basis_universal: Sync with upstream 646a9f826

https://github.com/BinomialLLC/basis_universal/commit/646a9f826131cb0b9e14b5e4740874808315f83a

10 files changed, 52 insertions, 62 deletions

diff --git a/thirdparty/README.md b/thirdparty/README.md
index ee3fbb7adc..e1c0aaec89 100644
--- a/thirdparty/README.md
+++ b/thirdparty/README.md
@@ -8,7 +8,7 @@ readability.
 ## basis_universal
 
 - Upstream: https://github.com/BinomialLLC/basis_universal
-- Version: git (ba1c3e40f1d434ebaf9a167b44e9b11d2bf0f765, 2021)
+- Version: git (646a9f826131cb0b9e14b5e4740874808315f83a, 2021)
 - License: Apache 2.0
 
 Files extracted from upstream source:
diff --git a/thirdparty/basis_universal/encoder/apg_bmp.c b/thirdparty/basis_universal/encoder/apg_bmp.c
index ef3d015e40..d342b20fc8 100644
--- a/thirdparty/basis_universal/encoder/apg_bmp.c
+++ b/thirdparty/basis_universal/encoder/apg_bmp.c
@@ -247,7 +247,7 @@ unsigned char* apg_bmp_read( const char* filename, int* w, int* h, unsigned int*
   }
 
   // allocate memory for the output pixels block. cast to size_t in case width and height are both the max of 65536 and n_dst_chans > 1
-  unsigned char* dst_img_ptr = malloc( (size_t)width * (size_t)height * (size_t)n_dst_chans );
+  unsigned char* dst_img_ptr = (unsigned char*)malloc( (size_t)width * (size_t)height * (size_t)n_dst_chans );
   if ( !dst_img_ptr ) {
     free( record.data );
     return NULL;
@@ -480,7 +480,7 @@ unsigned int apg_bmp_write( const char* filename, unsigned char* pixels_ptr, int
     dib_hdr.bitmask_b = 0x0000FF00;
   }
 
-  uint8_t* dst_pixels_ptr = malloc( dst_pixels_padded_sz );
+  uint8_t* dst_pixels_ptr = (uint8_t*)malloc( dst_pixels_padded_sz );
   if ( !dst_pixels_ptr ) { return 0; }
   {
     size_t dst_byte_idx = 0;
diff --git a/thirdparty/basis_universal/encoder/basisu_bc7enc.cpp b/thirdparty/basis_universal/encoder/basisu_bc7enc.cpp
index 06aa7eb8b1..22fdfa603f 100644
--- a/thirdparty/basis_universal/encoder/basisu_bc7enc.cpp
+++ b/thirdparty/basis_universal/encoder/basisu_bc7enc.cpp
@@ -174,9 +174,8 @@ static void astc_init()
 	} // range
 }
 
-static inline uint32_t astc_interpolate(uint32_t l, uint32_t h, uint32_t w)
+static inline uint32_t astc_interpolate_linear(uint32_t l, uint32_t h, uint32_t w)
 {
-	// This is for linear values, not sRGB.
 	l = (l << 8) | l;
 	h = (h << 8) | h;
 	uint32_t k = (l * (64 - w) + h * w + 32) >> 6;
@@ -230,7 +229,7 @@ void bc7enc_compress_block_init()
 			{
 				uint32_t high = (h << 4) | h;
 				
-				const int k = astc_interpolate(low, high, g_bc7_weights3[BC7ENC_ASTC_4BIT_3BIT_OPTIMAL_INDEX]);
+				const int k = astc_interpolate_linear(low, high, g_bc7_weights3[BC7ENC_ASTC_4BIT_3BIT_OPTIMAL_INDEX]);
 				const int err = (k - c) * (k - c);
 
 				if (err < best.m_error)
@@ -259,7 +258,7 @@ void bc7enc_compress_block_init()
 			{
 				uint32_t high = (h << 4) | h;
 				
-				const int k = astc_interpolate(low, high, g_bc7_weights2[BC7ENC_ASTC_4BIT_2BIT_OPTIMAL_INDEX]);
+				const int k = astc_interpolate_linear(low, high, g_bc7_weights2[BC7ENC_ASTC_4BIT_2BIT_OPTIMAL_INDEX]);
 				const int err = (k - c) * (k - c);
 
 				if (err < best.m_error)
@@ -288,7 +287,7 @@ void bc7enc_compress_block_init()
 			{
 				uint32_t high = g_astc_sorted_order_unquant[7][h].m_unquant;
 				
