diff --git a/etc/shaders/lovr.glsl b/etc/shaders/lovr.glsl
index 370f3b5b..ba0e7748 100644
--- a/etc/shaders/lovr.glsl
+++ b/etc/shaders/lovr.glsl
@@ -28,14 +28,13 @@ struct Camera {
 };
 
 struct Draw {
-  mat4 transform;
-  mat3 normalMatrix;
+  mat4x3 transform;
   vec4 color;
 };
 
 layout(set = 0, binding = 0) uniform Globals { vec2 Resolution; float Time; };
 layout(set = 0, binding = 1) uniform CameraBuffer { Camera Cameras[6]; };
-layout(set = 0, binding = 2) uniform DrawBuffer { Draw Draws[256]; };
+layout(set = 0, binding = 2) uniform DrawBuffer { layout(row_major) Draw Draws[256]; };
 layout(set = 0, binding = 3) uniform sampler Sampler;
 
 layout(set = 1, binding = 0) uniform MaterialBuffer {
@@ -154,8 +153,8 @@ layout(location = 14) in vec3 Tangent;
 
 #ifdef GL_VERTEX_SHADER
 #define DrawID gl_BaseInstance
-#define Transform Draws[DrawID].transform
-#define NormalMatrix Draws[DrawID].normalMatrix
+#define Transform mat4(Draws[DrawID].transform)
+#define NormalMatrix (cofactor3(Draws[DrawID].transform))
 #define PassColor Draws[DrawID].color
 #define ClipFromLocal (ViewProjection * Transform)
 #define ClipFromWorld (ViewProjection)
@@ -167,6 +166,20 @@ layout(location = 14) in vec3 Tangent;
 #define WorldFromView (inverse(View))
 #define WorldFromClip (inverse(ViewProjection))
 #define DefaultPosition (ClipFromLocal * VertexPosition)
+
+mat3 cofactor3(mat4x3 m) {
+  return mat3(vec3(
+     (m[1][1] * m[2][2] - m[2][1] * m[1][2]),
+    -(m[1][0] * m[2][2] - m[2][0] * m[1][2]),
+     (m[1][0] * m[2][1] - m[2][0] * m[1][1])), vec3(
+    -(m[0][1] * m[2][2] - m[2][1] * m[0][2]),
+     (m[0][0] * m[2][2] - m[2][0] * m[0][2]),
+    -(m[0][0] * m[2][1] - m[2][0] * m[0][1])), vec3(
+     (m[0][1] * m[1][2] - m[1][1] * m[0][2]),
+    -(m[0][0] * m[1][2] - m[1][0] * m[0][2]),
+     (m[0][0] * m[1][1] - m[1][0] * m[0][1])
+  ));
+}
 #endif
 
 #ifdef GL_FRAGMENT_SHADER
diff --git a/src/core/maf.h b/src/core/maf.h
index da52a931..0949d4ee 100644
--- a/src/core/maf.h
+++ b/src/core/maf.h
@@ -616,30 +616,6 @@ MAF mat4 mat4_invert(mat4 m) {
   return m;
 }
 
