Optimize WelsDctT4_c and WelsDctFourT4_c with LASX (#3483)
diff --git a/codec/encoder/core/inc/encode_mb_aux.h b/codec/encoder/core/inc/encode_mb_aux.h
index 1dd5252..def5d03 100644
--- a/codec/encoder/core/inc/encode_mb_aux.h
+++ b/codec/encoder/core/inc/encode_mb_aux.h
@@ -179,6 +179,14 @@
void WelsQuantFour4x4Max_lsx (int16_t* pDct, const int16_t* pFF, const int16_t* pMF, int16_t* pMax);
#endif//HAVE_LSX
+#ifdef HAVE_LASX
+/****************************************************************************
+ * * DCT functions
+ * ****************************************************************************/
+void WelsDctT4_lasx (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2);
+void WelsDctFourT4_lasx (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2);
+#endif
+
#if defined(__cplusplus)
}
#endif//__cplusplus
diff --git a/codec/encoder/core/loongarch/dct_lasx.c b/codec/encoder/core/loongarch/dct_lasx.c
new file mode 100644
index 0000000..49bb888
--- /dev/null
+++ b/codec/encoder/core/loongarch/dct_lasx.c
@@ -0,0 +1,224 @@
+/*!
+ * \copy
+ * Copyright (c) 2022, Cisco Systems
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * * Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in
+ * the documentation and/or other materials provided with the
+ * distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+ * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+ * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ *
+ *
+ * \file dct_lasx.c
+ *
+ * \brief Loongson optimization
+ *
+ * \date 15/02/2022 Created
+ *
+ *************************************************************************************
+ */
+#include <stdint.h>
+#include "loongson_intrinsics.h"
+
+#define CALC_TEMPS_AND_PDCT(src0, src1, src2, src3, \
+ dst0, dst1, dst2, dst3) \
+do { \
+ __m256i tms0, tms1, tms2, tms3; \
+ tms0 = __lasx_xvadd_h(src0, src3); \
+ tms1 = __lasx_xvadd_h(src1, src2); \
+ tms2 = __lasx_xvsub_h(src1, src2); \
+ tms3 = __lasx_xvsub_h(src0, src3); \
+ dst0 = __lasx_xvadd_h(tms0, tms1); \
+ dst1 = __lasx_xvslli_h(tms3, 1); \
+ dst1 = __lasx_xvadd_h(dst1, tms2); \
+ dst2 = __lasx_xvsub_h(tms0, tms1); \
+ dst3 = __lasx_xvslli_h(tms2, 1); \
+ dst3 = __lasx_xvsub_h(tms3, dst3); \
+}while(0)
+
+/****************************************************************************
+ * DCT functions
+ ****************************************************************************/
+void WelsDctT4_lasx (int16_t* pDct, uint8_t* pPixel1,
+ int32_t iStride1, uint8_t* pPixel2,
+ int32_t iStride2) {
+ int32_t iStride0 = 0;
+ int32_t iStride1_x2 = iStride1 << 1;
+ int32_t iStride1_x3 = iStride1_x2 + iStride1;
+ int32_t iStride2_x2 = iStride2 << 1;
+ int32_t iStride2_x3 = iStride2_x2 + iStride2;
+
+ __m256i src0, src1, src2, src3, src4, src5, src6, src7;
