This prevents those symbols from being exposed in applications or shared
libraries that link statically with libjpeg-turbo.
+13. Added Loongson MMI SIMD implementations of the RGB-to-YCbCr and
+YCbCr-to-RGB colorspace conversion, 4:2:0 chroma downsampling, 4:2:0 fancy
+chroma upsampling, integer quantization, and slow integer DCT/IDCT algorithms.
+When using the slow integer DCT/IDCT, this speeds up the compression of RGB
+images by approximately 70-100% and the decompression of RGB images by
+approximately 2-3.5x.
+
1.5.3
=====
set_target_properties(simd PROPERTIES POSITION_INDEPENDENT_CODE 1)
endif()
+###############################################################################
+# Loongson (Intrinsics)
+###############################################################################
+
+elseif(CPU_TYPE STREQUAL "loongson")
+
+set(SIMD_SOURCES loongson/jccolor-mmi.c loongson/jcsample-mmi.c
+ loongson/jdcolor-mmi.c loongson/jdsample-mmi.c loongson/jfdctint-mmi.c
+ loongson/jidctint-mmi.c loongson/jquanti-mmi.c)
+
+if(CMAKE_COMPILER_IS_GNUCC)
+ foreach(file ${SIMD_SOURCES})
+ set_property(SOURCE ${file} APPEND_STRING PROPERTY COMPILE_FLAGS
+ " -fno-strict-aliasing")
+ endforeach()
+endif()
+
+add_library(simd OBJECT ${SIMD_SOURCES} loongson/jsimd.c)
+
+if(CMAKE_POSITION_INDEPENDENT_CODE OR ENABLE_SHARED)
+ set_target_properties(simd PROPERTIES POSITION_INDEPENDENT_CODE 1)
+endif()
###############################################################################
# PowerPC (Intrinsics)
* Copyright (C) 2013-2014, MIPS Technologies, Inc., California.
* Copyright (C) 2014, Linaro Limited.
* Copyright (C) 2015-2016, Matthieu Darbois.
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
*
* Based on the x86 SIMD extension for IJG JPEG library,
* Copyright (C) 1999-2006, MIYASAKA Masaru.
#define JSIMD_DSPR2 0x20
#define JSIMD_ALTIVEC 0x40
#define JSIMD_AVX2 0x80
+#define JSIMD_MMI 0x100
/* SIMD Ext: retrieve SIMD/CPU information */
EXTERN(unsigned int) jpeg_simd_cpu_support (void);
(JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_rgb_ycc_convert_mmi
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgb_ycc_convert_mmi
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgbx_ycc_convert_mmi
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgr_ycc_convert_mmi
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgrx_ycc_convert_mmi
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxbgr_ycc_convert_mmi
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxrgb_ycc_convert_mmi
+ (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows);
+
EXTERN(void) jsimd_rgb_ycc_convert_altivec
(JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
JDIMENSION output_row, int num_rows);
(JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_rgb_convert_mmi
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extrgb_convert_mmi
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extrgbx_convert_mmi
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extbgr_convert_mmi
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extbgrx_convert_mmi
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extxbgr_convert_mmi
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+EXTERN(void) jsimd_ycc_extxrgb_convert_mmi
+ (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows);
+
EXTERN(void) jsimd_ycc_rgb_convert_altivec
(JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row,
JSAMPARRAY output_buf, int num_rows);
JDIMENSION v_samp_factor, JDIMENSION width_in_blocks,
JSAMPARRAY input_data, JSAMPARRAY output_data);
+EXTERN(void) jsimd_h2v2_downsample_mmi
+ (JDIMENSION image_width, int max_v_samp_factor,
+ JDIMENSION v_samp_factor, JDIMENSION width_in_blocks,
+ JSAMPARRAY input_data, JSAMPARRAY output_data);
+
EXTERN(void) jsimd_h2v2_downsample_altivec
(JDIMENSION image_width, int max_v_samp_factor,
JDIMENSION v_samp_factor, JDIMENSION width_in_blocks,
(int max_v_samp_factor, JDIMENSION downsampled_width,
JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
+EXTERN(void) jsimd_h2v2_fancy_upsample_mmi
+ (int max_v_samp_factor, JDIMENSION downsampled_width,
+ JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
+
EXTERN(void) jsimd_h2v1_fancy_upsample_altivec
(int max_v_samp_factor, JDIMENSION downsampled_width,
JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr);
EXTERN(void) jsimd_fdct_islow_dspr2 (DCTELEM *data);
+EXTERN(void) jsimd_fdct_islow_mmi (DCTELEM *data);
+
EXTERN(void) jsimd_fdct_islow_altivec (DCTELEM *data);
/* Fast Integer Forward DCT */
EXTERN(void) jsimd_quantize_dspr2
(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace);
+EXTERN(void) jsimd_quantize_mmi
+ (JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace);
+
EXTERN(void) jsimd_quantize_altivec
(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace);
(void *dct_table, JCOEFPTR coef_block, int *output_buf,
JSAMPLE *output_col);
+EXTERN(void) jsimd_idct_islow_mmi
+ (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf,
+ JDIMENSION output_col);
+
EXTERN(void) jsimd_idct_islow_altivec
(void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf,
JDIMENSION output_col);
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright (C) 2014-2015, D. R. Commander. All Rights Reserved.
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ * Authors: ZhuChen <zhuchen@loongson.cn>
+ * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
+ * CaiWanwei <caiwanwei@loongson.cn>
+ *
+ * Based on the x86 SIMD extension for IJG JPEG library
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* This file is included by jccolor-mmi.c */
+
+
+#if RGB_RED == 0
+#define mmA mm0
+#define mmB mm1
+#elif RGB_GREEN == 0
+#define mmA mm2
+#define mmB mm3
+#elif RGB_BLUE == 0
+#define mmA mm4
+#define mmB mm5
+#else
+#define mmA mm6
+#define mmB mm7
+#endif
+
+#if RGB_RED == 1
+#define mmC mm0
+#define mmD mm1
+#elif RGB_GREEN == 1
+#define mmC mm2
+#define mmD mm3
+#elif RGB_BLUE == 1
+#define mmC mm4
+#define mmD mm5
+#else
+#define mmC mm6
+#define mmD mm7
+#endif
+
+#if RGB_RED == 2
+#define mmE mm0
+#define mmF mm1
+#elif RGB_GREEN == 2
+#define mmE mm2
+#define mmF mm3
+#elif RGB_BLUE == 2
+#define mmE mm4
+#define mmF mm5
+#else
+#define mmE mm6
+#define mmF mm7
+#endif
+
+#if RGB_RED == 3
+#define mmG mm0
+#define mmH mm1
+#elif RGB_GREEN == 3
+#define mmG mm2
+#define mmH mm3
+#elif RGB_BLUE == 3
+#define mmG mm4
+#define mmH mm5
+#else
+#define mmG mm6
+#define mmH mm7
+#endif
+
+
+void
+jsimd_rgb_ycc_convert_mmi (JDIMENSION image_width, JSAMPARRAY input_buf,
+ JSAMPIMAGE output_buf, JDIMENSION output_row,
+ int num_rows)
+{
+ JSAMPROW inptr, outptr0, outptr1, outptr2;
+ int num_cols, col;
+ __m64 mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7;
+ __m64 wk[7];
+ __m64 Y_BG, Cb_RG, Cr_BG;
+
+ while (--num_rows >= 0) {
+ inptr = *input_buf++;
+ outptr0 = output_buf[0][output_row];
+ outptr1 = output_buf[1][output_row];
+ outptr2 = output_buf[2][output_row];
+ output_row++;
+
+ for (num_cols = image_width; num_cols > 0; num_cols -= 8,
+ outptr0 += 8, outptr1 += 8, outptr2 += 8) {
+
+#if RGB_PIXELSIZE == 3
+
+ if (num_cols < 8) {
+ col = num_cols * 3;
+ asm(".set noreorder\r\n"
+
+ "li $8, 1\r\n"
+ "move $9, %3\r\n"
+ "and $10, $9, $8\r\n"
+ "beqz $10, 1f\r\n"
+ "nop \r\n"
+ "subu $9, $9, 1\r\n"
+ "xor $12, $12, $12\r\n"
+ "move $13, %5\r\n"
+ "dadd $13, $13, $9\r\n"
+ "lbu $12, 0($13)\r\n"
+
+ "1: \r\n"
+ "li $8, 2\r\n"
+ "and $10, $9, $8\r\n"
+ "beqz $10, 2f\r\n"
+ "nop \r\n"
+ "subu $9, $9, 2\r\n"
+ "xor $11, $11, $11\r\n"
+ "move $13, %5\r\n"
+ "dadd $13, $13, $9\r\n"
+ "lhu $11, 0($13)\r\n"
+ "sll $12, $12, 16\r\n"
+ "or $12, $12, $11\r\n"
+
+ "2: \r\n"
+ "dmtc1 $12, %0\r\n"
+ "li $8, 4\r\n"
+ "and $10, $9, $8\r\n"
+ "beqz $10, 3f\r\n"
+ "nop \r\n"
+ "subu $9, $9, 4\r\n"
+ "move $13, %5\r\n"
+ "dadd $13, $13, $9\r\n"
+ "lwu $14, 0($13)\r\n"
+ "dmtc1 $14, %1\r\n"
+ "dsll32 $12, $12, 0\r\n"
+ "or $12, $12, $14\r\n"
+ "dmtc1 $12, %0\r\n"
+
+ "3: \r\n"
+ "li $8, 8\r\n"
+ "and $10, $9, $8\r\n"
+ "beqz $10, 4f\r\n"
+ "nop \r\n"
+ "mov.s %1, %0\r\n"
+ "ldc1 %0, 0(%5)\r\n"
+ "li $9, 8\r\n"
+ "j 5f\r\n"
+ "nop \r\n"
+
+ "4: \r\n"
+ "li $8, 16\r\n"
+ "and $10, $9, $8\r\n"
+ "beqz $10, 5f\r\n"
+ "nop \r\n"
+ "mov.s %2, %0\r\n"
+ "ldc1 %0, 0(%5)\r\n"
+ "ldc1 %1, 8(%5)\r\n"
+
+ "5: \r\n"
+ "nop \r\n"
+ ".set reorder\r\n"
+
+ : "=f" (mmA), "=f" (mmG), "=f" (mmF)
+ : "r" (col), "r" (num_rows), "r" (inptr)
+ : "$f0", "$f2", "$f4", "$8", "$9", "$10", "$11", "$12", "$13",
+ "$14", "memory"
+ );
+ } else {
+ mmA = _mm_load_si64((__m64 *)&inptr[0]);
+ mmG = _mm_load_si64((__m64 *)&inptr[8]);
+ mmF = _mm_load_si64((__m64 *)&inptr[16]);
+ inptr += RGB_PIXELSIZE*8;
+ }
+ mmD = mmA;
+ mmA = _mm_slli_si64(mmA, 4*BYTE_BIT);
+ mmD = _mm_srli_si64(mmD, 4*BYTE_BIT);
+
+ mmA = _mm_unpackhi_pi8(mmA, mmG);
+ mmG = _mm_slli_si64(mmG, 4*BYTE_BIT);
+
+ mmD = _mm_unpacklo_pi8(mmD, mmF);
+ mmG = _mm_unpackhi_pi8(mmG, mmF);
+
+ mmE = mmA;
+ mmA = _mm_slli_si64(mmA, 4*BYTE_BIT);
+ mmE = _mm_srli_si64(mmE, 4*BYTE_BIT);
+
+ mmA = _mm_unpackhi_pi8(mmA, mmD);
+ mmD = _mm_slli_si64(mmD, 4*BYTE_BIT);
+
+ mmE = _mm_unpacklo_pi8(mmE, mmG);
+ mmD = _mm_unpackhi_pi8(mmD, mmG);
+ mmC = mmA;
+ mmA = _mm_loadlo_pi8_f(mmA);
+ mmC = _mm_loadhi_pi8_f(mmC);
+
+ mmB = mmE;
+ mmE = _mm_loadlo_pi8_f(mmE);
+ mmB = _mm_loadhi_pi8_f(mmB);
+
+ mmF = mmD;
+ mmD = _mm_loadlo_pi8_f(mmD);
+ mmF = _mm_loadhi_pi8_f(mmF);
+
+#else /* RGB_PIXELSIZE == 4 */
+
+ if (num_cols < 8) {
+ col = num_cols;
+ asm(".set noreorder\r\n"
+
+ "li $8, 1\r\n"
+ "move $9, %4\r\n"
+ "and $10, $9, $8\r\n"
+ "beqz $10, 1f\r\n"
+ "nop \r\n"
+ "subu $9, $9, 1\r\n"
+ "dsll $11, $9, 2\r\n"
+ "move $13, %5\r\n"
+ "daddu $13, $13, $11\r\n"
+ "lwc1 %0, 0($13)\r\n"
+
+ "1: \r\n"
+ "li $8, 2\r\n"
+ "and $10, $9, $8\r\n"
+ "beqz $10, 2f\r\n"
+ "nop \r\n"
+ "subu $9, $9, 2\r\n"
+ "dsll $11, $9, 2\r\n"
+ "move $13, %5\r\n"
+ "daddu $13, $13, $11\r\n"
+ "mov.s %1, %0\r\n"
+ "ldc1 %0, 0($13)\r\n"
+
+ "2: \r\n"
+ "li $8, 4\r\n"
+ "and $10, $9, $8\r\n"
+ "beqz $10, 3f\r\n"
+ "nop \r\n"
+ "mov.s %2, %0\r\n"
+ "mov.s %3, %1\r\n"
+ "ldc1 %0, 0(%5)\r\n"
+ "ldc1 %1, 8(%5)\r\n"
+
+ "3: \r\n"
+ "nop \r\n"
+ ".set reorder\r\n"
+
+ : "=f" (mmA), "=f" (mmF), "=f" (mmD), "=f" (mmC)
+ : "r" (col), "r" (inptr)
+ : "$f0", "$f2", "$8", "$9", "$10", "$11", "$13", "memory"
+ );
+ } else {
+ mmA = _mm_load_si64((__m64 *)&inptr[0]);
+ mmF = _mm_load_si64((__m64 *)&inptr[8]);
+ mmD = _mm_load_si64((__m64 *)&inptr[16]);
+ mmC = _mm_load_si64((__m64 *)&inptr[24]);
+ inptr += RGB_PIXELSIZE*8;
+ }
+ mmB = mmA;
+ mmA = _mm_unpacklo_pi8(mmA, mmF);
+ mmB = _mm_unpackhi_pi8(mmB, mmF);
+
+ mmG = mmD;
+ mmD = _mm_unpacklo_pi8(mmD, mmC);
+ mmG = _mm_unpackhi_pi8(mmG, mmC);
+
+ mmE = mmA;
+ mmA = _mm_unpacklo_pi16(mmA, mmD);
+ mmE = _mm_unpackhi_pi16(mmE, mmD);
+
+ mmH = mmB;
+ mmB = _mm_unpacklo_pi16(mmB, mmG);
+ mmH = _mm_unpackhi_pi16(mmH, mmG);
+
+ mmC = mmA;
+ mmA = _mm_loadlo_pi8_f(mmA);
+ mmC = _mm_loadhi_pi8_f(mmC);
+
+ mmD = mmB;
+ mmB = _mm_loadlo_pi8_f(mmB);
+ mmD = _mm_loadhi_pi8_f(mmD);
+
+ mmG = mmE;
+ mmE = _mm_loadlo_pi8_f(mmE);
+ mmG = _mm_loadhi_pi8_f(mmG);
+
+ mmF = mmH;
+ mmF = _mm_unpacklo_pi8(mmF, mmH);
+ mmH = _mm_unpackhi_pi8(mmH, mmH);
+ mmF = _mm_srli_pi16(mmF, BYTE_BIT);
+ mmH = _mm_srli_pi16(mmH, BYTE_BIT);
+
+#endif
+
+ wk[0] = mm0;
+ wk[1] = mm1;
+ wk[2] = mm4;
+ wk[3] = mm5;
+
+ mm6 = mm1;
+ mm1 = _mm_unpacklo_pi16(mm1, mm3);
+ mm6 = _mm_unpackhi_pi16(mm6, mm3);
+ mm7 = mm1;
+ mm4 = mm6;
+ mm1 = _mm_madd_pi16(mm1, PW_F0299_F0337);
+ mm6 = _mm_madd_pi16(mm6, PW_F0299_F0337);
+ mm7 = _mm_madd_pi16(mm7, PW_MF016_MF033);
+ mm4 = _mm_madd_pi16(mm4, PW_MF016_MF033);
+
+ wk[4] = mm1;
+ wk[5] = mm6;
+
+ mm1 = _mm_loadlo_pi16_f(mm5);
+ mm6 = _mm_loadhi_pi16_f(mm5);
+ mm1 = _mm_srli_pi32(mm1, 1);
+ mm6 = _mm_srli_pi32(mm6, 1);
+
+ mm5 = PD_ONEHALFM1_CJ;
+ mm7 = _mm_add_pi32(mm7, mm1);
+ mm4 = _mm_add_pi32(mm4, mm6);
+ mm7 = _mm_add_pi32(mm7, mm5);
+ mm4 = _mm_add_pi32(mm4, mm5);
+ mm7 = _mm_srli_pi32(mm7, SCALEBITS);
+ mm4 = _mm_srli_pi32(mm4, SCALEBITS);
+ mm7 = _mm_packs_pi32(mm7, mm4);
+
+ mm1 = wk[2];
+ mm6 = mm0;
+ mm0 = _mm_unpacklo_pi16(mm0, mm2);
+ mm6 = _mm_unpackhi_pi16(mm6, mm2);
+ mm5 = mm0;
+ mm4 = mm6;
+ mm0 = _mm_madd_pi16(mm0, PW_F0299_F0337);
+ mm6 = _mm_madd_pi16(mm6, PW_F0299_F0337);
+ mm5 = _mm_madd_pi16(mm5, PW_MF016_MF033);
+ mm4 = _mm_madd_pi16(mm4, PW_MF016_MF033);
+
+ wk[6] = mm0;
+ wk[7] = mm6;
+ mm0 = _mm_loadlo_pi16_f(mm1);
+ mm6 = _mm_loadhi_pi16_f(mm1);
+ mm0 = _mm_srli_pi32(mm0, 1);
+ mm6 = _mm_srli_pi32(mm6, 1);
+
+ mm1 = PD_ONEHALFM1_CJ;
+ mm5 = _mm_add_pi32(mm5, mm0);
+ mm4 = _mm_add_pi32(mm4, mm6);
+ mm5 = _mm_add_pi32(mm5, mm1);
+ mm4 = _mm_add_pi32(mm4, mm1);
+ mm5 = _mm_srli_pi32(mm5, SCALEBITS);
+ mm4 = _mm_srli_pi32(mm4, SCALEBITS);
+ mm5 = _mm_packs_pi32(mm5, mm4);
+
+ mm7 = _mm_slli_pi16(mm7, BYTE_BIT);
+ mm5 = _mm_or_si64(mm5, mm7);
+ Cb_RG = mm5;
+
+ mm0 = wk[3];
+ mm6 = wk[2];
+ mm1 = wk[1];
+
+ mm4 = mm0;
+ mm0 = _mm_unpacklo_pi16(mm0, mm3);
+ mm4 = _mm_unpackhi_pi16(mm4, mm3);
+ mm7 = mm0;
+ mm5 = mm4;
+ mm0 = _mm_madd_pi16(mm0, PW_F0114_F0250);
+ mm4 = _mm_madd_pi16(mm4, PW_F0114_F0250);
+ mm7 = _mm_madd_pi16(mm7, PW_MF008_MF041);
+ mm5 = _mm_madd_pi16(mm5, PW_MF008_MF041);
+
+ mm3 = PD_ONEHALF;
+ mm0 = _mm_add_pi32(mm0, wk[4]);
+ mm4 = _mm_add_pi32(mm4, wk[5]);
+ mm0 = _mm_add_pi32(mm0, mm3);
+ mm4 = _mm_add_pi32(mm4, mm3);
+ mm0 = _mm_srli_pi32(mm0, SCALEBITS);
+ mm4 = _mm_srli_pi32(mm4, SCALEBITS);
+ mm0 = _mm_packs_pi32(mm0,mm4);
+
+ mm3 = _mm_loadlo_pi16_f(mm1);
+ mm4 = _mm_loadhi_pi16_f(mm1);
+ mm3 = _mm_srli_pi32(mm3, 1);
+ mm4 = _mm_srli_pi32(mm4, 1);
+
+ mm1 = PD_ONEHALFM1_CJ;
+ mm7 = _mm_add_pi32(mm7, mm3);
+ mm5 = _mm_add_pi32(mm5, mm4);
+ mm7 = _mm_add_pi32(mm7, mm1);
+ mm5 = _mm_add_pi32(mm5, mm1);
+ mm7 = _mm_srli_pi32(mm7, SCALEBITS);
+ mm5 = _mm_srli_pi32(mm5, SCALEBITS);
+ mm7 = _mm_packs_pi32(mm7, mm5);
+
+ mm3 = wk[0];
+ mm4 = mm6;
+ mm6 = _mm_unpacklo_pi16(mm6, mm2);
+ mm4 = _mm_unpackhi_pi16(mm4, mm2);
+ mm1 = mm6;
+ mm5 = mm4;
+ mm6 = _mm_madd_pi16(mm6, PW_F0114_F0250);
+ mm4 = _mm_madd_pi16(mm4, PW_F0114_F0250);
+ mm1 = _mm_madd_pi16(mm1, PW_MF008_MF041);
+ mm5 = _mm_madd_pi16(mm5, PW_MF008_MF041);
+
+ mm2 = PD_ONEHALF;
+ mm6 = _mm_add_pi32(mm6, wk[6]);
+ mm4 = _mm_add_pi32(mm4, wk[7]);
+ mm6 = _mm_add_pi32(mm6, mm2);
+ mm4 = _mm_add_pi32(mm4, mm2);
+ mm6 = _mm_srli_pi32(mm6, SCALEBITS);
+ mm4 = _mm_srli_pi32(mm4, SCALEBITS);
+ mm6 = _mm_packs_pi32(mm6, mm4);
+
+ mm0 = _mm_slli_pi16(mm0, BYTE_BIT);
+ mm6 = _mm_or_si64(mm6, mm0);
+ Y_BG = mm6;
+
+ mm2 = _mm_loadlo_pi16_f(mm3);
+ mm4 = _mm_loadhi_pi16_f(mm3);
+ mm2 = _mm_srli_pi32(mm2, 1);
+ mm4 = _mm_srli_pi32(mm4, 1);
+
+ mm0 = PD_ONEHALFM1_CJ;
+ mm1 = _mm_add_pi32(mm1, mm2);
+ mm5 = _mm_add_pi32(mm5, mm4);
+ mm1 = _mm_add_pi32(mm1, mm0);
+ mm5 = _mm_add_pi32(mm5, mm0);
+ mm1 = _mm_srli_pi32(mm1, SCALEBITS);
+ mm5 = _mm_srli_pi32(mm5, SCALEBITS);
+ mm1 = _mm_packs_pi32(mm1, mm5);
+
+ mm7 = _mm_slli_pi16(mm7, BYTE_BIT);
+ mm1 = _mm_or_si64(mm1, mm7);
+ Cr_BG = mm1;
+
+ _mm_store_si64((__m64 *)&outptr0[0], Y_BG);
+ _mm_store_si64((__m64 *)&outptr1[0], Cb_RG);
+ _mm_store_si64((__m64 *)&outptr2[0], Cr_BG);
+ }
+ }
+}
+
+#undef mmA
+#undef mmB
+#undef mmC
+#undef mmD
+#undef mmE
+#undef mmF
+#undef mmG
+#undef mmH
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2011, 2014, D. R. Commander. All Rights Reserved.
