From ab540ae0c57a79c8829f0d515160bd90e4b0be83 Mon Sep 17 00:00:00 2001 From: James Zern Date: Wed, 3 Aug 2022 19:36:41 -0700 Subject: [PATCH] dsp,x86: normalize types w/_mm_cvtsi32_si128 calls fixes integer sanitizer warnings of the form: implicit conversion from type 'uint32_t' (aka 'unsigned int') of value 3724541952 (32-bit, unsigned) to type 'int' changed the value to -570425344 (32-bit, signed) Bug: b/229626362 Change-Id: Ie4d599aba88226e4e047250464ac37ca11d2cd3b --- src/dsp/dec_sse2.c | 2 +- src/dsp/lossless_sse2.c | 34 +++++++++++++++++----------------- src/dsp/yuv_sse2.c | 6 ++++-- src/dsp/yuv_sse41.c | 6 ++++-- 4 files changed, 26 insertions(+), 22 deletions(-) diff --git a/src/dsp/dec_sse2.c b/src/dsp/dec_sse2.c index 77c08c33..15a96c1b 100644 --- a/src/dsp/dec_sse2.c +++ b/src/dsp/dec_sse2.c @@ -990,7 +990,7 @@ static void RD4_SSE2(uint8_t* dst) { // Down-right const uint32_t K = dst[-1 + 2 * BPS]; const uint32_t L = dst[-1 + 3 * BPS]; const __m128i LKJI_____ = - _mm_cvtsi32_si128(L | (K << 8) | (J << 16) | (I << 24)); + _mm_cvtsi32_si128((int)(L | (K << 8) | (J << 16) | (I << 24))); const __m128i LKJIXABCD = _mm_or_si128(LKJI_____, ____XABCD); const __m128i KJIXABCD_ = _mm_srli_si128(LKJIXABCD, 1); const __m128i JIXABCD__ = _mm_srli_si128(LKJIXABCD, 2); diff --git a/src/dsp/lossless_sse2.c b/src/dsp/lossless_sse2.c index 8c6a94f7..e980806b 100644 --- a/src/dsp/lossless_sse2.c +++ b/src/dsp/lossless_sse2.c @@ -27,9 +27,9 @@ static WEBP_INLINE uint32_t ClampedAddSubtractFull_SSE2(uint32_t c0, uint32_t c1, uint32_t c2) { const __m128i zero = _mm_setzero_si128(); - const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero); - const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero); - const __m128i C2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c2), zero); + const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c0), zero); + const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c1), zero); + const __m128i C2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c2), zero); const __m128i V1 = _mm_add_epi16(C0, C1); const __m128i V2 = _mm_sub_epi16(V1, C2); const __m128i b = _mm_packus_epi16(V2, V2); @@ -41,9 +41,9 @@ static WEBP_INLINE uint32_t ClampedAddSubtractHalf_SSE2(uint32_t c0, uint32_t c1, uint32_t c2) { const __m128i zero = _mm_setzero_si128(); - const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero); - const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero); - const __m128i B0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c2), zero); + const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c0), zero); + const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c1), zero); + const __m128i B0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c2), zero); const __m128i avg = _mm_add_epi16(C1, C0); const __m128i A0 = _mm_srli_epi16(avg, 1); const __m128i A1 = _mm_sub_epi16(A0, B0); @@ -59,9 +59,9 @@ static WEBP_INLINE uint32_t ClampedAddSubtractHalf_SSE2(uint32_t c0, static WEBP_INLINE uint32_t Select_SSE2(uint32_t a, uint32_t b, uint32_t c) { int pa_minus_pb; const __m128i zero = _mm_setzero_si128(); - const __m128i A0 = _mm_cvtsi32_si128(a); - const __m128i B0 = _mm_cvtsi32_si128(b); - const __m128i C0 = _mm_cvtsi32_si128(c); + const __m128i A0 = _mm_cvtsi32_si128((int)a); + const __m128i B0 = _mm_cvtsi32_si128((int)b); + const __m128i C0 = _mm_cvtsi32_si128((int)c); const __m128i AC0 = _mm_subs_epu8(A0, C0); const __m128i CA0 = _mm_subs_epu8(C0, A0); const __m128i BC0 = _mm_subs_epu8(B0, C0); @@ -94,8 +94,8 @@ static WEBP_INLINE void Average2_uint32_SSE2(const uint32_t a0, __m128i* const avg) { // (a + b) >> 1 = ((a + b + 1) >> 1) - ((a ^ b) & 1) const __m128i ones = _mm_set1_epi8(1); - const __m128i A0 = _mm_cvtsi32_si128(a0); - const __m128i A1 = _mm_cvtsi32_si128(a1); + const __m128i A0 = _mm_cvtsi32_si128((int)a0); + const __m128i A1 = _mm_cvtsi32_si128((int)a1); const __m128i avg1 = _mm_avg_epu8(A0, A1); const __m128i one = _mm_and_si128(_mm_xor_si128(A0, A1), ones); *avg = _mm_sub_epi8(avg1, one); @@ -103,8 +103,8 @@ static WEBP_INLINE void