+ __m256i dst0, dst1, dst2, dst3;
+
+ DUP4_ARG2(__lasx_xvldx,
+ pPixel1, iStride0,
+ pPixel1, iStride1,
+ pPixel1, iStride1_x2,
+ pPixel1, iStride1_x3,
+ src0, src1, src2, src3);
+ DUP4_ARG2(__lasx_xvldx,
+ pPixel2, iStride0,
+ pPixel2, iStride2,
+ pPixel2, iStride2_x2,
+ pPixel2, iStride2_x3,
+ src4, src5, src6, src7);
+ DUP4_ARG2(__lasx_xvilvl_b,
+ src0, src4,
+ src1, src5,
+ src2, src6,
+ src3, src7,
+ src0, src1, src2, src3);
+ DUP4_ARG2(__lasx_xvhsubw_hu_bu,
+ src0, src0,
+ src1, src1,
+ src2, src2,
+ src3, src3,
+ src0, src1, src2, src3);
+ LASX_TRANSPOSE4x4_H(src0, src1, src2, src3,
+ src0, src1, src2, src3);
+ CALC_TEMPS_AND_PDCT(src0, src1, src2, src3,
+ dst0, dst1, dst2, dst3);
+ LASX_TRANSPOSE4x4_H(dst0, dst1, dst2, dst3,
+ src0, src1, src2, src3);
+ CALC_TEMPS_AND_PDCT(src0, src1, src2, src3,
+ dst0, dst1, dst2, dst3);
+ dst0 = __lasx_xvpackev_d(dst1, dst0);
+ dst2 = __lasx_xvpackev_d(dst3, dst2);
+ dst0 = __lasx_xvpermi_q(dst2, dst0, 0x20);
+ __lasx_xvst(dst0, pDct, 0);
+}
+
+void WelsDctFourT4_lasx (int16_t* pDct, uint8_t* pPixel1,
+ int32_t iStride1, uint8_t* pPixel2,
+ int32_t iStride2) {
+ int32_t stride_1 = iStride1 << 2;
+ int32_t stride_2 = iStride2 << 2;
+ int32_t iStride0 = 0;
+ int32_t iStride1_x2 = iStride1 << 1;
+ int32_t iStride1_x3 = iStride1_x2 + iStride1;
+ int32_t iStride2_x2 = iStride2 << 1;
+ int32_t iStride2_x3 = iStride2_x2 + iStride2;
+ uint8_t *psrc10 = pPixel1, *psrc11 = pPixel2;
+ uint8_t *psrc20 = pPixel1 + stride_1, *psrc21 = pPixel2 + stride_2;
+
+ __m256i src0, src1, src2, src3, src4, src5, src6, src7,
+ src8, src9, src10, src11, src12, src13, src14 ,src15;
+ __m256i tmp0, tmp1, tmp2, tmp3, dst0, dst1, dst2, dst3, dst4,
+ dst5, dst6, dst7;
+
+ DUP4_ARG2(__lasx_xvldx,
+ psrc10, iStride0,
+ psrc10, iStride1,
+ psrc10, iStride1_x2,
+ psrc10, iStride1_x3,
+ src0, src1, src2, src3);
+ DUP4_ARG2(__lasx_xvldx,
+ psrc11, iStride0,
+ psrc11, iStride2,
+ psrc11, iStride2_x2,
+ psrc11, iStride2_x3,
+ src4, src5, src6, src7);
+ DUP4_ARG2(__lasx_xvldx,
+ psrc20, iStride0,
+ psrc20, iStride1,
+ psrc20, iStride1_x2,
+ psrc20, iStride1_x3,
+ src8, src9, src10, src11);
+ DUP4_ARG2(__lasx_xvldx,
+ psrc21, iStride0,
+ psrc21, iStride2,
+ psrc21, iStride2_x2,
+ psrc21, iStride2_x3,
+ src12, src13, src14, src15);
+ DUP4_ARG2(__lasx_xvilvl_b,
+ src0, src4,
+ src1, src5,
+ src2, src6,
+ src3, src7,
+ src0, src1, src2, src3);
+ DUP4_ARG2(__lasx_xvilvl_b,
+ src8, src12,
+ src9, src13,
+ src10, src14,
+ src11, src15,
+ src8, src9, src10, src11);
+ DUP4_ARG2(__lasx_xvhsubw_hu_bu,
+ src0, src0,
+ src1, src1,
+ src2, src2,
+ src3, src3,
+ src0, src1, src2, src3);
+ DUP4_ARG2(__lasx_xvhsubw_hu_bu,
+ src8, src8,
+ src9, src9,
+ src10, src10,
+ src11, src11,
+ src8, src9, src10 ,src11);
