Swiched from FPU to SIMD math.

src/bgfx.cpp

@@ -235,22 +235,6 @@ namespace bgfx
 		g_callback->fatal(_code, temp);
 	}
 
-	inline void vec4MulMtx(float* __restrict _result, const float* __restrict _vec, const float* __restrict _mat)
-	{
-		_result[0] = _vec[0] * _mat[ 0] + _vec[1] * _mat[4] + _vec[2] * _mat[ 8] + _vec[3] * _mat[12];
-		_result[1] = _vec[0] * _mat[ 1] + _vec[1] * _mat[5] + _vec[2] * _mat[ 9] + _vec[3] * _mat[13];
-		_result[2] = _vec[0] * _mat[ 2] + _vec[1] * _mat[6] + _vec[2] * _mat[10] + _vec[3] * _mat[14];
-		_result[3] = _vec[0] * _mat[ 3] + _vec[1] * _mat[7] + _vec[2] * _mat[11] + _vec[3] * _mat[15];
-	}
-
-	void mtxMul(float* __restrict _result, const float* __restrict _a, const float* __restrict _b)
-	{
-		vec4MulMtx(&_result[ 0], &_a[ 0], _b);
-		vec4MulMtx(&_result[ 4], &_a[ 4], _b);
-		vec4MulMtx(&_result[ 8], &_a[ 8], _b);
-		vec4MulMtx(&_result[12], &_a[12], _b);
-	}
-
 	void mtxOrtho(float* _result, float _left, float _right, float _bottom, float _top, float _near, float _far)
 	{
 		const float aa = 2.0f/(_right - _left);

src/bgfx_p.h

@@ -61,6 +61,7 @@ namespace bgfx
 
 #include <bx/bx.h>
 #include <bx/debug.h>
+#include <bx/float4x4_t.h>
 #include <bx/blockalloc.h>
 #include <bx/endian.h>
 #include <bx/handlealloc.h>
@@ -639,16 +640,19 @@ namespace bgfx
 
 	BX_ALIGN_STRUCT_16(struct) Matrix4
 	{
-		float val[16];
+		union
+		{
+			bx::float4x4_t f4x4;
+			float val[16];
+		} un;
 
 		void setIdentity()
 		{
-			memset(val, 0, sizeof(val) );
-			val[0] = val[5] = val[10] = val[15] = 1.0f;
+			memset(un.val, 0, sizeof(un.val) );
+			un.val[0] = un.val[5] = un.val[10] = un.val[15] = 1.0f;
 		}
 	};
 
-	void mtxMul(float* __restrict _result, const float* __restrict _a, const float* __restrict _b);
 	void mtxOrtho(float* _result, float _left, float _right, float _bottom, float _top, float _near, float _far);
 
 	struct MatrixCache
@@ -2416,7 +2420,7 @@ namespace bgfx
 
 			if (NULL != _view)
 			{
-				memcpy(m_view[_id].val, _view, sizeof(Matrix4) );
+				memcpy(m_view[_id].un.val, _view, sizeof(Matrix4) );
 			}
 			else
 			{
@@ -2425,7 +2429,7 @@ namespace bgfx
 
 			if (NULL != _proj)
 			{
-				memcpy(m_proj[_id].val, _proj, sizeof(Matrix4) );
+				memcpy(m_proj[_id].un.val, _proj, sizeof(Matrix4) );
 			}
 			else
 			{

src/image.cpp

@@ -4,7 +4,6 @@
  */
 
 #include "bgfx_p.h"
-#include <bx/float4_t.h>
 #include <math.h> // powf, sqrtf
 
 #include "image.h"

src/renderer_d3d11.cpp

@@ -176,6 +176,14 @@ namespace bgfx
 		},
 	};
 