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ * Authors: ZhuChen <zhuchen@loongson.cn>
+ * CaiWanwei <caiwanwei@loongson.cn>
+ * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* RGB --> YCC CONVERSION */
+
+#include "jsimd_mmi.h"
+
+
+#define F_0_081 ((short) 5329) /* FIX(0.08131) */
+#define F_0_114 ((short) 7471) /* FIX(0.11400) */
+#define F_0_168 ((short)11059) /* FIX(0.16874) */
+#define F_0_250 ((short)16384) /* FIX(0.25000) */
+#define F_0_299 ((short)19595) /* FIX(0.29900) */
+#define F_0_331 ((short)21709) /* FIX(0.33126) */
+#define F_0_418 ((short)27439) /* FIX(0.41869) */
+#define F_0_587 ((short)38470) /* FIX(0.58700) */
+#define F_0_337 ((short)(F_0_587 - F_0_250)) /* FIX(0.58700) - FIX(0.25000) */
+
+enum const_index {
+ index_PD_ONEHALF,
+ index_PW_F0299_F0337,
+ index_PW_F0114_F0250,
+ index_PW_MF016_MF033,
+ index_PW_MF008_MF041,
+ index_PD_ONEHALFM1_CJ
+};
+
+static uint64_t const_value[] = {
+ _uint64_set_pi32((int)(1 << (SCALEBITS-1)), (int)(1 << (SCALEBITS-1))),
+ _uint64_set_pi16(F_0_337, F_0_299, F_0_337, F_0_299),
+ _uint64_set_pi16(F_0_250, F_0_114, F_0_250, F_0_114),
+ _uint64_set_pi16(-F_0_331, -F_0_168, -F_0_331, -F_0_168),
+ _uint64_set_pi16(-F_0_418, -F_0_081, -F_0_418, -F_0_081),
+ _uint64_set_pi32(((1 << (SCALEBITS-1)) - 1 + (CENTERJSAMPLE << SCALEBITS)),
+ ((1 << (SCALEBITS-1)) - 1 + (CENTERJSAMPLE << SCALEBITS)))
+};
+
+#define get_const_value(index) (*(__m64 *)&const_value[index])
+
+#define PD_ONEHALF get_const_value(index_PD_ONEHALF)
+#define PW_F0299_F0337 get_const_value(index_PW_F0299_F0337)
+#define PW_F0114_F0250 get_const_value(index_PW_F0114_F0250)
+#define PW_MF016_MF033 get_const_value(index_PW_MF016_MF033)
+#define PW_MF008_MF041 get_const_value(index_PW_MF008_MF041)
+#define PD_ONEHALFM1_CJ get_const_value(index_PD_ONEHALFM1_CJ)
+
+
+#include "jccolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+
+#define RGB_RED EXT_RGB_RED
+#define RGB_GREEN EXT_RGB_GREEN
+#define RGB_BLUE EXT_RGB_BLUE
+#define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+#define jsimd_rgb_ycc_convert_mmi jsimd_extrgb_ycc_convert_mmi
+#include "jccolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_ycc_convert_mmi
+
+#define RGB_RED EXT_RGBX_RED
+#define RGB_GREEN EXT_RGBX_GREEN
+#define RGB_BLUE EXT_RGBX_BLUE
+#define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+#define jsimd_rgb_ycc_convert_mmi jsimd_extrgbx_ycc_convert_mmi
+#include "jccolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_ycc_convert_mmi
+
+#define RGB_RED EXT_BGR_RED
+#define RGB_GREEN EXT_BGR_GREEN
+#define RGB_BLUE EXT_BGR_BLUE
+#define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+#define jsimd_rgb_ycc_convert_mmi jsimd_extbgr_ycc_convert_mmi
+#include "jccolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_ycc_convert_mmi
+
+#define RGB_RED EXT_BGRX_RED
+#define RGB_GREEN EXT_BGRX_GREEN
+#define RGB_BLUE EXT_BGRX_BLUE
+#define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+#define jsimd_rgb_ycc_convert_mmi jsimd_extbgrx_ycc_convert_mmi
+#include "jccolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_ycc_convert_mmi
+
+#define RGB_RED EXT_XBGR_RED
+#define RGB_GREEN EXT_XBGR_GREEN
+#define RGB_BLUE EXT_XBGR_BLUE
+#define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+#define jsimd_rgb_ycc_convert_mmi jsimd_extxbgr_ycc_convert_mmi
+#include "jccolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_ycc_convert_mmi
+
+#define RGB_RED EXT_XRGB_RED
+#define RGB_GREEN EXT_XRGB_GREEN
+#define RGB_BLUE EXT_XRGB_BLUE
+#define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+#define jsimd_rgb_ycc_convert_mmi jsimd_extxrgb_ycc_convert_mmi
+#include "jccolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_ycc_convert_mmi
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2015, 2018, D. R. Commander. All Rights Reserved.
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ * Authors: ZhuChen <zhuchen@loongson.cn>
+ * CaiWanwei <caiwanwei@loongson.cn>
+ * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
+ *
+ * Based on the x86 SIMD extension for IJG JPEG library
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* CHROMA DOWNSAMPLING */
+
+#include "jsimd_mmi.h"
+#include "jcsample.h"
+
+
+void
+jsimd_h2v2_downsample_mmi (JDIMENSION image_width, int max_v_samp_factor,
+ JDIMENSION v_samp_factor,
+ JDIMENSION width_in_blocks, JSAMPARRAY input_data,
+ JSAMPARRAY output_data)
+{
+ int inrow, outrow, outcol, bias;
+ JDIMENSION output_cols = width_in_blocks * DCTSIZE;
+ JSAMPROW inptr0, inptr1, outptr;
+ __m64 mm0, mm1, mm2, mm3, mm4, mm5, mm6 = 0.0, mm7;
+
+ expand_right_edge(input_data, max_v_samp_factor, image_width,
+ output_cols * 2);
+
+ bias = (1 << 17) + 1; /* 0x00020001 (bias pattern) */
+ mm7 = _mm_set1_pi32(bias); /* mm7={1, 2, 1, 2} */
+ mm6 = _mm_cmpeq_pi16(mm6, mm6);
+ mm6 = _mm_srli_pi16(mm6, BYTE_BIT); /* mm6={0xFF 0x00 0xFF 0x00 ..} */
+
+ for (inrow = 0, outrow = 0; outrow < v_samp_factor;
+ inrow += 2, outrow++) {
+
+ inptr0 = input_data[inrow];
+ inptr1 = input_data[inrow + 1];
+ outptr = output_data[outrow];
+
+ for (outcol = output_cols; outcol > 0;
+ outcol -= 8, inptr0 += 16, inptr1 += 16, outptr += 8) {
+
+ mm0 = _mm_load_si64((__m64 *)&inptr0[0]);
+ mm1 = _mm_load_si64((__m64 *)&inptr1[0]);
+ mm2 = _mm_load_si64((__m64 *)&inptr0[8]);
+ mm3 = _mm_load_si64((__m64 *)&inptr1[8]);
+
+ mm4 = mm0;
+ mm5 = mm1;
+ mm0 = _mm_and_si64(mm0, mm6);
+ mm4 = _mm_srli_pi16(mm4, BYTE_BIT);
+ mm1 = _mm_and_si64(mm1, mm6);
+ mm5 = _mm_srli_pi16(mm5, BYTE_BIT);
+ mm0 = _mm_add_pi16(mm0, mm4);
+ mm1 = _mm_add_pi16(mm1, mm5);
+
+ mm4 = mm2;
+ mm5 = mm3;
+ mm2 = _mm_and_si64(mm2, mm6);
+ mm4 = _mm_srli_pi16(mm4, BYTE_BIT);
+ mm3 = _mm_and_si64(mm3, mm6);
+ mm5 = _mm_srli_pi16(mm5, BYTE_BIT);
+ mm2 = _mm_add_pi16(mm2, mm4);
+ mm3 = _mm_add_pi16(mm3, mm5);
+
+ mm0 = _mm_add_pi16(mm0, mm1);
+ mm2 = _mm_add_pi16(mm2, mm3);
+ mm0 = _mm_add_pi16(mm0, mm7);
+ mm2 = _mm_add_pi16(mm2, mm7);
+ mm0 = _mm_srli_pi16(mm0, 2);
+ mm2 = _mm_srli_pi16(mm2, 2);
+
+ mm0 = _mm_packs_pu16(mm0, mm2);
+
+ _mm_store_si64((__m64 *)&outptr[0], mm0);
+ }
+ }
+}
--- /dev/null
+/*
+ * jcsample.h
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Copyright (C) 1991-1996, Thomas G. Lane.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ */
+
+LOCAL(void)
+expand_right_edge (JSAMPARRAY image_data, int num_rows,
+ JDIMENSION input_cols, JDIMENSION output_cols)
+{
+ register JSAMPROW ptr;
+ register JSAMPLE pixval;
+ register int count;
+ int row;
+ int numcols = (int) (output_cols - input_cols);
+
+ if (numcols > 0) {
+ for (row = 0; row < num_rows; row++) {
+ ptr = image_data[row] + input_cols;
+ pixval = ptr[-1]; /* don't need GETJSAMPLE() here */
+ for (count = numcols; count > 0; count--)
+ *ptr++ = pixval;
+ }
+ }
+}
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright (C) 2015, D. R. Commander. All Rights Reserved.
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ * Authors: ZhuChen <zhuchen@loongson.cn>
+ * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
+ * CaiWanwei <caiwanwei@loongson.cn>
+ *
+ * Based on the x86 SIMD extension for IJG JPEG library
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* This file is included by jdcolor-mmi.c */
+
+
+#if RGB_RED == 0
+#define mmA mm0
+#define mmB mm1
+#elif RGB_GREEN == 0
+#define mmA mm2
+#define mmB mm3
+#elif RGB_BLUE == 0
+#define mmA mm4
+#define mmB mm5
+#else
+#define mmA mm6
+#define mmB mm7
+#endif
+
+#if RGB_RED == 1
+#define mmC mm0
+#define mmD mm1
+#elif RGB_GREEN == 1
+#define mmC mm2
+#define mmD mm3
+#elif RGB_BLUE == 1
+#define mmC mm4
+#define mmD mm5
+#else
+#define mmC mm6
+#define mmD mm7
+#endif
+
+#if RGB_RED == 2
+#define mmE mm0
+#define mmF mm1
+#elif RGB_GREEN == 2
+#define mmE mm2
+#define mmF mm3
+#elif RGB_BLUE == 2
+#define mmE mm4
+#define mmF mm5
+#else
+#define mmE mm6
+#define mmF mm7
+#endif
+
+#if RGB_RED == 3
+#define mmG mm0
+#define mmH mm1
+#elif RGB_GREEN == 3
+#define mmG mm2
+#define mmH mm3
+#elif RGB_BLUE == 3
+#define mmG mm4
+#define mmH mm5
+#else
+#define mmG mm6
+#define mmH mm7
+#endif
+
+
+void
+jsimd_ycc_rgb_convert_mmi (JDIMENSION out_width, JSAMPIMAGE input_buf,
+ JDIMENSION input_row, JSAMPARRAY output_buf,
+ int num_rows)
+{
+ JSAMPROW outptr, inptr0, inptr1, inptr2;
+ int num_cols, col;
+ __m64 mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7;
+ __m64 mm8, wk[2];
+
+ while (--num_rows >= 0) {
+ inptr0 = input_buf[0][input_row];
+ inptr1 = input_buf[1][input_row];
+ inptr2 = input_buf[2][input_row];
+ input_row++;
+ outptr = *output_buf++;
+
+ for (num_cols = out_width; num_cols > 0; num_cols -= 8,
+ inptr0 += 8, inptr1 += 8, inptr2 += 8) {
+
+ mm5 = _mm_load_si64((__m64 *)inptr1);
+ mm1 = _mm_load_si64((__m64 *)inptr2);
+ mm8 = _mm_load_si64((__m64 *)inptr0);
+ mm4 = 0;
+ mm7 = 0;
+ mm4 = _mm_cmpeq_pi16(mm4, mm4);
+ mm7 = _mm_cmpeq_pi16(mm7, mm7);
+ mm4 = _mm_srli_pi16(mm4, BYTE_BIT);
+ mm7 = _mm_slli_pi16(mm7, 7); /* mm7={0xFF80 0xFF80 0xFF80 0xFF80} */
+ mm0 = mm4; /* mm0=mm4={0xFF 0x00 0xFF 0x00 ..} */
+
+ mm4 = _mm_and_si64(mm4, mm5); /* mm4=Cb(0246)=CbE */
+ mm5 = _mm_srli_pi16(mm5, BYTE_BIT); /* mm5=Cb(1357)=CbO */
+ mm0 = _mm_and_si64(mm0, mm1); /* mm0=Cr(0246)=CrE */
+ mm1 = _mm_srli_pi16(mm1, BYTE_BIT); /* mm1=Cr(1357)=CrO */
+ mm4 = _mm_add_pi16(mm4, mm7);
+ mm5 = _mm_add_pi16(mm5, mm7);
+ mm0 = _mm_add_pi16(mm0, mm7);
+ mm1 = _mm_add_pi16(mm1, mm7);
+
+ /* (Original)
+ * R = Y + 1.40200 * Cr
+ * G = Y - 0.34414 * Cb - 0.71414 * Cr
+ * B = Y + 1.77200 * Cb
+ *
+ * (This implementation)
+ * R = Y + 0.40200 * Cr + Cr
+ * G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
+ * B = Y - 0.22800 * Cb + Cb + Cb
+ */
+
+ mm2 = mm4; /* mm2 = CbE */
+ mm3 = mm5; /* mm3 = CbO */
+ mm4 = _mm_add_pi16(mm4, mm4); /* mm4 = 2*CbE */
+ mm5 = _mm_add_pi16(mm5, mm5); /* mm5 = 2*CbO */
+ mm6 = mm0; /* mm6 = CrE */
+ mm7 = mm1; /* mm7 = CrO */
+ mm0 = _mm_add_pi16(mm0, mm0); /* mm0 = 2*CrE */
+ mm1 = _mm_add_pi16(mm1, mm1); /* mm1 = 2*CrO */
+
+ mm4 = _mm_mulhi_pi16(mm4, PW_MF0228); /* mm4=(2*CbE * -FIX(0.22800) */
+ mm5 = _mm_mulhi_pi16(mm5, PW_MF0228); /* mm5=(2*CbO * -FIX(0.22800) */
+ mm0 = _mm_mulhi_pi16(mm0, PW_F0402); /* mm0=(2*CrE * FIX(0.40200)) */
+ mm1 = _mm_mulhi_pi16(mm1, PW_F0402); /* mm1=(2*CrO * FIX(0.40200)) */
+
+ mm4 = _mm_add_pi16(mm4, PW_ONE);
+ mm5 = _mm_add_pi16(mm5, PW_ONE);
+ mm4 = _mm_srai_pi16(mm4, 1); /* mm4=(CbE * -FIX(0.22800)) */
+ mm5 = _mm_srai_pi16(mm5, 1); /* mm5=(CbO * -FIX(0.22800)) */
+ mm0 = _mm_add_pi16(mm0, PW_ONE);
+ mm1 = _mm_add_pi16(mm1, PW_ONE);
+ mm0 = _mm_srai_pi16(mm0, 1); /* mm0=(CrE * FIX(0.40200)) */
+ mm1 = _mm_srai_pi16(mm1, 1); /* mm1=(CrO * FIX(0.40200)) */
+
+ mm4 = _mm_add_pi16(mm4, mm2);
+ mm5 = _mm_add_pi16(mm5, mm3);
+ mm4 = _mm_add_pi16(mm4, mm2); /* mm4=(CbE * FIX(1.77200))=(B-Y)E */
+ mm5 = _mm_add_pi16(mm5, mm3); /* mm5=(CbO * FIX(1.77200))=(B-Y)O */
+ mm0 = _mm_add_pi16(mm0, mm6); /* mm0=(CrE * FIX(1.40200))=(R-Y)E */
+ mm1 = _mm_add_pi16(mm1, mm7); /* mm1=(CrO * FIX(1.40200))=(R-Y)O */