+ LASX_TRANSPOSE8x8_H(src0, src1, src2, src3, src8, src9, src10, src11,
+ src0, src1, src2, src3, src8, src9, src10, src11);
+ DUP4_ARG3(__lasx_xvpermi_q,
+ src8, src0, 0x20,
+ src9, src1, 0x20,
+ src10,src2, 0x20,
+ src11,src3, 0x20,
+ src0, src1, src2, src3);
+ CALC_TEMPS_AND_PDCT(src0, src1, src2, src3,
+ dst0, dst1, dst2, dst3);
+ DUP4_ARG3(__lasx_xvpermi_q,
+ dst0, dst0, 0x31,
+ dst1, dst1, 0x31,
+ dst2, dst2, 0x31,
+ dst3, dst3, 0x31,
+ dst4, dst5, dst6, dst7);
+ LASX_TRANSPOSE8x8_H(dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7,
+ dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
+ DUP4_ARG3(__lasx_xvpermi_q,
+ dst4, dst0, 0x20,
+ dst5, dst1, 0x20,
+ dst6, dst2, 0x20,
+ dst7, dst3, 0x20,
+ dst0, dst1, dst2, dst3);
+ CALC_TEMPS_AND_PDCT(dst0, dst1, dst2, dst3,
+ dst0, dst1, dst2, dst3);
+ DUP2_ARG2(__lasx_xvpackev_d,
+ dst1, dst0,
+ dst3, dst2,
+ tmp0, tmp1);
+ DUP2_ARG2(__lasx_xvpackod_d,
+ dst1, dst0,
+ dst3, dst2,
+ tmp2, tmp3);
+ DUP2_ARG3(__lasx_xvpermi_q,
+ tmp1, tmp0, 0x20,
+ tmp3, tmp2, 0x20,
+ dst0, dst1);
+ DUP2_ARG3(__lasx_xvpermi_q,
+ tmp1, tmp0, 0x31,
+ tmp3, tmp2, 0x31,
+ dst2, dst3);
+ __lasx_xvst(dst0, pDct, 0);
+ __lasx_xvst(dst1, pDct, 32);
+ __lasx_xvst(dst2, pDct, 64);
+ __lasx_xvst(dst3, pDct, 96);
+}
diff --git a/codec/encoder/core/src/encode_mb_aux.cpp b/codec/encoder/core/src/encode_mb_aux.cpp
index 62d753b..e1bdb6d 100644
--- a/codec/encoder/core/src/encode_mb_aux.cpp
+++ b/codec/encoder/core/src/encode_mb_aux.cpp
@@ -632,5 +632,11 @@
pFuncList->pfQuantizationFour4x4Max = WelsQuantFour4x4Max_lsx;
}
#endif
+#if defined(HAVE_LASX)
+ if (uiCpuFlag & WELS_CPU_LASX) {
+ pFuncList->pfDctT4 = WelsDctT4_lasx;
+ pFuncList->pfDctFourT4 = WelsDctFourT4_lasx;
+}
+#endif
}
}
diff --git a/codec/encoder/targets.mk b/codec/encoder/targets.mk
index 014088f..012839a 100644
--- a/codec/encoder/targets.mk
+++ b/codec/encoder/targets.mk
@@ -109,6 +109,7 @@
ENCODER_OBJSLOONGARCH_LSX += $(ENCODER_ASM_LOONGARCH_LSX_SRCS:.c=.$(OBJ))
ENCODER_ASM_LOONGARCH_LASX_SRCS=\
+ $(ENCODER_SRCDIR)/core/loongarch/dct_lasx.c\
ENCODER_OBJSLOONGARCH_LASX += $(ENCODER_ASM_LOONGARCH_LASX_SRCS:.c=.$(OBJ))
ifeq ($(ASM_ARCH), loongarch)
diff --git a/test/encoder/EncUT_EncoderMbAux.cpp b/test/encoder/EncUT_EncoderMbAux.cpp
index 8580da0..418e775 100644
--- a/test/encoder/EncUT_EncoderMbAux.cpp
+++ b/test/encoder/EncUT_EncoderMbAux.cpp
@@ -292,6 +292,15 @@
FREE_MEMORY (iDctS);
}
#endif
+#ifdef HAVE_LASX
+TEST (EncodeMbAuxTest, WelsDctT4_lasx) {
+ TestDctT4 (WelsDctT4_lasx);
+}
+
+TEST (EncodeMbAuxTest, WelsDctFourT4_lasx) {
+ TestDctFourT4 (WelsDctFourT4_lasx);
+}
+#endif
void copy (uint8_t* pDst, int32_t iDStride, uint8_t* pSrc, int32_t iSStride, int32_t iWidth, int32_t iHeight) {
for (int i = 0; i < iHeight; i++)
