aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--src/pcm/PcmFormat.cxx48
1 files changed, 20 insertions, 28 deletions
diff --git a/src/pcm/PcmFormat.cxx b/src/pcm/PcmFormat.cxx
index 7c6eff790..f2d71a6fb 100644
--- a/src/pcm/PcmFormat.cxx
+++ b/src/pcm/PcmFormat.cxx
@@ -25,17 +25,9 @@
#include "FloatConvert.hxx"
#include "ShiftConvert.hxx"
#include "util/ConstBuffer.hxx"
-#include "util/WritableBuffer.hxx"
#include "PcmDither.cxx" // including the .cxx file to get inlined templates
-template<typename T>
-static inline ConstBuffer<T>
-ToConst(WritableBuffer<T> b)
-{
- return { b.data, b.size };
-}
-
/**
* Wrapper for a class that converts one sample at a time into one
* that converts a buffer at a time.
@@ -88,7 +80,7 @@ struct FloatToInteger
: PerSampleConvert<FloatToIntegerSampleConvert<F, Traits>> {};
template<class C>
-static WritableBuffer<typename C::DstTraits::value_type>
+static ConstBuffer<typename C::DstTraits::value_type>
AllocateConvert(PcmBuffer &buffer, C convert,
ConstBuffer<typename C::SrcTraits::value_type> src)
{
@@ -98,7 +90,7 @@ AllocateConvert(PcmBuffer &buffer, C convert,
}
template<SampleFormat F, class Traits=SampleTraits<F>>
-static WritableBuffer<typename Traits::value_type>
+static ConstBuffer<typename Traits::value_type>
AllocateFromFloat(PcmBuffer &buffer, ConstBuffer<float> src)
{
return AllocateConvert(buffer, FloatToInteger<F, Traits>(), src);
@@ -107,27 +99,27 @@ AllocateFromFloat(PcmBuffer &buffer, ConstBuffer<float> src)
static ConstBuffer<int16_t>
pcm_allocate_8_to_16(PcmBuffer &buffer, ConstBuffer<int8_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert8To16(), src));
+ return AllocateConvert(buffer, Convert8To16(), src);
}
static ConstBuffer<int16_t>
pcm_allocate_24p32_to_16(PcmBuffer &buffer, PcmDither &dither,
ConstBuffer<int32_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert24To16(dither), src));
+ return AllocateConvert(buffer, Convert24To16(dither), src);
}
static ConstBuffer<int16_t>
pcm_allocate_32_to_16(PcmBuffer &buffer, PcmDither &dither,
ConstBuffer<int32_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert32To16(dither), src));
+ return AllocateConvert(buffer, Convert32To16(dither), src);
}
static ConstBuffer<int16_t>
pcm_allocate_float_to_16(PcmBuffer &buffer, ConstBuffer<float> src)
{
- return ToConst(AllocateFromFloat<SampleFormat::S16>(buffer, src));
+ return AllocateFromFloat<SampleFormat::S16>(buffer, src);
}
ConstBuffer<int16_t>
@@ -173,13 +165,13 @@ struct Convert16To24
static ConstBuffer<int32_t>
pcm_allocate_8_to_24(PcmBuffer &buffer, ConstBuffer<int8_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert8To24(), src));
+ return AllocateConvert(buffer, Convert8To24(), src);
}
static ConstBuffer<int32_t>
pcm_allocate_16_to_24(PcmBuffer &buffer, ConstBuffer<int16_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert16To24(), src));
+ return AllocateConvert(buffer, Convert16To24(), src);
}
struct Convert32To24
@@ -189,10 +181,10 @@ struct Convert32To24
static ConstBuffer<int32_t>
pcm_allocate_32_to_24(PcmBuffer &buffer, ConstBuffer<int32_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert32To24(), src));
+ return AllocateConvert(buffer, Convert32To24(), src);
}
-static WritableBuffer<int32_t>
+static ConstBuffer<int32_t>
pcm_allocate_float_to_24(PcmBuffer &buffer, ConstBuffer<float> src)
{
return AllocateFromFloat<SampleFormat::S24_P32>(buffer, src);
@@ -223,8 +215,8 @@ pcm_convert_to_24(PcmBuffer &buffer,
ConstBuffer<int32_t>::FromVoid(src));
case SampleFormat::FLOAT:
- return ToConst(pcm_allocate_float_to_24(buffer,
- ConstBuffer<float>::FromVoid(src)));
+ return pcm_allocate_float_to_24(buffer,
+ ConstBuffer<float>::FromVoid(src));
}
return nullptr;
@@ -245,25 +237,25 @@ struct Convert24To32
static ConstBuffer<int32_t>
pcm_allocate_8_to_32(PcmBuffer &buffer, ConstBuffer<int8_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert8To32(), src));
+ return AllocateConvert(buffer, Convert8To32(), src);
}
static ConstBuffer<int32_t>
pcm_allocate_16_to_32(PcmBuffer &buffer, ConstBuffer<int16_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert16To32(), src));
+ return AllocateConvert(buffer, Convert16To32(), src);
}
static ConstBuffer<int32_t>
pcm_allocate_24p32_to_32(PcmBuffer &buffer, ConstBuffer<int32_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert24To32(), src));
+ return AllocateConvert(buffer, Convert24To32(), src);
}
static ConstBuffer<int32_t>
pcm_allocate_float_to_32(PcmBuffer &buffer, ConstBuffer<float> src)
{
- return ToConst(AllocateFromFloat<SampleFormat::S32>(buffer, src));
+ return AllocateFromFloat<SampleFormat::S32>(buffer, src);
}
ConstBuffer<int32_t>
@@ -313,25 +305,25 @@ struct Convert32ToFloat
static ConstBuffer<float>
pcm_allocate_8_to_float(PcmBuffer &buffer, ConstBuffer<int8_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert8ToFloat(), src));
+ return AllocateConvert(buffer, Convert8ToFloat(), src);
}
static ConstBuffer<float>
pcm_allocate_16_to_float(PcmBuffer &buffer, ConstBuffer<int16_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert16ToFloat(), src));
+ return AllocateConvert(buffer, Convert16ToFloat(), src);
}
static ConstBuffer<float>
pcm_allocate_24p32_to_float(PcmBuffer &buffer, ConstBuffer<int32_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert24ToFloat(), src));
+ return AllocateConvert(buffer, Convert24ToFloat(), src);
}
static ConstBuffer<float>
pcm_allocate_32_to_float(PcmBuffer &buffer, ConstBuffer<int32_t> src)
{
- return ToConst(AllocateConvert(buffer, Convert32ToFloat(), src));
+ return AllocateConvert(buffer, Convert32ToFloat(), src);
}
ConstBuffer<float>