Transform: support multi-threading for most modes

Multi-threading support for transform modes: bevel-weight, crease,
push-pull, rotate, shear, shrink-fatten, skin-resize, to-sphere,
trackball & translate.

This is done using a parallel loop over transform data.

From testing a 1.5million polygon mesh on a 32 core system
the overall performance gains were between ~20-28%

To ensure the code is thread-safe arguments to shared data are const.

Reviewed By: mano-wii

12 files changed, 934 insertions, 299 deletions

diff --git a/source/blender/editors/transform/transform_data.h b/source/blender/editors/transform/transform_data.h
index 59b76c2eec5..606453e356b 100644
--- a/source/blender/editors/transform/transform_data.h
+++ b/source/blender/editors/transform/transform_data.h
@@ -149,6 +149,8 @@ typedef struct TransData {
   short protectflag;
 } TransData;
 
+#define TRANSDATA_THREAD_LIMIT 1024
+
 /** #TransData.flag */
 enum {
   TD_SELECTED = 1 << 0,
diff --git a/source/blender/editors/transform/transform_mode_edge_bevelweight.c b/source/blender/editors/transform/transform_mode_edge_bevelweight.c
index 3ce52ed3296..425bfec241e 100644
--- a/source/blender/editors/transform/transform_mode_edge_bevelweight.c
+++ b/source/blender/editors/transform/transform_mode_edge_bevelweight.c
@@ -25,6 +25,7 @@
 
 #include "BLI_math.h"
 #include "BLI_string.h"
+#include "BLI_task.h"
 
 #include "BKE_context.h"
 #include "BKE_unit.h"
@@ -40,6 +41,50 @@
 #include "transform_snap.h"
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (Bevel Weight) Element
+ * \{ */
+
+/**
+ * \note Small arrays / data-structures should be stored copied for faster memory access.
+ */
+struct TransDataArgs_BevelWeight {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+  float weight;
+};
+
+static void transdata_elem_bevel_weight(const TransInfo *UNUSED(t),
+                                        const TransDataContainer *UNUSED(tc),
+                                        TransData *td,
+                                        const float weight)
+{
+  if (td->val == NULL) {
+    return;
+  }
+  *td->val = td->ival + weight * td->factor;
+  if (*td->val < 0.0f) {
+    *td->val = 0.0f;
+  }
+  if (*td->val > 1.0f) {
+    *td->val = 1.0f;
+  }
+}
+
+static void transdata_elem_bevel_weight_fn(void *__restrict iter_data_v,
+                                           const int iter,
+                                           const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  struct TransDataArgs_BevelWeight *data = iter_data_v;
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  transdata_elem_bevel_weight(data->t, data->tc, td, data->weight);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (Bevel Weight)
  * \{ */
 
@@ -83,18 +128,25 @@ static void applyBevelWeight(TransInfo *t, const int UNUSED(mval[2]))
   }
 
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *td = tc->data;
-    for (i = 0; i < tc->data_len; i++, td++) {
-      if (td->val) {
-        *td->val = td->ival + weight * td->factor;
-        if (*td->val < 0.0f) {
-          *td->val = 0.0f;
-        }
-        if (*td->val > 1.0f) {
-          *td->val = 1.0f;
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
         }
+        transdata_elem_bevel_weight(t, tc, td, weight);
       }
     }
+    else {
+      struct TransDataArgs_BevelWeight data = {
+          .t = t,
+          .tc = tc,
+          .weight = weight,
+      };
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      BLI_task_parallel_range(0, tc->data_len, &data, transdata_elem_bevel_weight_fn, &settings);
+    }
   }
 
   recalcData(t);
diff --git a/source/blender/editors/transform/transform_mode_edge_crease.c b/source/blender/editors/transform/transform_mode_edge_crease.c
index 23fa20b68ff..91e2507e544 100644
--- a/source/blender/editors/transform/transform_mode_edge_crease.c
+++ b/source/blender/editors/transform/transform_mode_edge_crease.c
@@ -25,6 +25,7 @@
 
 #include "BLI_math.h"
 #include "BLI_string.h"
+#include "BLI_task.h"
 
 #include "BKE_context.h"
 #include "BKE_unit.h"
@@ -40,6 +41,51 @@
 #include "transform_snap.h"
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (Crease) Element
+ * \{ */
+
+/**
+ * \note Small arrays / data-structures should be stored copied for faster memory access.
+ */
+struct TransDataArgs_Crease {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+  float crease;
+};
+
+static void transdata_elem_crease(const TransInfo *UNUSED(t),
+                                  const TransDataContainer *UNUSED(tc),
+                                  TransData *td,
+                                  const float crease)
+{
+  if (td->val == NULL) {
+    return;
+  }
+
+  *td->val = td->ival + crease * td->factor;
+  if (*td->val < 0.0f) {
+    *td->val = 0.0f;
+  }
+  if (*td->val > 1.0f) {
+    *td->val = 1.0f;
+  }
+}
+
+static void transdata_elem_crease_fn(void *__restrict iter_data_v,
+                                     const int iter,
+                                     const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  struct TransDataArgs_Crease *data = iter_data_v;
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  transdata_elem_crease(data->t, data->tc, td, data->crease);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (Crease)
  * \{ */
 
@@ -83,22 +129,25 @@ static void applyCrease(TransInfo *t, const int UNUSED(mval[2]))
   }
 
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *td = tc->data;
-    for (i = 0; i < tc->data_len; i++, td++) {
-      if (td->flag & TD_SKIP) {
-        continue;
-      }
-
-      if (td->val) {
-        *td->val = td->ival + crease * td->factor;
-        if (*td->val < 0.0f) {
-          *td->val = 0.0f;
-        }
-        if (*td->val > 1.0f) {
-          *td->val = 1.0f;
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
         }
+        transdata_elem_crease(t, tc, td, crease);
       }
     }
+    else {
+      struct TransDataArgs_Crease data = {
+          .t = t,
+          .tc = tc,
+          .crease = crease,
+      };
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      BLI_task_parallel_range(0, tc->data_len, &data, transdata_elem_crease_fn, &settings);
+    }
   }
 
   recalcData(t);
@@ -124,4 +173,5 @@ void initCrease(TransInfo *t)
 
   t->flag |= T_NO_CONSTRAINT | T_NO_PROJECT;
 }
+
 /** \} */
diff --git a/source/blender/editors/transform/transform_mode_push_pull.c b/source/blender/editors/transform/transform_mode_push_pull.c
index b08e479a3d0..098698b0e79 100644
--- a/source/blender/editors/transform/transform_mode_push_pull.c
+++ b/source/blender/editors/transform/transform_mode_push_pull.c
@@ -25,6 +25,7 @@
 