+
+ wk[0] = mm4; /* wk(0)=(B-Y)E */
+ wk[1] = mm5; /* wk(1)=(B-Y)O */
+
+ mm4 = mm2;
+ mm5 = mm3;
+ mm2 = _mm_unpacklo_pi16(mm2, mm6);
+ mm4 = _mm_unpackhi_pi16(mm4, mm6);
+ mm2 = _mm_madd_pi16(mm2, PW_MF0344_F0285);
+ mm4 = _mm_madd_pi16(mm4, PW_MF0344_F0285);
+ mm3 = _mm_unpacklo_pi16(mm3, mm7);
+ mm5 = _mm_unpackhi_pi16(mm5, mm7);
+ mm3 = _mm_madd_pi16(mm3, PW_MF0344_F0285);
+ mm5 = _mm_madd_pi16(mm5, PW_MF0344_F0285);
+
+ mm2 = _mm_add_pi32(mm2, PD_ONEHALF);
+ mm4 = _mm_add_pi32(mm4, PD_ONEHALF);
+ mm2 = _mm_srai_pi32(mm2, SCALEBITS);
+ mm4 = _mm_srai_pi32(mm4, SCALEBITS);
+ mm3 = _mm_add_pi32(mm3, PD_ONEHALF);
+ mm5 = _mm_add_pi32(mm5, PD_ONEHALF);
+ mm3 = _mm_srai_pi32(mm3, SCALEBITS);
+ mm5 = _mm_srai_pi32(mm5, SCALEBITS);
+
+ mm2 = _mm_packs_pi32(mm2, mm4); /* mm2=CbE*-FIX(0.344)+CrE*FIX(0.285) */
+ mm3 = _mm_packs_pi32(mm3, mm5); /* mm3=CbO*-FIX(0.344)+CrO*FIX(0.285) */
+ mm2 = _mm_sub_pi16(mm2, mm6); /* mm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E */
+ mm3 = _mm_sub_pi16(mm3, mm7); /* mm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O */
+
+ mm5 = mm8; /* mm5=Y(01234567) */
+
+ mm4 = _mm_cmpeq_pi16(mm4, mm4);
+ mm4 = _mm_srli_pi16(mm4, BYTE_BIT); /* mm4={0xFF 0x00 0xFF 0x00 ..} */
+ mm4 = _mm_and_si64(mm4, mm5); /* mm4=Y(0246)=YE */
+ mm5 = _mm_srli_pi16(mm5, BYTE_BIT); /* mm5=Y(1357)=YO */
+
+ mm0 = _mm_add_pi16(mm0, mm4); /* mm0=((R-Y)E+YE)=RE=(R0 R2 R4 R6) */
+ mm1 = _mm_add_pi16(mm1, mm5); /* mm1=((R-Y)O+YO)=RO=(R1 R3 R5 R7) */
+ mm0 = _mm_packs_pu16(mm0, mm0); /* mm0=(R0 R2 R4 R6 ** ** ** **) */
+ mm1 = _mm_packs_pu16(mm1, mm1); /* mm1=(R1 R3 R5 R7 ** ** ** **) */
+
+ mm2 = _mm_add_pi16(mm2, mm4); /* mm2=((G-Y)E+YE)=GE=(G0 G2 G4 G6) */
+ mm3 = _mm_add_pi16(mm3, mm5); /* mm3=((G-Y)O+YO)=GO=(G1 G3 G5 G7) */
+ mm2 = _mm_packs_pu16(mm2, mm2); /* mm2=(G0 G2 G4 G6 ** ** ** **) */
+ mm3 = _mm_packs_pu16(mm3, mm3); /* mm3=(G1 G3 G5 G7 ** ** ** **) */
+
+ mm4 = _mm_add_pi16(mm4, wk[0]); /* mm4=(YE+(B-Y)E)=BE=(B0 B2 B4 B6) */
+ mm5 = _mm_add_pi16(mm5, wk[1]); /* mm5=(YO+(B-Y)O)=BO=(B1 B3 B5 B7) */
+ mm4 = _mm_packs_pu16(mm4, mm4); /* mm4=(B0 B2 B4 B6 ** ** ** **) */
+ mm5 = _mm_packs_pu16(mm5, mm5); /* mm5=(B1 B3 B5 B7 ** ** ** **) */
+
+#if RGB_PIXELSIZE == 3
+
+ /* mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **) */
+ /* mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **) */
+ mmA = _mm_unpacklo_pi8(mmA, mmC); /* mmA=(00 10 02 12 04 14 06 16) */
+ mmE = _mm_unpacklo_pi8(mmE, mmB); /* mmE=(20 01 22 03 24 05 26 07) */
+ mmD = _mm_unpacklo_pi8(mmD, mmF); /* mmD=(11 21 13 23 15 25 17 27) */
+
+ mmG = mmA;
+ mmH = mmA;
+ mmA = _mm_unpacklo_pi16(mmA, mmE); /* mmA=(00 10 20 01 02 12 22 03) */
+ mmG = _mm_unpackhi_pi16(mmG, mmE); /* mmG=(04 14 24 05 06 16 26 07) */
+
+ mmH = _mm_srli_si64(mmH, 2*BYTE_BIT);
+ mmE = _mm_srli_si64(mmE, 2*BYTE_BIT);
+
+ mmC = mmD;
+ mmB = mmD;
+ mmD = _mm_unpacklo_pi16(mmD, mmH); /* mmD=(11 21 02 12 13 23 04 14) */
+ mmC = _mm_unpackhi_pi16(mmC, mmH); /* mmC=(15 25 06 16 17 27 -- --) */
+
+ mmB = _mm_srli_si64(mmB, 2*BYTE_BIT); /* mmB=(13 23 15 25 17 27 -- --) */
+
+ mmF = mmE;
+ mmE = _mm_unpacklo_pi16(mmE, mmB); /* mmE=(22 03 13 23 24 05 15 25) */
+ mmF = _mm_unpackhi_pi16(mmF, mmB); /* mmF=(26 07 17 27 -- -- -- --) */
+
+ mmA = _mm_unpacklo_pi32(mmA, mmD); /* mmA=(00 10 20 01 11 21 02 12) */
+ mmE = _mm_unpacklo_pi32(mmE, mmG); /* mmE=(22 03 13 23 04 14 24 05) */
+ mmC = _mm_unpacklo_pi32(mmC, mmF); /* mmC=(15 25 06 16 26 07 17 27) */
+
+ if (num_cols >= 8) {
+ _mm_store_si64((__m64 *)outptr, mmA);
+ _mm_store_si64((__m64 *)(outptr+8), mmE);
+ _mm_store_si64((__m64 *)(outptr+16), mmC);
+ outptr += RGB_PIXELSIZE*8;
+ } else {
+ col = num_cols * 3;
+ asm(".set noreorder\r\n"
+
+ "li $8, 16\r\n"
+ "move $9, %4\r\n"
+ "mov.s $f4, %1\r\n"
+ "mov.s $f6, %3\r\n"
+ "move $10, %5\r\n"
+ "bltu $9, $8, 1f\r\n"
+ "nop \r\n"
+ "gssdlc1 $f4, 7($10)\r\n"
+ "gssdrc1 $f4, 0($10)\r\n"
+ "gssdlc1 $f6, 7+8($10)\r\n"
+ "gssdrc1 $f6, 8($10)\r\n"
+ "mov.s $f4, %2\r\n"
+ "subu $9, $9, 16\r\n"
+ "daddu $10, $10, 16\r\n"
+ "b 2f\r\n"
+ "nop \r\n"
+
+ "1: \r\n"
+ "li $8, 8\r\n" /* st8 */
+ "bltu $9, $8, 2f\r\n"
+ "nop \r\n"
+ "gssdlc1 $f4, 7($10)\r\n"
+ "gssdrc1 $f4, ($10)\r\n"
+ "mov.s $f4, %3\r\n"
+ "subu $9, $9, 8\r\n"
+ "daddu $10, $10, 8\r\n"
+
+ "2: \r\n"
+ "li $8, 4\r\n" /* st4 */
+ "mfc1 $11, $f4\r\n"
+ "bltu $9, $8, 3f\r\n"
+ "nop \r\n"
+ "swl $11, 3($10)\r\n"
+ "swr $11, 0($10)\r\n"
+ "li $8, 32\r\n"
+ "mtc1 $8, $f6\r\n"
+ "dsrl $f4, $f4, $f6\r\n"
+ "mfc1 $11, $f4\r\n"
+ "subu $9, $9, 4\r\n"
+ "daddu $10, $10, 4\r\n"
+
+ "3: \r\n"
+ "li $8, 2\r\n" /* st2 */
+ "bltu $9, $8, 4f\r\n"
+ "nop \r\n"
+ "ush $11, 0($10)\r\n"
+ "srl $11, 16\r\n"
+ "subu $9, $9, 2\r\n"
+ "daddu $10, $10, 2\r\n"
+
+ "4: \r\n"
+ "li $8, 1\r\n" /* st1 */
+ "bltu $9, $8, 5f\r\n"
+ "nop \r\n"
+ "sb $11, 0($10)\r\n"
+
+ "5: \r\n"
+ "nop \r\n" /* end */
+ : "=m" (*outptr)
+ : "f" (mmA), "f" (mmC), "f" (mmE), "r" (col), "r" (outptr)
+ : "$f4", "$f6", "$8", "$9", "$10", "$11", "memory"
+ );
+ }
+
+#else /* RGB_PIXELSIZE == 4 */
+
+#ifdef RGBX_FILLER_0XFF
+ mm6 = _mm_cmpeq_pi8(mm6, mm6);
+ mm7 = _mm_cmpeq_pi8(mm7, mm7);
+#else
+ mm6 = _mm_xor_si64(mm6, mm6);
+ mm7 = _mm_xor_si64(mm7, mm7);
+#endif
+ /* mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **) */
+ /* mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **) */
+ /* mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **) */
+ /* mmG=(30 32 34 36 ** ** ** **), mmH=(31 33 35 37 ** ** ** **) */
+
+ mmA = _mm_unpacklo_pi8(mmA, mmC); /* mmA=(00 10 02 12 04 14 06 16) */
+ mmE = _mm_unpacklo_pi8(mmE, mmG); /* mmE=(20 30 22 32 24 34 26 36) */
+ mmB = _mm_unpacklo_pi8(mmB, mmD); /* mmB=(01 11 03 13 05 15 07 17) */
+ mmF = _mm_unpacklo_pi8(mmF, mmH); /* mmF=(21 31 23 33 25 35 27 37) */
+
+ mmC = mmA;
+ mmA = _mm_unpacklo_pi16(mmA, mmE); /* mmA=(00 10 20 30 02 12 22 32) */
+ mmC = _mm_unpackhi_pi16(mmC, mmE); /* mmC=(04 14 24 34 06 16 26 36) */
+ mmG = mmB;
+ mmB = _mm_unpacklo_pi16(mmB, mmF); /* mmB=(01 11 21 31 03 13 23 33) */
+ mmG = _mm_unpackhi_pi16(mmG, mmF); /* mmG=(05 15 25 35 07 17 27 37) */
+
+ mmD = mmA;
+ mmA = _mm_unpacklo_pi32(mmA, mmB); /* mmA=(00 10 20 30 01 11 21 31) */
+ mmD = _mm_unpackhi_pi32(mmD, mmB); /* mmD=(02 12 22 32 03 13 23 33) */
+ mmH = mmC;
+ mmC = _mm_unpacklo_pi32(mmC, mmG); /* mmC=(04 14 24 34 05 15 25 35) */
+ mmH = _mm_unpackhi_pi32(mmH, mmG); /* mmH=(06 16 26 36 07 17 27 37) */
+
+ if (num_cols >= 8) {
+ _mm_store_si64((__m64 *)outptr, mmA);
+ _mm_store_si64((__m64 *)(outptr+8), mmD);
+ _mm_store_si64((__m64 *)(outptr+16), mmC);
+ _mm_store_si64((__m64 *)(outptr+24), mmH);
+ outptr += RGB_PIXELSIZE*8;
+ } else {
+ col = num_cols;
+ asm(".set noreorder\r\n" /* st16 */
+
+ "li $8, 4\r\n"
+ "move $9, %6\r\n"
+ "move $10, %7\r\n"
+ "mov.s $f4, %2\r\n"
+ "mov.s $f6, %4\r\n"
+ "bltu $9, $8, 1f\r\n"
+ "nop \r\n"
+ "gssdlc1 $f4, 7($10)\r\n"
+ "gssdrc1 $f4, ($10)\r\n"
+ "gssdlc1 $f6, 7+8($10)\r\n"
+ "gssdrc1 $f6, 8($10)\r\n"
+ "mov.s $f4, %3\r\n"
+ "mov.s $f6, %5\r\n"
+ "subu $9, $9, 4\r\n"
+ "daddu $10, $10, 16\r\n"
+
+ "1: \r\n"
+ "li $8, 2\r\n" /* st8 */
+ "bltu $9, $8, 2f\r\n"
+ "nop \r\n"
+ "gssdlc1 $f4, 7($10)\r\n"
+ "gssdrc1 $f4, 0($10)\r\n"
+ "mov.s $f4, $f6\r\n"
+ "subu $9, $9, 2\r\n"
+ "daddu $10, $10, 8\r\n"
+
+ "2: \r\n"
+ "li $8, 1\r\n" /* st4 */
+ "bltu $9, $8, 3f\r\n"
+ "nop \r\n"
+ "gsswlc1 $f4, 3($10)\r\n"
+ "gsswrc1 $f4, 0($10)\r\n"
+
+ "3: \r\n"
+ "li %1, 0\r\n" /* end */
+ : "=m" (*outptr), "=r" (col)
+ : "f" (mmA), "f" (mmC), "f" (mmD), "f" (mmH), "r" (col),
+ "r" (outptr)
+ : "$f4", "$f6", "$8", "$9", "$10", "memory"
+ );
+ }
+
+#endif
+
+ }
+ }
+}
+
+#undef mmA
+#undef mmB
+#undef mmC
+#undef mmD
+#undef mmE
+#undef mmF
+#undef mmG
+#undef mmH
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2011, 2015, D. R. Commander. All Rights Reserved.
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ * Authors: ZhuChen <zhuchen@loongson.cn>
+ * CaiWanwei <caiwanwei@loongson.cn>
+ * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* YCC --> RGB CONVERSION */
+
+#include "jsimd_mmi.h"
+
+
+#define F_0_344 ((short)22554) /* FIX(0.34414) */
+#define F_0_402 ((short)26345) /* FIX(1.40200) - FIX(1) */
+#define F_0_285 ((short)18734) /* FIX(1) - FIX(0.71414) */
+#define F_0_228 ((short)14942) /* FIX(2) - FIX(1.77200) */
+
+enum const_index {
+ index_PW_ONE,
+ index_PW_F0402,
+ index_PW_MF0228,
+ index_PW_MF0344_F0285,
+ index_PD_ONEHALF
+};
+
+static uint64_t const_value[] = {
+ _uint64_set_pi16(1, 1, 1, 1),
+ _uint64_set_pi16(F_0_402, F_0_402, F_0_402, F_0_402),
+ _uint64_set_pi16(-F_0_228, -F_0_228, -F_0_228, -F_0_228),
+ _uint64_set_pi16(F_0_285, -F_0_344, F_0_285, -F_0_344),
+ _uint64_set_pi32((int)(1 << (SCALEBITS-1)), (int)(1 << (SCALEBITS-1)))
+};
+
+#define PW_ONE get_const_value(index_PW_ONE)
+#define PW_F0402 get_const_value(index_PW_F0402)
+#define PW_MF0228 get_const_value(index_PW_MF0228)
+#define PW_MF0344_F0285 get_const_value(index_PW_MF0344_F0285)
+#define PD_ONEHALF get_const_value(index_PD_ONEHALF)
+
+#define RGBX_FILLER_0XFF 1
+
+
+#include "jdcolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+
+#define RGB_RED EXT_RGB_RED
+#define RGB_GREEN EXT_RGB_GREEN
+#define RGB_BLUE EXT_RGB_BLUE
+#define RGB_PIXELSIZE EXT_RGB_PIXELSIZE
+#define jsimd_ycc_rgb_convert_mmi jsimd_ycc_extrgb_convert_mmi
+#include "jdcolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_mmi
+
+#define RGB_RED EXT_RGBX_RED
+#define RGB_GREEN EXT_RGBX_GREEN
+#define RGB_BLUE EXT_RGBX_BLUE
+#define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE
+#define jsimd_ycc_rgb_convert_mmi jsimd_ycc_extrgbx_convert_mmi
+#include "jdcolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_mmi
+
+#define RGB_RED EXT_BGR_RED
+#define RGB_GREEN EXT_BGR_GREEN
+#define RGB_BLUE EXT_BGR_BLUE
+#define RGB_PIXELSIZE EXT_BGR_PIXELSIZE
+#define jsimd_ycc_rgb_convert_mmi jsimd_ycc_extbgr_convert_mmi
+#include "jdcolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_mmi
+
+#define RGB_RED EXT_BGRX_RED
+#define RGB_GREEN EXT_BGRX_GREEN
+#define RGB_BLUE EXT_BGRX_BLUE
+#define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE
+#define jsimd_ycc_rgb_convert_mmi jsimd_ycc_extbgrx_convert_mmi
+#include "jdcolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_mmi
+
+#define RGB_RED EXT_XBGR_RED
+#define RGB_GREEN EXT_XBGR_GREEN
+#define RGB_BLUE EXT_XBGR_BLUE
+#define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE
+#define jsimd_ycc_rgb_convert_mmi jsimd_ycc_extxbgr_convert_mmi
+#include "jdcolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_mmi
+
+#define RGB_RED EXT_XRGB_RED
+#define RGB_GREEN EXT_XRGB_GREEN
+#define RGB_BLUE EXT_XRGB_BLUE
+#define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE
+#define jsimd_ycc_rgb_convert_mmi jsimd_ycc_extxrgb_convert_mmi
+#include "jdcolext-mmi.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_ycc_rgb_convert_mmi
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2015, 2018, D. R. Commander. All Rights Reserved.