+	static const Matrix4 s_bias =
+	{
+		0.5f, 0.0f, 0.0f, 0.0f,
+		0.0f, 0.5f, 0.0f, 0.0f,
+		0.0f, 0.0f, 0.5f, 0.0f,
+		0.5f, 0.5f, 0.5f, 1.0f,
+	};
+
 	struct TextureFormatInfo
 	{
 		DXGI_FORMAT m_fmt;
@@ -2206,7 +2214,7 @@ namespace bgfx
 		Matrix4 viewProj[BGFX_CONFIG_MAX_VIEWS];
 		for (uint32_t ii = 0; ii < BGFX_CONFIG_MAX_VIEWS; ++ii)
 		{
-			mtxMul(viewProj[ii].val, m_render->m_view[ii].val, m_render->m_proj[ii].val);
+			bx::float4x4_mul(&viewProj[ii].un.f4x4, &m_render->m_view[ii].un.f4x4, &m_render->m_proj[ii].un.f4x4);
 		}
 
 		bool wireframe = !!(m_render->m_debug&BGFX_DEBUG_WIREFRAME);
@@ -2445,20 +2453,20 @@ namespace bgfx
 
 						case PredefinedUniform::View:
 							{
-								s_renderCtx->setShaderConstant(flags, predefined.m_loc, m_render->m_view[view].val, bx::uint32_min(4, predefined.m_count) );
+								s_renderCtx->setShaderConstant(flags, predefined.m_loc, m_render->m_view[view].un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 
 						case PredefinedUniform::ViewProj:
 							{
-								s_renderCtx->setShaderConstant(flags, predefined.m_loc, viewProj[view].val, bx::uint32_min(4, predefined.m_count) );
+								s_renderCtx->setShaderConstant(flags, predefined.m_loc, viewProj[view].un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 
 						case PredefinedUniform::Model:
 							{
 								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
-								s_renderCtx->setShaderConstant(flags, predefined.m_loc, model.val, bx::uint32_min(state.m_num*4, predefined.m_count) );
+								s_renderCtx->setShaderConstant(flags, predefined.m_loc, model.un.val, bx::uint32_min(state.m_num*4, predefined.m_count) );
 							}
 							break;
 
@@ -2466,8 +2474,8 @@ namespace bgfx
 							{
 								Matrix4 modelView;
 								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
-								mtxMul(modelView.val, model.val, m_render->m_view[view].val);
-								s_renderCtx->setShaderConstant(flags, predefined.m_loc, modelView.val, bx::uint32_min(4, predefined.m_count) );
+								bx::float4x4_mul(&modelView.un.f4x4, &model.un.f4x4, &m_render->m_view[view].un.f4x4);
+								s_renderCtx->setShaderConstant(flags, predefined.m_loc, modelView.un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 
@@ -2475,8 +2483,8 @@ namespace bgfx
 							{
 								Matrix4 modelViewProj;
 								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
-								mtxMul(modelViewProj.val, model.val, viewProj[view].val);
-								s_renderCtx->setShaderConstant(flags, predefined.m_loc, modelViewProj.val, bx::uint32_min(4, predefined.m_count) );
+								bx::float4x4_mul(&modelViewProj.un.f4x4, &model.un.f4x4, &viewProj[view].un.f4x4);
+								s_renderCtx->setShaderConstant(flags, predefined.m_loc, modelViewProj.un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 
@@ -2484,40 +2492,24 @@ namespace bgfx
 							{
 								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
 
-								static const BX_ALIGN_STRUCT_16(float) s_bias[16] =
-								{
-									0.5f, 0.0f, 0.0f, 0.0f,
-									0.0f, 0.5f, 0.0f, 0.0f,
-									0.0f, 0.0f, 0.5f, 0.0f,
-									0.5f, 0.5f, 0.5f, 1.0f,
-								};
-
 								uint8_t other = m_render->m_other[view];
 								Matrix4 viewProjBias;
-								mtxMul(viewProjBias.val, viewProj[other].val, s_bias);
+								bx::float4x4_mul(&viewProjBias.un.f4x4, &viewProj[other].un.f4x4, &s_bias.un.f4x4);
 
 								Matrix4 modelViewProj;
-								mtxMul(modelViewProj.val, model.val, viewProjBias.val);
+								bx::float4x4_mul(&modelViewProj.un.f4x4, &model.un.f4x4, &viewProjBias.un.f4x4);
 