 #include "BLI_math.h"
 #include "BLI_string.h"
+#include "BLI_task.h"
 
 #include "BKE_context.h"
 #include "BKE_unit.h"
@@ -41,12 +42,75 @@
 #include "transform_snap.h"
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (Push/Pull) Element
+ * \{ */
+
+/**
+ * \note Small arrays / data-structures should be stored copied for faster memory access.
+ */
+struct TransDataArgs_PushPull {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+
+  float distance;
+  const float axis_global[3];
+  bool is_data_space;
+};
+
+static void transdata_elem_push_pull(const TransInfo *t,
+                                     const TransDataContainer *tc,
+                                     TransData *td,
+                                     const float distance,
+                                     const float axis_global[3],
+                                     const bool is_data_space)
+{
+  float vec[3];
+  sub_v3_v3v3(vec, tc->center_local, td->center);
+  if (t->con.applyRot && t->con.mode & CON_APPLY) {
+    float axis[3];
+    copy_v3_v3(axis, axis_global);
+    t->con.applyRot(t, tc, td, axis, NULL);
+
+    mul_m3_v3(td->smtx, axis);
+    if (isLockConstraint(t)) {
+      float dvec[3];
+      project_v3_v3v3(dvec, vec, axis);
+      sub_v3_v3(vec, dvec);
+    }
+    else {
+      project_v3_v3v3(vec, vec, axis);
+    }
+  }
+  normalize_v3_length(vec, distance * td->factor);
+  if (is_data_space) {
+    mul_m3_v3(td->smtx, vec);
+  }
+
+  add_v3_v3v3(td->loc, td->iloc, vec);
+}
+
+static void transdata_elem_push_pull_fn(void *__restrict iter_data_v,
+                                        const int iter,
+                                        const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  struct TransDataArgs_PushPull *data = iter_data_v;
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  transdata_elem_push_pull(
+      data->t, data->tc, td, data->distance, data->axis_global, data->is_data_space);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (Push/Pull)
  * \{ */
 
 static void applyPushPull(TransInfo *t, const int UNUSED(mval[2]))
 {
-  float vec[3], axis_global[3];
+  float axis_global[3];
   float distance;
   int i;
   char str[UI_MAX_DRAW_STR];
@@ -80,34 +144,26 @@ static void applyPushPull(TransInfo *t, const int UNUSED(mval[2]))
   const bool is_data_space = (t->options & CTX_POSE_BONE) != 0;
 
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *td = tc->data;
-    for (i = 0; i < tc->data_len; i++, td++) {
-      if (td->flag & TD_SKIP) {
-        continue;
-      }
-
-      sub_v3_v3v3(vec, tc->center_local, td->center);
-      if (t->con.applyRot && t->con.mode & CON_APPLY) {
-        float axis[3];
-        copy_v3_v3(axis, axis_global);
-        t->con.applyRot(t, tc, td, axis, NULL);
-
-        mul_m3_v3(td->smtx, axis);
-        if (isLockConstraint(t)) {
-          float dvec[3];
-          project_v3_v3v3(dvec, vec, axis);
-          sub_v3_v3(vec, dvec);
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
         }
-        else {
-          project_v3_v3v3(vec, vec, axis);
-        }
-      }
-      normalize_v3_length(vec, distance * td->factor);
-      if (is_data_space) {
-        mul_m3_v3(td->smtx, vec);
-      }
 
-      add_v3_v3v3(td->loc, td->iloc, vec);
+        transdata_elem_push_pull(t, tc, td, distance, axis_global, is_data_space);
+      }
+    }
+    else {
+      struct TransDataArgs_PushPull data = {
+          .t = t,
+          .tc = tc,
+          .axis_global = {UNPACK3(axis_global)},
+          .is_data_space = is_data_space,
+      };
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      BLI_task_parallel_range(0, tc->data_len, &data, transdata_elem_push_pull_fn, &settings);
     }
   }
 
diff --git a/source/blender/editors/transform/transform_mode_resize.c b/source/blender/editors/transform/transform_mode_resize.c
index 6af16dee9b7..65f4623b3be 100644
--- a/source/blender/editors/transform/transform_mode_resize.c
+++ b/source/blender/editors/transform/transform_mode_resize.c
@@ -24,6 +24,7 @@
 #include <stdlib.h>
 
 #include "BLI_math.h"
+#include "BLI_task.h"
 
 #include "BKE_context.h"
 #include "BKE_unit.h"
@@ -39,6 +40,30 @@
 #include "transform_snap.h"
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (Resize) Element
+ * \{ */
+
+struct ElemResizeData {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+  float mat[3][3];
+};
+
+static void element_resize_fn(void *__restrict iter_data_v,
+                              const int iter,
+                              const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  struct ElemResizeData *data = iter_data_v;
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  ElementResize(data->t, data->tc, td, data->mat);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (Resize)
  * \{ */
 
@@ -123,13 +148,27 @@ static void applyResize(TransInfo *t, const int UNUSED(mval[2]))
   copy_m3_m3(t->mat, mat); /* used in gizmo */
 
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *td = tc->data;
-    for (i = 0; i < tc->data_len; i++, td++) {
-      if (td->flag & TD_SKIP) {
-        continue;
-      }
 