-				const int k = astc_interpolate(low, high, g_bc7_weights2[BC7ENC_ASTC_RANGE7_2BIT_OPTIMAL_INDEX]);
+				const int k = astc_interpolate_linear(low, high, g_bc7_weights2[BC7ENC_ASTC_RANGE7_2BIT_OPTIMAL_INDEX]);
 				const int err = (k - c) * (k - c);
 
 				if (err < best.m_error)
@@ -317,7 +316,7 @@ void bc7enc_compress_block_init()
 			{
 				uint32_t high = g_astc_sorted_order_unquant[13][h].m_unquant;
 				
-				const int k = astc_interpolate(low, high, g_astc_weights4[BC7ENC_ASTC_RANGE13_4BIT_OPTIMAL_INDEX]);
+				const int k = astc_interpolate_linear(low, high, g_astc_weights4[BC7ENC_ASTC_RANGE13_4BIT_OPTIMAL_INDEX]);
 				const int err = (k - c) * (k - c);
 
 				if (err < best.m_error)
@@ -346,7 +345,7 @@ void bc7enc_compress_block_init()
 			{
 				uint32_t high = g_astc_sorted_order_unquant[13][h].m_unquant;
 				
-				const int k = astc_interpolate(low, high, g_bc7_weights2[BC7ENC_ASTC_RANGE13_2BIT_OPTIMAL_INDEX]);
+				const int k = astc_interpolate_linear(low, high, g_bc7_weights2[BC7ENC_ASTC_RANGE13_2BIT_OPTIMAL_INDEX]);
 				const int err = (k - c) * (k - c);
 
 				if (err < best.m_error)
@@ -375,7 +374,7 @@ void bc7enc_compress_block_init()
 			{
 				uint32_t high = g_astc_sorted_order_unquant[11][h].m_unquant;
 
-				const int k = astc_interpolate(low, high, g_astc_weights5[BC7ENC_ASTC_RANGE11_5BIT_OPTIMAL_INDEX]);
+				const int k = astc_interpolate_linear(low, high, g_astc_weights5[BC7ENC_ASTC_RANGE11_5BIT_OPTIMAL_INDEX]);
 				const int err = (k - c) * (k - c);
 
 				if (err < best.m_error)
@@ -650,7 +649,7 @@ static uint64_t pack_astc_4bit_3bit_to_one_color(const color_cell_compressor_par
 		uint32_t low = (pResults->m_low_endpoint.m_c[i] << 4) | pResults->m_low_endpoint.m_c[i];
 		uint32_t high = (pResults->m_high_endpoint.m_c[i] << 4) | pResults->m_high_endpoint.m_c[i];
 		
-		p.m_c[i] = (uint8_t)astc_interpolate(low, high, g_bc7_weights3[BC7ENC_ASTC_4BIT_3BIT_OPTIMAL_INDEX]);
+		p.m_c[i] = (uint8_t)astc_interpolate_linear(low, high, g_bc7_weights3[BC7ENC_ASTC_4BIT_3BIT_OPTIMAL_INDEX]);
 	}
 	p.m_c[3] = 255;
 
@@ -689,7 +688,7 @@ static uint64_t pack_astc_4bit_2bit_to_one_color_rgba(const color_cell_compresso
 		uint32_t low = (pResults->m_low_endpoint.m_c[i] << 4) | pResults->m_low_endpoint.m_c[i];
 		uint32_t high = (pResults->m_high_endpoint.m_c[i] << 4) | pResults->m_high_endpoint.m_c[i];
 		
-		p.m_c[i] = (uint8_t)astc_interpolate(low, high, g_bc7_weights2[BC7ENC_ASTC_4BIT_2BIT_OPTIMAL_INDEX]);
+		p.m_c[i] = (uint8_t)astc_interpolate_linear(low, high, g_bc7_weights2[BC7ENC_ASTC_4BIT_2BIT_OPTIMAL_INDEX]);
 	}
 	
 	uint64_t total_err = 0;
@@ -728,7 +727,7 @@ static uint64_t pack_astc_range7_2bit_to_one_color(const color_cell_compressor_p
 		uint32_t low = g_astc_sorted_order_unquant[7][pResults->m_low_endpoint.m_c[i]].m_unquant;
 		uint32_t high = g_astc_sorted_order_unquant[7][pResults->m_high_endpoint.m_c[i]].m_unquant;
 		
