Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(527)

Issues Search
    My Issues | Starred     Open | Closed | All

Unified Diff: third_party/libwebp/dsp/lossless_sse2.c

Issue 1546003002: libwebp: update to 0.5.0 (Closed) Base URL: https://chromium.googlesource.com/chromium/src.git@master
Patch Set: rebase Created 5 years ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « third_party/libwebp/dsp/lossless_neon.c ('k') | third_party/libwebp/dsp/mips_macro.h » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: third_party/libwebp/dsp/lossless_sse2.c
diff --git a/third_party/libwebp/dsp/lossless_sse2.c b/third_party/libwebp/dsp/lossless_sse2.c
index 713090980958fe2e1f6450ce1a55f08339fc2455..2d016c2911b7db88cbd93cb0a026be588c27b6d0 100644
--- a/third_party/libwebp/dsp/lossless_sse2.c
+++ b/third_party/libwebp/dsp/lossless_sse2.c
@@ -13,9 +13,8 @@
#include "./dsp.h"
-#include <assert.h>
-
#if defined(WEBP_USE_SSE2)
+#include <assert.h>
#include <emmintrin.h>
#include "./lossless.h"
@@ -156,32 +155,14 @@ static uint32_t Predictor13(uint32_t left, const uint32_t* const top) {
//------------------------------------------------------------------------------
// Subtract-Green Transform
-static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
- const __m128i mask = _mm_set1_epi32(0x0000ff00);
- int i;
- for (i = 0; i + 4 <= num_pixels; i += 4) {
- const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]);
- const __m128i in_00g0 = _mm_and_si128(in, mask); // 00g0|00g0|...
- const __m128i in_0g00 = _mm_slli_epi32(in_00g0, 8); // 0g00|0g00|...
- const __m128i in_000g = _mm_srli_epi32(in_00g0, 8); // 000g|000g|...
- const __m128i in_0g0g = _mm_or_si128(in_0g00, in_000g);
- const __m128i out = _mm_sub_epi8(in, in_0g0g);
- _mm_storeu_si128((__m128i*)&argb_data[i], out);
- }
- // fallthrough and finish off with plain-C
- VP8LSubtractGreenFromBlueAndRed_C(argb_data + i, num_pixels - i);
-}
-
static void AddGreenToBlueAndRed(uint32_t* argb_data, int num_pixels) {
- const __m128i mask = _mm_set1_epi32(0x0000ff00);
int i;
for (i = 0; i + 4 <= num_pixels; i += 4) {
- const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]);
- const __m128i in_00g0 = _mm_and_si128(in, mask); // 00g0|00g0|...
- const __m128i in_0g00 = _mm_slli_epi32(in_00g0, 8); // 0g00|0g00|...
- const __m128i in_000g = _mm_srli_epi32(in_00g0, 8); // 000g|000g|...
- const __m128i in_0g0g = _mm_or_si128(in_0g00, in_000g);
- const __m128i out = _mm_add_epi8(in, in_0g0g);
+ const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]); // argb
+ const __m128i A = _mm_srli_epi16(in, 8); // 0 a 0 g
+ const __m128i B = _mm_shufflelo_epi16(A, _MM_SHUFFLE(2, 2, 0, 0));
+ const __m128i C = _mm_shufflehi_epi16(B, _MM_SHUFFLE(2, 2, 0, 0)); // 0g0g
+ const __m128i out = _mm_add_epi8(in, C);
_mm_storeu_si128((__m128i*)&argb_data[i], out);
}
// fallthrough and finish off with plain-C
@@ -191,94 +172,36 @@ static void AddGreenToBlueAndRed(uint32_t* argb_data, int num_pixels) {
//------------------------------------------------------------------------------
// Color Transform
-static WEBP_INLINE __m128i ColorTransformDelta(__m128i color_pred,
- __m128i color) {
- // We simulate signed 8-bit multiplication as:
- // * Left shift the two (8-bit) numbers by 8 bits,
- // * Perform a 16-bit signed multiplication and retain the higher 16-bits.
- const __m128i color_pred_shifted = _mm_slli_epi32(color_pred, 8);
- const __m128i color_shifted = _mm_slli_epi32(color, 8);
- // Note: This performs multiplication on 8 packed 16-bit numbers, 4 of which
- // happen to be zeroes.
