I422ToAR30Row_SSSE3(const uint8_t* y_buf,
                                 uint8_t* dst_ar30,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "pcmpeqb     %%xmm5,%%xmm5                 \n"  // AR30 constants
@@ -2822,7 +2822,7 @@ void OMITFP I210ToARGBRow_SSSE3(const uint16_t* y_buf,
                                 uint8_t* dst_argb,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
@@ -2852,7 +2852,7 @@ void OMITFP I212ToARGBRow_SSSE3(const uint16_t* y_buf,
                                 uint8_t* dst_argb,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
@@ -2882,7 +2882,7 @@ void OMITFP I210ToAR30Row_SSSE3(const uint16_t* y_buf,
                                 uint8_t* dst_ar30,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
@@ -2917,7 +2917,7 @@ void OMITFP I212ToAR30Row_SSSE3(const uint16_t* y_buf,
                                 uint8_t* dst_ar30,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
@@ -2952,7 +2952,7 @@ void OMITFP I410ToARGBRow_SSSE3(const uint16_t* y_buf,
                                 uint8_t* dst_argb,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
@@ -3045,7 +3045,7 @@ void OMITFP I410ToAR30Row_SSSE3(const uint16_t* y_buf,
                                 uint8_t* dst_ar30,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
@@ -3238,7 +3238,7 @@ void OMITFP P210ToAR30Row_SSSE3(const uint16_t* y_buf,
                                 uint8_t* dst_ar30,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
       "psrlw       $14,%%xmm5                    \n"
@@ -3269,7 +3269,7 @@ void OMITFP P410ToAR30Row_SSSE3(const uint16_t* y_buf,
                                 uint8_t* dst_ar30,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
       "psrlw       $14,%%xmm5                    \n"
@@ -3301,7 +3301,7 @@ void OMITFP I422ToRGBARow_SSSE3(const uint8_t* y_buf,
                                 uint8_t* dst_rgba,
                                 const struct YuvConstants* yuvconstants,
                                 int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "pcmpeqb     %%xmm5,%%xmm5                 \n"
@@ -3712,7 +3712,7 @@ void OMITFP I444ToARGBRow_AVX2(const uint8_t* y_buf,
                                uint8_t* dst_argb,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"
@@ -3746,7 +3746,7 @@ void OMITFP I422ToARGBRow_AVX2(const uint8_t* y_buf,
                                uint8_t* dst_argb,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"
@@ -3786,7 +3786,7 @@ void OMITFP I422ToARGBRow_AVX512BW(const uint8_t* y_buf,
                                    uint8_t* dst_argb,
                                    const struct YuvConstants* yuvconstants,
                                    int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX512BW(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%xmm5,%%xmm5,%%xmm5          \n"
@@ -3825,7 +3825,7 @@ void OMITFP I422ToAR30Row_AVX2(const uint8_t* y_buf,
                                uint8_t* dst_ar30,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"  // AR30 constants
@@ -3865,7 +3865,7 @@ void OMITFP I210ToARGBRow_AVX2(const uint16_t* y_buf,
                                uint8_t* dst_argb,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"
@@ -3900,7 +3900,7 @@ void OMITFP I212ToARGBRow_AVX2(const uint16_t* y_buf,
                                uint8_t* dst_argb,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"
@@ -3935,7 +3935,7 @@ void OMITFP I210ToAR30Row_AVX2(const uint16_t* y_buf,
                                uint8_t* dst_ar30,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"  // AR30 constants
@@ -3975,7 +3975,7 @@ void OMITFP I212ToAR30Row_AVX2(const uint16_t* y_buf,
                                uint8_t* dst_ar30,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"  // AR30 constants
