[4.0] ResourceImporterWAV: Add support for 64-bit IEEE float

This commit is contained in:
Douglas Leão 2022-05-18 14:21:07 -03:00 committed by Douglas
parent 838cb598e3
commit 1c0d9eef7a

View file

@ -162,7 +162,7 @@ Error ResourceImporterWAV::import(const String &p_source_file, const String &p_s
//Consider revision for engine version 3.0
compression_code = file->get_16();
if (compression_code != 1 && compression_code != 3) {
ERR_FAIL_V_MSG(ERR_INVALID_DATA, "Format not supported for WAVE file (not PCM). Save WAVE files as uncompressed PCM instead.");
ERR_FAIL_V_MSG(ERR_INVALID_DATA, "Format not supported for WAVE file (not PCM). Save WAVE files as uncompressed PCM or IEEE float instead.");
}
format_channels = file->get_16();
@ -180,6 +180,10 @@ Error ResourceImporterWAV::import(const String &p_source_file, const String &p_s
ERR_FAIL_V_MSG(ERR_INVALID_DATA, "Invalid amount of bits in the sample (should be one of 8, 16, 24 or 32).");
}
if (compression_code == 3 && format_bits % 32) {
ERR_FAIL_V_MSG(ERR_INVALID_DATA, "Invalid amount of bits in the IEEE float sample (should be 32 or 64).");
}
/* Don't need anything else, continue */
format_found = true;
}
@ -208,36 +212,46 @@ Error ResourceImporterWAV::import(const String &p_source_file, const String &p_s
data.resize(frames * format_channels);
if (format_bits == 8) {
for (int i = 0; i < frames * format_channels; i++) {
// 8 bit samples are UNSIGNED
if (compression_code == 1) {
if (format_bits == 8) {
for (int i = 0; i < frames * format_channels; i++) {
// 8 bit samples are UNSIGNED
data.write[i] = int8_t(file->get_8() - 128) / 128.f;
}
} else if (format_bits == 32 && compression_code == 3) {
for (int i = 0; i < frames * format_channels; i++) {
//32 bit IEEE Float
data.write[i] = file->get_float();
}
} else if (format_bits == 16) {
for (int i = 0; i < frames * format_channels; i++) {
//16 bit SIGNED
data.write[i] = int16_t(file->get_16()) / 32768.f;
}
} else {
for (int i = 0; i < frames * format_channels; i++) {
//16+ bits samples are SIGNED
// if sample is > 16 bits, just read extra bytes
uint32_t s = 0;
for (int b = 0; b < (format_bits >> 3); b++) {
s |= ((uint32_t)file->get_8()) << (b * 8);
data.write[i] = int8_t(file->get_8() - 128) / 128.f;
}
s <<= (32 - format_bits);
} else if (format_bits == 16) {
for (int i = 0; i < frames * format_channels; i++) {
//16 bit SIGNED
data.write[i] = (int32_t(s) >> 16) / 32768.f;
data.write[i] = int16_t(file->get_16()) / 32768.f;
}
} else {
for (int i = 0; i < frames * format_channels; i++) {
//16+ bits samples are SIGNED
// if sample is > 16 bits, just read extra bytes
uint32_t s = 0;
for (int b = 0; b < (format_bits >> 3); b++) {
s |= ((uint32_t)file->get_8()) << (b * 8);
}
s <<= (32 - format_bits);
data.write[i] = (int32_t(s) >> 16) / 32768.f;
}
}
} else if (compression_code == 3) {
if (format_bits == 32) {
for (int i = 0; i < frames * format_channels; i++) {
//32 bit IEEE Float
data.write[i] = file->get_float();
}
} else if (format_bits == 64) {
for (int i = 0; i < frames * format_channels; i++) {
//64 bit IEEE Float
data.write[i] = file->get_double();
}
}
}