From 4e97181cd991043776777d1d1d6d7b1078821714 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=D0=91=D1=80=D0=B0=D0=BD=D0=B8=D0=BC=D0=B8=D1=80=20=D0=9A?=
 =?UTF-8?q?=D0=B0=D1=80=D0=B0=D1=9F=D0=B8=D1=9B?=
 <branimirkaradzic@gmail.com>
Date: Sat, 4 Apr 2020 17:55:38 -0700
Subject: [PATCH] Updated meshoptimizer.

---
 3rdparty/meshoptimizer/src/meshoptimizer.h  |  4 ++
 3rdparty/meshoptimizer/src/vertexfilter.cpp | 57 +++++++++++++++++++++
 2 files changed, 61 insertions(+)

diff --git a/3rdparty/meshoptimizer/src/meshoptimizer.h b/3rdparty/meshoptimizer/src/meshoptimizer.h
index 12160951a..d75fa1417 100644
--- a/3rdparty/meshoptimizer/src/meshoptimizer.h
+++ b/3rdparty/meshoptimizer/src/meshoptimizer.h
@@ -219,9 +219,13 @@ MESHOPTIMIZER_API int meshopt_decodeVertexBuffer(void* destination, size_t verte
  *
  * meshopt_decodeFilterQuat decodes 3-component quaternion encoding with K-bit (4 <= K <= 16) component encoding and a 2-bit component index indicating which component to reconstruct.
  * Each component is stored as an 16-bit integer; stride must be equal to 8.
+ *
+ * meshopt_decodeFilterExp decodes exponential encoding of floating-point data with 8-bit exponent and 24-bit integer mantissa as 2^E*M.
+ * Each 32-bit component is decoded in isolation; stride must be divisible by 4.
  */
 MESHOPTIMIZER_EXPERIMENTAL void meshopt_decodeFilterOct(void* buffer, size_t vertex_count, size_t vertex_size);
 MESHOPTIMIZER_EXPERIMENTAL void meshopt_decodeFilterQuat(void* buffer, size_t vertex_count, size_t vertex_size);
+MESHOPTIMIZER_EXPERIMENTAL void meshopt_decodeFilterExp(void* buffer, size_t vertex_count, size_t vertex_size);
 
 /**
  * Experimental: Mesh simplifier
diff --git a/3rdparty/meshoptimizer/src/vertexfilter.cpp b/3rdparty/meshoptimizer/src/vertexfilter.cpp
index 35bfdd958..cedc270b8 100644
--- a/3rdparty/meshoptimizer/src/vertexfilter.cpp
+++ b/3rdparty/meshoptimizer/src/vertexfilter.cpp
@@ -96,6 +96,29 @@ static void decodeFilterQuat(short* data, size_t count)
 		data[i * 4 + order[qc][3]] = short(wf);
 	}
 }
+
+static void decodeFilterExp(unsigned int* data, size_t count)
+{
+	for (size_t i = 0; i < count; ++i)
+	{
+		unsigned int v = data[i];
+
+		// decode mantissa and exponent
+		int m = int(v << 8) >> 8;
+		int e = int(v) >> 24;
+
+		union {
+			float f;
+			unsigned int ui;
+		} u;
+
+		// optimized version of ldexp(float(m), e)
+		u.ui = unsigned(e + 127) << 23;
+		u.f = u.f * float(m);
+
+		data[i] = u.ui;
+	}
+}
 #endif
 
 #ifdef SIMD_WASM
@@ -277,6 +300,26 @@ static void decodeFilterQuatSimd(short* data, size_t count)
 		out[3] = __builtin_rotateleft64(wasm_i64x2_extract_lane(res_1, 1), wasm_i32x4_extract_lane(cm, 3));
 	}
 }
+
+static void decodeFilterExpSimd(unsigned int* data, size_t count)
+{
+	for (size_t i = 0; i < count; i += 4)
+	{
+		v128_t v = wasm_v128_load(&data[i]);
+
+		// decode exponent into 2^x directly
+		v128_t ef = wasm_i32x4_shr(v, 24);
+		v128_t es = wasm_i32x4_shl(wasm_i32x4_add(ef, wasm_i32x4_splat(127)), 23);
+
+		// decode 24-bit mantissa into floating-point value
+		v128_t mf = wasm_i32x4_shr(wasm_i32x4_shl(v, 8), 8);
+		v128_t m = wasm_f32x4_convert_i32x4(mf);
+
+		v128_t r = wasm_f32x4_mul(es, m);
+
+		wasm_v128_store(&data[i], r);
+	}
+}
 #endif
 
 } // namespace meshopt
@@ -316,4 +359,18 @@ void meshopt_decodeFilterQuat(void* buffer, size_t vertex_count, size_t vertex_s
 #endif
 }
 
+void meshopt_decodeFilterExp(void* buffer, size_t vertex_count, size_t vertex_size)
+{
+	using namespace meshopt;
+
+	assert(vertex_count % 4 == 0);
+	assert(vertex_size % 4 == 0);
+
+#if defined(SIMD_WASM)
+	decodeFilterExpSimd(static_cast<unsigned int*>(buffer), vertex_count * (vertex_size / 4));
+#else
+	decodeFilterExp(static_cast<unsigned int*>(buffer), vertex_count * (vertex_size / 4));
+#endif
+}
+
 #undef SIMD_WASM