@@ -4015,7 +4015,7 @@ void OMITFP I410ToARGBRow_AVX2(const uint16_t* y_buf,
                                uint8_t* dst_argb,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"
@@ -4120,7 +4120,7 @@ void OMITFP I410ToAR30Row_AVX2(const uint16_t* y_buf,
                                uint8_t* dst_ar30,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"  // AR30 constants
@@ -4228,7 +4228,7 @@ void OMITFP I422ToRGBARow_AVX2(const uint8_t* y_buf,
                                uint8_t* dst_argb,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "sub         %[u_buf],%[v_buf]             \n"
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"
@@ -4430,7 +4430,7 @@ void OMITFP P210ToAR30Row_AVX2(const uint16_t* y_buf,
                                uint8_t* dst_ar30,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"  // AR30 constants
       "vpsrlw      $14,%%ymm5,%%ymm5             \n"
@@ -4467,7 +4467,7 @@ void OMITFP P410ToAR30Row_AVX2(const uint16_t* y_buf,
                                uint8_t* dst_ar30,
                                const struct YuvConstants* yuvconstants,
                                int width) {
-  asm (
+  asm volatile (
     YUVTORGB_SETUP_AVX2(yuvconstants)
       "vpcmpeqb    %%ymm5,%%ymm5,%%ymm5          \n"  // AR30 constants
       "vpsrlw      $14,%%ymm5,%%ymm5             \n"
@@ -5681,9 +5681,7 @@ void MergeXRGBRow_AVX2(const uint8_t* src_r,
                        const uint8_t* src_b,
                        uint8_t* dst_argb,
                        int width) {
-  asm(
-
-      LABELALIGN
+  asm volatile (
       "1:                                        \n"
 
       "vmovdqu     (%2),%%xmm0                   \n"  // B
@@ -7381,7 +7379,7 @@ void ARGBUnattenuateRow_SSE2(const uint8_t* src_argb,
                              uint8_t* dst_argb,
                              int width) {
   uintptr_t alpha;
-  asm(
+  asm volatile (
       // 4 pixel loop.
       LABELALIGN
       "1:                                        \n"
@@ -7841,7 +7839,7 @@ void ARGBAddRow_SSE2(const uint8_t* src_argb,
                      const uint8_t* src_argb1,
                      uint8_t* dst_argb,
                      int width) {
-  asm(
+  asm volatile (
       // 4 pixel loop.
       LABELALIGN
       "1:                                        \n"
@@ -7869,7 +7867,7 @@ void ARGBAddRow_AVX2(const uint8_t* src_argb,
                      const uint8_t* src_argb1,
                      uint8_t* dst_argb,
                      int width) {
-  asm(
+  asm volatile (
       // 4 pixel loop.
       LABELALIGN
       "1:                                        \n"
@@ -7897,7 +7895,7 @@ void ARGBSubtractRow_SSE2(const uint8_t* src_argb,
                           const uint8_t* src_argb1,
                           uint8_t* dst_argb,
                           int width) {
-  asm(
+  asm volatile (
       // 4 pixel loop.
       LABELALIGN
       "1:                                        \n"
@@ -7925,7 +7923,7 @@ void ARGBSubtractRow_AVX2(const uint8_t* src_argb,
                           const uint8_t* src_argb1,
                           uint8_t* dst_argb,
                           int width) {
-  asm(
+  asm volatile (
       // 4 pixel loop.
       LABELALIGN
       "1:                                        \n"
@@ -9099,7 +9097,7 @@ void ARGBColorTableRow_X86(uint8_t* dst_argb,
                            const uint8_t* table_argb,
                            int width) {
   uintptr_t pixel_temp;
-  asm(
+  asm volatile (
       // 1 pixel loop.
       LABELALIGN
       "1:                                        \n"
@@ -9132,7 +9130,7 @@ void RGBColorTableRow_X86(uint8_t* dst_argb,
                           const uint8_t* table_argb,
                           int width) {
   uintptr_t pixel_temp;
-  asm(
+  asm volatile (
       // 1 pixel loop.