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ * Authors: ZhuChen <zhuchen@loongson.cn>
+ * CaiWanwei <caiwanwei@loongson.cn>
+ * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
+ *
+ * Based on the x86 SIMD extension for IJG JPEG library
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* CHROMA UPSAMPLING */
+
+#include "jsimd_mmi.h"
+
+
+enum const_index {
+ index_PW_THREE,
+ index_PW_SEVEN,
+ index_PW_EIGHT,
+};
+
+static uint64_t const_value[] = {
+ _uint64_set_pi16(3, 3, 3, 3),
+ _uint64_set_pi16(7, 7, 7, 7),
+ _uint64_set_pi16(8, 8, 8, 8),
+};
+
+#define PW_THREE get_const_value(index_PW_THREE)
+#define PW_SEVEN get_const_value(index_PW_SEVEN)
+#define PW_EIGHT get_const_value(index_PW_EIGHT)
+
+
+#define PROCESS_ROW(r) \
+{ \
+ mm7 = _mm_load_si64((__m64 *)outptr##r); /* mm7=IntrL=( 0 1 2 3) */ \
+ mm3 = _mm_load_si64((__m64 *)outptr##r+1); /* mm3=IntrH=( 4 5 6 7) */ \
+ \
+ mm0 = mm7; \
+ mm4 = mm3; \
+ mm0 = _mm_srli_si64(mm0, 2*BYTE_BIT); /* mm0=( 1 2 3 -) */ \
+ mm4 = _mm_slli_si64(mm4, (SIZEOF_MMWORD-2)*BYTE_BIT); /* mm4=( - - - 4) */ \
+ mm5 = mm7; \
+ mm6 = mm3; \
+ mm5 = _mm_srli_si64(mm5, (SIZEOF_MMWORD-2)*BYTE_BIT); /* mm5=( 3 - - -) */ \
+ mm6 = _mm_slli_si64(mm6, 2*BYTE_BIT); /* mm6=( - 4 5 6) */ \
+ \
+ mm0 = _mm_or_si64(mm0, mm4); /* mm0=( 1 2 3 4) */ \
+ mm5 = _mm_or_si64(mm5, mm6); /* mm5=( 3 4 5 6) */ \
+ \
+ mm1 = mm7; \
+ mm2 = mm3; \
+ mm1 = _mm_slli_si64(mm1, 2*BYTE_BIT); /* mm1=( - 0 1 2) */ \
+ mm2 = _mm_srli_si64(mm2, 2*BYTE_BIT); /* mm2=( 5 6 7 -) */ \
+ mm4 = mm3; \
+ mm4 = _mm_srli_si64(mm4, (SIZEOF_MMWORD-2)*BYTE_BIT); /* mm4=( 7 - - -) */ \
+ \
+ mm1 = _mm_or_si64(mm1, wk[r]); /* mm1=(-1 0 1 2) */ \
+ mm2 = _mm_or_si64(mm2, wk[r+2]); /* mm2=( 5 6 6 8) */ \
+ \
+ wk[r] = mm4; \
+ \
+ mm7 = _mm_mullo_pi16(mm7, PW_THREE); \
+ mm3 = _mm_mullo_pi16(mm3, PW_THREE); \
+ mm1 = _mm_add_pi16(mm1, PW_EIGHT); \
+ mm5 = _mm_add_pi16(mm5, PW_EIGHT); \
+ mm0 = _mm_add_pi16(mm0, PW_SEVEN); \
+ mm2 = _mm_add_pi16(mm2, PW_SEVEN); \
+ \
+ mm1 = _mm_add_pi16(mm1, mm7); \
+ mm5 = _mm_add_pi16(mm5, mm3); \
+ mm1 = _mm_srli_pi16(mm1, 4); /* mm1=OutrLE=( 0 2 4 6) */ \
+ mm5 = _mm_srli_pi16(mm5, 4); /* mm5=OutrHE=( 8 10 12 14) */ \
+ mm0 = _mm_add_pi16(mm0, mm7); \
+ mm2 = _mm_add_pi16(mm2, mm3); \
+ mm0 = _mm_srli_pi16(mm0, 4); /* mm0=OutrLO=( 1 3 5 7) */ \
+ mm2 = _mm_srli_pi16(mm2, 4); /* mm2=OutrHO=( 9 11 13 15) */ \
+ \
+ mm0 = _mm_slli_pi16(mm0, BYTE_BIT); \
+ mm2 = _mm_slli_pi16(mm2, BYTE_BIT); \
+ mm1 = _mm_or_si64(mm1, mm0); /* mm1=OutrL=( 0 1 2 3 4 5 6 7) */ \
+ mm5 = _mm_or_si64(mm5, mm2); /* mm5=OutrH=( 8 9 10 11 12 13 14 15) */ \
+ \
+ _mm_store_si64((__m64 *)outptr##r, mm1); \
+ _mm_store_si64((__m64 *)outptr##r+1, mm5); \
+}
+
+void
+jsimd_h2v2_fancy_upsample_mmi (int max_v_samp_factor,
+ JDIMENSION downsampled_width,
+ JSAMPARRAY input_data,
+ JSAMPARRAY *output_data_ptr)
+{
+ JSAMPARRAY output_data = *output_data_ptr;
+ JSAMPROW inptr_1, inptr0, inptr1, outptr0, outptr1;
+ int inrow, outrow, incol, tmp, tmp1;
+ __m64 mm0, mm1, mm2, mm3 = 0.0, mm4, mm5, mm6, mm7 = 0.0;
+ __m64 wk[4], mm_tmp;
+
+ for (inrow = 0, outrow = 0; outrow < max_v_samp_factor; inrow++) {
+
+ inptr_1 = input_data[inrow - 1];
+ inptr0 = input_data[inrow];
+ inptr1 = input_data[inrow + 1];
+ outptr0 = output_data[outrow++];
+ outptr1 = output_data[outrow++];
+
+ if (downsampled_width & 7) {
+ tmp = (downsampled_width - 1) * sizeof(JSAMPLE);
+ tmp1 = downsampled_width * sizeof(JSAMPLE);
+ asm("daddu $8, %3, %6\r\n"
+ "lb $9, ($8)\r\n"
+ "daddu $8, %3, %7\r\n"
+ "sb $9, ($8)\r\n"
+ "daddu $8, %4, %6\r\n"
+ "lb $9, ($8)\r\n"
+ "daddu $8, %4, %7\r\n"
+ "sb $9, ($8)\r\n"
+ "daddu $8, %5, %6\r\n"
+ "lb $9, ($8)\r\n"
+ "daddu $8, %5, %7\r\n"
+ "sb $9, ($8)\r\n"
+ : "=m" (*inptr_1), "=m" (*inptr0), "=m" (*inptr1)
+ : "r" (inptr_1), "r" (inptr0), "r" (inptr1), "r" (tmp), "r" (tmp1)
+ : "$8", "$9"
+ );
+ }
+
+ /* process the first column block */
+ mm0 = _mm_load_si64((__m64 *)inptr0); /* mm0 = row[ 0][0] */
+ mm1 = _mm_load_si64((__m64 *)inptr_1); /* mm1 = row[-1][0] */
+ mm2 = _mm_load_si64((__m64 *)inptr1); /* mm2 = row[ 1][0] */
+
+ mm3 = _mm_xor_si64(mm3, mm3); /* mm3 = (all 0's) */
+ mm4 = mm0;
+ mm0 = _mm_unpacklo_pi8(mm0, mm3); /* mm0 = row[ 0][0]( 0 1 2 3) */
+ mm4 = _mm_unpackhi_pi8(mm4, mm3); /* mm4 = row[ 0][0]( 4 5 6 7) */
+ mm5 = mm1;
+ mm1 = _mm_unpacklo_pi8(mm1, mm3); /* mm1 = row[-1][0]( 0 1 2 3) */
+ mm5 = _mm_unpackhi_pi8(mm5, mm3); /* mm5 = row[-1][0]( 4 5 6 7) */
+ mm6 = mm2;
+ mm2 = _mm_unpacklo_pi8(mm2, mm3); /* mm2 = row[+1][0]( 0 1 2 3) */
+ mm6 = _mm_unpackhi_pi8(mm6, mm3); /* mm6 = row[+1][0]( 4 5 6 7) */
+
+ mm0 = _mm_mullo_pi16(mm0, PW_THREE);
+ mm4 = _mm_mullo_pi16(mm4, PW_THREE);
+
+ mm7 = _mm_cmpeq_pi8(mm7, mm7);
+ mm7 = _mm_srli_si64(mm7, (SIZEOF_MMWORD-2)*BYTE_BIT);
+
+ mm1 = _mm_add_pi16(mm1, mm0); /* mm1=Int0L=( 0 1 2 3) */
+ mm5 = _mm_add_pi16(mm5, mm4); /* mm5=Int0H=( 4 5 6 7) */
+ mm2 = _mm_add_pi16(mm2, mm0); /* mm2=Int1L=( 0 1 2 3) */
+ mm6 = _mm_add_pi16(mm6, mm4); /* mm6=Int1H=( 4 5 6 7) */
+
+ _mm_store_si64((__m64*)outptr0, mm1); /* temporarily save */
+ _mm_store_si64((__m64*)outptr0+1, mm5); /* the intermediate data */
+ _mm_store_si64((__m64*)outptr1, mm2);
+ _mm_store_si64((__m64*)outptr1+1, mm6);
+
+ mm1 = _mm_and_si64(mm1, mm7); /* mm1=( 0 - - -) */
+ mm2 = _mm_and_si64(mm2, mm7); /* mm2=( 0 - - -) */
+
+ wk[0] = mm1;
+ wk[1] = mm2;
+
+ for (incol = downsampled_width; incol > 0;
+ incol -= 8, inptr_1 += 8, inptr0 += 8, inptr1 += 8,
+ outptr0 += 16, outptr1 += 16) {
+
+ if (incol > 8) {
+ /* process the next column block */
+ mm0 = _mm_load_si64((__m64 *)inptr0+1); /* mm0 = row[ 0][1] */
+ mm1 = _mm_load_si64((__m64 *)inptr_1+1); /* mm1 = row[-1][1] */
+ mm2 = _mm_load_si64((__m64 *)inptr1+1); /* mm2 = row[+1][1] */
+
+ mm3 = _mm_setzero_si64(); /* mm3 = (all 0's) */
+ mm4 = mm0;
+ mm0 = _mm_unpacklo_pi8(mm0, mm3); /* mm0 = row[ 0][1]( 0 1 2 3) */
+ mm4 = _mm_unpackhi_pi8(mm4, mm3); /* mm4 = row[ 0][1]( 4 5 6 7) */
+ mm5 = mm1;
+ mm1 = _mm_unpacklo_pi8(mm1, mm3); /* mm1 = row[-1][1]( 0 1 2 3) */
+ mm5 = _mm_unpackhi_pi8(mm5, mm3); /* mm5 = row[-1][1]( 4 5 6 7) */
+ mm6 = mm2;
+ mm2 = _mm_unpacklo_pi8(mm2, mm3); /* mm2 = row[+1][1]( 0 1 2 3) */
+ mm6 = _mm_unpackhi_pi8(mm6, mm3); /* mm6 = row[+1][1]( 4 5 6 7) */
+
+ mm0 = _mm_mullo_pi16(mm0, PW_THREE);
+ mm4 = _mm_mullo_pi16(mm4, PW_THREE);
+
+ mm1 = _mm_add_pi16(mm1, mm0); /* mm1 = Int0L = ( 0 1 2 3) */
+ mm5 = _mm_add_pi16(mm5, mm4); /* mm5 = Int0H = ( 4 5 6 7) */
+ mm2 = _mm_add_pi16(mm2, mm0); /* mm2 = Int1L = ( 0 1 2 3) */
+ mm6 = _mm_add_pi16(mm6, mm4); /* mm6 = Int1H = ( 4 5 6 7) */
+
+ _mm_store_si64((__m64*)outptr0+2, mm1); /* temporarily save */
+ _mm_store_si64((__m64*)outptr0+3, mm5); /* the intermediate data */
+ _mm_store_si64((__m64*)outptr1+2, mm2);
+ _mm_store_si64((__m64*)outptr1+3, mm6);
+
+ mm1 = _mm_slli_si64(mm1, (SIZEOF_MMWORD-2)*BYTE_BIT); /* mm1=( - - - 0) */
+ mm2 = _mm_slli_si64(mm2, (SIZEOF_MMWORD-2)*BYTE_BIT); /* mm2=( - - - 0) */
+
+ wk[2] = mm1;
+ wk[3] = mm2;
+ } else {
+ /* process the last column block */
+ mm1 = _mm_cmpeq_pi8(mm1, mm1);
+ mm1 = _mm_slli_si64(mm1, (SIZEOF_MMWORD-2)*BYTE_BIT);
+ mm2 = mm1;
+
+ mm_tmp = _mm_load_si64((__m64 *)outptr0+1);
+ mm1 = _mm_and_si64(mm1, mm_tmp); /* mm1=( - - - 7) */
+ mm_tmp = _mm_load_si64((__m64 *)outptr1+1);
+ mm2 = _mm_and_si64(mm2, mm_tmp); /* mm2=( - - - 7) */
+
+ wk[2] = mm1;
+ wk[3] = mm2;
+ }
+
+ /* process the upper row */
+ PROCESS_ROW(0)
+
+ /* process the lower row */
+ PROCESS_ROW(1)
+ }
+ }
+}
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014, 2018, D. R. Commander. All Rights Reserved.
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ * Authors: ZhuChen <zhuchen@loongson.cn>
+ * CaiWanwei <caiwanwei@loongson.cn>
+ * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
+ *
+ * Based on the x86 SIMD extension for IJG JPEG library
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* SLOW INTEGER FORWARD DCT */
+
+#include "jsimd_mmi.h"
+
+
+#define CONST_BITS 13
+#define PASS1_BITS 2
+#define DESCALE_P1 (CONST_BITS-PASS1_BITS)
+#define DESCALE_P2 (CONST_BITS+PASS1_BITS)
+
+#define FIX_0_298 ((short)2446) /* FIX(0.298631336) */
+#define FIX_0_390 ((short)3196) /* FIX(0.390180644) */
+#define FIX_0_541 ((short)4433) /* FIX(0.541196100) */
+#define FIX_0_765 ((short)6270) /* FIX(0.765366865) */
+#define FIX_0_899 ((short)7373) /* FIX(0.899976223) */
+#define FIX_1_175 ((short)9633) /* FIX(1.175875602) */
+#define FIX_1_501 ((short)12299) /* FIX(1.501321110) */
+#define FIX_1_847 ((short)15137) /* FIX(1.847759065) */
+#define FIX_1_961 ((short)16069) /* FIX(1.961570560) */
+#define FIX_2_053 ((short)16819) /* FIX(2.053119869) */
+#define FIX_2_562 ((short)20995) /* FIX(2.562915447) */
+#define FIX_3_072 ((short)25172) /* FIX(3.072711026) */
+
+enum const_index{
+ index_PW_F130_F054,
+ index_PW_F054_MF130,
+ index_PW_MF078_F117,
+ index_PW_F117_F078,
+ index_PW_MF060_MF089,
+ index_PW_MF089_F060,
+ index_PW_MF050_MF256,
+ index_PW_MF256_F050,
+ index_PD_DESCALE_P1,
+ index_PD_DESCALE_P2,
+ index_PW_DESCALE_P2X
+};
+
+static uint64_t const_value[]={
+ _uint64_set_pi16(FIX_0_541, (FIX_0_541+FIX_0_765),
+ FIX_0_541, (FIX_0_541+FIX_0_765)),
+ _uint64_set_pi16((FIX_0_541-FIX_1_847), FIX_0_541,
+ (FIX_0_541-FIX_1_847), FIX_0_541),
+ _uint64_set_pi16(FIX_1_175, (FIX_1_175-FIX_1_961),
+ FIX_1_175, (FIX_1_175-FIX_1_961)),
+ _uint64_set_pi16((FIX_1_175-FIX_0_390), FIX_1_175,
+ (FIX_1_175-FIX_0_390), FIX_1_175),
+ _uint64_set_pi16(-FIX_0_899, (FIX_0_298-FIX_0_899),
+ -FIX_0_899, (FIX_0_298-FIX_0_899)),
+ _uint64_set_pi16((FIX_1_501-FIX_0_899), -FIX_0_899,
+ (FIX_1_501-FIX_0_899), -FIX_0_899),
+ _uint64_set_pi16(-FIX_2_562, (FIX_2_053-FIX_2_562),
+ -FIX_2_562, (FIX_2_053-FIX_2_562)),
+ _uint64_set_pi16((FIX_3_072-FIX_2_562), -FIX_2_562,
+ (FIX_3_072-FIX_2_562), -FIX_2_562),
+ _uint64_set_pi32((1 << (DESCALE_P1-1)), (1 << (DESCALE_P1-1))),
+ _uint64_set_pi32((1 << (DESCALE_P2-1)), (1 << (DESCALE_P2-1))),
+ _uint64_set_pi16((1 << (PASS1_BITS-1)), (1 << (PASS1_BITS-1)),
+ (1 << (PASS1_BITS-1)), (1 << (PASS1_BITS-1)))
+};
+
+#define PW_F130_F054 get_const_value(index_PW_F130_F054)
+#define PW_F054_MF130 get_const_value(index_PW_F054_MF130)
+#define PW_MF078_F117 get_const_value(index_PW_MF078_F117)
+#define PW_F117_F078 get_const_value(index_PW_F117_F078)
+#define PW_MF060_MF089 get_const_value(index_PW_MF060_MF089)
+#define PW_MF089_F060 get_const_value(index_PW_MF089_F060)
+#define PW_MF050_MF256 get_const_value(index_PW_MF050_MF256)
+#define PW_MF256_F050 get_const_value(index_PW_MF256_F050)
+#define PD_DESCALE_P1 get_const_value(index_PD_DESCALE_P1)
+#define PD_DESCALE_P2 get_const_value(index_PD_DESCALE_P2)
+#define PW_DESCALE_P2X get_const_value(index_PW_DESCALE_P2X)
+
+
+#define DO_FDCT_COMMON(PASS) \
+{ \
+ __m64 tmp1312l, tmp1312h, tmp47l, tmp47h, tmp4l, tmp4h, tmp7l, tmp7h; \
+ __m64 tmp56l, tmp56h, tmp5l, tmp5h, tmp6l, tmp6h; \
+ __m64 out1l, out1h, out2l, out2h, out3l, out3h; \
+ __m64 out5l, out5h, out6l, out6h, out7l, out7h; \
+ __m64 z34l, z34h, z3l, z3h, z4l, z4h, z3, z4; \
+ \
+ /* (Original) \
+ * z1 = (tmp12 + tmp13) * 0.541196100; \
+ * out2 = z1 + tmp13 * 0.765366865; \
+ * out6 = z1 + tmp12 * -1.847759065; \
+ * \
+ * (This implementation) \
+ * out2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; \
+ * out6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); \
+ */ \
+ \
+ tmp1312l = _mm_unpacklo_pi16(tmp13, tmp12); \
+ tmp1312h = _mm_unpackhi_pi16(tmp13, tmp12); \
+ \
+ out2l = _mm_madd_pi16(tmp1312l, PW_F130_F054); \
+ out2h = _mm_madd_pi16(tmp1312h, PW_F130_F054); \
+ out6l = _mm_madd_pi16(tmp1312l, PW_F054_MF130); \
+ out6h = _mm_madd_pi16(tmp1312h, PW_F054_MF130); \
+ \
+ out2l = _mm_add_pi32(out2l, PD_DESCALE_P##PASS); \
+ out2h = _mm_add_pi32(out2h, PD_DESCALE_P##PASS); \
+ out2l = _mm_srai_pi32(out2l, DESCALE_P##PASS); \
+ out2h = _mm_srai_pi32(out2h, DESCALE_P##PASS); \
+ \
+ out6l = _mm_add_pi32(out6l, PD_DESCALE_P##PASS); \
+ out6h = _mm_add_pi32(out6h, PD_DESCALE_P##PASS); \
+ out6l = _mm_srai_pi32(out6l, DESCALE_P##PASS); \
+ out6h = _mm_srai_pi32(out6h, DESCALE_P##PASS); \
+ \
+ out2 = _mm_packs_pi32(out2l, out2h); \
+ out6 = _mm_packs_pi32(out6l, out6h); \
+ \
+ /* Odd part */ \
+ \
+ z3 = _mm_add_pi16(tmp4, tmp6); \
+ z4 = _mm_add_pi16(tmp5, tmp7); \
+ \
+ /* (Original) \
+ * z5 = (z3 + z4) * 1.175875602; \
+ * z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; \
+ * z3 += z5; z4 += z5; \
+ * \
+ * (This implementation) \
+ * z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; \
+ * z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); \
+ */ \
+ \
+ z34l = _mm_unpacklo_pi16(z3, z4); \
+ z34h = _mm_unpackhi_pi16(z3, z4); \
+ z3l = _mm_madd_pi16(z34l, PW_MF078_F117); \
+ z3h = _mm_madd_pi16(z34h, PW_MF078_F117); \
+ z4l = _mm_madd_pi16(z34l, PW_F117_F078); \
+ z4h = _mm_madd_pi16(z34h, PW_F117_F078); \
+ \
+ /* (Original) \
+ * z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; \
+ * tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; \
+ * tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; \
+ * z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; \
+ * out7 = tmp4 + z1 + z3; out5 = tmp5 + z2 + z4; \
+ * out3 = tmp6 + z2 + z3; out1 = tmp7 + z1 + z4; \
+ * \
+ * (This implementation) \
+ * tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; \
+ * tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; \
+ * tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); \
+ * tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); \
+ * out7 = tmp4 + z3; out5 = tmp5 + z4; \
+ * out3 = tmp6 + z3; out1 = tmp7 + z4; \
+ */ \
+ \
+ tmp47l = _mm_unpacklo_pi16(tmp4, tmp7); \
+ tmp47h = _mm_unpackhi_pi16(tmp4, tmp7); \
+ \
+ tmp4l = _mm_madd_pi16(tmp47l, PW_MF060_MF089); \
+ tmp4h = _mm_madd_pi16(tmp47h, PW_MF060_MF089); \
+ tmp7l = _mm_madd_pi16(tmp47l, PW_MF089_F060); \
+ tmp7h = _mm_madd_pi16(tmp47h, PW_MF089_F060); \
+ \
+ out7l = _mm_add_pi32(tmp4l, z3l); \
+ out7h = _mm_add_pi32(tmp4h, z3h); \
+ out1l = _mm_add_pi32(tmp7l, z4l); \
+ out1h = _mm_add_pi32(tmp7h, z4h); \
+ \
+ out7l = _mm_add_pi32(out7l, PD_DESCALE_P##PASS); \
+ out7h = _mm_add_pi32(out7h, PD_DESCALE_P##PASS); \
+ out7l = _mm_srai_pi32(out7l, DESCALE_P##PASS); \
+ out7h = _mm_srai_pi32(out7h, DESCALE_P##PASS); \
+ \
+ out1l = _mm_add_pi32(out1l, PD_DESCALE_P##PASS); \
+ out1h = _mm_add_pi32(out1h, PD_DESCALE_P##PASS); \
+ out1l = _mm_srai_pi32(out1l, DESCALE_P##PASS); \
+ out1h = _mm_srai_pi32(out1h, DESCALE_P##PASS); \
+ \
+ out7 = _mm_packs_pi32(out7l, out7h); \
+ out1 = _mm_packs_pi32(out1l, out1h); \
+ \
+ tmp56l = _mm_unpacklo_pi16(tmp5, tmp6); \
+ tmp56h = _mm_unpackhi_pi16(tmp5, tmp6); \
+ \
+ tmp5l = _mm_madd_pi16(tmp56l, PW_MF050_MF256); \
+ tmp5h = _mm_madd_pi16(tmp56h, PW_MF050_MF256); \
+ tmp6l = _mm_madd_pi16(tmp56l, PW_MF256_F050); \
+ tmp6h = _mm_madd_pi16(tmp56h, PW_MF256_F050); \
+ \
+ out5l = _mm_add_pi32(tmp5l, z4l); \
+ out5h = _mm_add_pi32(tmp5h, z4h); \
+ out3l = _mm_add_pi32(tmp6l, z3l); \
+ out3h = _mm_add_pi32(tmp6h, z3h); \
+ \
+ out5l = _mm_add_pi32(out5l, PD_DESCALE_P##PASS); \
+ out5h = _mm_add_pi32(out5h, PD_DESCALE_P##PASS); \
+ out5l = _mm_srai_pi32(out5l, DESCALE_P##PASS); \
+ out5h = _mm_srai_pi32(out5h, DESCALE_P##PASS); \
+ \
+ out3l = _mm_add_pi32(out3l, PD_DESCALE_P##PASS); \
+ out3h = _mm_add_pi32(out3h, PD_DESCALE_P##PASS); \
+ out3l = _mm_srai_pi32(out3l, DESCALE_P##PASS); \
+ out3h = _mm_srai_pi32(out3h, DESCALE_P##PASS); \
+ \
+ out5 = _mm_packs_pi32(out5l, out5h); \
