diff --git a/include/share/compat.h b/include/share/compat.h index a17cc63..65387b0 100644 --- a/include/share/compat.h +++ b/include/share/compat.h @@ -199,12 +199,34 @@ int flac_snprintf(char *str, size_t size, const char *fmt, ...); }; #endif -/* SSSE3, SSE4 support: MSVS 2008, GCC 4.3 -- currently disabled, Intel Compiler 10.0 */ -#if ( defined _MSC_VER && _MSC_VER >= 1500 ) \ - || ( 0 && defined __GNUC__ && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3)) ) \ - || ( defined __INTEL_COMPILER && __INTEL_COMPILER >= 1000 ) -#define FLAC__SSSE3_SUPPORTED 1 -#define FLAC__SSE4_SUPPORTED 1 +/* SSSE3, SSE4 support: MSVS 2008, Intel Compiler 10.0, GCC 4.3 */ +#if defined __INTEL_COMPILER + #define FLAC__SSE_SUPPORTED 1 + #define FLAC__SSE2_SUPPORTED 1 + #if (__INTEL_COMPILER >= 1000) + #define FLAC__SSSE3_SUPPORTED 1 + #define FLAC__SSE4_1_SUPPORTED 1 + #endif +#elif defined _MSC_VER + #define FLAC__SSE_SUPPORTED 1 + #define FLAC__SSE2_SUPPORTED 1 + #if (_MSC_VER >= 1500) + #define FLAC__SSSE3_SUPPORTED 1 + #define FLAC__SSE4_1_SUPPORTED 1 + #endif +#elif defined __GNUC__ + #ifdef __SSE__ + #define FLAC__SSE_SUPPORTED 1 + #endif + #ifdef __SSE2__ + #define FLAC__SSE2_SUPPORTED 1 + #endif + #ifdef __SSSE3__ + #define FLAC__SSSE3_SUPPORTED 1 + #endif + #if defined __SSE4_1__ + #define FLAC__SSE4_1_SUPPORTED 1 + #endif #endif #endif /* FLAC__SHARE__COMPAT_H */ diff --git a/src/libFLAC/include/private/lpc.h b/src/libFLAC/include/private/lpc.h index 27760b4..cf2061e 100644 --- a/src/libFLAC/include/private/lpc.h +++ b/src/libFLAC/include/private/lpc.h @@ -37,6 +37,7 @@ #include #endif +#include "share/compat.h" #include "private/float.h" #include "FLAC/format.h" @@ -80,10 +81,12 @@ void FLAC__lpc_compute_autocorrelation_asm_ia32_3dnow(const FLAC__real data[], u # endif # endif # if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && defined FLAC__HAS_X86INTRIN +# ifdef FLAC__SSE_SUPPORTED void FLAC__lpc_compute_autocorrelation_intrin_sse_lag_4(const FLAC__real data[], unsigned data_len, unsigned lag, FLAC__real autoc[]); void FLAC__lpc_compute_autocorrelation_intrin_sse_lag_8(const FLAC__real data[], unsigned data_len, unsigned lag, FLAC__real autoc[]); void FLAC__lpc_compute_autocorrelation_intrin_sse_lag_12(const FLAC__real data[], unsigned data_len, unsigned lag, FLAC__real autoc[]); void FLAC__lpc_compute_autocorrelation_intrin_sse_lag_16(const FLAC__real data[], unsigned data_len, unsigned lag, FLAC__real autoc[]); +# endif # endif #endif @@ -156,9 +159,11 @@ void FLAC__lpc_compute_residual_from_qlp_coefficients_wide_asm_ia32(const FLAC__ # endif # endif # if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && defined FLAC__HAS_X86INTRIN +# ifdef FLAC__SSE2_SUPPORTED void FLAC__lpc_compute_residual_from_qlp_coefficients_16_intrin_sse2(const FLAC__int32 *data, unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 residual[]); void FLAC__lpc_compute_residual_from_qlp_coefficients_intrin_sse2(const FLAC__int32 *data, unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 residual[]); -# ifdef FLAC__SSE4_SUPPORTED +# endif +# ifdef FLAC__SSE4_1_SUPPORTED void FLAC__lpc_compute_residual_from_qlp_coefficients_wide_intrin_sse41(const FLAC__int32 *data, unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 residual[]); # endif # endif @@ -195,7 +200,7 @@ void FLAC__lpc_restore_signal_asm_ppc_altivec_16(const FLAC__int32 residual[], u void FLAC__lpc_restore_signal_asm_ppc_altivec_16_order8(const FLAC__int32 residual[], unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 data[]); # endif /* FLAC__CPU_IA32 || FLAC__CPU_PPC */ # if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && defined FLAC__HAS_X86INTRIN -# ifdef FLAC__SSE4_SUPPORTED +# ifdef FLAC__SSE4_1_SUPPORTED void FLAC__lpc_restore_signal_wide_intrin_sse41(const FLAC__int32 residual[], unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 data[]); # endif # endif diff --git a/src/libFLAC/include/private/stream_encoder.h b/src/libFLAC/include/private/stream_encoder.h index ee7d978..aeedccf 100644 --- a/src/libFLAC/include/private/stream_encoder.h +++ b/src/libFLAC/include/private/stream_encoder.h @@ -41,8 +41,10 @@ #include "share/compat.h" #include "FLAC/format.h" +#ifdef FLAC__SSE2_SUPPORTED extern void FLAC__precompute_partition_info_sums_intrin_sse2(const FLAC__int32 residual[], FLAC__uint64 abs_residual_partition_sums[], unsigned residual_samples, unsigned predictor_order, unsigned min_partition_order, unsigned max_partition_order, unsigned bps); +#endif #ifdef FLAC__SSSE3_SUPPORTED extern void FLAC__precompute_partition_info_sums_intrin_ssse3(const FLAC__int32 residual[], FLAC__uint64 abs_residual_partition_sums[], diff --git a/src/libFLAC/lpc_intrin_sse.c b/src/libFLAC/lpc_intrin_sse.c index e8f9f57..64d33bd 100644 --- a/src/libFLAC/lpc_intrin_sse.c +++ b/src/libFLAC/lpc_intrin_sse.c @@ -34,9 +34,12 @@ # include #endif +#include "share/compat.h" + #ifndef FLAC__INTEGER_ONLY_LIBRARY #ifndef FLAC__NO_ASM #if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && defined FLAC__HAS_X86INTRIN +#ifdef FLAC__SSE_SUPPORTED #include "FLAC/assert.h" #include "FLAC/format.h" @@ -241,6 +244,7 @@ void FLAC__lpc_compute_autocorrelation_intrin_sse_lag_16(const FLAC__real data[] _mm_storeu_ps(autoc+12,xmm9); } +#endif /* FLAC__SSE_SUPPORTED */ #endif /* (FLAC__CPU_IA32 || FLAC__CPU_X86_64) && FLAC__HAS_X86INTRIN */ #endif /* FLAC__NO_ASM */ #endif /* FLAC__INTEGER_ONLY_LIBRARY */ diff --git a/src/libFLAC/lpc_intrin_sse2.c b/src/libFLAC/lpc_intrin_sse2.c index 9311151..c429225 100644 --- a/src/libFLAC/lpc_intrin_sse2.c +++ b/src/libFLAC/lpc_intrin_sse2.c @@ -34,9 +34,12 @@ # include #endif +#include "share/compat.h" + #ifndef FLAC__INTEGER_ONLY_LIBRARY #ifndef FLAC__NO_ASM #if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && defined FLAC__HAS_X86INTRIN +#ifdef FLAC__SSE2_SUPPORTED #include "FLAC/assert.h" #include "FLAC/format.h" @@ -1313,6 +1316,7 @@ void FLAC__lpc_compute_residual_from_qlp_coefficients_intrin_sse2(const FLAC__in } } +#endif /* FLAC__SSE2_SUPPORTED */ #endif /* (FLAC__CPU_IA32 || FLAC__CPU_X86_64) && FLAC__HAS_X86INTRIN */ #endif /* FLAC__NO_ASM */ #endif /* FLAC__INTEGER_ONLY_LIBRARY */ diff --git a/src/libFLAC/lpc_intrin_sse41.c b/src/libFLAC/lpc_intrin_sse41.c index ea8eb37..2c6985c 100644 --- a/src/libFLAC/lpc_intrin_sse41.c +++ b/src/libFLAC/lpc_intrin_sse41.c @@ -39,7 +39,7 @@ #ifndef FLAC__INTEGER_ONLY_LIBRARY #ifndef FLAC__NO_ASM #if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && defined FLAC__HAS_X86INTRIN -#ifdef FLAC__SSE4_SUPPORTED +#ifdef FLAC__SSE4_1_SUPPORTED #include "FLAC/assert.h" #include "FLAC/format.h" @@ -1120,7 +1120,7 @@ void FLAC__lpc_restore_signal_wide_intrin_sse41(const FLAC__int32 residual[], un } } -#endif /* FLAC__SSE4_SUPPORTED */ +#endif /* FLAC__SSE4_1_SUPPORTED */ #endif /* (FLAC__CPU_IA32 || FLAC__CPU_X86_64) && FLAC__HAS_X86INTRIN */ #endif /* FLAC__NO_ASM */ #endif /* FLAC__INTEGER_ONLY_LIBRARY */ diff --git a/src/libFLAC/stream_decoder.c b/src/libFLAC/stream_decoder.c index d8cd714..633dcdc 100644 --- a/src/libFLAC/stream_decoder.c +++ b/src/libFLAC/stream_decoder.c @@ -417,7 +417,7 @@ static FLAC__StreamDecoderInitStatus init_stream_internal_( } #endif #ifdef FLAC__HAS_X86INTRIN -# if defined FLAC__SSE4_SUPPORTED && 0 /* now we have FLAC__lpc_restore_signal_wide_asm_ia32() which is slightly faster */ +# if defined FLAC__SSE4_1_SUPPORTED && 0 /* now we have FLAC__lpc_restore_signal_wide_asm_ia32() which is slightly faster */ if(decoder->private_->cpuinfo.ia32.sse41) decoder->private_->local_lpc_restore_signal_64bit = FLAC__lpc_restore_signal_wide_intrin_sse41; # endif diff --git a/src/libFLAC/stream_encoder.c b/src/libFLAC/stream_encoder.c index cbf2815..68ece5c 100644 --- a/src/libFLAC/stream_encoder.c +++ b/src/libFLAC/stream_encoder.c @@ -920,11 +920,13 @@ static FLAC__StreamEncoderInitStatus init_stream_internal_( encoder->private_->local_fixed_compute_best_predictor = FLAC__fixed_compute_best_predictor_asm_ia32_mmx_cmov; # endif /* FLAC__HAS_NASM */ # ifdef FLAC__HAS_X86INTRIN +# ifdef FLAC__SSE2_SUPPORTED if(encoder->private_->cpuinfo.ia32.sse2) { encoder->private_->local_lpc_compute_residual_from_qlp_coefficients = FLAC__lpc_compute_residual_from_qlp_coefficients_intrin_sse2; encoder->private_->local_lpc_compute_residual_from_qlp_coefficients_16bit = FLAC__lpc_compute_residual_from_qlp_coefficients_16_intrin_sse2; } -# ifdef FLAC__SSE4_SUPPORTED +# endif +# ifdef FLAC__SSE4_1_SUPPORTED if(encoder->private_->cpuinfo.ia32.sse41) encoder->private_->local_lpc_compute_residual_from_qlp_coefficients_64bit = FLAC__lpc_compute_residual_from_qlp_coefficients_wide_intrin_sse41; # endif @@ -956,8 +958,10 @@ static FLAC__StreamEncoderInitStatus init_stream_internal_( encoder->private_->local_precompute_partition_info_sums = FLAC__precompute_partition_info_sums_intrin_ssse3; else # endif +# ifdef FLAC__SSE2_SUPPORTED if(encoder->private_->cpuinfo.ia32.sse2) encoder->private_->local_precompute_partition_info_sums = FLAC__precompute_partition_info_sums_intrin_sse2; +# endif # elif defined FLAC__CPU_X86_64 # ifdef FLAC__SSSE3_SUPPORTED if(encoder->private_->cpuinfo.x86_64.ssse3) diff --git a/src/libFLAC/stream_encoder_intrin_sse2.c b/src/libFLAC/stream_encoder_intrin_sse2.c index ac2e69f..403c107 100644 --- a/src/libFLAC/stream_encoder_intrin_sse2.c +++ b/src/libFLAC/stream_encoder_intrin_sse2.c @@ -34,8 +34,11 @@ # include #endif +#include "share/compat.h" + #ifndef FLAC__NO_ASM #if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && defined FLAC__HAS_X86INTRIN +#ifdef FLAC__SSE2_SUPPORTED #include /* for abs() */ #include /* SSE2 */ @@ -157,5 +160,6 @@ void FLAC__precompute_partition_info_sums_intrin_sse2(const FLAC__int32 residual } } +#endif /* FLAC__SSE2_SUPPORTED */ #endif /* (FLAC__CPU_IA32 || FLAC__CPU_X86_64) && FLAC__HAS_X86INTRIN */ #endif /* FLAC__NO_ASM */