-      ElementResize(t, tc, td, mat);
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
+        }
+
+        ElementResize(t, tc, td, mat);
+      }
+    }
+    else {
+      struct ElemResizeData data = {
+          .t = t,
+          .tc = tc,
+      };
+      copy_m3_m3(data.mat, mat);
+
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      BLI_task_parallel_range(0, tc->data_len, &data, element_resize_fn, &settings);
     }
   }
 
diff --git a/source/blender/editors/transform/transform_mode_rotate.c b/source/blender/editors/transform/transform_mode_rotate.c
index 066cdf20e93..44a29cfac45 100644
--- a/source/blender/editors/transform/transform_mode_rotate.c
+++ b/source/blender/editors/transform/transform_mode_rotate.c
@@ -24,6 +24,7 @@
 #include <stdlib.h>
 
 #include "BLI_math.h"
+#include "BLI_task.h"
 
 #include "BKE_context.h"
 #include "BKE_unit.h"
@@ -37,6 +38,140 @@
 #include "transform_snap.h"
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (Rotation) Matrix Cache
+ * \{ */
+
+struct RotateMatrixCache {
+  /**
+   * Counter for needed updates (when we need to update to non-default matrix,
+   * we also need another update on next iteration to go back to default matrix,
+   * hence the '2' value used here, instead of a mere boolean).
+   */
+  short do_update_matrix;
+  float mat[3][3];
+};
+
+static void rmat_cache_init(struct RotateMatrixCache *rmc, const float angle, const float axis[3])
+{
+  axis_angle_normalized_to_mat3(rmc->mat, axis, angle);
+  rmc->do_update_matrix = 0;
+}
+
+static void rmat_cache_reset(struct RotateMatrixCache *rmc)
+{
+  rmc->do_update_matrix = 2;
+}
+
+static void rmat_cache_update(struct RotateMatrixCache *rmc,
+                              const float axis[3],
+                              const float angle)
+{
+  if (rmc->do_update_matrix > 0) {
+    axis_angle_normalized_to_mat3(rmc->mat, axis, angle);
+    rmc->do_update_matrix--;
+  }
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
+/** \name Transform (Rotation) Element
+ * \{ */
+
+/**
+ * \note Small arrays / data-structures should be stored copied for faster memory access.
+ */
+struct TransDataArgs_Rotate {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+  const float axis[3];
+  float angle;
+  float angle_step;
+  bool is_large_rotation;
+};
+
+struct TransDataArgs_RotateTLS {
+  struct RotateMatrixCache rmc;
+};
+
+static void transdata_elem_rotate(const TransInfo *t,
+                                  const TransDataContainer *tc,
+                                  TransData *td,
+                                  const float axis[3],
+                                  const float angle,
+                                  const float angle_step,
+                                  const bool is_large_rotation,
+                                  struct RotateMatrixCache *rmc)
+{
+  float axis_buffer[3];
+  const float *axis_final = axis;
+
+  float angle_final = angle;
+  if (t->con.applyRot) {
+    copy_v3_v3(axis_buffer, axis);
+    axis_final = axis_buffer;
+    t->con.applyRot(t, tc, td, axis_buffer, NULL);
+    angle_final = angle * td->factor;
+    /* Even though final angle might be identical to orig value,
+     * we have to update the rotation matrix in that case... */
+    rmat_cache_reset(rmc);
+  }
+  else if (t->flag & T_PROP_EDIT) {
+    angle_final = angle * td->factor;
+  }
+
+  /* Rotation is very likely to be above 180°, we need to do rotation by steps.
+   * Note that this is only needed when doing 'absolute' rotation
+   * (i.e. from initial rotation again, typically when using numinput).
+   * regular incremental rotation (from mouse/widget/...) will be called often enough,
+   * hence steps are small enough to be properly handled without that complicated trick.
+   * Note that we can only do that kind of stepped rotation if we have initial rotation values
+   * (and access to some actual rotation value storage).
+   * Otherwise, just assume it's useless (e.g. in case of mesh/UV/etc. editing).
+   * Also need to be in Euler rotation mode, the others never allow more than one turn anyway.
+   */
+  if (is_large_rotation && td->ext != NULL && td->ext->rotOrder == ROT_MODE_EUL) {
+    copy_v3_v3(td->ext->rot, td->ext->irot);
+    for (float angle_progress = angle_step; fabsf(angle_progress) < fabsf(angle_final);
+         angle_progress += angle_step) {
+      axis_angle_normalized_to_mat3(rmc->mat, axis_final, angle_progress);
+      ElementRotation(t, tc, td, rmc->mat, t->around);
+    }
+    rmat_cache_reset(rmc);
+  }
+  else if (angle_final != angle) {
+    rmat_cache_reset(rmc);
+  }
+
+  rmat_cache_update(rmc, axis_final, angle_final);
+
+  ElementRotation(t, tc, td, rmc->mat, t->around);
+}
+
+static void transdata_elem_rotate_fn(void *__restrict iter_data_v,
+                                     const int iter,
+                                     const TaskParallelTLS *__restrict tls)
+{
+  struct TransDataArgs_Rotate *data = iter_data_v;
+  struct TransDataArgs_RotateTLS *tls_data = tls->userdata_chunk;
+
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  transdata_elem_rotate(data->t,
+                        data->tc,
+                        td,
+                        data->axis,
+                        data->angle,
+                        data->angle_step,
+                        data->is_large_rotation,
+                        &tls_data->rmc);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (Rotation)
  * \{ */
 
@@ -115,12 +250,9 @@ static float large_rotation_limit(float angle)
 
 static void applyRotationValue(TransInfo *t,
                                float angle,
-                               float axis[3],
+                               const float axis[3],
                                const bool is_large_rotation)
 {
-  float mat[3][3];
-  int i;
-
   const float angle_sign = angle < 0.0f ? -1.0f : 1.0f;
   /* We cannot use something too close to 180°, or 'continuous' rotation may fail
    * due to computing error... */
@@ -132,60 +264,37 @@ static void applyRotationValue(TransInfo *t,
     angle = large_rotation_limit(angle);
   }
 
-  axis_angle_normalized_to_mat3(mat, axis, angle);
-  /* Counter for needed updates (when we need to update to non-default matrix,
-   * we also need another update on next iteration to go back to default matrix,
-   * hence the '2' value used here, instead of a mere boolean). */
-  short do_update_matrix = 0;
+  struct RotateMatrixCache rmc = {0};
+  rmat_cache_init(&rmc, angle, axis);
 