-		p.m_c[i] = (uint8_t)astc_interpolate(low, high, g_bc7_weights2[BC7ENC_ASTC_RANGE7_2BIT_OPTIMAL_INDEX]);
+		p.m_c[i] = (uint8_t)astc_interpolate_linear(low, high, g_bc7_weights2[BC7ENC_ASTC_RANGE7_2BIT_OPTIMAL_INDEX]);
 	}
 	p.m_c[3] = 255;
 
@@ -768,7 +767,7 @@ static uint64_t pack_astc_range13_2bit_to_one_color(const color_cell_compressor_
 		uint32_t low = g_astc_sorted_order_unquant[13][pResults->m_low_endpoint.m_c[i]].m_unquant;
 		uint32_t high = g_astc_sorted_order_unquant[13][pResults->m_high_endpoint.m_c[i]].m_unquant;
 		
-		p.m_c[i] = (uint8_t)astc_interpolate(low, high, g_bc7_weights2[BC7ENC_ASTC_RANGE13_2BIT_OPTIMAL_INDEX]);
+		p.m_c[i] = (uint8_t)astc_interpolate_linear(low, high, g_bc7_weights2[BC7ENC_ASTC_RANGE13_2BIT_OPTIMAL_INDEX]);
 	}
 	
 	uint64_t total_err = 0;
@@ -807,7 +806,7 @@ static uint64_t pack_astc_range11_5bit_to_one_color(const color_cell_compressor_
 		uint32_t low = g_astc_sorted_order_unquant[11][pResults->m_low_endpoint.m_c[i]].m_unquant;
 		uint32_t high = g_astc_sorted_order_unquant[11][pResults->m_high_endpoint.m_c[i]].m_unquant;
 
-		p.m_c[i] = (uint8_t)astc_interpolate(low, high, g_astc_weights5[BC7ENC_ASTC_RANGE11_5BIT_OPTIMAL_INDEX]);
+		p.m_c[i] = (uint8_t)astc_interpolate_linear(low, high, g_astc_weights5[BC7ENC_ASTC_RANGE11_5BIT_OPTIMAL_INDEX]);
 	}
 
 	uint64_t total_err = 0;
@@ -863,7 +862,7 @@ static uint64_t evaluate_solution(const color_quad_u8 *pLow, const color_quad_u8
 		for (uint32_t i = 1; i < (N - 1); i++)
 		{
 			for (uint32_t j = 0; j < nc; j++)
-				weightedColors[i].m_c[j] = (uint8_t)(astc_interpolate(actualMinColor.m_c[j], actualMaxColor.m_c[j], pParams->m_pSelector_weights[i]));
+				weightedColors[i].m_c[j] = (uint8_t)(astc_interpolate_linear(actualMinColor.m_c[j], actualMaxColor.m_c[j], pParams->m_pSelector_weights[i]));
 		}
 	}
 	else
@@ -1300,7 +1299,7 @@ void check_best_overall_error(const color_cell_compressor_params *pParams, color
 	
 	for (uint32_t i = 1; i < pParams->m_num_selector_weights - 1; i++)
 		for (uint32_t c = 0; c < 4; c++)
-			colors[i].m_c[c] = (uint8_t)astc_interpolate(colors[0].m_c[c], colors[n - 1].m_c[c], pParams->m_pSelector_weights[i]);
+			colors[i].m_c[c] = (uint8_t)astc_interpolate_linear(colors[0].m_c[c], colors[n - 1].m_c[c], pParams->m_pSelector_weights[i]);
 
