From 4667992d8ba51d60045d6fed705635a6455eb4f8 Mon Sep 17 00:00:00 2001 From: James Zern Date: Wed, 27 Jul 2022 15:22:37 -0700 Subject: [PATCH] x86: normalize type with _mm_cvtsi128_si32 prefer int in most cases w/clang -fsanitize=integer fixes warnings of the form: implicit conversion from type 'int' of value -809931979 (32-bit, signed) to type 'uint32_t' (aka 'unsigned int') changed the value to 3485035317 (32-bit, unsigned) Bug: b/229626362 Change-Id: I0c6604efc188f2660c531eddfc7aa10060637813 --- vp8/encoder/x86/denoising_sse2.c | 2 +- vpx_dsp/x86/avg_intrin_sse2.c | 2 +- vpx_dsp/x86/convolve_avx2.h | 5 ++--- vpx_dsp/x86/mem_sse2.h | 2 +- vpx_dsp/x86/sad_avx2.c | 15 ++++++--------- vpx_dsp/x86/variance_sse2.c | 2 +- vpx_dsp/x86/vpx_subpixel_4t_intrin_sse2.c | 6 +++--- vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c | 2 +- vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c | 6 +++--- 9 files changed, 19 insertions(+), 23 deletions(-) diff --git a/vp8/encoder/x86/denoising_sse2.c b/vp8/encoder/x86/denoising_sse2.c index 89cad53..f35b930 100644 --- a/vp8/encoder/x86/denoising_sse2.c +++ b/vp8/encoder/x86/denoising_sse2.c @@ -30,7 +30,7 @@ static INLINE unsigned int abs_sum_diff_16x1(__m128i acc_diff) { _mm_add_epi32(hg_fe_dc_ba, _mm_srli_si128(hg_fe_dc_ba, 8)); const __m128i hgfedcba = _mm_add_epi32(hgfe_dcba, _mm_srli_si128(hgfe_dcba, 4)); - unsigned int sum_diff = abs(_mm_cvtsi128_si32(hgfedcba)); + unsigned int sum_diff = (unsigned int)abs(_mm_cvtsi128_si32(hgfedcba)); return sum_diff; } diff --git a/vpx_dsp/x86/avg_intrin_sse2.c b/vpx_dsp/x86/avg_intrin_sse2.c index 0c4919f..015c11a 100644 --- a/vpx_dsp/x86/avg_intrin_sse2.c +++ b/vpx_dsp/x86/avg_intrin_sse2.c @@ -164,7 +164,7 @@ unsigned int vpx_highbd_avg_8x8_sse2(const uint8_t *s8, int p) { s0 = _mm_add_epi32(s0, s1); s0 = _mm_add_epi32(s0, _mm_srli_si128(s0, 8)); s0 = _mm_add_epi32(s0, _mm_srli_si128(s0, 4)); - avg = _mm_cvtsi128_si32(s0); + avg = (unsigned int)_mm_cvtsi128_si32(s0); return (avg + 32) >> 6; } diff --git a/vpx_dsp/x86/convolve_avx2.h b/vpx_dsp/x86/convolve_avx2.h index 99bc963..ebee964 100644 --- a/vpx_dsp/x86/convolve_avx2.h +++ b/vpx_dsp/x86/convolve_avx2.h @@ -129,9 +129,8 @@ static INLINE void mm256_storeu2_epi64(__m128i *const dst_ptr_1, static INLINE void mm256_storeu2_epi32(__m128i *const dst_ptr_1, __m128i *const dst_ptr_2, const __m256i *const src) { - *((uint32_t *)(dst_ptr_1)) = _mm_cvtsi128_si32(_mm256_castsi256_si128(*src)); - *((uint32_t *)(dst_ptr_2)) = - _mm_cvtsi128_si32(_mm256_extractf128_si256(*src, 1)); + *((int *)(dst_ptr_1)) = _mm_cvtsi128_si32(_mm256_castsi256_si128(*src)); + *((int *)(dst_ptr_2)) = _mm_cvtsi128_si32(_mm256_extractf128_si256(*src, 1)); } static INLINE __m256i mm256_round_epi32(const __m256i *const src, diff --git a/vpx_dsp/x86/mem_sse2.h b/vpx_dsp/x86/mem_sse2.h index 8b6d4d1..6df4773 100644 --- a/vpx_dsp/x86/mem_sse2.h +++ b/vpx_dsp/x86/mem_sse2.h @@ -33,7 +33,7 @@ static INLINE __m128i load_unaligned_u32(const void *a) { } static INLINE void store_unaligned_u32(void *const a, const __m128i v) { - const uint32_t val = _mm_cvtsi128_si32(v); + const int val = _mm_cvtsi128_si32(v); memcpy(a, &val, sizeof(val)); } diff --git a/vpx_dsp/x86/sad_avx2.c b/vpx_dsp/x86/sad_avx2.c index 3b48acd..29bedb0 100644 --- a/vpx_dsp/x86/sad_avx2.c +++ b/vpx_dsp/x86/sad_avx2.c @@ -14,7 +14,7 @@ #define FSAD64_H(h) \ unsigned int vpx_sad64x##h##_avx2(const uint8_t *src_ptr, int src_stride, \ const uint8_t *ref_ptr, int ref_stride) { \ - int i, res; \ + int i; \ __m256i sad1_reg, sad2_reg, ref1_reg, ref2_reg; \ __m256i sum_sad = _mm256_setzero_si256(); \ __m256i sum_sad_h; \ @@ -35,8 +35,7 @@ sum_sad = _mm256_add_epi32(sum_sad, sum_sad_h); \ sum_sad128 = _mm256_extracti128_si256(sum_sad, 1); \ sum_sad128 = _mm_add_epi32(_mm256_castsi256_si128(sum_sad), sum_sad128); \ - res = _mm_cvtsi128_si32(sum_sad128); \ - return res; \ + return (unsigned int)_mm_cvtsi128_si32(sum_sad128); \ } #define FSAD32_H(h) \ @@ -92,7 +91,7 @@ FSAD32 unsigned int vpx_sad64x##h##_avg_avx2( \ const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred) { \ - int i, res; \ + int i; \ __m256i sad1_reg, sad2_reg, ref1_reg, ref2_reg; \ __m256i sum_sad = _mm256_setzero_si256(); \ __m256i sum_sad_h; \ @@ -118,15 +117,14 @@ FSAD32 sum_sad = _mm256_add_epi32(sum_sad, sum_sad_h); \ sum_sad128 = _mm256_extracti128_si256(sum_sad, 1); \ sum_sad128 = _mm_add_epi32(_mm256_castsi256_si128(sum_sad), sum_sad128); \ - res = _mm_cvtsi128_si32(sum_sad128); \ - return res; \ + return (unsigned int)_mm_cvtsi128_si32(sum_sad128); \ } #define FSADAVG32_H(h) \ unsigned int vpx_sad32x##h##_avg_avx2( \ const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred) { \ - int i, res; \ + int i; \ __m256i sad1_reg, sad2_reg, ref1_reg, ref2_reg; \ __m256i sum_sad = _mm256_setzero_si256(); \ __m256i sum_sad_h; \ @@ -156,8 +154,7 @@ FSAD32 sum_sad = _mm256_add_epi32(sum_sad, sum_sad_h); \ sum_sad128 = _mm256_extracti128_si256(sum_sad, 1); \ sum_sad128 = _mm_add_epi32(_mm256_castsi256_si128(sum_sad), sum_sad128); \ - res = _mm_cvtsi128_si32(sum_sad128); \ - return res; \ + return (unsigned int)_mm_cvtsi128_si32(sum_sad128); \ } #define FSADAVG64 \ diff --git a/vpx_dsp/x86/variance_sse2.c b/vpx_dsp/x86/variance_sse2.c index a67c92a..fedc8b8 100644 --- a/vpx_dsp/x86/variance_sse2.c +++ b/vpx_dsp/x86/variance_sse2.c @@ -19,7 +19,7 @@ static INLINE unsigned int add32x4_sse2(__m128i val) { val = _mm_add_epi32(val, _mm_srli_si128(val, 8)); val = _mm_add_epi32(val, _mm_srli_si128(val, 4)); - return _mm_cvtsi128_si32(val); + return (unsigned int)_mm_cvtsi128_si32(val); } unsigned int vpx_get_mb_ss_sse2(const int16_t *src_ptr) { diff --git a/vpx_dsp/x86/vpx_subpixel_4t_intrin_sse2.c b/vpx_dsp/x86/vpx_subpixel_4t_intrin_sse2.c index 0cbd151..21a35ae 100644 --- a/vpx_dsp/x86/vpx_subpixel_4t_intrin_sse2.c +++ b/vpx_dsp/x86/vpx_subpixel_4t_intrin_sse2.c @@ -485,7 +485,7 @@ static void vpx_filter_block1d4_h4_sse2(const uint8_t *src_ptr, // Saturate and convert to 8-bit words dst_first = _mm_packus_epi16(dst_first, _mm_setzero_si128()); - *((uint32_t *)(dst_ptr)) = _mm_cvtsi128_si32(dst_first); + *((int *)(dst_ptr)) = _mm_cvtsi128_si32(dst_first); src_ptr += src_stride; dst_ptr += dst_stride; @@ -589,8 +589,8 @@ static void vpx_filter_block1d4_v4_sse2(const uint8_t *src_ptr, res_reg_0123 = _mm_packus_epi16(res_reg_0123_lo, reg_zero); // Save only half of the register (8 words) - *((uint32_t *)(dst_ptr)) = _mm_cvtsi128_si32(res_reg_m1012); - *((uint32_t *)(dst_ptr + dst_stride)) = _mm_cvtsi128_si32(res_reg_0123); + *((int *)(dst_ptr)) = _mm_cvtsi128_si32(res_reg_m1012); + *((int *)(dst_ptr + dst_stride)) = _mm_cvtsi128_si32(res_reg_0123); // Update the source by two rows src_ptr += src_stride_unrolled; diff --git a/vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c b/vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c index 6f2983a..db3c39d 100644 --- a/vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c +++ b/vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c @@ -798,7 +798,7 @@ static void vpx_filter_block1d4_h4_avx2(const uint8_t *src_ptr, // Pack to 8-bits dst = _mm_packus_epi16(dst, _mm_setzero_si128()); - *((uint32_t *)(dst_ptr)) = _mm_cvtsi128_si32(dst); + *((int *)(dst_ptr)) = _mm_cvtsi128_si32(dst); } } diff --git a/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c b/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c index ed46d62..4ea2752 100644 --- a/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c +++ b/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c @@ -580,7 +580,7 @@ static void vpx_filter_block1d4_h4_ssse3(const uint8_t *src_ptr, // Pack to 8-bits dst_first = _mm_packus_epi16(dst_first, _mm_setzero_si128()); - *((uint32_t *)(dst_ptr)) = _mm_cvtsi128_si32(dst_first); + *((int *)(dst_ptr)) = _mm_cvtsi128_si32(dst_first); src_ptr += src_stride; dst_ptr += dst_stride; @@ -666,8 +666,8 @@ static void vpx_filter_block1d4_v4_ssse3(const uint8_t *src_ptr, reg_1 = _mm_packus_epi16(reg_1, reg_1); // Save the result - *((uint32_t *)(dst_ptr)) = _mm_cvtsi128_si32(reg_0); - *((uint32_t *)(dst_ptr + dst_stride)) = _mm_cvtsi128_si32(reg_1); + *((int *)(dst_ptr)) = _mm_cvtsi128_si32(reg_0); + *((int *)(dst_ptr + dst_stride)) = _mm_cvtsi128_si32(reg_1); // Update the source by two rows src_ptr += src_stride_unrolled; -- 2.7.4