-								s_renderCtx->setShaderConstant(flags, predefined.m_loc, modelViewProj.val, bx::uint32_min(4, predefined.m_count) );
+								s_renderCtx->setShaderConstant(flags, predefined.m_loc, modelViewProj.un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 
 						case PredefinedUniform::ViewProjX:
 							{
-								static const BX_ALIGN_STRUCT_16(float) s_bias[16] =
-								{
-									0.5f, 0.0f, 0.0f, 0.0f,
-									0.0f, 0.5f, 0.0f, 0.0f,
-									0.0f, 0.0f, 0.5f, 0.0f,
-									0.5f, 0.5f, 0.5f, 1.0f,
-								};
-
 								uint8_t other = m_render->m_other[view];
 								Matrix4 viewProjBias;
-								mtxMul(viewProjBias.val, viewProj[other].val, s_bias);
+								bx::float4x4_mul(&viewProjBias.un.f4x4, &viewProj[other].un.f4x4, &s_bias.un.f4x4);
 
-								s_renderCtx->setShaderConstant(flags, predefined.m_loc, viewProjBias.val, bx::uint32_min(4, predefined.m_count) );
+								s_renderCtx->setShaderConstant(flags, predefined.m_loc, viewProjBias.un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 

src/renderer_d3d9.cpp

@@ -250,6 +250,14 @@ namespace bgfx
 		{ D3DFMT_RAWZ, D3DUSAGE_DEPTHSTENCIL, D3DRTYPE_SURFACE, false },
 	};
 
+	static const Matrix4 s_bias =
+	{
+		0.5f, 0.0f, 0.0f, 0.0f,
+		0.0f, 0.5f, 0.0f, 0.0f,
+		0.0f, 0.0f, 0.5f, 0.0f,
+		0.5f, 0.5f, 0.5f, 1.0f,
+	};
+
 	static const GUID IID_IDirect3D9         = { 0x81bdcbca, 0x64d4, 0x426d, { 0xae, 0x8d, 0xad, 0x1, 0x47, 0xf4, 0x27, 0x5c } };
 	static const GUID IID_IDirect3DDevice9Ex = { 0xb18b10ce, 0x2649, 0x405a, { 0x87, 0xf, 0x95, 0xf7, 0x77, 0xd4, 0x31, 0x3a } };
 
@@ -2245,7 +2253,7 @@ namespace bgfx
 		Matrix4 viewProj[BGFX_CONFIG_MAX_VIEWS];
 		for (uint32_t ii = 0; ii < BGFX_CONFIG_MAX_VIEWS; ++ii)
 		{
-			mtxMul(viewProj[ii].val, m_render->m_view[ii].val, m_render->m_proj[ii].val);
+			bx::float4x4_mul(&viewProj[ii].un.f4x4, &m_render->m_view[ii].un.f4x4, &m_render->m_proj[ii].un.f4x4);
 		}
 
 		DX_CHECK(device->SetRenderState(D3DRS_FILLMODE, m_render->m_debug&BGFX_DEBUG_WIREFRAME ? D3DFILL_WIREFRAME : D3DFILL_SOLID) );
@@ -2609,20 +2617,20 @@ namespace bgfx
 
 						case PredefinedUniform::View:
 							{
-								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, m_render->m_view[view].val, bx::uint32_min(4, predefined.m_count) );
+								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, m_render->m_view[view].un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 
 						case PredefinedUniform::ViewProj:
 							{
-								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, viewProj[view].val, bx::uint32_min(4, predefined.m_count) );
+								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, viewProj[view].un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 
 						case PredefinedUniform::Model:
 							{
  								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
-								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, model.val, bx::uint32_min(state.m_num*4, predefined.m_count) );
+								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, model.un.val, bx::uint32_min(state.m_num*4, predefined.m_count) );
 							}
 							break;
 