- const __m128i signed_mult =
- _mm_mulhi_epi16(color_pred_shifted, color_shifted);
- return _mm_srli_epi32(signed_mult, 5);
-}
-
-static WEBP_INLINE void TransformColor(const VP8LMultipliers* const m,
- uint32_t* argb_data,
- int num_pixels) {
- const __m128i g_to_r = _mm_set1_epi32(m->green_to_red_); // multipliers
- const __m128i g_to_b = _mm_set1_epi32(m->green_to_blue_);
- const __m128i r_to_b = _mm_set1_epi32(m->red_to_blue_);
-
- int i;
-
- for (i = 0; i + 4 <= num_pixels; i += 4) {
- const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]);
- const __m128i alpha_green_mask = _mm_set1_epi32(0xff00ff00); // masks
- const __m128i red_mask = _mm_set1_epi32(0x00ff0000);
- const __m128i green_mask = _mm_set1_epi32(0x0000ff00);
- const __m128i lower_8bit_mask = _mm_set1_epi32(0x000000ff);
- const __m128i ag = _mm_and_si128(in, alpha_green_mask); // alpha, green
- const __m128i r = _mm_srli_epi32(_mm_and_si128(in, red_mask), 16);
- const __m128i g = _mm_srli_epi32(_mm_and_si128(in, green_mask), 8);
- const __m128i b = in;
-
- const __m128i r_delta = ColorTransformDelta(g_to_r, g); // red
- const __m128i r_new =
- _mm_and_si128(_mm_sub_epi32(r, r_delta), lower_8bit_mask);
- const __m128i r_new_shifted = _mm_slli_epi32(r_new, 16);
-
- const __m128i b_delta_1 = ColorTransformDelta(g_to_b, g); // blue
- const __m128i b_delta_2 = ColorTransformDelta(r_to_b, r);
- const __m128i b_delta = _mm_add_epi32(b_delta_1, b_delta_2);
- const __m128i b_new =
- _mm_and_si128(_mm_sub_epi32(b, b_delta), lower_8bit_mask);
-
- const __m128i out = _mm_or_si128(_mm_or_si128(ag, r_new_shifted), b_new);
- _mm_storeu_si128((__m128i*)&argb_data[i], out);
- }
-
- // Fall-back to C-version for left-overs.
- VP8LTransformColor_C(m, argb_data + i, num_pixels - i);
-}
-
-static WEBP_INLINE void TransformColorInverse(const VP8LMultipliers* const m,
- uint32_t* argb_data,
- int num_pixels) {
- const __m128i g_to_r = _mm_set1_epi32(m->green_to_red_); // multipliers
- const __m128i g_to_b = _mm_set1_epi32(m->green_to_blue_);
- const __m128i r_to_b = _mm_set1_epi32(m->red_to_blue_);
-
+static void TransformColorInverse(const VP8LMultipliers* const m,
+ uint32_t* argb_data, int num_pixels) {
+ // sign-extended multiplying constants, pre-shifted by 5.
+#define CST(X) (((int16_t)(m->X << 8)) >> 5) // sign-extend
+ const __m128i mults_rb = _mm_set_epi16(
+ CST(green_to_red_), CST(green_to_blue_),
+ CST(green_to_red_), CST(green_to_blue_),
+ CST(green_to_red_), CST(green_to_blue_),
+ CST(green_to_red_), CST(green_to_blue_));
+ const __m128i mults_b2 = _mm_set_epi16(
+ CST(red_to_blue_), 0, CST(red_to_blue_), 0,
+ CST(red_to_blue_), 0, CST(red_to_blue_), 0);
+#undef CST
+ const __m128i mask_ag = _mm_set1_epi32(0xff00ff00); // alpha-green masks
int i;
-
for (i = 0; i + 4 <= num_pixels; i += 4) {
- const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]);
- const __m128i alpha_green_mask = _mm_set1_epi32(0xff00ff00); // masks
- const __m128i red_mask = _mm_set1_epi32(0x00ff0000);
- const __m128i green_mask = _mm_set1_epi32(0x0000ff00);
- const __m128i lower_8bit_mask = _mm_set1_epi32(0x000000ff);
- const __m128i ag = _mm_and_si128(in, alpha_green_mask); // alpha, green
- const __m128i r = _mm_srli_epi32(_mm_and_si128(in, red_mask), 16);
- const __m128i g = _mm_srli_epi32(_mm_and_si128(in, green_mask), 8);
- const __m128i b = in;
-
- const __m128i r_delta = ColorTransformDelta(g_to_r, g); // red
- const __m128i r_new =
- _mm_and_si128(_mm_add_epi32(r, r_delta), lower_8bit_mask);
- const __m128i r_new_shifted = _mm_slli_epi32(r_new, 16);
-