-MAF mat4 mat4_cofactor(mat4 m) {
-  float m00 = m[0], m04 = m[4], m08 = m[8], m12 = m[12];
-  float m01 = m[1], m05 = m[5], m09 = m[9], m13 = m[13];
-  float m02 = m[2], m06 = m[6], m10 = m[10], m14 = m[14];
-  float m03 = m[3], m07 = m[7], m11 = m[11], m15 = m[15];
-  m[0]  =  (m05 * (m10 * m15 - m11 * m14) - m09 * (m06 * m15 - m07 * m14) + m13 * (m06 * m11 - m07 * m10));
-  m[1]  = -(m04 * (m10 * m15 - m11 * m14) - m08 * (m06 * m15 - m07 * m14) + m12 * (m06 * m11 - m07 * m10));
-  m[2]  =  (m04 * (m09 * m15 - m11 * m13) - m08 * (m05 * m15 - m07 * m13) + m12 * (m05 * m11 - m07 * m09));
-  m[3]  = -(m04 * (m09 * m14 - m10 * m13) - m08 * (m05 * m14 - m06 * m13) + m12 * (m05 * m10 - m06 * m09));
-  m[4]  = -(m01 * (m10 * m15 - m11 * m14) - m09 * (m02 * m15 - m03 * m14) + m13 * (m02 * m11 - m03 * m10));
-  m[5]  =  (m00 * (m10 * m15 - m11 * m14) - m08 * (m02 * m15 - m03 * m14) + m12 * (m02 * m11 - m03 * m10));
-  m[6]  = -(m00 * (m09 * m15 - m11 * m13) - m08 * (m01 * m15 - m03 * m13) + m12 * (m01 * m11 - m03 * m09));
-  m[7]  =  (m00 * (m09 * m14 - m10 * m13) - m08 * (m01 * m14 - m02 * m13) + m12 * (m01 * m10 - m02 * m09));
-  m[8]  =  (m01 * (m06 * m15 - m07 * m14) - m05 * (m02 * m15 - m03 * m14) + m13 * (m02 * m07 - m03 * m06));
-  m[9]  = -(m00 * (m06 * m15 - m07 * m14) - m04 * (m02 * m15 - m03 * m14) + m12 * (m02 * m07 - m03 * m06));
-  m[10] =  (m00 * (m05 * m15 - m07 * m13) - m04 * (m01 * m15 - m03 * m13) + m12 * (m01 * m07 - m03 * m05));
-  m[11] = -(m00 * (m05 * m14 - m06 * m13) - m04 * (m01 * m14 - m02 * m13) + m12 * (m01 * m06 - m02 * m05));
-  m[12] = -(m01 * (m06 * m11 - m07 * m10) - m05 * (m02 * m11 - m03 * m10) + m09 * (m02 * m07 - m03 * m06));
-  m[13] =  (m00 * (m06 * m11 - m07 * m10) - m04 * (m02 * m11 - m03 * m10) + m08 * (m02 * m07 - m03 * m06));
-  m[14] = -(m00 * (m05 * m11 - m07 * m09) - m04 * (m01 * m11 - m03 * m09) + m08 * (m01 * m07 - m03 * m05));
-  m[15] =  (m00 * (m05 * m10 - m06 * m09) - m04 * (m01 * m10 - m02 * m09) + m08 * (m01 * m06 - m02 * m05));
-  return m;
-}
-
 // Calculate matrix equivalent to "apply n, then m"
 MAF mat4 mat4_mul(mat4 m, mat4 n) {
   float m00 = m[0], m01 = m[1], m02 = m[2], m03 = m[3],
diff --git a/src/core/spv.c b/src/core/spv.c
index 1e3279bb..f55a3a75 100644
--- a/src/core/spv.c
+++ b/src/core/spv.c
@@ -610,8 +610,8 @@ static spv_result spv_parse_field(spv_context* spv, const uint32_t* word, spv_fi
   }
 
   if (OP_CODE(word) == 22 && word[2] == 32) { // OpTypeFloat
-    if (columnCount >= 2 && columnCount <= 4 && componentCount == columnCount) {
-      field->type = SPV_MAT2 + columnCount - 2;
+    if (columnCount >= 2 && columnCount <= 4 && componentCount >= 2 && componentCount <= 4) {
+      field->type = SPV_MAT2x2 + (columnCount - 2) * 3 + (componentCount - 2);
     } else if (columnCount == 1 && componentCount >= 2 && componentCount <= 4) {
       field->type = SPV_F32x2 + componentCount - 2;
     } else if (columnCount == 1 && componentCount == 1) {
diff --git a/src/core/spv.h b/src/core/spv.h
index c5c87ba0..c08a29be 100644
--- a/src/core/spv.h
+++ b/src/core/spv.h
@@ -17,9 +17,15 @@ typedef enum {
   SPV_F32x2,
   SPV_F32x3,
   SPV_F32x4,
-  SPV_MAT2,
-  SPV_MAT3,
-  SPV_MAT4,
+  SPV_MAT2x2,
+  SPV_MAT2x3,
+  SPV_MAT2x4,
+  SPV_MAT3x2,
+  SPV_MAT3x3,
+  SPV_MAT3x4,
+  SPV_MAT4x2,
+  SPV_MAT4x3,
+  SPV_MAT4x4,
   SPV_STRUCT
 } spv_type;
 
diff --git a/src/modules/graphics/graphics.c b/src/modules/graphics/graphics.c
index 1b80b6f6..68543e32 100644
--- a/src/modules/graphics/graphics.c
+++ b/src/modules/graphics/graphics.c
@@ -194,8 +194,7 @@ struct Mesh {
 };
 
 typedef struct {
-  float transform[16];
-  float cofactor[12];
+  float transform[12];
   float color[4];
 } DrawData;
 
@@ -1277,27 +1276,29 @@ static void recordRenderPass(Pass* pass, gpu_stream* stream) {
 