       LABELALIGN
       "1:                                        \n"
diff --git a/source/row_lsx.cc b/source/row_lsx.cc
index 09f206cab..ee74cad9f 100644
--- a/source/row_lsx.cc
+++ b/source/row_lsx.cc
@@ -2805,7 +2805,8 @@ static void ARGBToYMatrixRow_LSX(const uint8_t* src_argb,
                                  uint8_t* dst_y,
                                  int width,
                                  const struct RgbConstants* rgbconstants) {
-  asm("vldrepl.b      $vr0,  %3,    0             \n\t"  // load rgbconstants
+  asm volatile (
+      "vldrepl.b      $vr0,  %3,    0             \n\t"  // load rgbconstants
       "vldrepl.b      $vr1,  %3,    1             \n\t"  // load rgbconstants
       "vldrepl.b      $vr2,  %3,    2             \n\t"  // load rgbconstants
       "vldrepl.h      $vr3,  %3,    4             \n\t"  // load rgbconstants
@@ -2863,7 +2864,8 @@ static void RGBAToYMatrixRow_LSX(const uint8_t* src_rgba,
                                  uint8_t* dst_y,
                                  int width,
                                  const struct RgbConstants* rgbconstants) {
-  asm("vldrepl.b      $vr0,  %3,    0             \n\t"  // load rgbconstants
+  asm volatile (
+      "vldrepl.b      $vr0,  %3,    0             \n\t"  // load rgbconstants
       "vldrepl.b      $vr1,  %3,    1             \n\t"  // load rgbconstants
       "vldrepl.b      $vr2,  %3,    2             \n\t"  // load rgbconstants
       "vldrepl.h      $vr3,  %3,    4             \n\t"  // load rgbconstants
@@ -2920,7 +2922,8 @@ static void RGBToYMatrixRow_LSX(const uint8_t* src_rgba,
                       7,  9,  10, 12, 13, 15, 1,  0,  4,  0,  7,  0,  10,
                       0,  13, 0,  16, 0,  19, 0,  22, 0,  25, 0,  28, 0,
                       31, 0,  2,  0,  5,  0,  8,  0,  11, 0,  14, 0};
-  asm("vldrepl.b      $vr0,  %3,    0             \n\t"  // load rgbconstants
+  asm volatile (
+      "vldrepl.b      $vr0,  %3,    0             \n\t"  // load rgbconstants
       "vldrepl.b      $vr1,  %3,    1             \n\t"  // load rgbconstants
       "vldrepl.b      $vr2,  %3,    2             \n\t"  // load rgbconstants
       "vldrepl.h      $vr3,  %3,    4             \n\t"  // load rgbconstants
diff --git a/source/row_neon64.cc b/source/row_neon64.cc
index 8a4193196..bbe33842c 100644
--- a/source/row_neon64.cc
+++ b/source/row_neon64.cc
@@ -331,7 +331,8 @@ void I212ToAR30Row_NEON(const uint16_t* src_y,
   const uvec8* uv_coeff = &yuvconstants->kUVCoeff;
   const vec16* rgb_coeff = &yuvconstants->kRGBCoeffBias;
   const uint16_t limit = 0x3ff0;
-  asm(YUVTORGB_SETUP
+  asm volatile (
+      YUVTORGB_SETUP
       "dup      v22.8h, %w[limit]                  \n"
       "movi     v23.8h, #0xc0, lsl #8              \n"  // A
       "1:                                          \n" READYUV212 NVTORGB
@@ -400,7 +401,8 @@ void I212ToARGBRow_NEON(const uint16_t* src_y,
                         int width) {
   const uvec8* uv_coeff = &yuvconstants->kUVCoeff;
   const vec16* rgb_coeff = &yuvconstants->kRGBCoeffBias;
-  asm(YUVTORGB_SETUP
+  asm volatile (
+      YUVTORGB_SETUP
       "movi        v19.8b, #255             \n"
       "1:                                   \n" READYUV212 NVTORGB RGBTORGB8
       "subs        %w[width], %w[width], #8 \n"