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *td = tc->data;
-    for (i = 0; i < tc->data_len; i++, td++) {
-      if (td->flag & TD_SKIP) {
-        continue;
-      }
-
-      float angle_final = angle;
-      if (t->con.applyRot) {
-        t->con.applyRot(t, tc, td, axis, NULL);
-        angle_final = angle * td->factor;
-        /* Even though final angle might be identical to orig value,
-         * we have to update the rotation matrix in that case... */
-        do_update_matrix = 2;
-      }
-      else if (t->flag & T_PROP_EDIT) {
-        angle_final = angle * td->factor;
-      }
-
-      /* Rotation is very likely to be above 180°, we need to do rotation by steps.
-       * Note that this is only needed when doing 'absolute' rotation
-       * (i.e. from initial rotation again, typically when using numinput).
-       * regular incremental rotation (from mouse/widget/...) will be called often enough,
-       * hence steps are small enough to be properly handled without that complicated trick.
-       * Note that we can only do that kind of stepped rotation if we have initial rotation values
-       * (and access to some actual rotation value storage).
-       * Otherwise, just assume it's useless (e.g. in case of mesh/UV/etc. editing).
-       * Also need to be in Euler rotation mode, the others never allow more than one turn anyway.
-       */
-      if (is_large_rotation && td->ext != NULL && td->ext->rotOrder == ROT_MODE_EUL) {
-        copy_v3_v3(td->ext->rot, td->ext->irot);
-        for (float angle_progress = angle_step; fabsf(angle_progress) < fabsf(angle_final);
-             angle_progress += angle_step) {
-          axis_angle_normalized_to_mat3(mat, axis, angle_progress);
-          ElementRotation(t, tc, td, mat, t->around);
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (int i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
         }
-        do_update_matrix = 2;
+        transdata_elem_rotate(t, tc, td, axis, angle, angle_step, is_large_rotation, &rmc);
       }
-      else if (angle_final != angle) {
-        do_update_matrix = 2;
-      }
-
-      if (do_update_matrix > 0) {
-        axis_angle_normalized_to_mat3(mat, axis, angle_final);
-        do_update_matrix--;
-      }
-
-      ElementRotation(t, tc, td, mat, t->around);
+    }
+    else {
+      struct TransDataArgs_Rotate data = {
+          .t = t,
+          .tc = tc,
+          .axis = {UNPACK3(axis)},
+          .angle = angle,
+          .angle_step = angle_step,
+          .is_large_rotation = is_large_rotation,
+      };
+      struct TransDataArgs_RotateTLS tls_data = {
+          .rmc = rmc,
+      };
+
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      settings.userdata_chunk = &tls_data;
+      settings.userdata_chunk_size = sizeof(tls_data);
+      BLI_task_parallel_range(0, tc->data_len, &data, transdata_elem_rotate_fn, &settings);
     }
   }
 }
diff --git a/source/blender/editors/transform/transform_mode_shear.c b/source/blender/editors/transform/transform_mode_shear.c
index 23ee55bf6c5..ab5dd11a926 100644
--- a/source/blender/editors/transform/transform_mode_shear.c
+++ b/source/blender/editors/transform/transform_mode_shear.c
@@ -27,6 +27,7 @@
 
 #include "BLI_math.h"
 #include "BLI_string.h"
+#include "BLI_task.h"
 
 #include "BKE_context.h"
 #include "BKE_unit.h"
@@ -44,6 +45,79 @@
 #include "transform_snap.h"
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (Shear) Element
+ * \{ */
+
+/**
+ * \note Small arrays / data-structures should be stored copied for faster memory access.
+ */
+struct TransDataArgs_Shear {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+  float totmat[3][3];
+  bool is_local_center;
+};
+
+static void transdata_elem_shear(const TransInfo *t,
+                                 const TransDataContainer *tc,
+                                 TransData *td,
+                                 const float totmat[3][3],
+                                 const bool is_local_center)
+{
+  float tmat[3][3];
+  const float *center;
+  if (t->flag & T_EDIT) {
+    mul_m3_series(tmat, td->smtx, totmat, td->mtx);
+  }
+  else {
+    copy_m3_m3(tmat, totmat);
+  }
+
+  if (is_local_center) {
+    center = td->center;
+  }
+  else {
+    center = tc->center_local;
+  }
+
+  float vec[3];
+  sub_v3_v3v3(vec, td->iloc, center);
+  mul_m3_v3(tmat, vec);
+  add_v3_v3(vec, center);
+  sub_v3_v3(vec, td->iloc);
+
+  if (t->options & CTX_GPENCIL_STROKES) {
+    /* Grease pencil multi-frame falloff. */
+    bGPDstroke *gps = (bGPDstroke *)td->extra;
+    if (gps != NULL) {
+      mul_v3_fl(vec, td->factor * gps->runtime.multi_frame_falloff);
+    }
+    else {
+      mul_v3_fl(vec, td->factor);
+    }
+  }
+  else {
+    mul_v3_fl(vec, td->factor);
+  }
+
+  add_v3_v3v3(td->loc, td->iloc, vec);
+}
+
+static void transdata_elem_shear_fn(void *__restrict iter_data_v,
+                                    const int iter,
+                                    const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  struct TransDataArgs_Shear *data = iter_data_v;
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  transdata_elem_shear(data->t, data->tc, td, data->totmat, data->is_local_center);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (Shear)
  * \{ */
 
@@ -117,8 +191,7 @@ static eRedrawFlag handleEventShear(TransInfo *t, const wmEvent *event)
 
 static void applyShear(TransInfo *t, const int UNUSED(mval[2]))
 {
-  float vec[3];
-  float smat[3][3], tmat[3][3], totmat[3][3], axismat[3][3], axismat_inv[3][3];
+  float smat[3][3], totmat[3][3], axismat[3][3], axismat_inv[3][3];
   float value;
   int i;
   char str[UI_MAX_DRAW_STR];
@@ -160,47 +233,26 @@ static void applyShear(TransInfo *t, const int UNUSED(mval[2]))
   mul_m3_series(totmat, axismat_inv, smat, axismat);
 