@@ -2630,8 +2638,8 @@ namespace bgfx
 							{
 								Matrix4 modelView;
 								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
-								mtxMul(modelView.val, model.val, m_render->m_view[view].val);
-								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, modelView.val, bx::uint32_min(4, predefined.m_count) );
+								bx::float4x4_mul(&modelView.un.f4x4, &model.un.f4x4, &m_render->m_view[view].un.f4x4);
+								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, modelView.un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 
@@ -2639,8 +2647,8 @@ namespace bgfx
 							{
 								Matrix4 modelViewProj;
 								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
-								mtxMul(modelViewProj.val, model.val, viewProj[view].val);
-								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, modelViewProj.val, bx::uint32_min(4, predefined.m_count) );
+								bx::float4x4_mul(&modelViewProj.un.f4x4, &model.un.f4x4, &viewProj[view].un.f4x4);
+								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, modelViewProj.un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 
@@ -2648,40 +2656,24 @@ namespace bgfx
 							{
 								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
 
-								static const BX_ALIGN_STRUCT_16(float) s_bias[16] =
-								{
-									0.5f, 0.0f, 0.0f, 0.0f,
-									0.0f, 0.5f, 0.0f, 0.0f,
-									0.0f, 0.0f, 0.5f, 0.0f,
-									0.5f, 0.5f, 0.5f, 1.0f,
-								};
-
 								uint8_t other = m_render->m_other[view];
 								Matrix4 viewProjBias;
-								mtxMul(viewProjBias.val, viewProj[other].val, s_bias);
+								bx::float4x4_mul(&viewProjBias.un.f4x4, &viewProj[other].un.f4x4, &s_bias.un.f4x4);
 
 								Matrix4 modelViewProj;
-								mtxMul(modelViewProj.val, model.val, viewProjBias.val);
+								bx::float4x4_mul(&modelViewProj.un.f4x4, &model.un.f4x4, &viewProjBias.un.f4x4);
 
-								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, modelViewProj.val, bx::uint32_min(4, predefined.m_count) );
+								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, modelViewProj.un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 
 						case PredefinedUniform::ViewProjX:
 							{
-								static const BX_ALIGN_STRUCT_16(float) s_bias[16] =
-								{
-									0.5f, 0.0f, 0.0f, 0.0f,
-									0.0f, 0.5f, 0.0f, 0.0f,
-									0.0f, 0.0f, 0.5f, 0.0f,
-									0.5f, 0.5f, 0.5f, 1.0f,
-								};
-
 								uint8_t other = m_render->m_other[view];
 								Matrix4 viewProjBias;
-								mtxMul(viewProjBias.val, viewProj[other].val, s_bias);
+								bx::float4x4_mul(&viewProjBias.un.f4x4, &viewProj[other].un.f4x4, &s_bias.un.f4x4);
 
-								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, viewProjBias.val, bx::uint32_min(4, predefined.m_count) );
+								s_renderCtx->setShaderConstantF(flags, predefined.m_loc, viewProjBias.un.val, bx::uint32_min(4, predefined.m_count) );
 							}
 							break;
 

src/renderer_gl.cpp

@@ -242,6 +242,14 @@ namespace bgfx
 		{ GL_STENCIL_INDEX8,                           GL_DEPTH_STENCIL,                            GL_UNSIGNED_BYTE,               false }, // D0S8
 	};
 
+	static const Matrix4 s_bias =
+	{
+		0.5f, 0.0f, 0.0f, 0.0f,
+		0.0f, 0.5f, 0.0f, 0.0f,
+		0.0f, 0.0f, 0.5f, 0.0f,
+		0.5f, 0.5f, 0.5f, 1.0f,
+	};
+
 	struct Extension
 	{
 		enum Enum
@@ -3087,7 +3095,7 @@ namespace bgfx
 		Matrix4 viewProj[BGFX_CONFIG_MAX_VIEWS];
 		for (uint32_t ii = 0; ii < BGFX_CONFIG_MAX_VIEWS; ++ii)
 		{
-			mtxMul(viewProj[ii].val, m_render->m_view[ii].val, m_render->m_proj[ii].val);
+			float4x4_mul(&viewProj[ii].un.f4x4, &m_render->m_view[ii].un.f4x4, &m_render->m_proj[ii].un.f4x4);
 		}
 