+ out3 = _mm_packs_pi32(out3l, out3h); \
+}
+
+#define DO_FDCT_PASS1() \
+{ \
+ __m64 row0l, row0h, row1l, row1h, row2l, row2h, row3l, row3h; \
+ __m64 row01a, row01b, row01c, row01d, row23a, row23b, row23c, row23d; \
+ __m64 col0, col1, col2, col3, col4, col5, col6, col7; \
+ __m64 tmp10, tmp11; \
+ \
+ row0l = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*0]); /* (00 01 02 03) */ \
+ row0h = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*0+4]); /* (04 05 06 07) */ \
+ row1l = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*1]); /* (10 11 12 13) */ \
+ row1h = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*1+4]); /* (14 15 16 17) */ \
+ row2l = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*2]); /* (20 21 22 23) */ \
+ row2h = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*2+4]); /* (24 25 26 27) */ \
+ row3l = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*3]); /* (30 31 32 33) */ \
+ row3h = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*3+4]); /* (34 35 36 37) */ \
+ \
+ /* Transpose coefficients */ \
+ \
+ row23a = _mm_unpacklo_pi16(row2l, row3l); /* row23a=(20 30 21 31) */ \
+ row23b = _mm_unpackhi_pi16(row2l, row3l); /* row23b=(22 32 23 33) */ \
+ row23c = _mm_unpacklo_pi16(row2h, row3h); /* row23c=(24 34 25 35) */ \
+ row23d = _mm_unpackhi_pi16(row2h, row3h); /* row23d=(26 36 27 37) */ \
+ \
+ row01a = _mm_unpacklo_pi16(row0l, row1l); /* row01a=(00 10 01 11) */ \
+ row01b = _mm_unpackhi_pi16(row0l, row1l); /* row01b=(02 12 03 13) */ \
+ row01c = _mm_unpacklo_pi16(row0h, row1h); /* row01c=(04 14 05 15) */ \
+ row01d = _mm_unpackhi_pi16(row0h, row1h); /* row01d=(06 16 07 17) */ \
+ \
+ col0 = _mm_unpacklo_pi32(row01a, row23a); /* col0=(00 10 20 30) */ \
+ col1 = _mm_unpackhi_pi32(row01a, row23a); /* col1=(01 11 21 31) */ \
+ col6 = _mm_unpacklo_pi32(row01d, row23d); /* col6=(06 16 26 36) */ \
+ col7 = _mm_unpackhi_pi32(row01d, row23d); /* col7=(07 17 27 37) */ \
+ \
+ tmp6 = _mm_sub_pi16(col1, col6); /* tmp6=col1-col6 */ \
+ tmp7 = _mm_sub_pi16(col0, col7); /* tmp7=col0-col7 */ \
+ tmp1 = _mm_add_pi16(col1, col6); /* tmp1=col1+col6 */ \
+ tmp0 = _mm_add_pi16(col0, col7); /* tmp0=col0+col7 */ \
+ \
+ col2 = _mm_unpacklo_pi32(row01b, row23b); /* col2=(02 12 22 32) */ \
+ col3 = _mm_unpackhi_pi32(row01b, row23b); /* col3=(03 13 23 33) */ \
+ col4 = _mm_unpacklo_pi32(row01c, row23c); /* col4=(04 14 24 34) */ \
+ col5 = _mm_unpackhi_pi32(row01c, row23c); /* col5=(05 15 25 35) */ \
+ \
+ tmp3 = _mm_add_pi16(col3, col4); /* tmp3=col3+col4 */ \
+ tmp2 = _mm_add_pi16(col2, col5); /* tmp2=col2+col5 */ \
+ tmp4 = _mm_sub_pi16(col3, col4); /* tmp4=col3-col4 */ \
+ tmp5 = _mm_sub_pi16(col2, col5); /* tmp5=col2-col5 */ \
+ \
+ /* Even part */ \
+ \
+ tmp10 = _mm_add_pi16(tmp0, tmp3); /* tmp10=tmp0+tmp3 */ \
+ tmp13 = _mm_sub_pi16(tmp0, tmp3); /* tmp13=tmp0-tmp3 */ \
+ tmp11 = _mm_add_pi16(tmp1, tmp2); /* tmp11=tmp1+tmp2 */ \
+ tmp12 = _mm_sub_pi16(tmp1, tmp2); /* tmp12=tmp1-tmp2 */ \
+ \
+ out0 = _mm_add_pi16(tmp10, tmp11); /* out0=tmp10+tmp11 */ \
+ out4 = _mm_sub_pi16(tmp10, tmp11); /* out4=tmp10-tmp11 */ \
+ out0 = _mm_slli_pi16(out0, PASS1_BITS); \
+ out4 = _mm_slli_pi16(out4, PASS1_BITS); \
+ \
+ DO_FDCT_COMMON(1) \
+ \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*0], out0); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*0+4], out4); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*1], out1); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*1+4], out5); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*2], out2); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*2+4], out6); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*3], out3); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*3+4], out7); \
+}
+
+#define DO_FDCT_PASS2() \
+{ \
+ __m64 col0l, col0h, col1l, col1h, col2l, col2h, col3l, col3h; \
+ __m64 col01a, col01b, col01c, col01d, col23a, col23b, col23c, col23d; \
+ __m64 row0, row1, row2, row3, row4, row5, row6, row7; \
+ __m64 tmp10, tmp11; \
+ \
+ col0l = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*0]); /* (00 10 20 30) */ \
+ col1l = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*1]); /* (01 11 21 31) */ \
+ col2l = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*2]); /* (02 12 22 32) */ \
+ col3l = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*3]); /* (03 13 23 33) */ \
+ col0h = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*4]); /* (40 50 60 70) */ \
+ col1h = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*5]); /* (41 51 61 71) */ \
+ col2h = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*6]); /* (42 52 62 72) */ \
+ col3h = _mm_load_si64((__m64 *)&dataptr[DCTSIZE*7]); /* (43 53 63 73) */ \
+ \
+ /* Transpose coefficients */ \
+ \
+ col23a = _mm_unpacklo_pi16(col2l, col3l); /* col23a=(02 03 12 13) */ \
+ col23b = _mm_unpackhi_pi16(col2l, col3l); /* col23b=(22 23 32 33) */ \
+ col23c = _mm_unpacklo_pi16(col2h, col3h); /* col23c=(42 43 52 53) */ \
+ col23d = _mm_unpackhi_pi16(col2h, col3h); /* col23d=(62 63 72 73) */ \
+ \
+ col01a = _mm_unpacklo_pi16(col0l, col1l); /* col01a=(00 01 10 11) */ \
+ col01b = _mm_unpackhi_pi16(col0l, col1l); /* col01b=(20 21 30 31) */ \
+ col01c = _mm_unpacklo_pi16(col0h, col1h); /* col01c=(40 41 50 51) */ \
+ col01d = _mm_unpackhi_pi16(col0h, col1h); /* col01d=(60 61 70 71) */ \
+ \
+ row0 = _mm_unpacklo_pi32(col01a, col23a); /* row0=(00 01 02 03) */ \
+ row1 = _mm_unpackhi_pi32(col01a, col23a); /* row1=(10 11 12 13) */ \
+ row6 = _mm_unpacklo_pi32(col01d, col23d); /* row6=(60 61 62 63) */ \
+ row7 = _mm_unpackhi_pi32(col01d, col23d); /* row7=(70 71 72 73) */ \
+ \
+ tmp6 = _mm_sub_pi16(row1, row6); /* tmp6=row1-row6 */ \
+ tmp7 = _mm_sub_pi16(row0, row7); /* tmp7=row0-row7 */ \
+ tmp1 = _mm_add_pi16(row1, row6); /* tmp1=row1+row6 */ \
+ tmp0 = _mm_add_pi16(row0, row7); /* tmp0=row0+row7 */ \
+ \
+ row2 = _mm_unpacklo_pi32(col01b, col23b); /* row2=(20 21 22 23) */ \
+ row3 = _mm_unpackhi_pi32(col01b, col23b); /* row3=(30 31 32 33) */ \
+ row4 = _mm_unpacklo_pi32(col01c, col23c); /* row4=(40 41 42 43) */ \
+ row5 = _mm_unpackhi_pi32(col01c, col23c); /* row5=(50 51 52 53) */ \
+ \
+ tmp3 = _mm_add_pi16(row3, row4); /* tmp3=row3+row4 */ \
+ tmp2 = _mm_add_pi16(row2, row5); /* tmp2=row2+row5 */ \
+ tmp4 = _mm_sub_pi16(row3, row4); /* tmp4=row3-row4 */ \
+ tmp5 = _mm_sub_pi16(row2, row5); /* tmp5=row2-row5 */ \
+ \
+ /* Even part */ \
+ \
+ tmp10 = _mm_add_pi16(tmp0, tmp3); /* tmp10=tmp0+tmp3 */ \
+ tmp13 = _mm_sub_pi16(tmp0, tmp3); /* tmp13=tmp0-tmp3 */ \
+ tmp11 = _mm_add_pi16(tmp1, tmp2); /* tmp11=tmp1+tmp2 */ \
+ tmp12 = _mm_sub_pi16(tmp1, tmp2); /* tmp12=tmp1-tmp2 */ \
+ \
+ out0 = _mm_add_pi16(tmp10, tmp11); /* out0=tmp10+tmp11 */ \
+ out4 = _mm_sub_pi16(tmp10, tmp11); /* out4=tmp10-tmp11 */ \
+ \
+ out0 = _mm_add_pi16(out0, PW_DESCALE_P2X); \
+ out4 = _mm_add_pi16(out4, PW_DESCALE_P2X); \
+ out0 = _mm_srai_pi16(out0, PASS1_BITS); \
+ out4 = _mm_srai_pi16(out4, PASS1_BITS); \
+ \
+ DO_FDCT_COMMON(2) \
+ \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*0], out0); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*1], out1); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*2], out2); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*3], out3); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*4], out4); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*5], out5); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*6], out6); \
+ _mm_store_si64((__m64*)&dataptr[DCTSIZE*7], out7); \
+}
+
+void
+jsimd_fdct_islow_mmi (DCTELEM *data)
+{
+ __m64 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+ __m64 out0, out1, out2, out3, out4, out5, out6, out7;
+ __m64 tmp12, tmp13;
+ DCTELEM *dataptr = data;
+
+ /* Pass 1: process rows. */
+
+ DO_FDCT_PASS1()
+ dataptr += DCTSIZE*4;
+ DO_FDCT_PASS1()
+
+ /* Pass 2: process columns. */
+
+ dataptr = data;
+ DO_FDCT_PASS2()
+ dataptr += 4;
+ DO_FDCT_PASS2()
+}
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2014-2015, 2018, D. R. Commander. All Rights Reserved.
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ * Authors: ZhuChen <zhuchen@loongson.cn>
+ * CaiWanwei <caiwanwei@loongson.cn>
+ * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
+ *
+ * Based on the x86 SIMD extension for IJG JPEG library
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* SLOW INTEGER INVERSE DCT */
+
+#include "jsimd_mmi.h"
+
+
+#define CONST_BITS 13
+#define PASS1_BITS 2
+#define DESCALE_P1 (CONST_BITS-PASS1_BITS)
+#define DESCALE_P2 (CONST_BITS+PASS1_BITS+3)
+#define CENTERJSAMPLE 128
+
+#define FIX_0_298 ((short)2446) /* FIX(0.298631336) */
+#define FIX_0_390 ((short)3196) /* FIX(0.390180644) */
+#define FIX_0_899 ((short)7373) /* FIX(0.899976223) */
+#define FIX_0_541 ((short)4433) /* FIX(0.541196100) */
+#define FIX_0_765 ((short)6270) /* FIX(0.765366865) */
+#define FIX_1_175 ((short)9633) /* FIX(1.175875602) */
+#define FIX_1_501 ((short)12299) /* FIX(1.501321110) */
+#define FIX_1_847 ((short)15137) /* FIX(1.847759065) */
+#define FIX_1_961 ((short)16069) /* FIX(1.961570560) */
+#define FIX_2_053 ((short)16819) /* FIX(2.053119869) */
+#define FIX_2_562 ((short)20995) /* FIX(2.562915447) */
+#define FIX_3_072 ((short)25172) /* FIX(3.072711026) */
+
+enum const_index {
+ index_PW_F130_F054,
+ index_PW_F054_MF130,
+ index_PW_MF078_F117,
+ index_PW_F117_F078,
+ index_PW_MF060_MF089,
+ index_PW_MF089_F060,
+ index_PW_MF050_MF256,
+ index_PW_MF256_F050,
+ index_PD_DESCALE_P1,
+ index_PD_DESCALE_P2,
+ index_PB_CENTERJSAMP
+};
+
+static uint64_t const_value[] = {
+ _uint64_set_pi16(FIX_0_541, (FIX_0_541+FIX_0_765),
+ FIX_0_541, (FIX_0_541+FIX_0_765)),
+ _uint64_set_pi16((FIX_0_541-FIX_1_847), FIX_0_541,
+ (FIX_0_541-FIX_1_847), FIX_0_541),
+ _uint64_set_pi16(FIX_1_175, (FIX_1_175-FIX_1_961),
+ FIX_1_175, (FIX_1_175-FIX_1_961)),
+ _uint64_set_pi16((FIX_1_175-FIX_0_390), FIX_1_175,
+ (FIX_1_175-FIX_0_390), FIX_1_175),
+ _uint64_set_pi16(-FIX_0_899, (FIX_0_298-FIX_0_899),
+ -FIX_0_899, (FIX_0_298-FIX_0_899)),
+ _uint64_set_pi16((FIX_1_501-FIX_0_899), -FIX_0_899,
+ (FIX_1_501-FIX_0_899), -FIX_0_899),
+ _uint64_set_pi16(-FIX_2_562, (FIX_2_053-FIX_2_562),
+ -FIX_2_562, (FIX_2_053-FIX_2_562)),
+ _uint64_set_pi16((FIX_3_072-FIX_2_562), -FIX_2_562,
+ (FIX_3_072-FIX_2_562), -FIX_2_562),
+ _uint64_set_pi32((1 << (DESCALE_P1-1)), (1 << (DESCALE_P1-1))),
+ _uint64_set_pi32((1 << (DESCALE_P2-1)), (1 << (DESCALE_P2-1))),
+ _uint64_set_pi8(CENTERJSAMPLE, CENTERJSAMPLE, CENTERJSAMPLE, CENTERJSAMPLE,
+ CENTERJSAMPLE, CENTERJSAMPLE, CENTERJSAMPLE, CENTERJSAMPLE)
+};
+
+#define PW_F130_F054 get_const_value(index_PW_F130_F054)
+#define PW_F054_MF130 get_const_value(index_PW_F054_MF130)
+#define PW_MF078_F117 get_const_value(index_PW_MF078_F117)
+#define PW_F117_F078 get_const_value(index_PW_F117_F078)
+#define PW_MF060_MF089 get_const_value(index_PW_MF060_MF089)
+#define PW_MF089_F060 get_const_value(index_PW_MF089_F060)
+#define PW_MF050_MF256 get_const_value(index_PW_MF050_MF256)
+#define PW_MF256_F050 get_const_value(index_PW_MF256_F050)
+#define PD_DESCALE_P1 get_const_value(index_PD_DESCALE_P1)
+#define PD_DESCALE_P2 get_const_value(index_PD_DESCALE_P2)
+#define PB_CENTERJSAMP get_const_value(index_PB_CENTERJSAMP)
+
+
+#define test_m32_zero(mm32) (!(*(uint32_t *)&mm32))
+#define test_m64_zero(mm64) (!(*(uint64_t *)&mm64))
+
+
+#define DO_IDCT_COMMON(PASS) \
+{ \
+ __m64 tmp0_3l, tmp0_3h, tmp1_2l, tmp1_2h; \
+ __m64 tmp0l, tmp0h, tmp1l, tmp1h, tmp2l, tmp2h, tmp3l, tmp3h; \
+ __m64 z34l, z34h, z3l, z3h, z4l, z4h, z3, z4; \
+ __m64 out0l, out0h, out1l, out1h, out2l, out2h, out3l, out3h; \
+ __m64 out4l, out4h, out5l, out5h, out6l, out6h, out7l, out7h; \
+ \
+ z3 = _mm_add_pi16(tmp0, tmp2); \
+ z4 = _mm_add_pi16(tmp1, tmp3); \
+ \
+ /* (Original) \
+ * z5 = (z3 + z4) * 1.175875602; \
+ * z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; \
+ * z3 += z5; z4 += z5; \
+ * \
+ * (This implementation) \
+ * z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; \
+ * z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); \
+ */ \
+ \
+ z34l = _mm_unpacklo_pi16(z3, z4); \
+ z34h = _mm_unpackhi_pi16(z3, z4); \
+ z3l = _mm_madd_pi16(z34l, PW_MF078_F117); \
+ z3h = _mm_madd_pi16(z34h, PW_MF078_F117); \
+ z4l = _mm_madd_pi16(z34l, PW_F117_F078); \
+ z4h = _mm_madd_pi16(z34h, PW_F117_F078); \
+ \
+ /* (Original) \
+ * z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; \
+ * tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; \
+ * tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; \
+ * z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; \
+ * tmp0 += z1 + z3; tmp1 += z2 + z4; \
+ * tmp2 += z2 + z3; tmp3 += z1 + z4; \
+ * \
+ * (This implementation) \
+ * tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; \
+ * tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; \
+ * tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); \
+ * tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); \
+ * tmp0 += z3; tmp1 += z4; \
+ * tmp2 += z3; tmp3 += z4; \
+ */ \
+ \
+ tmp0_3l = _mm_unpacklo_pi16(tmp0, tmp3); \
+ tmp0_3h = _mm_unpackhi_pi16(tmp0, tmp3); \
+ \
+ tmp0l = _mm_madd_pi16(tmp0_3l, PW_MF060_MF089); \
+ tmp0h = _mm_madd_pi16(tmp0_3h, PW_MF060_MF089); \
+ tmp3l = _mm_madd_pi16(tmp0_3l, PW_MF089_F060); \
+ tmp3h = _mm_madd_pi16(tmp0_3h, PW_MF089_F060); \
+ \
+ tmp0l = _mm_add_pi32(tmp0l, z3l); \
+ tmp0h = _mm_add_pi32(tmp0h, z3h); \
+ tmp3l = _mm_add_pi32(tmp3l, z4l); \
+ tmp3h = _mm_add_pi32(tmp3h, z4h); \
+ \
+ tmp1_2l = _mm_unpacklo_pi16(tmp1, tmp2); \
+ tmp1_2h = _mm_unpackhi_pi16(tmp1, tmp2); \
+ \
+ tmp1l = _mm_madd_pi16(tmp1_2l, PW_MF050_MF256); \
+ tmp1h = _mm_madd_pi16(tmp1_2h, PW_MF050_MF256); \
+ tmp2l = _mm_madd_pi16(tmp1_2l, PW_MF256_F050); \
+ tmp2h = _mm_madd_pi16(tmp1_2h, PW_MF256_F050); \
+ \
+ tmp1l = _mm_add_pi32(tmp1l, z4l); \
+ tmp1h = _mm_add_pi32(tmp1h, z4h); \
+ tmp2l = _mm_add_pi32(tmp2l, z3l); \
+ tmp2h = _mm_add_pi32(tmp2h, z3h); \
+ \
+ /* Final output stage */ \
+ \
+ out0l = _mm_add_pi32(tmp10l, tmp3l); \
+ out0h = _mm_add_pi32(tmp10h, tmp3h); \
+ out7l = _mm_sub_pi32(tmp10l, tmp3l); \
+ out7h = _mm_sub_pi32(tmp10h, tmp3h); \
+ \
+ out0l = _mm_add_pi32(out0l, PD_DESCALE_P##PASS); \
+ out0h = _mm_add_pi32(out0h, PD_DESCALE_P##PASS); \
+ out0l = _mm_srai_pi32(out0l, DESCALE_P##PASS); \
+ out0h = _mm_srai_pi32(out0h, DESCALE_P##PASS); \
+ \
+ out7l = _mm_add_pi32(out7l, PD_DESCALE_P##PASS); \
+ out7h = _mm_add_pi32(out7h, PD_DESCALE_P##PASS); \
+ out7l = _mm_srai_pi32(out7l, DESCALE_P##PASS); \
+ out7h = _mm_srai_pi32(out7h, DESCALE_P##PASS); \
+ \
+ out0 = _mm_packs_pi32(out0l, out0h); \
+ out7 = _mm_packs_pi32(out7l, out7h); \
+ \
+ out1l = _mm_add_pi32(tmp11l, tmp2l); \
+ out1h = _mm_add_pi32(tmp11h, tmp2h); \
+ out6l = _mm_sub_pi32(tmp11l, tmp2l); \
+ out6h = _mm_sub_pi32(tmp11h, tmp2h); \
+ \
+ out1l = _mm_add_pi32(out1l, PD_DESCALE_P##PASS); \
+ out1h = _mm_add_pi32(out1h, PD_DESCALE_P##PASS); \
+ out1l = _mm_srai_pi32(out1l, DESCALE_P##PASS); \
+ out1h = _mm_srai_pi32(out1h, DESCALE_P##PASS); \
+ \
+ out6l = _mm_add_pi32(out6l, PD_DESCALE_P##PASS); \
+ out6h = _mm_add_pi32(out6h, PD_DESCALE_P##PASS); \
+ out6l = _mm_srai_pi32(out6l, DESCALE_P##PASS); \
+ out6h = _mm_srai_pi32(out6h, DESCALE_P##PASS); \
+ \
+ out1 = _mm_packs_pi32(out1l, out1h); \
+ out6 = _mm_packs_pi32(out6l, out6h); \
+ \
+ out2l = _mm_add_pi32(tmp12l, tmp1l); \
+ out2h = _mm_add_pi32(tmp12h, tmp1h); \
+ out5l = _mm_sub_pi32(tmp12l, tmp1l); \
+ out5h = _mm_sub_pi32(tmp12h, tmp1h); \
+ \
+ out2l = _mm_add_pi32(out2l, PD_DESCALE_P##PASS); \