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *td = tc->data;
-    for (i = 0; i < tc->data_len; i++, td++) {
-      const float *center;
-      if (td->flag & TD_SKIP) {
-        continue;
-      }
-
-      if (t->flag & T_EDIT) {
-        mul_m3_series(tmat, td->smtx, totmat, td->mtx);
-      }
-      else {
-        copy_m3_m3(tmat, totmat);
-      }
-
-      if (is_local_center) {
-        center = td->center;
-      }
-      else {
-        center = tc->center_local;
-      }
-
-      sub_v3_v3v3(vec, td->iloc, center);
-      mul_m3_v3(tmat, vec);
-      add_v3_v3(vec, center);
-      sub_v3_v3(vec, td->iloc);
-
-      if (t->options & CTX_GPENCIL_STROKES) {
-        /* grease pencil multiframe falloff */
-        bGPDstroke *gps = (bGPDstroke *)td->extra;
-        if (gps != NULL) {
-          mul_v3_fl(vec, td->factor * gps->runtime.multi_frame_falloff);
-        }
-        else {
-          mul_v3_fl(vec, td->factor);
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
         }
+        transdata_elem_shear(t, tc, td, totmat, is_local_center);
       }
-      else {
-        mul_v3_fl(vec, td->factor);
-      }
-
-      add_v3_v3v3(td->loc, td->iloc, vec);
+    }
+    else {
+      struct TransDataArgs_Shear data = {
+          .t = t,
+          .tc = tc,
+          .is_local_center = is_local_center,
+      };
+      copy_m3_m3(data.totmat, totmat);
+
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      BLI_task_parallel_range(0, tc->data_len, &data, transdata_elem_shear_fn, &settings);
     }
   }
 
diff --git a/source/blender/editors/transform/transform_mode_shrink_fatten.c b/source/blender/editors/transform/transform_mode_shrink_fatten.c
index d2d73a14396..4cdaab599b4 100644
--- a/source/blender/editors/transform/transform_mode_shrink_fatten.c
+++ b/source/blender/editors/transform/transform_mode_shrink_fatten.c
@@ -25,6 +25,7 @@
 
 #include "BLI_math.h"
 #include "BLI_string.h"
+#include "BLI_task.h"
 
 #include "BKE_context.h"
 #include "BKE_unit.h"
@@ -43,6 +44,47 @@
 #include "transform_snap.h"
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (Shrink-Fatten) Element
+ * \{ */
+
+/**
+ * \note Small arrays / data-structures should be stored copied for faster memory access.
+ */
+struct TransDataArgs_ShrinkFatten {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+  float distance;
+};
+
+static void transdata_elem_shrink_fatten(const TransInfo *t,
+                                         const TransDataContainer *UNUSED(tc),
+                                         TransData *td,
+                                         const float distance)
+{
+  /* Get the final offset. */
+  float tdistance = distance * td->factor;
+  if (td->ext && (t->flag & T_ALT_TRANSFORM) != 0) {
+    tdistance *= td->ext->isize[0]; /* shell factor */
+  }
+
+  madd_v3_v3v3fl(td->loc, td->iloc, td->axismtx[2], tdistance);
+}
+
+static void transdata_elem_shrink_fatten_fn(void *__restrict iter_data_v,
+                                            const int iter,
+                                            const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  struct TransDataArgs_ShrinkFatten *data = iter_data_v;
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  transdata_elem_shrink_fatten(data->t, data->tc, td, data->distance);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (Shrink-Fatten)
  * \{ */
 
@@ -114,20 +156,24 @@ static void applyShrinkFatten(TransInfo *t, const int UNUSED(mval[2]))
   /* done with header string */
 
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *td = tc->data;
-    for (i = 0; i < tc->data_len; i++, td++) {
-      float tdistance; /* temp dist */
-      if (td->flag & TD_SKIP) {
-        continue;
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
+        }
+        transdata_elem_shrink_fatten(t, tc, td, distance);
       }
-
-      /* get the final offset */
-      tdistance = distance * td->factor;
-      if (td->ext && (t->flag & T_ALT_TRANSFORM) != 0) {
-        tdistance *= td->ext->isize[0]; /* shell factor */
-      }
-
-      madd_v3_v3v3fl(td->loc, td->iloc, td->axismtx[2], tdistance);
+    }
+    else {
+      struct TransDataArgs_ShrinkFatten data = {
+          .t = t,
+          .tc = tc,
+          .distance = distance,
+      };
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      BLI_task_parallel_range(0, tc->data_len, &data, transdata_elem_shrink_fatten_fn, &settings);
     }
   }
 
diff --git a/source/blender/editors/transform/transform_mode_skin_resize.c b/source/blender/editors/transform/transform_mode_skin_resize.c
index 75ad83b0787..45336602c5a 100644
--- a/source/blender/editors/transform/transform_mode_skin_resize.c
+++ b/source/blender/editors/transform/transform_mode_skin_resize.c
@@ -24,6 +24,7 @@
 #include <stdlib.h>
 
 #include "BLI_math.h"
+#include "BLI_task.h"
 
 #include "BKE_context.h"
 #include "BKE_unit.h"
@@ -38,6 +39,58 @@
 #include "transform_snap.h"
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (Skin) Element
+ * \{ */
+
+/**
+ * \note Small arrays / data-structures should be stored copied for faster memory access.
+ */
+struct TransDataArgs_SkinResize {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+  float mat[3][3];
+};
+
+static void transdata_elem_skin_resize(const TransInfo *t,
+                                       const TransDataContainer *UNUSED(tc),
+                                       TransData *td,
+                                       const float mat[3][3])
+{
+  float tmat[3][3], smat[3][3];
+  float fsize[3];
+
+  if (t->flag & T_EDIT) {
+    mul_m3_m3m3(smat, mat, td->mtx);
+    mul_m3_m3m3(tmat, td->smtx, smat);
+  }
+  else {
+    copy_m3_m3(tmat, mat);
+  }
+
+  if (t->con.applySize) {
+    t->con.applySize(t, NULL, NULL, tmat);
+  }
+
+  mat3_to_size(fsize, tmat);
+  td->loc[0] = td->iloc[0] * (1 + (fsize[0] - 1) * td->factor);
+  td->loc[1] = td->iloc[1] * (1 + (fsize[1] - 1) * td->factor);
+}
+
+static void transdata_elem_skin_resize_fn(void *__restrict iter_data_v,
+                                          const int iter,
+                                          const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  struct TransDataArgs_SkinResize *data = iter_data_v;
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  transdata_elem_skin_resize(data->t, data->tc, td, data->mat);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (Skin)
  * \{ */
 