 	uint64_t total_err = 0;
 	for (uint32_t p = 0; p < pParams->m_num_pixels; p++)
@@ -1815,10 +1814,10 @@ uint64_t color_cell_compression_est_astc(
 	weightedColors[num_weights - 1] = highColor;
 	for (uint32_t i = 1; i < (num_weights - 1); i++)
 	{
-		weightedColors[i].m_c[0] = (uint8_t)astc_interpolate(lowColor.m_c[0], highColor.m_c[0], pWeight_table[i]);
-		weightedColors[i].m_c[1] = (uint8_t)astc_interpolate(lowColor.m_c[1], highColor.m_c[1], pWeight_table[i]);
-		weightedColors[i].m_c[2] = (uint8_t)astc_interpolate(lowColor.m_c[2], highColor.m_c[2], pWeight_table[i]);
-		weightedColors[i].m_c[3] = (num_comps == 4) ? (uint8_t)astc_interpolate(lowColor.m_c[3], highColor.m_c[3], pWeight_table[i]) : 255;
+		weightedColors[i].m_c[0] = (uint8_t)astc_interpolate_linear(lowColor.m_c[0], highColor.m_c[0], pWeight_table[i]);
+		weightedColors[i].m_c[1] = (uint8_t)astc_interpolate_linear(lowColor.m_c[1], highColor.m_c[1], pWeight_table[i]);
+		weightedColors[i].m_c[2] = (uint8_t)astc_interpolate_linear(lowColor.m_c[2], highColor.m_c[2], pWeight_table[i]);
+		weightedColors[i].m_c[3] = (num_comps == 4) ? (uint8_t)astc_interpolate_linear(lowColor.m_c[3], highColor.m_c[3], pWeight_table[i]) : 255;
 	}
 
 	// Compute dots and thresholds
diff --git a/thirdparty/basis_universal/encoder/basisu_bc7enc.h b/thirdparty/basis_universal/encoder/basisu_bc7enc.h
index 23469912e2..8d8b7888ca 100644
--- a/thirdparty/basis_universal/encoder/basisu_bc7enc.h
+++ b/thirdparty/basis_universal/encoder/basisu_bc7enc.h
@@ -12,6 +12,7 @@
 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 // See the License for the specific language governing permissions and
 // limitations under the License.
+#pragma once
 #include "basisu_enc.h"
 #include "../transcoder/basisu_transcoder_uastc.h"
 
diff --git a/thirdparty/basis_universal/encoder/basisu_comp.cpp b/thirdparty/basis_universal/encoder/basisu_comp.cpp
index dc4ae11539..10f96cec4a 100644
--- a/thirdparty/basis_universal/encoder/basisu_comp.cpp
+++ b/thirdparty/basis_universal/encoder/basisu_comp.cpp
@@ -467,7 +467,10 @@ namespace basisu
 					return false;
 				}
 
-				printf("Read source image \"%s\", %ux%u\n", pSource_filename, file_image.get_width(), file_image.get_height());
+				if (m_params.m_status_output)
+				{
+					printf("Read source image \"%s\", %ux%u\n", pSource_filename, file_image.get_width(), file_image.get_height());
+				}
 
 				// Optionally load another image and put a grayscale version of it into the alpha channel.
 				if ((source_file_index < m_params.m_source_alpha_filenames.size()) && (m_params.m_source_alpha_filenames[source_file_index].size()))
@@ -1427,7 +1430,10 @@ namespace basisu
 				return false;
 			}
 
-			printf("Wrote output .basis/.ktx2 file \"%s\"\n", output_filename.c_str());
+			if (m_params.m_status_output)
+			{
+				printf("Wrote output .basis/.ktx2 file \"%s\"\n", output_filename.c_str());
+			}
 		}
 
 		size_t comp_size = 0;
diff --git a/thirdparty/basis_universal/encoder/basisu_enc.cpp b/thirdparty/basis_universal/encoder/basisu_enc.cpp
index f02fb62c11..daaf65badc 100644
--- a/thirdparty/basis_universal/encoder/basisu_enc.cpp
+++ b/thirdparty/basis_universal/encoder/basisu_enc.cpp
@@ -195,7 +195,7 @@ namespace basisu
 	{
 		QueryPerformanceFrequency(reinterpret_cast<LARGE_INTEGER*>(pTicks));
 	}
-#elif defined(__APPLE__)
+#elif defined(__APPLE__) || defined(__OpenBSD__)
 #include <sys/time.h>
 	inline void query_counter(timer_ticks* pTicks)
 	{
@@ -1779,8 +1779,6 @@ namespace basisu
 			return nullptr;
 		}
 