+ out2h = _mm_add_pi32(out2h, PD_DESCALE_P##PASS); \
+ out2l = _mm_srai_pi32(out2l, DESCALE_P##PASS); \
+ out2h = _mm_srai_pi32(out2h, DESCALE_P##PASS); \
+ \
+ out5l = _mm_add_pi32(out5l, PD_DESCALE_P##PASS); \
+ out5h = _mm_add_pi32(out5h, PD_DESCALE_P##PASS); \
+ out5l = _mm_srai_pi32(out5l, DESCALE_P##PASS); \
+ out5h = _mm_srai_pi32(out5h, DESCALE_P##PASS); \
+ \
+ out2 = _mm_packs_pi32(out2l, out2h); \
+ out5 = _mm_packs_pi32(out5l, out5h); \
+ \
+ out3l = _mm_add_pi32(tmp13l, tmp0l); \
+ out3h = _mm_add_pi32(tmp13h, tmp0h); \
+ \
+ out4l = _mm_sub_pi32(tmp13l, tmp0l); \
+ out4h = _mm_sub_pi32(tmp13h, tmp0h); \
+ \
+ out3l = _mm_add_pi32(out3l, PD_DESCALE_P##PASS); \
+ out3h = _mm_add_pi32(out3h, PD_DESCALE_P##PASS); \
+ out3l = _mm_srai_pi32(out3l, DESCALE_P##PASS); \
+ out3h = _mm_srai_pi32(out3h, DESCALE_P##PASS); \
+ \
+ out4l = _mm_add_pi32(out4l, PD_DESCALE_P##PASS); \
+ out4h = _mm_add_pi32(out4h, PD_DESCALE_P##PASS); \
+ out4l = _mm_srai_pi32(out4l, DESCALE_P##PASS); \
+ out4h = _mm_srai_pi32(out4h, DESCALE_P##PASS); \
+ \
+ out3 = _mm_packs_pi32(out3l, out3h); \
+ out4 = _mm_packs_pi32(out4l, out4h); \
+}
+
+#define DO_IDCT_PASS1(iter) \
+{ \
+ __m64 col0l, col1l, col2l, col3l, col4l, col5l, col6l, col7l; \
+ __m64 quant0l, quant1l, quant2l, quant3l; \
+ __m64 quant4l, quant5l, quant6l, quant7l; \
+ __m64 z23, z2, z3, z23l, z23h; \
+ __m64 row01a, row01b, row01c, row01d, row23a, row23b, row23c, row23d; \
+ __m64 row0l, row0h, row1l, row1h, row2l, row2h, row3l, row3h; \
+ __m64 tmp0l, tmp0h, tmp1l, tmp1h, tmp2l, tmp2h, tmp3l, tmp3h; \
+ __m64 tmp10l, tmp10h, tmp11l, tmp11h, tmp12l, tmp12h, tmp13l, tmp13h; \
+ __m32 col0a, col1a, mm0; \
+ \
+ col0a = _mm_load_si32((__m32 *)&inptr[DCTSIZE*1]); \
+ col1a = _mm_load_si32((__m32 *)&inptr[DCTSIZE*2]); \
+ mm0 = _mm_or_si32(col0a, col1a); \
+ \
+ if (test_m32_zero(mm0)) { \
+ __m64 mm1, mm2; \
+ \
+ col0l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*0]); \
+ col1l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*1]); \
+ col2l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*2]); \
+ col3l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*3]); \
+ col4l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*4]); \
+ col5l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*5]); \
+ col6l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*6]); \
+ col7l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*7]); \
+ \
+ mm1 = _mm_or_si64(col1l, col3l); \
+ mm2 = _mm_or_si64(col2l, col4l); \
+ mm1 = _mm_or_si64(mm1, col5l); \
+ mm2 = _mm_or_si64(mm2, col6l); \
+ mm1 = _mm_or_si64(mm1, col7l); \
+ mm1 = _mm_or_si64(mm1, mm2); \
+ \
+ if (test_m64_zero(mm1)) { \
+ __m64 dcval, dcvall, dcvalh, row0, row1, row2, row3; \
+ \
+ /* AC terms all zero */ \
+ \
+ quant0l = _mm_load_si64((__m64 *)&quantptr[DCTSIZE*0]); \
+ \
+ dcval = _mm_mullo_pi16(col0l, quant0l); \
+ dcval = _mm_slli_pi16(dcval, PASS1_BITS); /* dcval=(00 10 20 30) */ \
+ \
+ dcvall = _mm_unpacklo_pi16(dcval, dcval); /* dcvall=(00 00 10 10) */ \
+ dcvalh = _mm_unpackhi_pi16(dcval, dcval); /* dcvalh=(20 20 30 30) */ \
+ \
+ row0 = _mm_unpacklo_pi32(dcvall, dcvall); /* row0=(00 00 00 00) */ \
+ row1 = _mm_unpackhi_pi32(dcvall, dcvall); /* row1=(10 10 10 10) */ \
+ row2 = _mm_unpacklo_pi32(dcvalh, dcvalh); /* row2=(20 20 20 20) */ \
+ row3 = _mm_unpackhi_pi32(dcvalh, dcvalh); /* row3=(30 30 30 30) */ \
+ \
+ _mm_store_si64((__m64 *)&wsptr[DCTSIZE*0], row0); \
+ _mm_store_si64((__m64 *)&wsptr[DCTSIZE*0+4], row0); \
+ _mm_store_si64((__m64 *)&wsptr[DCTSIZE*1], row1); \
+ _mm_store_si64((__m64 *)&wsptr[DCTSIZE*1+4], row1); \
+ _mm_store_si64((__m64 *)&wsptr[DCTSIZE*2], row2); \
+ _mm_store_si64((__m64 *)&wsptr[DCTSIZE*2+4], row2); \
+ _mm_store_si64((__m64 *)&wsptr[DCTSIZE*3], row3); \
+ _mm_store_si64((__m64 *)&wsptr[DCTSIZE*3+4], row3); \
+ \
+ goto nextcolumn##iter; \
+ } \
+ } \
+ \
+ /* Even part \
+ * \
+ * (Original) \
+ * z1 = (z2 + z3) * 0.541196100; \
+ * tmp2 = z1 + z3 * -1.847759065; \
+ * tmp3 = z1 + z2 * 0.765366865; \
+ * \
+ * (This implementation) \
+ * tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); \
+ * tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; \
+ */ \
+ \
+ col0l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*0]); /* (00 10 20 30) */ \
+ col2l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*2]); /* (02 12 22 32) */ \
+ col4l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*4]); /* (04 14 24 34) */ \
+ col6l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*6]); /* (06 16 26 36) */ \
+ \
+ quant0l = _mm_load_si64((__m64 *)&quantptr[DCTSIZE*0]); \
+ quant2l = _mm_load_si64((__m64 *)&quantptr[DCTSIZE*2]); \
+ quant4l = _mm_load_si64((__m64 *)&quantptr[DCTSIZE*4]); \
+ quant6l = _mm_load_si64((__m64 *)&quantptr[DCTSIZE*6]); \
+ \
+ z2 = _mm_mullo_pi16(col2l, quant2l); \
+ z3 = _mm_mullo_pi16(col6l, quant6l); \
+ \
+ z23l = _mm_unpacklo_pi16(z2, z3); \
+ z23h = _mm_unpackhi_pi16(z2, z3); \
+ tmp3l = _mm_madd_pi16(z23l, PW_F130_F054); \
+ tmp3h = _mm_madd_pi16(z23h, PW_F130_F054); \
+ tmp2l = _mm_madd_pi16(z23l, PW_F054_MF130); \
+ tmp2h = _mm_madd_pi16(z23h, PW_F054_MF130); \
+ \
+ z2 = _mm_mullo_pi16(col0l, quant0l); \
+ z3 = _mm_mullo_pi16(col4l, quant4l); \
+ \
+ z23 = _mm_add_pi16(z2, z3); \
+ tmp0l = _mm_loadlo_pi16_f(z23); \
+ tmp0h = _mm_loadhi_pi16_f(z23); \
+ tmp0l = _mm_srai_pi32(tmp0l, (16-CONST_BITS)); \
+ tmp0h = _mm_srai_pi32(tmp0h, (16-CONST_BITS)); \
+ \
+ tmp10l = _mm_add_pi32(tmp0l, tmp3l); \
+ tmp10h = _mm_add_pi32(tmp0h, tmp3h); \
+ tmp13l = _mm_sub_pi32(tmp0l, tmp3l); \
+ tmp13h = _mm_sub_pi32(tmp0h, tmp3h); \
+ \
+ z23 = _mm_sub_pi16(z2, z3); \
+ tmp1l = _mm_loadlo_pi16_f(z23); \
+ tmp1h = _mm_loadhi_pi16_f(z23); \
+ tmp1l = _mm_srai_pi32(tmp1l, (16-CONST_BITS)); \
+ tmp1h = _mm_srai_pi32(tmp1h, (16-CONST_BITS)); \
+ \
+ tmp11l = _mm_add_pi32(tmp1l, tmp2l); \
+ tmp11h = _mm_add_pi32(tmp1h, tmp2h); \
+ tmp12l = _mm_sub_pi32(tmp1l, tmp2l); \
+ tmp12h = _mm_sub_pi32(tmp1h, tmp2h); \
+ \
+ /* Odd part */ \
+ \
+ col1l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*1]); /* (01 11 21 31) */ \
+ col3l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*3]); /* (03 13 23 33) */ \
+ col5l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*5]); /* (05 15 25 35) */ \
+ col7l = _mm_load_si64((__m64 *)&inptr[DCTSIZE*7]); /* (07 17 27 37) */ \
+ \
+ quant1l = _mm_load_si64((__m64 *)&quantptr[DCTSIZE*1]); \
+ quant3l = _mm_load_si64((__m64 *)&quantptr[DCTSIZE*3]); \
+ quant5l = _mm_load_si64((__m64 *)&quantptr[DCTSIZE*5]); \
+ quant7l = _mm_load_si64((__m64 *)&quantptr[DCTSIZE*7]); \
+ \
+ tmp0 = _mm_mullo_pi16(col7l, quant7l); \
+ tmp1 = _mm_mullo_pi16(col5l, quant5l); \
+ tmp2 = _mm_mullo_pi16(col3l, quant3l); \
+ tmp3 = _mm_mullo_pi16(col1l, quant1l); \
+ \
+ DO_IDCT_COMMON(1) \
+ \
+ /* out0=(00 10 20 30), out1=(01 11 21 31) */ \
+ /* out2=(02 12 22 32), out3=(03 13 23 33) */ \
+ /* out4=(04 14 24 34), out5=(05 15 25 35) */ \
+ /* out6=(06 16 26 36), out7=(07 17 27 37) */ \
+ \
+ /* Transpose coefficients */ \
+ \
+ row01a = _mm_unpacklo_pi16(out0, out1); /* row01a=(00 01 10 11) */ \
+ row23a = _mm_unpackhi_pi16(out0, out1); /* row23a=(20 21 30 31) */ \
+ row01d = _mm_unpacklo_pi16(out6, out7); /* row01d=(06 07 16 17) */ \
+ row23d = _mm_unpackhi_pi16(out6, out7); /* row23d=(26 27 36 37) */ \
+ \
+ row01b = _mm_unpacklo_pi16(out2, out3); /* row01b=(02 03 12 13) */ \
+ row23b = _mm_unpackhi_pi16(out2, out3); /* row23b=(22 23 32 33) */ \
+ row01c = _mm_unpacklo_pi16(out4, out5); /* row01c=(04 05 14 15) */ \
+ row23c = _mm_unpackhi_pi16(out4, out5); /* row23c=(24 25 34 35) */ \
+ \
+ row0l = _mm_unpacklo_pi32(row01a, row01b); /* row0l=(00 01 02 03) */ \
+ row1l = _mm_unpackhi_pi32(row01a, row01b); /* row1l=(10 11 12 13) */ \
+ row2l = _mm_unpacklo_pi32(row23a, row23b); /* row2l=(20 21 22 23) */ \
+ row3l = _mm_unpackhi_pi32(row23a, row23b); /* row3l=(30 31 32 33) */ \
+ \
+ row0h = _mm_unpacklo_pi32(row01c, row01d); /* row0h=(04 05 06 07) */ \
+ row1h = _mm_unpackhi_pi32(row01c, row01d); /* row1h=(14 15 16 17) */ \
+ row2h = _mm_unpacklo_pi32(row23c, row23d); /* row2h=(24 25 26 27) */ \
+ row3h = _mm_unpackhi_pi32(row23c, row23d); /* row3h=(34 35 36 37) */ \
+ \
+ _mm_store_si64((__m64*)&wsptr[DCTSIZE*0], row0l); \
+ _mm_store_si64((__m64*)&wsptr[DCTSIZE*0+4], row0h); \
+ _mm_store_si64((__m64*)&wsptr[DCTSIZE*1], row1l); \
+ _mm_store_si64((__m64*)&wsptr[DCTSIZE*1+4], row1h); \
+ _mm_store_si64((__m64*)&wsptr[DCTSIZE*2], row2l); \
+ _mm_store_si64((__m64*)&wsptr[DCTSIZE*2+4], row2h); \
+ _mm_store_si64((__m64*)&wsptr[DCTSIZE*3], row3l); \
+ _mm_store_si64((__m64*)&wsptr[DCTSIZE*3+4], row3h); \
+}
+
+#define DO_IDCT_PASS2(ctr) \
+{ \
+ __m64 row0l, row1l, row2l, row3l, row4l, row5l, row6l, row7l; \
+ __m64 z23, z23l, z23h; \
+ __m64 col0123a, col0123b, col0123c, col0123d; \
+ __m64 col01l, col01h, col23l, col23h, row06, row17, row24, row35; \
+ __m64 col0, col1, col2, col3; \
+ __m64 tmp0l, tmp0h, tmp1l, tmp1h, tmp2l, tmp2h, tmp3l, tmp3h; \
+ __m64 tmp10l, tmp10h, tmp11l, tmp11h, tmp12l, tmp12h, tmp13l, tmp13h; \
+ \
+ row0l = _mm_load_si64((__m64 *)&wsptr[DCTSIZE*0]); /* (00 01 02 03) */ \
+ row1l = _mm_load_si64((__m64 *)&wsptr[DCTSIZE*1]); /* (10 11 12 13) */ \
+ row2l = _mm_load_si64((__m64 *)&wsptr[DCTSIZE*2]); /* (20 21 22 23) */ \
+ row3l = _mm_load_si64((__m64 *)&wsptr[DCTSIZE*3]); /* (30 31 32 33) */ \
+ row4l = _mm_load_si64((__m64 *)&wsptr[DCTSIZE*4]); /* (40 41 42 43) */ \
+ row5l = _mm_load_si64((__m64 *)&wsptr[DCTSIZE*5]); /* (50 51 52 53) */ \
+ row6l = _mm_load_si64((__m64 *)&wsptr[DCTSIZE*6]); /* (60 61 62 63) */ \
+ row7l = _mm_load_si64((__m64 *)&wsptr[DCTSIZE*7]); /* (70 71 72 73) */ \
+ \
+ /* Even part \
+ * \
+ * (Original) \
+ * z1 = (z2 + z3) * 0.541196100; \
+ * tmp2 = z1 + z3 * -1.847759065; \
+ * tmp3 = z1 + z2 * 0.765366865; \
+ * \
+ * (This implementation) \
+ * tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); \
+ * tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; \
+ */ \
+ \
+ z23l = _mm_unpacklo_pi16(row2l, row6l); \
+ z23h = _mm_unpackhi_pi16(row2l, row6l); \
+ \
+ tmp3l = _mm_madd_pi16(z23l, PW_F130_F054); \
+ tmp3h = _mm_madd_pi16(z23h, PW_F130_F054); \
+ tmp2l = _mm_madd_pi16(z23l, PW_F054_MF130); \
+ tmp2h = _mm_madd_pi16(z23h, PW_F054_MF130); \
+ \
+ z23 = _mm_add_pi16(row0l, row4l); \
+ tmp0l = _mm_loadlo_pi16_f(z23); \
+ tmp0h = _mm_loadhi_pi16_f(z23); \
+ tmp0l = _mm_srai_pi32(tmp0l, (16-CONST_BITS)); \
+ tmp0h = _mm_srai_pi32(tmp0h, (16-CONST_BITS)); \
+ \
+ tmp10l = _mm_add_pi32(tmp0l, tmp3l); \
+ tmp10h = _mm_add_pi32(tmp0h, tmp3h); \
+ tmp13l = _mm_sub_pi32(tmp0l, tmp3l); \
+ tmp13h = _mm_sub_pi32(tmp0h, tmp3h); \
+ \
+ z23 = _mm_sub_pi16(row0l, row4l); \
+ tmp1l = _mm_loadlo_pi16_f(z23); \
+ tmp1h = _mm_loadhi_pi16_f(z23); \
+ tmp1l = _mm_srai_pi32(tmp1l, (16-CONST_BITS)); \
+ tmp1h = _mm_srai_pi32(tmp1h, (16-CONST_BITS)); \
+ \
+ tmp11l = _mm_add_pi32(tmp1l, tmp2l); \
+ tmp11h = _mm_add_pi32(tmp1h, tmp2h); \
+ tmp12l = _mm_sub_pi32(tmp1l, tmp2l); \
+ tmp12h = _mm_sub_pi32(tmp1h, tmp2h); \
+ \
+ /* Odd part */ \
+ \
+ tmp0 = row7l; \
+ tmp1 = row5l; \
+ tmp2 = row3l; \
+ tmp3 = row1l; \
+ \
+ DO_IDCT_COMMON(2) \
+ \
+ /* out0=(00 01 02 03), out1=(10 11 12 13) */ \
+ /* out2=(20 21 22 23), out3=(30 31 32 33) */ \
+ /* out4=(40 41 42 43), out5=(50 51 52 53) */ \
+ /* out6=(60 61 62 63), out7=(70 71 72 73) */ \
+ \
+ row06 = _mm_packs_pi16(out0, out6); /* row06=(00 01 02 03 60 61 62 63) */ \
+ row17 = _mm_packs_pi16(out1, out7); /* row17=(10 11 12 13 70 71 72 73) */ \
+ row24 = _mm_packs_pi16(out2, out4); /* row24=(20 21 22 23 40 41 42 43) */ \
+ row35 = _mm_packs_pi16(out3, out5); /* row35=(30 31 32 33 50 51 52 53) */ \
+ \
+ row06 = _mm_add_pi8(row06, PB_CENTERJSAMP); \
+ row17 = _mm_add_pi8(row17, PB_CENTERJSAMP); \
+ row24 = _mm_add_pi8(row24, PB_CENTERJSAMP); \
+ row35 = _mm_add_pi8(row35, PB_CENTERJSAMP); \
+ \
+ /* Transpose coefficients */ \
+ \
+ col0123a = _mm_unpacklo_pi8(row06, row17); /* col0123a=(00 10 01 11 02 12 03 13) */ \
+ col0123d = _mm_unpackhi_pi8(row06, row17); /* col0123d=(60 70 61 71 62 72 63 73) */ \
+ col0123b = _mm_unpacklo_pi8(row24, row35); /* col0123b=(20 30 21 31 22 32 23 33) */ \
+ col0123c = _mm_unpackhi_pi8(row24, row35); /* col0123c=(40 50 41 51 42 52 43 53) */ \
+ \
+ col01l = _mm_unpacklo_pi16(col0123a, col0123b); /* col01l=(00 10 20 30 01 11 21 31) */ \
+ col23l = _mm_unpackhi_pi16(col0123a, col0123b); /* col23l=(02 12 22 32 03 13 23 33) */ \
+ col01h = _mm_unpacklo_pi16(col0123c, col0123d); /* col01h=(40 50 60 70 41 51 61 71) */ \
+ col23h = _mm_unpackhi_pi16(col0123c, col0123d); /* col23h=(42 52 62 72 43 53 63 73) */ \
+ \
+ col0 = _mm_unpacklo_pi32(col01l, col01h); /* col0=(00 10 20 30 40 50 60 70) */ \
+ col1 = _mm_unpackhi_pi32(col01l, col01h); /* col1=(01 11 21 31 41 51 61 71) */ \
+ col2 = _mm_unpacklo_pi32(col23l, col23h); /* col2=(02 12 22 32 42 52 62 72) */ \
+ col3 = _mm_unpackhi_pi32(col23l, col23h); /* col3=(03 13 23 33 43 53 63 73) */ \
+ \
+ _mm_store_si64((__m64*)(output_buf[ctr+0] + output_col), col0); \
+ _mm_store_si64((__m64*)(output_buf[ctr+1] + output_col), col1); \
+ _mm_store_si64((__m64*)(output_buf[ctr+2] + output_col), col2); \
+ _mm_store_si64((__m64*)(output_buf[ctr+3] + output_col), col3); \
+}
+
+void
+jsimd_idct_islow_mmi (void *dct_table, JCOEFPTR coef_block,
+ JSAMPARRAY output_buf, JDIMENSION output_col)
+{
+ __m64 tmp0, tmp1, tmp2, tmp3;
+ __m64 out0, out1, out2, out3, out4, out5, out6, out7;
+ JCOEFPTR inptr;
+ ISLOW_MULT_TYPE *quantptr;
+ JCOEF *wsptr;
+ JCOEF workspace[DCTSIZE2]; /* buffers data between passes */
+
+ /* Pass 1: process columns. */
+
+ inptr = coef_block;
+ quantptr = (ISLOW_MULT_TYPE *)dct_table;
+ wsptr = workspace;
+
+ DO_IDCT_PASS1(1)
+ nextcolumn1:
+ inptr += 4;
+ quantptr += 4;
+ wsptr += DCTSIZE*4;
+ DO_IDCT_PASS1(2)
+ nextcolumn2:
+
+ /* Pass 2: process rows. */
+
+ wsptr = workspace;
+
+ DO_IDCT_PASS2(0)
+ wsptr += 4;
+ DO_IDCT_PASS2(4)
+}
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ * Authors: ZhuChen <zhuchen@loongson.cn>
+ * CaiWanwei <caiwanwei@loongson.cn>
+ * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
+ * Copyright (C) 2018, D. R. Commander. All Rights Reserved.
+ *
+ * Based on the x86 SIMD extension for IJG JPEG library
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* INTEGER QUANTIZATION AND SAMPLE CONVERSION */
+
+#include "jsimd_mmi.h"
+
+
+#define DO_QUANT() \
+{ \
+ \
+ mm2 = _mm_load_si64((__m64 *)&workspace[0]); \
+ mm3 = _mm_load_si64((__m64 *)&workspace[4]); \
+ \
+ mm0 = mm2; \
+ mm1 = mm3; \
+ \
+ mm2 = _mm_srai_pi16(mm2, (WORD_BIT-1)); /* -1 if value < 0, */ \
+ /* 0 otherwise */ \
+ mm3 = _mm_srai_pi16(mm3, (WORD_BIT-1)); \
+ \
+ mm0 = _mm_xor_si64(mm0, mm2); /* val = -val */ \
+ mm1 = _mm_xor_si64(mm1, mm3); \
+ mm0 = _mm_sub_pi16(mm0, mm2); \
+ mm1 = _mm_sub_pi16(mm1, mm3); \
+ \
+ corr0 = _mm_load_si64((__m64 *)&divisors[DCTSIZE2*1]); /* correction */ \
+ corr1 = _mm_load_si64((__m64 *)&divisors[DCTSIZE2*1 + 4]); \
+ \
+ mm0 = _mm_add_pi16(mm0, corr0); /* correction + roundfactor */ \
+ mm1 = _mm_add_pi16(mm1, corr1); \
+ \
+ mm4 = mm0; \
+ mm5 = mm1; \
+ \
+ recip0 = _mm_load_si64((__m64 *)&divisors[DCTSIZE2*0]); /* reciprocal */ \
+ recip1 = _mm_load_si64((__m64 *)&divisors[DCTSIZE2*0 + 4]); \
+ \
+ mm0 = _mm_mulhi_pi16(mm0, recip0); \
+ mm1 = _mm_mulhi_pi16(mm1, recip1); \
+ \
+ mm0 = _mm_add_pi16(mm0, mm4); /* reciprocal is always negative */ \
+ mm1 = _mm_add_pi16(mm1, mm5); /* (MSB=1), so we always need to add the */ \