@@ -67,29 +120,24 @@ static void applySkinResize(TransInfo *t, const int UNUSED(mval[2]))
   headerResize(t, t->values_final, str, sizeof(str));
 
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *td = tc->data;
-    for (i = 0; i < tc->data_len; i++, td++) {
-      float tmat[3][3], smat[3][3];
-      float fsize[3];
-      if (td->flag & TD_SKIP) {
-        continue;
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
+        }
+        transdata_elem_skin_resize(t, tc, td, mat);
       }
-
-      if (t->flag & T_EDIT) {
-        mul_m3_m3m3(smat, mat, td->mtx);
-        mul_m3_m3m3(tmat, td->smtx, smat);
-      }
-      else {
-        copy_m3_m3(tmat, mat);
-      }
-
-      if (t->con.applySize) {
-        t->con.applySize(t, NULL, NULL, tmat);
-      }
-
-      mat3_to_size(fsize, tmat);
-      td->loc[0] = td->iloc[0] * (1 + (fsize[0] - 1) * td->factor);
-      td->loc[1] = td->iloc[1] * (1 + (fsize[1] - 1) * td->factor);
+    }
+    else {
+      struct TransDataArgs_SkinResize data = {
+          .t = t,
+          .tc = tc,
+      };
+      copy_m3_m3(data.mat, mat);
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      BLI_task_parallel_range(0, tc->data_len, &data, transdata_elem_skin_resize_fn, &settings);
     }
   }
 
diff --git a/source/blender/editors/transform/transform_mode_tosphere.c b/source/blender/editors/transform/transform_mode_tosphere.c
index 9bca9b2e9e6..8587d5ae140 100644
--- a/source/blender/editors/transform/transform_mode_tosphere.c
+++ b/source/blender/editors/transform/transform_mode_tosphere.c
@@ -25,6 +25,7 @@
 
 #include "BLI_math.h"
 #include "BLI_string.h"
+#include "BLI_task.h"
 
 #include "MEM_guardedalloc.h"
 
@@ -111,6 +112,74 @@ static void to_sphere_radius_update(TransInfo *t)
 /** \} */
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (ToSphere) Element
+ * \{ */
+
+/**
+ * \note Small arrays / data-structures should be stored copied for faster memory access.
+ */
+struct TransDataArgs_ToSphere {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+  float ratio;
+  const struct ToSphereInfo to_sphere_info;
+  bool is_local_center;
+  bool is_data_space;
+};
+
+static void transdata_elem_to_sphere(const TransInfo *UNUSED(t),
+                                     const TransDataContainer *tc,
+                                     TransData *td,
+                                     const float ratio,
+                                     const struct ToSphereInfo *to_sphere_info,
+                                     const bool is_local_center,
+                                     const bool is_data_space)
+{
+  float vec[3];
+  const float *center = is_local_center ? td->center : tc->center_local;
+  if (is_data_space) {
+    copy_v3_v3(vec, td->center);
+  }
+  else {
+    copy_v3_v3(vec, td->iloc);
+  }
+
+  sub_v3_v3(vec, center);
+  const float radius = normalize_v3(vec);
+  const float tratio = ratio * td->factor;
+  mul_v3_fl(vec, radius * (1.0f - tratio) + to_sphere_info->radius * tratio);
+  add_v3_v3(vec, center);
+
+  if (is_data_space) {
+    sub_v3_v3(vec, td->center);
+    mul_m3_v3(td->smtx, vec);
+    add_v3_v3(vec, td->iloc);
+  }
+
+  copy_v3_v3(td->loc, vec);
+}
+
+static void transdata_elem_to_sphere_fn(void *__restrict iter_data_v,
+                                        const int iter,
+                                        const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  struct TransDataArgs_ToSphere *data = iter_data_v;
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  transdata_elem_to_sphere(data->t,
+                           data->tc,
+                           td,
+                           data->ratio,
+                           &data->to_sphere_info,
+                           data->is_local_center,
+                           data->is_data_space);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (ToSphere)
  * \{ */
 
@@ -119,8 +188,7 @@ static void applyToSphere(TransInfo *t, const int UNUSED(mval[2]))
   const bool is_local_center = transdata_check_local_center(t, t->around);
   const bool is_data_space = (t->options & CTX_POSE_BONE) != 0;
 
-  float vec[3];
-  float ratio, radius;
+  float ratio;
   int i;
   char str[UI_MAX_DRAW_STR];
 
@@ -147,40 +215,33 @@ static void applyToSphere(TransInfo *t, const int UNUSED(mval[2]))
     BLI_snprintf(str, sizeof(str), TIP_("To Sphere: %.4f %s"), ratio, t->proptext);
   }
 
-  const struct ToSphereInfo *data = t->custom.mode.data;
-  if (data->prop_size_prev != t->prop_size) {
+  const struct ToSphereInfo *to_sphere_info = t->custom.mode.data;
+  if (to_sphere_info->prop_size_prev != t->prop_size) {
     to_sphere_radius_update(t);
   }
 