@@ -449,7 +451,8 @@ void I422ToAR30Row_NEON(const uint8_t* src_y,
   const uvec8* uv_coeff = &yuvconstants->kUVCoeff;
   const vec16* rgb_coeff = &yuvconstants->kRGBCoeffBias;
   const uint16_t limit = 0x3ff0;
-  asm(YUVTORGB_SETUP
+  asm volatile (
+      YUVTORGB_SETUP
       "dup      v22.8h, %w[limit]                  \n"
       "movi     v23.8h, #0xc0, lsl #8              \n"  // A
       "1:                                          \n" READYUV422 I4XXTORGB
@@ -4134,7 +4137,8 @@ void ARGBColorMatrixRow_NEON_I8MM(const uint8_t* src_argb,
                                   uint8_t* dst_argb,
                                   const int8_t* matrix_argb,
                                   int width) {
-  asm("ld1        {v31.16b}, [%[matrix_argb]]           \n"
+  asm volatile (
+      "ld1        {v31.16b}, [%[matrix_argb]]           \n"
 
       "1:                                               \n"
       "ld1        {v0.16b, v1.16b}, [%[src_argb]], #32  \n"
diff --git a/unit_test/cpu_test.cc b/unit_test/cpu_test.cc
index b551ddd52..ff353bba3 100644
--- a/unit_test/cpu_test.cc
+++ b/unit_test/cpu_test.cc
@@ -8,6 +8,7 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
+#include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 
@@ -66,18 +67,24 @@ TEST_F(LibYUVBaseTest, TestCpuHas) {
     printf("Has SVE2 0x%x\n", has_sve2);
     printf("Has SME 0x%x\n", has_sme);
 
-#if defined(__aarch64__)
+#if __aarch64__
     // Read and print the SVE and SME vector lengths.
     if (has_sve) {
       int sve_vl;
-      // rdvl x0, #1
-      asm(".inst 0x04bf5020; mov %w0, w0" : "=r"(sve_vl)::"x0");
+      asm(".inst 0x04bf5020    \n"  // rdvl x0, #1
+          "mov %w[sve_vl], w0  \n"
+          : [sve_vl] "=r"(sve_vl)  // %[sve_vl]
+          :
+          : "x0");
       printf("SVE vector length: %d bytes\n", sve_vl);
     }
     if (has_sme) {
       int sme_vl;
-      // rdsvl x0, #1
-      asm(".inst 0x04bf5820; mov %w0, w0" : "=r"(sme_vl)::"x0");
+      asm(".inst 0x04bf5820    \n"  // rdsvl x0, #1
+          "mov %w[sme_vl], w0  \n"
+          : [sme_vl] "=r"(sme_vl)  // %[sme_vl]
+          :
+          : "x0");
       printf("SME vector length: %d bytes\n", sme_vl);
     }
 #endif  // defined(__aarch64__)
diff --git a/util/cpuid.c b/util/cpuid.c
index 725bc928b..52fb6bd99 100644
--- a/util/cpuid.c
+++ b/util/cpuid.c
@@ -69,18 +69,24 @@ int main(int argc, const char* argv[]) {
     printf("Has SVE2 0x%x\n", has_sve2);
     printf("Has SME 0x%x\n", has_sme);
 
-#if defined(__aarch64__)
+#if __aarch64__
     // Read and print the SVE and SME vector lengths.
     if (has_sve) {
       int sve_vl;
-      // rdvl x0, #1
-      asm(".inst 0x04bf5020; mov %w0, w0" : "=r"(sve_vl)::"x0");
+      asm(".inst 0x04bf5020    \n"  // rdvl x0, #1
+          "mov %w[sve_vl], w0  \n"
+          : [sve_vl] "=r"(sve_vl)  // %[sve_vl]
+          :
+          : "x0");
       printf("SVE vector length: %d bytes\n", sve_vl);
     }
     if (has_sme) {
       int sme_vl;
-      // rdsvl x0, #1
-      asm(".inst 0x04bf5820; mov %w0, w0" : "=r"(sme_vl)::"x0");
+      asm(".inst 0x04bf5820    \n"  // rdsvl x0, #1
+          "mov %w[sme_vl], w0  \n"
+          : [sme_vl] "=r"(sme_vl)  // %[sme_vl]
+          :
+          : "x0");
       printf("SME vector length: %d bytes\n", sme_vl);
     }
 #endif  // defined(__aarch64__)