summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorShiyou Yin <yinshiyou-hf@loongson.cn>2021-04-12 23:37:05 +0800
committerMichael Niedermayer <michael@niedermayer.cc>2021-05-07 17:53:23 +0200
commitab04fedaaac34df4eeb71cff0e63aaad9e9ff711 (patch)
tree0ddbaeadc50a6dbeb77a73696dc224b8fb40166f
parent464d28c0708f23fb35271265f8a5c73fbb51db44 (diff)
downloadffmpeg-ab04fedaaac34df4eeb71cff0e63aaad9e9ff711.tar.gz
mips: Fix potential illegal instruction error.
MSA2 optimizations are attached to MSA macros in generic_macros_msa.h. It's difficult to do runtime check for them. Remove this part of code can make it more robust. H264 1080p decoding: 5.13x==>5.12x. Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
-rwxr-xr-xconfigure7
-rw-r--r--libavutil/mips/generic_macros_msa.h37
2 files changed, 1 insertions, 43 deletions
diff --git a/configure b/configure
index 9ece040c7b..2d7b1bb429 100755
--- a/configure
+++ b/configure
@@ -450,7 +450,6 @@ Optimization options (experts only):
--disable-mipsdsp disable MIPS DSP ASE R1 optimizations
--disable-mipsdspr2 disable MIPS DSP ASE R2 optimizations
--disable-msa disable MSA optimizations
- --disable-msa2 disable MSA2 optimizations
--disable-mipsfpu disable floating point MIPS optimizations
--disable-mmi disable Loongson SIMD optimizations
--disable-fast-unaligned consider unaligned accesses slow
@@ -2053,7 +2052,6 @@ ARCH_EXT_LIST_MIPS="
mipsdsp
mipsdspr2
msa
- msa2
"
ARCH_EXT_LIST_LOONGSON="
@@ -2592,7 +2590,6 @@ mipsdsp_deps="mips"
mipsdspr2_deps="mips"
mmi_deps_any="loongson2 loongson3"
msa_deps="mipsfpu"
-msa2_deps="msa"
cpunop_deps="i686"
x86_64_select="i686"
@@ -5932,9 +5929,8 @@ elif enabled mips; then
enabled mipsdsp && check_inline_asm_flags mipsdsp '"addu.qb $t0, $t1, $t2"' '-mdsp'
enabled mipsdspr2 && check_inline_asm_flags mipsdspr2 '"absq_s.qb $t0, $t1"' '-mdspr2'
- # MSA and MSA2 can be detected at runtime so we supply extra flags here
+ # MSA can be detected at runtime so we supply extra flags here
enabled mipsfpu && enabled msa && check_inline_asm msa '"addvi.b $w0, $w1, 1"' '-mmsa' && append MSAFLAGS '-mmsa'
- enabled msa && enabled msa2 && check_inline_asm msa2 '"nxbits.any.b $w0, $w0"' '-mmsa2' && append MSAFLAGS '-mmsa2'
# loongson2 have no switch cflag so we can only probe toolchain ability
enabled loongson2 && check_inline_asm loongson2 '"dmult.g $8, $9, $10"' && disable loongson3
@@ -7363,7 +7359,6 @@ if enabled mips; then
echo "MIPS DSP R1 enabled ${mipsdsp-no}"
echo "MIPS DSP R2 enabled ${mipsdspr2-no}"
echo "MIPS MSA enabled ${msa-no}"
- echo "MIPS MSA2 enabled ${msa2-no}"
echo "LOONGSON MMI enabled ${mmi-no}"
fi
if enabled ppc; then
diff --git a/libavutil/mips/generic_macros_msa.h b/libavutil/mips/generic_macros_msa.h
index bb25e9fd74..1486f7296e 100644
--- a/libavutil/mips/generic_macros_msa.h
+++ b/libavutil/mips/generic_macros_msa.h
@@ -25,10 +25,6 @@
#include <msa.h>
#include <config.h>
-#if HAVE_MSA2
-#include <msa2.h>
-#endif
-
#define ALIGNMENT 16
#define ALLOC_ALIGNED(align) __attribute__ ((aligned((align) << 1)))
@@ -1119,15 +1115,6 @@
unsigned absolute diff values, even-odd pairs are added
together to generate 8 halfword results.
*/
-#if HAVE_MSA2
-#define SAD_UB2_UH(in0, in1, ref0, ref1) \
-( { \
- v8u16 sad_m = { 0 }; \
- sad_m += __builtin_msa2_sad_adj2_u_w2x_b((v16u8) in0, (v16u8) ref0); \
- sad_m += __builtin_msa2_sad_adj2_u_w2x_b((v16u8) in1, (v16u8) ref1); \
- sad_m; \
-} )
-#else
#define SAD_UB2_UH(in0, in1, ref0, ref1) \
( { \
v16u8 diff0_m, diff1_m; \
@@ -1141,7 +1128,6 @@
\
sad_m; \
} )
-#endif // #if HAVE_MSA2
/* Description : Insert specified word elements from input vectors to 1
destination vector
@@ -2183,12 +2169,6 @@
extracted and interleaved with same vector 'in0' to generate
4 word elements keeping sign intact
*/
-#if HAVE_MSA2
-#define UNPCK_R_SH_SW(in, out) \
-{ \
- out = (v4i32) __builtin_msa2_w2x_lo_s_h((v8i16) in); \
-}
-#else
#define UNPCK_R_SH_SW(in, out) \
{ \
v8i16 sign_m; \
@@ -2196,7 +2176,6 @@
sign_m = __msa_clti_s_h((v8i16) in, 0); \
out = (v4i32) __msa_ilvr_h(sign_m, (v8i16) in); \
}
-#endif // #if HAVE_MSA2
/* Description : Sign extend byte elements from input vector and return
halfword results in pair of vectors
@@ -2209,13 +2188,6 @@
Then interleaved left with same vector 'in0' to
generate 8 signed halfword elements in 'out1'
*/
-#if HAVE_MSA2
-#define UNPCK_SB_SH(in, out0, out1) \
-{ \
- out0 = (v4i32) __builtin_msa2_w2x_lo_s_b((v16i8) in); \
- out1 = (v4i32) __builtin_msa2_w2x_hi_s_b((v16i8) in); \
-}
-#else
#define UNPCK_SB_SH(in, out0, out1) \
{ \
v16i8 tmp_m; \
@@ -2223,7 +2195,6 @@
tmp_m = __msa_clti_s_b((v16i8) in, 0); \
ILVRL_B2_SH(tmp_m, in, out0, out1); \
}
-#endif // #if HAVE_MSA2
/* Description : Zero extend unsigned byte elements to halfword elements
Arguments : Inputs - in (1 input unsigned byte vector)
@@ -2250,13 +2221,6 @@
Then interleaved left with same vector 'in0' to
generate 4 signed word elements in 'out1'
*/
-#if HAVE_MSA2
-#define UNPCK_SH_SW(in, out0, out1) \
-{ \
- out0 = (v4i32) __builtin_msa2_w2x_lo_s_h((v8i16) in); \
- out1 = (v4i32) __builtin_msa2_w2x_hi_s_h((v8i16) in); \
-}
-#else
#define UNPCK_SH_SW(in, out0, out1) \
{ \
v8i16 tmp_m; \
@@ -2264,7 +2228,6 @@
tmp_m = __msa_clti_s_h((v8i16) in, 0); \
ILVRL_H2_SW(tmp_m, in, out0, out1); \
}
-#endif // #if HAVE_MSA2
/* Description : Swap two variables
Arguments : Inputs - in0, in1