 		uint16_t programIdx = invalidHandle;
@@ -3422,7 +3430,7 @@ namespace bgfx
 								GL_CHECK(glUniformMatrix4fv(predefined.m_loc
 									, 1
 									, GL_FALSE
-									, m_render->m_view[view].val
+									, m_render->m_view[view].un.val
 									) );
 							}
 							break;
@@ -3432,7 +3440,7 @@ namespace bgfx
 								GL_CHECK(glUniformMatrix4fv(predefined.m_loc
 									, 1
 									, GL_FALSE
-									, viewProj[view].val
+									, viewProj[view].un.val
 									) );
 							}
 							break;
@@ -3443,7 +3451,7 @@ namespace bgfx
 								GL_CHECK(glUniformMatrix4fv(predefined.m_loc
 									, bx::uint32_min(predefined.m_count, state.m_num)
 									, GL_FALSE
-									, model.val
+									, model.un.val
 									) );
 							}
 							break;
@@ -3452,12 +3460,12 @@ namespace bgfx
 							{
 								Matrix4 modelView;
 								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
-								mtxMul(modelView.val, model.val, m_render->m_view[view].val);
+								bx::float4x4_mul(&modelView.un.f4x4, &model.un.f4x4, &m_render->m_view[view].un.f4x4);
 
 								GL_CHECK(glUniformMatrix4fv(predefined.m_loc
 									, 1
 									, GL_FALSE
-									, modelView.val
+									, modelView.un.val
 									) );
 							}
 							break;
@@ -3466,12 +3474,12 @@ namespace bgfx
 							{
 								Matrix4 modelViewProj;
 								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
-								mtxMul(modelViewProj.val, model.val, viewProj[view].val);
+								bx::float4x4_mul(&modelViewProj.un.f4x4, &model.un.f4x4, &viewProj[view].un.f4x4);
 
 								GL_CHECK(glUniformMatrix4fv(predefined.m_loc
 									, 1
 									, GL_FALSE
-									, modelViewProj.val
+									, modelViewProj.un.val
 									) );
 							}
 							break;
@@ -3480,47 +3488,31 @@ namespace bgfx
 							{
 								const Matrix4& model = m_render->m_matrixCache.m_cache[state.m_matrix];
 
-								static const BX_ALIGN_STRUCT_16(float) s_bias[16] =
-								{
-									0.5f, 0.0f, 0.0f, 0.0f,
-									0.0f, 0.5f, 0.0f, 0.0f,
-									0.0f, 0.0f, 0.5f, 0.0f,
-									0.5f, 0.5f, 0.5f, 1.0f,
-								};
-
 								uint8_t other = m_render->m_other[view];
 								Matrix4 viewProjBias;
-								mtxMul(viewProjBias.val, viewProj[other].val, s_bias);
+								bx::float4x4_mul(&viewProjBias.un.f4x4, &viewProj[other].un.f4x4, &s_bias.un.f4x4);
 
 								Matrix4 modelViewProj;
-								mtxMul(modelViewProj.val, model.val, viewProjBias.val);
+								bx::float4x4_mul(&modelViewProj.un.f4x4, &model.un.f4x4, &viewProjBias.un.f4x4);
 
 								GL_CHECK(glUniformMatrix4fv(predefined.m_loc
 									, 1
 									, GL_FALSE
-									, modelViewProj.val
+									, modelViewProj.un.val
 									) );
 							}
 							break;
 
 						case PredefinedUniform::ViewProjX:
 							{
-								static const BX_ALIGN_STRUCT_16(float) s_bias[16] =
-								{
-									0.5f, 0.0f, 0.0f, 0.0f,
-									0.0f, 0.5f, 0.0f, 0.0f,
-									0.0f, 0.0f, 0.5f, 0.0f,
-									0.5f, 0.5f, 0.5f, 1.0f,
-								};
-
 								uint8_t other = m_render->m_other[view];
 								Matrix4 viewProjBias;
-								mtxMul(viewProjBias.val, viewProj[other].val, s_bias);
+								bx::float4x4_mul(&viewProjBias.un.f4x4, &viewProj[other].un.f4x4, &s_bias.un.f4x4);
 
 								GL_CHECK(glUniformMatrix4fv(predefined.m_loc
 									, 1
 									, GL_FALSE
-									, viewProjBias.val
+									, viewProjBias.un.val
 									) );
 							}
 							break;