   // DrawData
 
-  uint32_t drawPageCount = (activeDrawCount + 255) / 256;
-  uint32_t drawPageSize = (uint32_t) ALIGN(256 * sizeof(DrawData), align);
-  mapped = mapBuffer(&state.streamBuffers, drawPageCount * drawPageSize, align);
-  builtins[2].buffer = (gpu_buffer_binding) { mapped.buffer, mapped.offset, drawPageSize };
+  mapped = mapBuffer(&state.streamBuffers, activeDrawCount * sizeof(DrawData), align);
+  builtins[2].buffer = (gpu_buffer_binding) { mapped.buffer, mapped.offset, 256 * sizeof(DrawData) };
   DrawData* data = mapped.pointer;
 
   for (uint32_t i = 0; i < activeDrawCount; i++, data++) {
     Draw* draw = &pass->draws[activeDraws[i] >> 8][activeDraws[i] & 0xff];
-
-    if ((i & 0xff) == 0) {
-      data = (DrawData*) ALIGN(data, align);
-    }
-
-    float cofactor[16];
-    mat4_init(cofactor, draw->transform);
-    memcpy(cofactor + 12, (float[4]) { 0.f, 0.f, 0.f, 1.f }, 4 * sizeof(float));
-    mat4_cofactor(cofactor);
-
-    memcpy(data->transform, draw->transform, 16 * sizeof(float));
-    memcpy(data->cofactor, cofactor, 12 * sizeof(float));
-    memcpy(data->color, draw->color, 4 * sizeof(float));
+    // transform is provided as 4x3 row-major matrix for packing reasons, need to transpose
+    data->transform[0] = draw->transform[0];
+    data->transform[1] = draw->transform[4];
+    data->transform[2] = draw->transform[8];
+    data->transform[3] = draw->transform[12];
+    data->transform[4] = draw->transform[1];
+    data->transform[5] = draw->transform[5];
+    data->transform[6] = draw->transform[9];
+    data->transform[7] = draw->transform[13];
+    data->transform[8] = draw->transform[2];
+    data->transform[9] = draw->transform[6];
+    data->transform[10] = draw->transform[10];
+    data->transform[11] = draw->transform[14];
+    data->color[0] = draw->color[0];
+    data->color[1] = draw->color[1];
+    data->color[2] = draw->color[2];
+    data->color[3] = draw->color[3];
   }
 
   gpu_bundle_write(&builtinBundle, &builtinfo, 1);
@@ -1426,7 +1427,7 @@ static void recordRenderPass(Pass* pass, gpu_stream* stream) {
     }
 
     if ((i & 0xff) == 0 || draw->camera != cameraIndex || constantsDirty) {
-      uint32_t dynamicOffsets[] = { draw->camera * canvas->views * sizeof(Camera), (i >> 8) * drawPageSize };
+      uint32_t dynamicOffsets[] = { draw->camera * canvas->views * sizeof(Camera), (i >> 8) * 256 * sizeof(DrawData) };
       gpu_bind_bundles(stream, draw->shader->gpu, &builtinBundle, 0, 1, dynamicOffsets, COUNTOF(dynamicOffsets));
       cameraIndex = draw->camera;
     }
@@ -2792,9 +2793,16 @@ Shader* lovrShaderCreate(const ShaderInfo* info) {
         [SPV_F32x2] = TYPE_F32x2,
         [SPV_F32x3] = TYPE_F32x3,
         [SPV_F32x4] = TYPE_F32x4,
-        [SPV_MAT2] = TYPE_MAT2,
-        [SPV_MAT3] = TYPE_MAT3,
-        [SPV_MAT4] = TYPE_MAT4
+        [SPV_MAT2x2] = TYPE_MAT2,
+        [SPV_MAT2x3] = ~0u,
+        [SPV_MAT2x4] = ~0u,
+        [SPV_MAT3x2] = ~0u,
+        [SPV_MAT3x3] = TYPE_MAT3,
+        [SPV_MAT3x4] = ~0u,
+        [SPV_MAT4x2] = ~0u,
+        [SPV_MAT4x3] = ~0u,
+        [SPV_MAT4x4] = TYPE_MAT4,
+        [SPV_STRUCT] = ~0u
       };
 
       spv_field* field = &spv[s].fields[i];
@@ -2802,7 +2810,7 @@ Shader* lovrShaderCreate(const ShaderInfo* info) {
       uint32_t base = s == 1 ? spv[0].fieldCount : 0;
 
       shader->fields[base + i] = (DataField) {
-        .type = field->type == SPV_STRUCT ? ~0u : dataTypes[field->type],
+        .type = dataTypes[field->type],
         .offset = field->offset,
         .length = field->arrayLength,
         .stride = field->arrayStride,