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *td = tc->data;
-    for (i = 0; i < tc->data_len; i++, td++) {
-      float tratio;
-      if (td->flag & TD_SKIP) {
-        continue;
-      }
-
-      const float *center = is_local_center ? td->center : tc->center_local;
-      if (is_data_space) {
-        copy_v3_v3(vec, td->center);
-      }
-      else {
-        copy_v3_v3(vec, td->iloc);
-      }
-
-      sub_v3_v3(vec, center);
-      radius = normalize_v3(vec);
-      tratio = ratio * td->factor;
-      mul_v3_fl(vec, radius * (1.0f - tratio) + data->radius * tratio);
-      add_v3_v3(vec, center);
-
-      if (is_data_space) {
-        sub_v3_v3(vec, td->center);
-        mul_m3_v3(td->smtx, vec);
-        add_v3_v3(vec, td->iloc);
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
+        }
+        transdata_elem_to_sphere(t, tc, td, ratio, to_sphere_info, is_local_center, is_data_space);
       }
-
-      copy_v3_v3(td->loc, vec);
+    }
+    else {
+      struct TransDataArgs_ToSphere data = {
+          .t = t,
+          .tc = tc,
+          .ratio = ratio,
+          .to_sphere_info = *to_sphere_info,
+          .is_local_center = is_local_center,
+          .is_data_space = is_data_space,
+      };
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      BLI_task_parallel_range(0, tc->data_len, &data, transdata_elem_to_sphere_fn, &settings);
     }
   }
 
diff --git a/source/blender/editors/transform/transform_mode_trackball.c b/source/blender/editors/transform/transform_mode_trackball.c
index d05077ef1ef..157283d08da 100644
--- a/source/blender/editors/transform/transform_mode_trackball.c
+++ b/source/blender/editors/transform/transform_mode_trackball.c
@@ -25,6 +25,7 @@
 
 #include "BLI_math.h"
 #include "BLI_string.h"
+#include "BLI_task.h"
 
 #include "BKE_context.h"
 #include "BKE_unit.h"
@@ -40,6 +41,51 @@
 #include "transform_snap.h"
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (Rotation - Trackball) Element
+ * \{ */
+
+/**
+ * \note Small arrays / data-structures should be stored copied for faster memory access.
+ */
+struct TransDataArgs_Trackball {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+  const float axis[3];
+  const float angle;
+  float mat[3][3];
+};
+
+static void transdata_elem_trackball(const TransInfo *t,
+                                     const TransDataContainer *tc,
+                                     TransData *td,
+                                     const float axis[3],
+                                     const float angle,
+                                     const float mat[3][3])
+{
+  float mat_buf[3][3];
+  const float(*mat_final)[3] = mat;
+  if (t->flag & T_PROP_EDIT) {
+    axis_angle_normalized_to_mat3(mat_buf, axis, td->factor * angle);
+    mat_final = mat_buf;
+  }
+  ElementRotation(t, tc, td, mat_final, t->around);
+}
+
+static void transdata_elem_trackball_fn(void *__restrict iter_data_v,
+                                        const int iter,
+                                        const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  struct TransDataArgs_Trackball *data = iter_data_v;
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  transdata_elem_trackball(data->t, data->tc, td, data->axis, data->angle, data->mat);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (Rotation - Trackball)
  * \{ */
 
@@ -59,17 +105,27 @@ static void applyTrackballValue(TransInfo *t,
   axis_angle_normalized_to_mat3(mat, axis, angle);
 
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-    TransData *td = tc->data;
-    for (i = 0; i < tc->data_len; i++, td++) {
-      if (td->flag & TD_SKIP) {
-        continue;
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
+        }
+        transdata_elem_trackball(t, tc, td, axis, angle, mat);
       }
-
-      if (t->flag & T_PROP_EDIT) {
-        axis_angle_normalized_to_mat3(mat, axis, td->factor * angle);
-      }
-
-      ElementRotation(t, tc, td, mat, t->around);
+    }
+    else {
+      struct TransDataArgs_Trackball data = {
+          .t = t,
+          .tc = tc,
+          .axis = {UNPACK3(axis)},
+          .angle = angle,
+      };
+      copy_m3_m3(data.mat, mat);
+
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      BLI_task_parallel_range(0, tc->data_len, &data, transdata_elem_trackball_fn, &settings);
     }
   }
 }
diff --git a/source/blender/editors/transform/transform_mode_translate.c b/source/blender/editors/transform/transform_mode_translate.c
index 49afd7f0421..55bfedf26b8 100644
--- a/source/blender/editors/transform/transform_mode_translate.c
+++ b/source/blender/editors/transform/transform_mode_translate.c
@@ -29,6 +29,7 @@
 
 #include "BLI_math.h"
 #include "BLI_string.h"
+#include "BLI_task.h"
 
 #include "BKE_context.h"
 #include "BKE_report.h"
@@ -49,6 +50,121 @@
 #include "transform_snap.h"
 