- const __m128i b_delta_1 = ColorTransformDelta(g_to_b, g); // blue
- const __m128i b_delta_2 = ColorTransformDelta(r_to_b, r_new);
- const __m128i b_delta = _mm_add_epi32(b_delta_1, b_delta_2);
- const __m128i b_new =
- _mm_and_si128(_mm_add_epi32(b, b_delta), lower_8bit_mask);
-
- const __m128i out = _mm_or_si128(_mm_or_si128(ag, r_new_shifted), b_new);
+ const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]); // argb
+ const __m128i A = _mm_and_si128(in, mask_ag); // a 0 g 0
+ const __m128i B = _mm_shufflelo_epi16(A, _MM_SHUFFLE(2, 2, 0, 0));
+ const __m128i C = _mm_shufflehi_epi16(B, _MM_SHUFFLE(2, 2, 0, 0)); // g0g0
+ const __m128i D = _mm_mulhi_epi16(C, mults_rb); // x dr x db1
+ const __m128i E = _mm_add_epi8(in, D); // x r' x b'
+ const __m128i F = _mm_slli_epi16(E, 8); // r' 0 b' 0
+ const __m128i G = _mm_mulhi_epi16(F, mults_b2); // x db2 0 0
+ const __m128i H = _mm_srli_epi32(G, 8); // 0 x db2 0
+ const __m128i I = _mm_add_epi8(H, F); // r' x b'' 0
+ const __m128i J = _mm_srli_epi16(I, 8); // 0 r' 0 b''
+ const __m128i out = _mm_or_si128(J, A);
_mm_storeu_si128((__m128i*)&argb_data[i], out);
}
-
// Fall-back to C-version for left-overs.
VP8LTransformColorInverse_C(m, argb_data + i, num_pixels - i);
}
@@ -418,95 +341,11 @@ static void ConvertBGRAToBGR(const uint32_t* src,
}
//------------------------------------------------------------------------------
-
-#define LINE_SIZE 16 // 8 or 16
-static void AddVector(const uint32_t* a, const uint32_t* b, uint32_t* out,
- int size) {
- int i;
- assert(size % LINE_SIZE == 0);
- for (i = 0; i < size; i += LINE_SIZE) {
- const __m128i a0 = _mm_loadu_si128((__m128i*)&a[i + 0]);
- const __m128i a1 = _mm_loadu_si128((__m128i*)&a[i + 4]);
-#if (LINE_SIZE == 16)
- const __m128i a2 = _mm_loadu_si128((__m128i*)&a[i + 8]);
- const __m128i a3 = _mm_loadu_si128((__m128i*)&a[i + 12]);
-#endif
- const __m128i b0 = _mm_loadu_si128((__m128i*)&b[i + 0]);
- const __m128i b1 = _mm_loadu_si128((__m128i*)&b[i + 4]);
-#if (LINE_SIZE == 16)
- const __m128i b2 = _mm_loadu_si128((__m128i*)&b[i + 8]);
- const __m128i b3 = _mm_loadu_si128((__m128i*)&b[i + 12]);
-#endif
- _mm_storeu_si128((__m128i*)&out[i + 0], _mm_add_epi32(a0, b0));
- _mm_storeu_si128((__m128i*)&out[i + 4], _mm_add_epi32(a1, b1));
-#if (LINE_SIZE == 16)
- _mm_storeu_si128((__m128i*)&out[i + 8], _mm_add_epi32(a2, b2));
- _mm_storeu_si128((__m128i*)&out[i + 12], _mm_add_epi32(a3, b3));
-#endif
- }
-}
-
-static void AddVectorEq(const uint32_t* a, uint32_t* out, int size) {
- int i;
- assert(size % LINE_SIZE == 0);
- for (i = 0; i < size; i += LINE_SIZE) {
- const __m128i a0 = _mm_loadu_si128((__m128i*)&a[i + 0]);
- const __m128i a1 = _mm_loadu_si128((__m128i*)&a[i + 4]);
-#if (LINE_SIZE == 16)
- const __m128i a2 = _mm_loadu_si128((__m128i*)&a[i + 8]);
- const __m128i a3 = _mm_loadu_si128((__m128i*)&a[i + 12]);
-#endif
- const __m128i b0 = _mm_loadu_si128((__m128i*)&out[i + 0]);
- const __m128i b1 = _mm_loadu_si128((__m128i*)&out[i + 4]);
-#if (LINE_SIZE == 16)
- const __m128i b2 = _mm_loadu_si128((__m128i*)&out[i + 8]);
- const __m128i b3 = _mm_loadu_si128((__m128i*)&out[i + 12]);
-#endif
- _mm_storeu_si128((__m128i*)&out[i + 0], _mm_add_epi32(a0, b0));
- _mm_storeu_si128((__m128i*)&out[i + 4], _mm_add_epi32(a1, b1));
-#if (LINE_SIZE == 16)
- _mm_storeu_si128((__m128i*)&out[i + 8], _mm_add_epi32(a2, b2));
- _mm_storeu_si128((__m128i*)&out[i + 12], _mm_add_epi32(a3, b3));
-#endif
- }
-}
-#undef LINE_SIZE
-
-// Note we are adding uint32_t's as *signed* int32's (using _mm_add_epi32). But
-// that's ok since the histogram values are less than 1<<28 (max picture size).