+ /* initial value (input value is never */ \
+ /* negative as we inverted it at the */ \
+ /* start of this routine) */ \
+ \
+ scale0 = _mm_load_si64((__m64 *)&divisors[DCTSIZE2*2]); /* scale */ \
+ scale1 = _mm_load_si64((__m64 *)&divisors[DCTSIZE2*2 + 4]); \
+ \
+ mm6 = scale0; \
+ mm7 = scale1; \
+ mm4 = mm0; \
+ mm5 = mm1; \
+ \
+ mm0 = _mm_mulhi_pi16(mm0, mm6); \
+ mm1 = _mm_mulhi_pi16(mm1, mm7); \
+ \
+ mm6 = _mm_srai_pi16(mm6, (WORD_BIT-1)); /* determine if scale... */ \
+ /* is negative */ \
+ mm7 = _mm_srai_pi16(mm7, (WORD_BIT-1)); \
+ \
+ mm6 = _mm_and_si64(mm6, mm4); /* and add input if it is */ \
+ mm7 = _mm_and_si64(mm7, mm5); \
+ mm0 = _mm_add_pi16(mm0, mm6); \
+ mm1 = _mm_add_pi16(mm1, mm7); \
+ \
+ mm4 = _mm_srai_pi16(mm4, (WORD_BIT-1)); /* then check if... */ \
+ mm5 = _mm_srai_pi16(mm5, (WORD_BIT-1)); /* negative input */ \
+ \
+ mm4 = _mm_and_si64(mm4, scale0); /* and add scale if it is */ \
+ mm5 = _mm_and_si64(mm5, scale1); \
+ mm0 = _mm_add_pi16(mm0, mm4); \
+ mm1 = _mm_add_pi16(mm1, mm5); \
+ \
+ mm0 = _mm_xor_si64(mm0, mm2); /* val = -val */ \
+ mm1 = _mm_xor_si64(mm1, mm3); \
+ mm0 = _mm_sub_pi16(mm0, mm2); \
+ mm1 = _mm_sub_pi16(mm1, mm3); \
+ \
+ _mm_store_si64((__m64 *)&output_ptr[0], mm0); \
+ _mm_store_si64((__m64 *)&output_ptr[4], mm1); \
+ \
+ workspace += DCTSIZE; \
+ divisors += DCTSIZE; \
+ output_ptr += DCTSIZE; \
+}
+
+
+void
+jsimd_quantize_mmi (JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace)
+{
+ JCOEFPTR output_ptr = coef_block;
+ __m64 mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7;
+ __m64 corr0, corr1, recip0, recip1, scale0, scale1;
+
+ DO_QUANT()
+ DO_QUANT()
+ DO_QUANT()
+ DO_QUANT()
+ DO_QUANT()
+ DO_QUANT()
+ DO_QUANT()
+ DO_QUANT()
+}
--- /dev/null
+/*
+ * jsimd_loongson.c
+ *
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright (C) 2009-2011, 2014, 2016, D. R. Commander.
+ * Copyright (C) 2013-2014, MIPS Technologies, Inc., California.
+ * Copyright (C) 2015, Matthieu Darbois.
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ *
+ * Based on the x86 SIMD extension for IJG JPEG library,
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ * For conditions of distribution and use, see copyright notice in jsimdext.inc
+ *
+ * This file contains the interface between the "normal" portions
+ * of the library and the SIMD implementations when running on a
+ * Loongson architecture.
+ */
+
+#define JPEG_INTERNALS
+#include "../../jinclude.h"
+#include "../../jpeglib.h"
+#include "../../jsimd.h"
+#include "../../jdct.h"
+#include "../../jsimddct.h"
+#include "../jsimd.h"
+
+static unsigned int simd_support = ~0;
+
+/*
+ * Check what SIMD accelerations are supported.
+ *
+ * FIXME: This code is racy under a multi-threaded environment.
+ */
+LOCAL(void)
+init_simd (void)
+{
+ char *env = NULL;
+
+ if (simd_support != ~0U)
+ return;
+
+ simd_support |= JSIMD_MMI;
+
+ /* Force different settings through environment variables */
+ env = getenv("JSIMD_FORCENONE");
+ if ((env != NULL) && (strcmp(env, "1") == 0))
+ simd_support = 0;
+}
+
+GLOBAL(int)
+jsimd_can_rgb_ycc (void)
+{
+ init_simd();
+
+ /* The code is optimised for these values only */
+ if (BITS_IN_JSAMPLE != 8)
+ return 0;
+ if (sizeof(JDIMENSION) != 4)
+ return 0;
+ if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+ return 0;
+
+ if (simd_support & JSIMD_MMI)
+ return 1;
+
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_rgb_gray (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_ycc_rgb (void)
+{
+ init_simd();
+
+ /* The code is optimised for these values only */
+ if (BITS_IN_JSAMPLE != 8)
+ return 0;
+ if (sizeof(JDIMENSION) != 4)
+ return 0;
+ if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+ return 0;
+
+ if (simd_support & JSIMD_MMI)
+ return 1;
+
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_ycc_rgb565 (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_c_can_null_convert (void)
+{
+ return 0;
+}
+
+GLOBAL(void)
+jsimd_rgb_ycc_convert (j_compress_ptr cinfo,
+ JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows)
+{
+ void (*mmifct)(JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
+
+ switch(cinfo->in_color_space) {
+ case JCS_EXT_RGB:
+ mmifct=jsimd_extrgb_ycc_convert_mmi;
+ break;
+ case JCS_EXT_RGBX:
+ case JCS_EXT_RGBA:
+ mmifct=jsimd_extrgbx_ycc_convert_mmi;
+ break;
+ case JCS_EXT_BGR:
+ mmifct=jsimd_extbgr_ycc_convert_mmi;
+ break;
+ case JCS_EXT_BGRX:
+ case JCS_EXT_BGRA:
+ mmifct=jsimd_extbgrx_ycc_convert_mmi;
+ break;
+ case JCS_EXT_XBGR:
+ case JCS_EXT_ABGR:
+ mmifct=jsimd_extxbgr_ycc_convert_mmi;
+ break;
+ case JCS_EXT_XRGB:
+ case JCS_EXT_ARGB:
+ mmifct=jsimd_extxrgb_ycc_convert_mmi;
+ break;
+ default:
+ mmifct=jsimd_rgb_ycc_convert_mmi;
+ break;
+ }
+
+ mmifct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
+}
+
+GLOBAL(void)
+jsimd_rgb_gray_convert (j_compress_ptr cinfo,
+ JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows)
+{
+}
+
+GLOBAL(void)
+jsimd_ycc_rgb_convert (j_decompress_ptr cinfo,
+ JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows)
+{
+ void (*mmifct)(JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
+
+ switch(cinfo->out_color_space) {
+ case JCS_EXT_RGB:
+ mmifct=jsimd_ycc_extrgb_convert_mmi;
+ break;
+ case JCS_EXT_RGBX:
+ case JCS_EXT_RGBA:
+ mmifct=jsimd_ycc_extrgbx_convert_mmi;
+ break;
+ case JCS_EXT_BGR:
+ mmifct=jsimd_ycc_extbgr_convert_mmi;
+ break;
+ case JCS_EXT_BGRX:
+ case JCS_EXT_BGRA:
+ mmifct=jsimd_ycc_extbgrx_convert_mmi;
+ break;
+ case JCS_EXT_XBGR:
+ case JCS_EXT_ABGR:
+ mmifct=jsimd_ycc_extxbgr_convert_mmi;
+ break;
+ case JCS_EXT_XRGB:
+ case JCS_EXT_ARGB:
+ mmifct=jsimd_ycc_extxrgb_convert_mmi;
+ break;
+ default:
+ mmifct=jsimd_ycc_rgb_convert_mmi;
+ break;
+ }
+
+ mmifct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
+}
+
+GLOBAL(void)
+jsimd_ycc_rgb565_convert (j_decompress_ptr cinfo,
+ JSAMPIMAGE input_buf, JDIMENSION input_row,
+ JSAMPARRAY output_buf, int num_rows)
+{
+}
+
+GLOBAL(void)
+jsimd_c_null_convert (j_compress_ptr cinfo,
+ JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+ JDIMENSION output_row, int num_rows)
+{
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_downsample (void)
+{
+ init_simd();
+
+ /* The code is optimised for these values only */
+ if (BITS_IN_JSAMPLE != 8)
+ return 0;
+ if (sizeof(JDIMENSION) != 4)
+ return 0;
+
+ if (simd_support & JSIMD_MMI)
+ return 1;
+
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_smooth_downsample (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_downsample (void)
+{
+ return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
+ JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+ jsimd_h2v2_downsample_mmi(cinfo->image_width, cinfo->max_v_samp_factor,
+ compptr->v_samp_factor, compptr->width_in_blocks,
+ input_data, output_data);
+}
+
+GLOBAL(void)
+jsimd_h2v2_smooth_downsample (j_compress_ptr cinfo,
+ jpeg_component_info *compptr,
+ JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+}
+
+GLOBAL(void)
+jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info *compptr,
+ JSAMPARRAY input_data, JSAMPARRAY output_data)
+{
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_upsample (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_upsample (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_int_upsample (void)
+{
+ return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_upsample (j_decompress_ptr cinfo,
+ jpeg_component_info *compptr,
+ JSAMPARRAY input_data,
+ JSAMPARRAY *output_data_ptr)
+{
+}
+
+GLOBAL(void)
+jsimd_h2v1_upsample (j_decompress_ptr cinfo,
+ jpeg_component_info *compptr,
+ JSAMPARRAY input_data,
+ JSAMPARRAY *output_data_ptr)
+{
+}
+
+GLOBAL(void)
+jsimd_int_upsample (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
+{
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_fancy_upsample (void)
+{
+ init_simd();
+
+ /* The code is optimised for these values only */
+ if (BITS_IN_JSAMPLE != 8)
+ return 0;
+ if (sizeof(JDIMENSION) != 4)
+ return 0;
+
+ if (simd_support & JSIMD_MMI)
+ return 1;
+
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_fancy_upsample (void)
+{
+ return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_fancy_upsample (j_decompress_ptr cinfo,
+ jpeg_component_info *compptr,
+ JSAMPARRAY input_data,
+ JSAMPARRAY *output_data_ptr)
+{
+ jsimd_h2v2_fancy_upsample_mmi(cinfo->max_v_samp_factor,
+ compptr->downsampled_width, input_data,
+ output_data_ptr);
+}
+
+GLOBAL(void)
+jsimd_h2v1_fancy_upsample (j_decompress_ptr cinfo,
+ jpeg_component_info *compptr,
+ JSAMPARRAY input_data,
+ JSAMPARRAY *output_data_ptr)
+{
+}
+
+GLOBAL(int)
+jsimd_can_h2v2_merged_upsample (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_h2v1_merged_upsample (void)
+{
+ return 0;
+}
+
+GLOBAL(void)
+jsimd_h2v2_merged_upsample (j_decompress_ptr cinfo,
+ JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr,
+ JSAMPARRAY output_buf)
+{
+}
+
+GLOBAL(void)
+jsimd_h2v1_merged_upsample (j_decompress_ptr cinfo,
+ JSAMPIMAGE input_buf,
+ JDIMENSION in_row_group_ctr,
+ JSAMPARRAY output_buf)
+{
+}
+
+GLOBAL(int)
+jsimd_can_convsamp (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_convsamp_float (void)
+{
+ return 0;
+}
+
+GLOBAL(void)
+jsimd_convsamp (JSAMPARRAY sample_data, JDIMENSION start_col,
+ DCTELEM *workspace)
+{
+}
+
+GLOBAL(void)
+jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col,
+ FAST_FLOAT *workspace)
+{
+}
+
+GLOBAL(int)
+jsimd_can_fdct_islow (void)
+{
+ init_simd();
+
+ /* The code is optimised for these values only */
+ if (DCTSIZE != 8)
+ return 0;
+ if (sizeof(DCTELEM) != 2)
+ return 0;
+
+ if (simd_support & JSIMD_MMI)
+ return 1;
+
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_fdct_ifast (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_fdct_float (void)
+{
+ return 0;
+}
+
+GLOBAL(void)
+jsimd_fdct_islow (DCTELEM *data)
+{
+ jsimd_fdct_islow_mmi(data);
+}
+
+GLOBAL(void)
+jsimd_fdct_ifast (DCTELEM *data)
+{
+}
+
+GLOBAL(void)
+jsimd_fdct_float (FAST_FLOAT *data)
+{
+}
+
+GLOBAL(int)
+jsimd_can_quantize (void)
+{
+ init_simd();
+
+ /* The code is optimised for these values only */
+ if (DCTSIZE != 8)
+ return 0;
+ if (sizeof(JCOEF) != 2)
+ return 0;
+ if (sizeof(DCTELEM) != 2)
+ return 0;
+
+ if (simd_support & JSIMD_MMI)
+ return 1;
+
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_quantize_float (void)
+{
+ return 0;
+}
+
+GLOBAL(void)
+jsimd_quantize (JCOEFPTR coef_block, DCTELEM *divisors,
+ DCTELEM *workspace)
+{
+ jsimd_quantize_mmi(coef_block, divisors, workspace);
+}
+
+GLOBAL(void)
+jsimd_quantize_float (JCOEFPTR coef_block, FAST_FLOAT *divisors,
+ FAST_FLOAT *workspace)
+{
+}
+
+GLOBAL(int)
+jsimd_can_idct_2x2 (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_4x4 (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_6x6 (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_12x12 (void)
+{
+ return 0;
+}
+
+GLOBAL(void)
+jsimd_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JCOEFPTR coef_block, JSAMPARRAY output_buf,
+ JDIMENSION output_col)
+{
+}
+
+GLOBAL(void)
+jsimd_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JCOEFPTR coef_block, JSAMPARRAY output_buf,
+ JDIMENSION output_col)
+{
+}
+
+GLOBAL(void)
+jsimd_idct_6x6 (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JCOEFPTR coef_block, JSAMPARRAY output_buf,
+ JDIMENSION output_col)
+{
+}
+
+GLOBAL(void)
+jsimd_idct_12x12 (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JCOEFPTR coef_block, JSAMPARRAY output_buf,
+ JDIMENSION output_col)
+{
+}
+
+GLOBAL(int)
+jsimd_can_idct_islow (void)
+{
+ init_simd();
+
+ /* The code is optimised for these values only */
+ if (DCTSIZE != 8)
+ return 0;
+ if (sizeof(JCOEF) != 2)
+ return 0;
+ if (BITS_IN_JSAMPLE != 8)
+ return 0;
+ if (sizeof(JDIMENSION) != 4)
+ return 0;
+ if (sizeof(ISLOW_MULT_TYPE) != 2)
+ return 0;
+
+ if (simd_support & JSIMD_MMI)
+ return 1;
+
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_ifast (void)
+{
+ return 0;
+}
+
+GLOBAL(int)
+jsimd_can_idct_float (void)
+{
+ return 0;
+}
+
+GLOBAL(void)
+jsimd_idct_islow (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JCOEFPTR coef_block, JSAMPARRAY output_buf,
+ JDIMENSION output_col)
+{
+ jsimd_idct_islow_mmi(compptr->dct_table, coef_block, output_buf, output_col);
+}
+
+GLOBAL(void)
+jsimd_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JCOEFPTR coef_block, JSAMPARRAY output_buf,
+ JDIMENSION output_col)
+{
+}
+
+GLOBAL(void)
+jsimd_idct_float (j_decompress_ptr cinfo, jpeg_component_info *compptr,
+ JCOEFPTR coef_block, JSAMPARRAY output_buf,
+ JDIMENSION output_col)
+{
+}
+
+GLOBAL(int)
+jsimd_can_huff_encode_one_block (void)
+{
+ return 0;
+}
+
+GLOBAL(JOCTET*)
+jsimd_huff_encode_one_block (void *state, JOCTET *buffer, JCOEFPTR block,
+ int last_dc_val, c_derived_tbl *dctbl,
+ c_derived_tbl *actbl)
+{
+ return NULL;
+}
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ * Authors: ZhuChen <zhuchen@loongson.cn>
+ * CaiWanwei <caiwanwei@loongson.cn>
+ * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+#define JPEG_INTERNALS
+#include "../../jinclude.h"
+#include "../../jpeglib.h"
+#include "../../jdct.h"
+#include "loongson-mmintrin.h"
+
+
+/* Common code */
+
+#define SIZEOF_MMWORD 8
+#define BYTE_BIT 8
+#define WORD_BIT 16
+#define SCALEBITS 16
+
+#define _uint64_set_pi8(a, b, c, d, e, f, g, h) \
+ (((uint64_t)(uint8_t)a << 56) | \
+ ((uint64_t)(uint8_t)b << 48) | \
+ ((uint64_t)(uint8_t)c << 40) | \
+ ((uint64_t)(uint8_t)d << 32) | \
+ ((uint64_t)(uint8_t)e << 24) | \
+ ((uint64_t)(uint8_t)f << 16) | \
+ ((uint64_t)(uint8_t)g << 8) | \
+ ((uint64_t)(uint8_t)h))
+#define _uint64_set_pi16(a, b, c, d) (((uint64_t)(uint16_t)a << 48) | \
+ ((uint64_t)(uint16_t)b << 32) | \
+ ((uint64_t)(uint16_t)c << 16) | \
+ ((uint64_t)(uint16_t)d))
+#define _uint64_set_pi32(a, b) (((uint64_t)(uint32_t)a << 32) | \
+ ((uint64_t)(uint32_t)b))
+
+#define get_const_value(index) (*(__m64 *)&const_value[index])
--- /dev/null
+/*
+ * Loongson MMI optimizations for libjpeg-turbo
+ *
+ * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
+ * All Rights Reserved.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty. In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ * claim that you wrote the original software. If you use this software
+ * in a product, an acknowledgment in the product documentation would be
+ * appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ * misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+#ifndef __LOONGSON_MMINTRIN_H__
+#define __LOONGSON_MMINTRIN_H__
+
+#include <stdint.h>
+
+
+#define FUNCTION_ATTRIBS \
+ __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+
+
+/* Vectors are stored in 64-bit floating-point registers. */
+typedef double __m64;
+
+/* Having a 32-bit datatype allows us to use 32-bit loads in places like
+ load8888. */
+typedef float __m32;
+
+
+/********** Set Operations **********/
+
+extern __inline __m64
+_mm_setzero_si64 (void)
+{
+ return 0.0;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_set_pi8 (uint8_t __b7, uint8_t __b6, uint8_t __b5, uint8_t __b4,
+ uint8_t __b3, uint8_t __b2, uint8_t __b1, uint8_t __b0)
+{
+ __m64 ret;
+ uint32_t lo = ((uint32_t)__b6 << 24) |
+ ((uint32_t)__b4 << 16) |
+ ((uint32_t)__b2 << 8) |
+ (uint32_t)__b0;
+ uint32_t hi = ((uint32_t)__b7 << 24) |
+ ((uint32_t)__b5 << 16) |
+ ((uint32_t)__b3 << 8) |
+ (uint32_t)__b1;
+
+ asm("mtc1 %1, %0\n\t"
+ "mtc1 %2, $f0\n\t"
+ "punpcklbh %0, %0, $f0\n\t"
+ : "=f" (ret)
+ : "r" (lo), "r" (hi)
+ : "$f0"
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_set_pi16 (uint16_t __h3, uint16_t __h2, uint16_t __h1, uint16_t __h0)