Average2_uint32_SSE2(const uint32_t a0, static WEBP_INLINE __m128i Average2_uint32_16_SSE2(uint32_t a0, uint32_t a1) { const __m128i zero = _mm_setzero_si128(); - const __m128i A0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a0), zero); - const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero); + const __m128i A0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)a0), zero); + const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)a1), zero); const __m128i sum = _mm_add_epi16(A1, A0); return _mm_srli_epi16(sum, 1); } @@ -119,7 +119,7 @@ static WEBP_INLINE uint32_t Average3_SSE2(uint32_t a0, uint32_t a1, uint32_t a2) { const __m128i zero = _mm_setzero_si128(); const __m128i avg1 = Average2_uint32_16_SSE2(a0, a2); - const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero); + const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)a1), zero); const __m128i sum = _mm_add_epi16(avg1, A1); const __m128i avg2 = _mm_srli_epi16(sum, 1); const __m128i A2 = _mm_packus_epi16(avg2, avg2); @@ -303,7 +303,7 @@ GENERATE_PREDICTOR_2(9, upper[i + 1]) static void PredictorAdd10_SSE2(const uint32_t* in, const uint32_t* upper, int num_pixels, uint32_t* out) { int i; - __m128i L = _mm_cvtsi32_si128(out[-1]); + __m128i L = _mm_cvtsi32_si128((int)out[-1]); for (i = 0; i + 4 <= num_pixels; i += 4) { __m128i src = _mm_loadu_si128((const __m128i*)&in[i]); __m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]); @@ -351,7 +351,7 @@ static void PredictorAdd11_SSE2(const uint32_t* in, const uint32_t* upper, int num_pixels, uint32_t* out) { int i; __m128i pa; - __m128i L = _mm_cvtsi32_si128(out[-1]); + __m128i L = _mm_cvtsi32_si128((int)out[-1]); for (i = 0; i + 4 <= num_pixels; i += 4) { __m128i T = _mm_loadu_si128((const __m128i*)&upper[i]); __m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]); @@ -402,7 +402,7 @@ static void PredictorAdd12_SSE2(const uint32_t* in, const uint32_t* upper, int num_pixels, uint32_t* out) { int i; const __m128i zero = _mm_setzero_si128(); - const __m128i L8 = _mm_cvtsi32_si128(out[-1]); + const __m128i L8 = _mm_cvtsi32_si128((int)out[-1]); __m128i L = _mm_unpacklo_epi8(L8, zero); for (i = 0; i + 4 <= num_pixels; i += 4) { // Load 4 pixels at a time. diff --git a/src/dsp/yuv_sse2.c b/src/dsp/yuv_sse2.c index cae435b8..01a48f9a 100644 --- a/src/dsp/yuv_sse2.c +++ b/src/dsp/yuv_sse2.c @@ -15,10 +15,12 @@ #if defined(WEBP_USE_SSE2) -#include "src/dsp/common_sse2.h" #include #include +#include "src/dsp/common_sse2.h" +#include "src/utils/utils.h" + //----------------------------------------------------------------------------- // Convert spans of 32 pixels to various RGB formats for the fancy upsampler. @@ -74,7 +76,7 @@ static WEBP_INLINE __m128i Load_HI_16_SSE2(const uint8_t* src) { // Load and replicate the U/V samples static WEBP_INLINE __m128i Load_UV_HI_8_SSE2(const uint8_t* src) { const __m128i zero = _mm_setzero_si128(); - const __m128i tmp0 = _mm_cvtsi32_si128(*(const uint32_t*)src); + const __m128i tmp0 = _mm_cvtsi32_si128(WebPMemToInt32(src)); const __m128i tmp1 = _mm_unpacklo_epi8(zero, tmp0); return _mm_unpacklo_epi16(tmp1, tmp1); // replicate samples } diff --git a/src/dsp/yuv_sse41.c b/src/dsp/yuv_sse41.c index 579d1f74..f79b802e 100644 --- a/src/dsp/yuv_sse41.c +++ b/src/dsp/yuv_sse41.c @@ -15,10 +15,12 @@ #if defined(WEBP_USE_SSE41) -#include "src/dsp/common_sse41.h" #include #include +#include "src/dsp/common_sse41.h" +#include "src/utils/utils.h" + //----------------------------------------------------------------------------- // Convert spans of 32 pixels to various RGB formats for the fancy upsampler. @@ -74,7 +76,7 @@ static WEBP_INLINE __m128i Load_HI_16_SSE41(const uint8_t* src) { // Load and replicate the U/V samples static WEBP_INLINE __m128i Load_UV_HI_8_SSE41(const uint8_t* src) { const __m128i zero = _mm_setzero_si128(); - const __m128i tmp0 = _mm_cvtsi32_si128(*(const uint32_t*)src); + const __m128i tmp0 = _mm_cvtsi32_si128(WebPMemToInt32(src)); const __m128i tmp1 = _mm_unpacklo_epi8(zero, tmp0); return _mm_unpacklo_epi16(tmp1, tmp1); // replicate samples }