 /* -------------------------------------------------------------------- */
+/** \name Transform (Translation) Element
+ * \{ */
+
+/**
+ * \note Small arrays / data-structures should be stored copied for faster memory access.
+ */
+struct TransDataArgs_Translate {
+  const TransInfo *t;
+  const TransDataContainer *tc;
+  const float tc_pivot[3];
+  const float vec[3];
+  bool apply_snap_align_rotation;
+  bool is_valid_snapping_normal;
+};
+
+static void transdata_elem_translate(const TransInfo *t,
+                                     const TransDataContainer *tc,
+                                     TransData *td,
+                                     const float pivot[3],
+                                     const float vec[3],
+                                     const bool apply_snap_align_rotation,
+                                     const bool is_valid_snapping_normal)
+{
+  float rotate_offset[3] = {0};
+  bool use_rotate_offset = false;
+
+  /* Handle snapping rotation before doing the translation. */
+  if (apply_snap_align_rotation) {
+    float mat[3][3];
+
+    if (is_valid_snapping_normal) {
+      const float *original_normal;
+
+      /* In pose mode, we want to align normals with Y axis of bones. */
+      if (t->options & CTX_POSE_BONE) {
+        original_normal = td->axismtx[1];
+      }
+      else {
+        original_normal = td->axismtx[2];
+      }
+
+      rotation_between_vecs_to_mat3(mat, original_normal, t->tsnap.snapNormal);
+    }
+    else {
+      unit_m3(mat);
+    }
+
+    ElementRotation_ex(t, tc, td, mat, pivot);
+
+    if (td->loc) {
+      use_rotate_offset = true;
+      sub_v3_v3v3(rotate_offset, td->loc, td->iloc);
+    }
+  }
+
+  float tvec[3];
+
+  if (t->con.applyVec) {
+    t->con.applyVec(t, tc, td, vec, tvec);
+  }
+  else {
+    copy_v3_v3(tvec, vec);
+  }
+
+  mul_m3_v3(td->smtx, tvec);
+
+  if (use_rotate_offset) {
+    add_v3_v3(tvec, rotate_offset);
+  }
+
+  if (t->options & CTX_GPENCIL_STROKES) {
+    /* Grease pencil multi-frame falloff. */
+    bGPDstroke *gps = (bGPDstroke *)td->extra;
+    if (gps != NULL) {
+      mul_v3_fl(tvec, td->factor * gps->runtime.multi_frame_falloff);
+    }
+    else {
+      mul_v3_fl(tvec, td->factor);
+    }
+  }
+  else {
+    /* Proportional editing falloff. */
+    mul_v3_fl(tvec, td->factor);
+  }
+
+  protectedTransBits(td->protectflag, tvec);
+
+  if (td->loc) {
+    add_v3_v3v3(td->loc, td->iloc, tvec);
+  }
+
+  constraintTransLim(t, td);
+}
+
+static void transdata_elem_translate_fn(void *__restrict iter_data_v,
+                                        const int iter,
+                                        const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  struct TransDataArgs_Translate *data = iter_data_v;
+  TransData *td = &data->tc->data[iter];
+  if (td->flag & TD_SKIP) {
+    return;
+  }
+  transdata_elem_translate(data->t,
+                           data->tc,
+                           td,
+                           data->tc_pivot,
+                           data->vec,
+                           data->apply_snap_align_rotation,
+                           data->is_valid_snapping_normal);
+}
+
+/** \} */
+
+/* -------------------------------------------------------------------- */
 /** \name Transform (Translation)
  * \{ */
 
@@ -242,14 +358,13 @@ static void ApplySnapTranslation(TransInfo *t, float vec[3])
 static void applyTranslationValue(TransInfo *t, const float vec[3])
 {
   const bool apply_snap_align_rotation = usingSnappingNormal(t);
-  float tvec[3];
+  const bool is_valid_snapping_normal = apply_snap_align_rotation && validSnappingNormal(t);
 
   /* Ideally "apply_snap_align_rotation" would only be used when a snap point is found:
    * `t->tsnap.status & POINT_INIT` - perhaps this function isn't the best place to apply rotation.
    * However snapping rotation needs to be handled before doing the translation
    * (unless the pivot is also translated). */
   FOREACH_TRANS_DATA_CONTAINER (t, tc) {
-
     float pivot[3];
     if (apply_snap_align_rotation) {
       copy_v3_v3(pivot, t->tsnap.snapTarget);
@@ -259,79 +374,28 @@ static void applyTranslationValue(TransInfo *t, const float vec[3])
       }
     }
 
-    TransData *td = tc->data;
-    for (int i = 0; i < tc->data_len; i++, td++) {
-      if (td->flag & TD_SKIP) {
-        continue;
-      }
-
-      float rotate_offset[3] = {0};
-      bool use_rotate_offset = false;
-
-      /* Handle snapping rotation before doing the translation. */
-      if (apply_snap_align_rotation) {
-        float mat[3][3];
-
-        if (validSnappingNormal(t)) {
-          const float *original_normal;
-
-          /* In pose mode, we want to align normals with Y axis of bones. */
-          if (t->options & CTX_POSE_BONE) {
-            original_normal = td->axismtx[1];
-          }
-          else {
-            original_normal = td->axismtx[2];
-          }
-
-          rotation_between_vecs_to_mat3(mat, original_normal, t->tsnap.snapNormal);
-        }
-        else {
-          unit_m3(mat);
-        }
-
-        ElementRotation_ex(t, tc, td, mat, pivot);
-
-        if (td->loc) {
-          use_rotate_offset = true;
-          sub_v3_v3v3(rotate_offset, td->loc, td->iloc);
+    if (tc->data_len < TRANSDATA_THREAD_LIMIT) {
+      TransData *td = tc->data;
+      for (int i = 0; i < tc->data_len; i++, td++) {
+        if (td->flag & TD_SKIP) {
+          continue;
         }
+        transdata_elem_translate(
+            t, tc, td, pivot, vec, apply_snap_align_rotation, is_valid_snapping_normal);
       }
-
-      if (t->con.applyVec) {
-        t->con.applyVec(t, tc, td, vec, tvec);
-      }
-      else {
-        copy_v3_v3(tvec, vec);
-      }
-
-      mul_m3_v3(td->smtx, tvec);
-
-      if (use_rotate_offset) {
-        add_v3_v3(tvec, rotate_offset);
-      }
-
-      if (t->options & CTX_GPENCIL_STROKES) {
-        /* Grease pencil multi-frame falloff. */
-        bGPDstroke *gps = (bGPDstroke *)td->extra;
-        if (gps != NULL) {
-          mul_v3_fl(tvec, td->factor * gps->runtime.multi_frame_falloff);
-        }
-        else {
-          mul_v3_fl(tvec, td->factor);
-        }
-      }
-      else {
-        /* Proportional editing falloff. */
-        mul_v3_fl(tvec, td->factor);
-      }
-
-      protectedTransBits(td->protectflag, tvec);
-
-      if (td->loc) {
-        add_v3_v3v3(td->loc, td->iloc, tvec);
-      }
-
-      constraintTransLim(t, td);
+    }
+    else {
+      struct TransDataArgs_Translate data = {
+          .t = t,
+          .tc = tc,
+          .tc_pivot = {UNPACK3(pivot)},
+          .vec = {UNPACK3(vec)},
+          .apply_snap_align_rotation = apply_snap_align_rotation,
+          .is_valid_snapping_normal = is_valid_snapping_normal,
+      };
+      TaskParallelSettings settings;
+      BLI_parallel_range_settings_defaults(&settings);
+      BLI_task_parallel_range(0, tc->data_len, &data, transdata_elem_translate_fn, &settings);
     }
   }
 }