-		const uint32_t bytes_per_line = hdr.m_width * tga_bytes_per_pixel;
-
 		const uint8_t *pSrc = pBuf + sizeof(tga_header);
 		uint32_t bytes_remaining = buf_size - sizeof(tga_header);
 
diff --git a/thirdparty/basis_universal/encoder/basisu_enc.h b/thirdparty/basis_universal/encoder/basisu_enc.h
index 05c95cbc3b..0ce011452d 100644
--- a/thirdparty/basis_universal/encoder/basisu_enc.h
+++ b/thirdparty/basis_universal/encoder/basisu_enc.h
@@ -1634,6 +1634,14 @@ namespace basisu
 
 				if ((!l_weight) || (!r_weight))
 				{
+					l_children.resize(0);
+					new_l_child.set(0.0f);
+					l_ttsum = 0.0f;
+					l_weight = 0;
+					r_children.resize(0);
+					new_r_child.set(0.0f);
+					r_ttsum = 0.0f;
+					r_weight = 0;
 					TrainingVectorType firstVec;
 					for (uint32_t i = 0; i < node.m_training_vecs.size(); i++)
 					{
@@ -1660,7 +1668,7 @@ namespace basisu
 						}
 					}
 
-					if (!l_weight)
+					if ((!l_weight) || (!r_weight))
 						return false;
 				}
 
diff --git a/thirdparty/basis_universal/encoder/basisu_resampler.cpp b/thirdparty/basis_universal/encoder/basisu_resampler.cpp
index e193ce83ff..f4cedf0031 100644
--- a/thirdparty/basis_universal/encoder/basisu_resampler.cpp
+++ b/thirdparty/basis_universal/encoder/basisu_resampler.cpp
@@ -15,14 +15,6 @@
 #include "basisu_resampler.h"
 #include "basisu_resampler_filters.h"
 
-#ifndef max
-#define max(a, b) (((a) > (b)) ? (a) : (b))
-#endif
-
-#ifndef min
-#define min(a, b) (((a) < (b)) ? (a) : (b))
-#endif
-
 #define RESAMPLER_DEBUG 0
 
 namespace basisu
diff --git a/thirdparty/basis_universal/encoder/cppspmd_sse.h b/thirdparty/basis_universal/encoder/cppspmd_sse.h
index b39cb82a5f..9a97eeb695 100644
--- a/thirdparty/basis_universal/encoder/cppspmd_sse.h
+++ b/thirdparty/basis_universal/encoder/cppspmd_sse.h
@@ -1327,33 +1327,15 @@ struct spmd_kernel
 	CPPSPMD_FORCE_INLINE float reduce_add(vfloat v)	
 	{ 
 		__m128 k3210 = _mm_castsi128_ps(blendv_mask_epi32(_mm_setzero_si128(), _mm_castps_si128(v.m_value), m_exec.m_mask));
-
-//#if CPPSPMD_SSE2
-#if 1
-		// See https://stackoverflow.com/questions/6996764/fastest-way-to-do-horizontal-sse-vector-sum-or-other-reduction/35270026#35270026
-		__m128 shuf   = _mm_shuffle_ps(k3210, k3210, _MM_SHUFFLE(2, 3, 0, 1));
-		__m128 sums   = _mm_add_ps(k3210, shuf);
-		shuf          = _mm_movehl_ps(shuf, sums);
-		sums          = _mm_add_ss(sums, shuf);
-		return _mm_cvtss_f32(sums);
-#else
-		// This is pretty slow.
-		__m128 a = _mm_hadd_ps(k3210, k3210);
-		__m128 b = _mm_hadd_ps(a, a);
-		return extractf_ps_x(b);
-#endif
+		__m128 temp = _mm_add_ps(_mm_shuffle_ps(k3210, k3210, _MM_SHUFFLE(0, 1, 2, 3)), k3210);
+		return _mm_cvtss_f32(_mm_add_ss(_mm_movehl_ps(temp, temp), temp));
 	}
-
+		
 	CPPSPMD_FORCE_INLINE int reduce_add(vint v)
 	{
 		__m128i k3210 = blendv_mask_epi32(_mm_setzero_si128(), v.m_value, m_exec.m_mask);
-
-		// See https://stackoverflow.com/questions/6996764/fastest-way-to-do-horizontal-sse-vector-sum-or-other-reduction/35270026#35270026
-		__m128i shuf = _mm_shuffle_epi32(k3210, _MM_SHUFFLE(2, 3, 0, 1));
-		__m128i sums = _mm_add_epi32(k3210, shuf);
-		shuf = _mm_castps_si128(_mm_movehl_ps(_mm_castsi128_ps(shuf), _mm_castsi128_ps(sums)));
-		sums = _mm_add_epi32(sums, shuf);
-		return extract_x(sums);
+		__m128i temp = _mm_add_epi32(_mm_shuffle_epi32(k3210, _MM_SHUFFLE(0, 1, 2, 3)), k3210);
+		return extract_x(_mm_add_epi32(_mm_castps_si128(_mm_movehl_ps(_mm_castsi128_ps(temp), _mm_castsi128_ps(temp))), temp));
 	}
 