+{
+ __m64 ret;
+ uint32_t lo = ((uint32_t)__h2 << 16) | (uint32_t)__h0;
+ uint32_t hi = ((uint32_t)__h3 << 16) | (uint32_t)__h1;
+
+ asm("mtc1 %1, %0\n\t"
+ "mtc1 %2, $f0\n\t"
+ "punpcklhw %0, %0, $f0\n\t"
+ : "=f" (ret)
+ : "r" (lo), "r" (hi)
+ : "$f0"
+ );
+
+ return ret;
+}
+
+#define _MM_SHUFFLE(fp3, fp2, fp1, fp0) \
+ (((fp3) << 6) | ((fp2) << 4) | ((fp1) << 2) | (fp0))
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_set_pi32 (uint32_t __i1, uint32_t __i0)
+{
+ if (__builtin_constant_p(__i1) &&
+ __builtin_constant_p(__i0))
+ {
+ uint64_t val = ((uint64_t)__i1 << 32) |
+ ((uint64_t)__i0 << 0);
+
+ return *(__m64 *)&val;
+ }
+ else if (__i1 == __i0)
+ {
+ uint64_t imm = _MM_SHUFFLE(1, 0, 1, 0);
+ __m64 ret;
+
+ asm("pshufh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (*(__m32 *)&__i1), "f" (*(__m64 *)&imm)
+ );
+
+ return ret;
+ }
+ else
+ {
+ uint64_t val = ((uint64_t)__i1 << 32) |
+ ((uint64_t)__i0 << 0);
+
+ return *(__m64 *)&val;
+ }
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_set1_pi8 (uint8_t __b0)
+{
+ __m64 ret;
+
+ asm("sll $8, %1, 8\n\t"
+ "or %1, %1, $8\n\t"
+ "mtc1 %1, %0\n\t"
+ "mtc1 $0, $f0\n\t"
+ "pshufh %0, %0, $f0\n\t"
+ : "=f" (ret)
+ : "r" (__b0)
+ : "$8", "$f0"
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_set1_pi16 (uint16_t __h0)
+{
+ __m64 ret;
+
+ asm("mtc1 %1, %0\n\t"
+ "mtc1 $0, $f0\n\t"
+ "pshufh %0, %0, $f0\n\t"
+ : "=f" (ret)
+ : "r" (__h0)
+ : "$8", "$f0"
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_set1_pi32 (unsigned __i0)
+{
+ return _mm_set_pi32(__i0, __i0);
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_setr_pi8 (uint8_t __h0, uint8_t __h1, uint8_t __h2, uint8_t __h3,
+ uint8_t __h4, uint8_t __h5, uint8_t __h6, uint8_t __h7)
+{
+ return _mm_set_pi8(__h7, __h6, __h5, __h4,
+ __h3, __h2, __h1, __h0);
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_setr_pi16 (uint16_t __w0, uint16_t __w1, uint16_t __w2, uint16_t __w3)
+{
+ return _mm_set_pi16(__w3, __w2, __w1, __w0);
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_setr_pi32 (uint32_t __i0, uint32_t __i1)
+{
+ return _mm_set_pi32(__i1, __i0);
+}
+
+
+/********** Arithmetic Operations **********/
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_add_pi8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("paddb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_add_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("paddh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_add_pi32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("paddw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_add_si64 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("paddd %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_adds_pi8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("paddsb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_adds_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("paddsh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_adds_pu8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("paddusb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_adds_pu16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("paddush %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_avg_pu8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pavgb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_avg_pu16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pavgh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_madd_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pmaddhw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_max_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pmaxsh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_max_pu8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pmaxub %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_min_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pminsh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_min_pu8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pminub %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline int FUNCTION_ATTRIBS
+_mm_movemask_pi8 (__m64 __m1)
+{
+ int ret;
+
+ asm("pmovmskb %0, %1\n\t"
+ : "=r" (ret)
+ : "y" (__m1)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_mulhi_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pmulhh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_mulhi_pu16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pmulhuh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_mullo_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pmullh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_mul_pu32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pmuluw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_sad_pu8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("psadbh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_asub_pu8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pasubub %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_biadd_pu8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("biadd %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_sub_pi8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("psubb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_sub_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("psubh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_sub_pi32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("psubw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_sub_si64 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("psubd %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_subs_pi8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("psubsb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_subs_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("psubsh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_subs_pu8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("psubusb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_subs_pu16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("psubush %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+
+/********** Logical Operations **********/
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_and_si64 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("and %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_andnot_si64 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("andn %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_or_si32 (__m32 __m1, __m32 __m2)
+{
+ __m32 ret;
+
+ asm("or %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_or_si64 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("or %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_xor_si64 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("xor %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+
+/********** Shift Operations **********/
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_slli_pi16 (__m64 __m, int64_t __count)
+{
+ __m64 ret;
+
+ asm("psllh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__count)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_slli_pi32 (__m64 __m, int64_t __count)
+{
+ __m64 ret;
+
+ asm("psllw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__count)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_slli_si64 (__m64 __m, int64_t __count)
+{
+ __m64 ret;
+
+ asm("dsll %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__count)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_srli_pi16 (__m64 __m, int64_t __count)
+{
+ __m64 ret;
+
+ asm("psrlh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__count)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_srli_pi32 (__m64 __m, int64_t __count)
+{
+ __m64 ret;
+
+ asm("psrlw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__count)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_srli_si64 (__m64 __m, int64_t __count)
+{
+ __m64 ret;
+
+ asm("dsrl %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__count)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_srai_pi16 (__m64 __m, int64_t __count)
+{
+ __m64 ret;
+
+ asm("psrah %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__count)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_srai_pi32 (__m64 __m, int64_t __count)
+{
+ __m64 ret;
+
+ asm("psraw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__count)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_srai_si64 (__m64 __m, int64_t __count)
+{
+ __m64 ret;
+
+ asm("dsra %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__count)
+ );
+
+ return ret;
+}
+
+
+/********** Conversion Intrinsics **********/
+
+extern __inline __m64 FUNCTION_ATTRIBS
+to_m64 (uint64_t x)
+{
+ return *(__m64 *)&x;
+}
+
+extern __inline uint64_t FUNCTION_ATTRIBS
+to_uint64 (__m64 x)
+{
+ return *(uint64_t *)&x;
+}
+
+
+/********** Comparison Intrinsics **********/
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_cmpeq_pi8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pcmpeqb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_cmpeq_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pcmpeqh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_cmpeq_pi32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pcmpeqw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_cmpgt_pi8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pcmpgtb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_cmpgt_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pcmpgth %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_cmpgt_pi32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pcmpgtw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_cmplt_pi8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pcmpltb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_cmplt_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pcmplth %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_cmplt_pi32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("pcmpltw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+
+/********** Miscellaneous Operations **********/
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_packs_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("packsshb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_packs_pi32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("packsswh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_packs_pi32_f (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("packsswh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_packs_pu16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("packushb %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_extract_pi16 (__m64 __m, int64_t __pos)
+{
+ __m64 ret;
+
+ asm("pextrh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__pos)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_insert_pi16 (__m64 __m1, __m64 __m2, int64_t __pos)
+{
+ __m64 ret;
+
+ switch (__pos)
+ {
+ case 0:
+
+ asm("pinsrh_0 %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2), "i" (__pos)
+ );
+
+ break;
+
+ case 1:
+
+ asm("pinsrh_1 %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2), "i" (__pos)
+ );
+
+ break;
+ case 2:
+
+ asm("pinsrh_2 %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2), "i" (__pos)
+ );
+
+ break;
+
+ case 3:
+
+ asm("pinsrh_3 %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2), "i" (__pos)
+ );
+
+ break;
+ }
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_shuffle_pi16 (__m64 __m, int64_t __n)
+{
+ __m64 ret;
+
+ asm("pshufh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m), "f" (*(__m64 *)&__n)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpackhi_pi8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpckhbh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpackhi_pi8_f (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpckhbh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpackhi_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpckhhw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpackhi_pi16_f (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpckhhw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpackhi_pi32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpckhwd %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpacklo_pi8 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpcklbh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+/* Since punpcklbh cares about the high 32-bits, we use the __m64 datatype,
+ which preserves the data. */
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpacklo_pi8_f64 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpcklbh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+/* Since punpcklbh doesn't care about the high 32-bits, we use the __m32,
+ datatype, which allows load8888 to use 32-bit loads. */
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpacklo_pi8_f (__m32 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpcklbh %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpacklo_pi16 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpcklhw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpacklo_pi16_f (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpcklhw %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpacklo_pi32 (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpcklwd %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_unpacklo_pi32_f (__m64 __m1, __m64 __m2)
+{
+ __m64 ret;
+
+ asm("punpcklwd %0, %1, %2\n\t"
+ : "=f" (ret)
+ : "f" (__m1), "f" (__m2)
+ );
+
+ return ret;
+}
+
+extern __inline void FUNCTION_ATTRIBS
+_mm_store_pi32 (__m32 *dest, __m64 src)
+{
+ src = _mm_packs_pu16(src, _mm_setzero_si64());
+
+ asm("swc1 %1, %0\n\t"
+ : "=m" (*dest)
+ : "f" (src)
+ : "memory"
+ );
+}
+
+extern __inline void FUNCTION_ATTRIBS
+_mm_store_si64 (__m64 *dest, __m64 src)
+{
+ asm("gssdlc1 %1, 7+%0\n\t"
+ "gssdrc1 %1, %0\n\t"
+ : "=m" (*dest)
+ : "f" (src)
+ : "memory"
+ );
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_load_si32 (const __m32 *src)
+{
+ __m32 ret;
+
+ asm("lwc1 %0, %1\n\t"
+ : "=f" (ret)
+ : "m" (*src)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_load_si64 (const __m64 *src)
+{
+ __m64 ret;
+
+ asm("ldc1 %0, %1\n\t"
+ : "=f" (ret)
+ : "m" (*src)
+ );
+
+ return ret;
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_loadlo_pi8 (const uint32_t *src)
+{
+ return _mm_unpacklo_pi8_f(*(__m32 *)src, _mm_setzero_si64());
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_loadlo_pi8_f (__m64 src)
+{
+ return _mm_unpacklo_pi8_f64(src, _mm_setzero_si64());
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_loadhi_pi8_f (__m64 src)
+{
+ return _mm_unpackhi_pi8_f(src, _mm_setzero_si64());
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_loadlo_pi16 (__m64 src)
+{
+ return _mm_unpacklo_pi16(src, _mm_setzero_si64());
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_loadlo_pi16_f (__m64 src)
+{
+ return _mm_unpacklo_pi16_f(_mm_setzero_si64(), src);
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_loadhi_pi16 (__m64 src)
+{
+ return _mm_unpackhi_pi16(src, _mm_setzero_si64());
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_loadhi_pi16_f (__m64 src)
+{
+ return _mm_unpackhi_pi16_f(_mm_setzero_si64(), src);
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_expand_alpha (__m64 pixel)
+{
+ return _mm_shuffle_pi16(pixel, _MM_SHUFFLE(3, 3, 3, 3));
+}
+
+extern __inline __m64 FUNCTION_ATTRIBS
+_mm_expand_alpha_rev (__m64 pixel)
+{
+ return _mm_shuffle_pi16(pixel, _MM_SHUFFLE(0, 0, 0, 0));
+}
+
+#endif /* __LOONGSON_MMINTRIN_H__ */
projects / libjpeg-turbo / commitdiff