-static void HistogramAdd(const VP8LHistogram* const a,
- const VP8LHistogram* const b,
- VP8LHistogram* const out) {
- int i;
- const int literal_size = VP8LHistogramNumCodes(a->palette_code_bits_);
- assert(a->palette_code_bits_ == b->palette_code_bits_);
- if (b != out) {
- AddVector(a->literal_, b->literal_, out->literal_, NUM_LITERAL_CODES);
- AddVector(a->red_, b->red_, out->red_, NUM_LITERAL_CODES);
- AddVector(a->blue_, b->blue_, out->blue_, NUM_LITERAL_CODES);
- AddVector(a->alpha_, b->alpha_, out->alpha_, NUM_LITERAL_CODES);
- } else {
- AddVectorEq(a->literal_, out->literal_, NUM_LITERAL_CODES);
- AddVectorEq(a->red_, out->red_, NUM_LITERAL_CODES);
- AddVectorEq(a->blue_, out->blue_, NUM_LITERAL_CODES);
- AddVectorEq(a->alpha_, out->alpha_, NUM_LITERAL_CODES);
- }
- for (i = NUM_LITERAL_CODES; i < literal_size; ++i) {
- out->literal_[i] = a->literal_[i] + b->literal_[i];
- }
- for (i = 0; i < NUM_DISTANCE_CODES; ++i) {
- out->distance_[i] = a->distance_[i] + b->distance_[i];
- }
-}
-
-#endif // WEBP_USE_SSE2
-
-//------------------------------------------------------------------------------
+// Entry point
extern void VP8LDspInitSSE2(void);
-void VP8LDspInitSSE2(void) {
-#if defined(WEBP_USE_SSE2)
+WEBP_TSAN_IGNORE_FUNCTION void VP8LDspInitSSE2(void) {
VP8LPredictors[5] = Predictor5;
VP8LPredictors[6] = Predictor6;
VP8LPredictors[7] = Predictor7;
@@ -517,19 +356,17 @@ void VP8LDspInitSSE2(void) {
VP8LPredictors[12] = Predictor12;
VP8LPredictors[13] = Predictor13;
- VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed;
VP8LAddGreenToBlueAndRed = AddGreenToBlueAndRed;
-
- VP8LTransformColor = TransformColor;
VP8LTransformColorInverse = TransformColorInverse;
VP8LConvertBGRAToRGBA = ConvertBGRAToRGBA;
VP8LConvertBGRAToRGBA4444 = ConvertBGRAToRGBA4444;
VP8LConvertBGRAToRGB565 = ConvertBGRAToRGB565;
VP8LConvertBGRAToBGR = ConvertBGRAToBGR;
-
- VP8LHistogramAdd = HistogramAdd;
-#endif // WEBP_USE_SSE2
}
-//------------------------------------------------------------------------------
+#else // !WEBP_USE_SSE2
+
+WEBP_DSP_INIT_STUB(VP8LDspInitSSE2)
+
+#endif // WEBP_USE_SSE2
« no previous file with comments | « third_party/libwebp/dsp/lossless_neon.c ('k') | third_party/libwebp/dsp/mips_macro.h » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698