 	#include "cppspmd_math_declares.h"
@@ -1686,6 +1668,12 @@ CPPSPMD_FORCE_INLINE vint uniform_shift_right_epi16(const vint& a, const vint& b
 CPPSPMD_FORCE_INLINE vint undefined_vint() { return vint{ _mm_undefined_si128() }; }
 CPPSPMD_FORCE_INLINE vfloat undefined_vfloat() { return vfloat{ _mm_undefined_ps() }; }
 
+CPPSPMD_FORCE_INLINE vint vint_lane_set(int v0, int v1, int v2, int v3) { return vint{ _mm_set_epi32(v3, v2, v1, v0) }; }
+CPPSPMD_FORCE_INLINE vfloat vfloat_lane_set(float v0, float v1, float v2, float v3) { return vfloat{ _mm_set_ps(v3, v2, v1, v0) }; }
+
+CPPSPMD_FORCE_INLINE vint vint_lane_set_r(int v3, int v2, int v1, int v0) { return vint{ _mm_set_epi32(v3, v2, v1, v0) }; }
+CPPSPMD_FORCE_INLINE vfloat vfloat_lane_set_r(float v3, float v2, float v1, float v0) { return vfloat{ _mm_set_ps(v3, v2, v1, v0) }; }
+
 // control is an 8-bit immediate value containing 4 2-bit indices which shuffles the int32's in each 128-bit lane.
 #define VINT_LANE_SHUFFLE_EPI32(a, control) vint(_mm_shuffle_epi32((a).m_value, control))
 
diff --git a/thirdparty/basis_universal/transcoder/basisu_transcoder.cpp b/thirdparty/basis_universal/transcoder/basisu_transcoder.cpp
index 29eb3c0d55..0b3733385d 100644
--- a/thirdparty/basis_universal/transcoder/basisu_transcoder.cpp
+++ b/thirdparty/basis_universal/transcoder/basisu_transcoder.cpp
@@ -10778,8 +10778,6 @@ namespace basist
 			return false;
 		}
 
-		const bool transcode_alpha_data_to_opaque_formats = (decode_flags & cDecodeFlagsTranscodeAlphaDataToOpaqueFormats) != 0;
-
 		if (decode_flags & cDecodeFlagsPVRTCDecodeToNextPow2)
 		{
 			BASISU_DEVEL_ERROR("basisu_transcoder::transcode_image_level: cDecodeFlagsPVRTCDecodeToNextPow2 currently unsupported\n");
@@ -17336,7 +17334,6 @@ namespace basist
 		
 	bool ktx2_transcoder::decompress_level_data(uint32_t level_index, basisu::uint8_vec& uncomp_data)
 	{
-		const uint8_t* pComp_data = m_levels[level_index].m_byte_offset + m_pData;
 		const uint64_t comp_size = m_levels[level_index].m_byte_length;
 		
 		const uint64_t uncomp_size = m_levels[level_index].m_uncompressed_byte_length;
@@ -17361,6 +17358,7 @@ namespace basist
 		if (m_header.m_supercompression_scheme == KTX2_SS_ZSTANDARD)
 		{
 #if BASISD_SUPPORT_KTX2_ZSTD
+			const uint8_t* pComp_data = m_levels[level_index].m_byte_offset + m_pData;
 			size_t actualUncompSize = ZSTD_decompress(uncomp_data.data(), (size_t)uncomp_size, pComp_data, (size_t)comp_size);
 			if (ZSTD_isError(actualUncompSize))
 			{