gfx_widgets+d3d: animate task-message hourglass on all drivers

The hourglass icon on the task-progress notification widget never
visibly rotated.  The animation system was correctly easing
msg->hourglass_rotation through 0 -> -2*pi every six seconds, but
the rendered icon stayed static across every backend.

Three independent bugs, each blocking a different family of
drivers:

  1. gfx/gfx_widgets.c - the call site that draws the hourglass
     forwarded the live `radians` value to gfx_widgets_draw_icon
     but left `cosine`/`sine` at their cos(0)/sin(0) defaults.
     The matrix-rotation path (gl, gl1, glcore, vulkan, d3d8)
     consumes those trig values via gfx_display_rotate_z, so it
     was always building an identity rotation regardless of
     `radians`.

     Fix: recompute cosf(radians)/sinf(radians) alongside `radians`
     in the in-progress branch.

  2. gfx/drivers/d3d_shaders/sprite_sm4.hlsl.h - the sprite shader
     used by the d3d10 / d3d11 / d3d12 menu display drivers.  The
     C code on those drivers dutifully writes
     `sprite->params.rotation = draw->rotation` into the vertex
     stream, but the shader pipeline never reads it: the GSInput
     struct doesn't have a `params` field, so rotation is dropped
     at the VS->GS boundary, and the geometry shader builds the
     four corner vertices straight from `position.xy + position.zw
     * cornerOffset` with no rotation applied.

     Fix: forward `params` from VS to GS (params.x is already
     consumed by the VS for scaling; we need params.y in the GS
     for rotation), and rewrite the GS to compute corner offsets
     by rotating per-corner pixel-space sign vectors `(sx, sy) in
     {+/-1}^2` and scaling them back by `position.z * 0.5` (X) and
     `position.w * 0.5` (Y).  Because position.z = 2W/vw and
     position.w = 2W/vh both encode the same pixel-space length W
     for a square icon, this preserves the icon's shape under
     rotation on a non-square viewport.  Non-square icons would
     skew, but the only caller of rotation today (the hourglass)
     is square.

  3. gfx/drivers/d3d9hlsl.c, gfx/drivers/d3d9cg.c - the d3d9
     menu display drivers' single-sprite path (no explicit vertex
     array supplied by caller, which is what gfx_widgets_draw_icon
     does) builds an axis-aligned quad from draw->x / draw->y /
     draw->width / draw->height directly, applies scale_factor,
     and never references draw->rotation or draw->matrix_data at
     all - so rotation was silently dropped on both shader
     paths.

     Fix: when draw->rotation is non-zero, rotate each of the
     four pixel-space corner offsets around the quad centre, then
     normalise back to [0,1] via 1/video_width and 1/video_height.
     Same aspect-correct approach as the d3d_shaders fix above.
     Add <math.h> to d3d9hlsl.c (d3d9cg.c already had it).

The d3d9 multi-vertex path uses the matrix_data route, so it's
handled by fix (1) alone and needs no driver-side change.

Tested on gl / gl1 / glcore / vulkan / d3d9_cg / d3d9_hlsl /
d3d10 / d3d11 / d3d12.

gfx/drivers/d3d9cg.c

@@ -870,17 +870,63 @@ static void gfx_display_d3d9_cg_draw(gfx_display_ctx_draw_t *draw,
          y2 = cy + hh;
       }
 
-      quad[0].x = x1; quad[0].y = y1; quad[0].z = 0.5f;
-      quad[0].u = 0.0f; quad[0].v = 0.0f; quad[0].color = col[0];
+      /* Apply draw->rotation around the quad center.  Rotation is
+       * computed in pixel space (using draw->width / draw->height as
+       * the icon's true square extents) and then converted back to
+       * normalised [0,1], so a non-square viewport does not skew the
+       * rotated icon. */
+      if (draw->rotation != 0.0f)
+      {
+         float cx     = (x1 + x2) * 0.5f;
+         float cy     = (y1 + y2) * 0.5f;
+         float half_w = draw->width  * 0.5f;
+         float half_h = draw->height * 0.5f;
+         if (draw->scale_factor && draw->scale_factor != 1.0f)
+         {
+            half_w *= draw->scale_factor;
+            half_h *= draw->scale_factor;
+         }
+         {
+            float c      = cosf(draw->rotation);
+            float s      = sinf(draw->rotation);
+            float inv_vw = 1.0f / (float)video_width;
+            float inv_vh = 1.0f / (float)video_height;
+            float ox[4]  = { -half_w,  half_w, -half_w,  half_w };
+            float oy[4]  = { -half_h, -half_h,  half_h,  half_h };
+            float rx[4], ry[4];
+            int   k;
+            for (k = 0; k < 4; k++)
+            {
+               rx[k] = (ox[k] * c - oy[k] * s) * inv_vw;
+               ry[k] = (ox[k] * s + oy[k] * c) * inv_vh;
+            }
+            quad[0].x = cx + rx[0]; quad[0].y = cy + ry[0]; quad[0].z = 0.5f;
+            quad[0].u = 0.0f; quad[0].v = 0.0f; quad[0].color = col[0];
 
-      quad[1].x = x2; quad[1].y = y1; quad[1].z = 0.5f;
-      quad[1].u = 1.0f; quad[1].v = 0.0f; quad[1].color = col[1];
+            quad[1].x = cx + rx[1]; quad[1].y = cy + ry[1]; quad[1].z = 0.5f;
+            quad[1].u = 1.0f; quad[1].v = 0.0f; quad[1].color = col[1];
 
-      quad[2].x = x1; quad[2].y = y2; quad[2].z = 0.5f;
-      quad[2].u = 0.0f; quad[2].v = 1.0f; quad[2].color = col[2];
+            quad[2].x = cx + rx[2]; quad[2].y = cy + ry[2]; quad[2].z = 0.5f;
+            quad[2].u = 0.0f; quad[2].v = 1.0f; quad[2].color = col[2];
 
-      quad[3].x = x2; quad[3].y = y2; quad[3].z = 0.5f;
-      quad[3].u = 1.0f; quad[3].v = 1.0f; quad[3].color = col[3];
+            quad[3].x = cx + rx[3]; quad[3].y = cy + ry[3]; quad[3].z = 0.5f;
+            quad[3].u = 1.0f; quad[3].v = 1.0f; quad[3].color = col[3];
+         }
+      }
+      else
+      {
+         quad[0].x = x1; quad[0].y = y1; quad[0].z = 0.5f;
+         quad[0].u = 0.0f; quad[0].v = 0.0f; quad[0].color = col[0];
+
+         quad[1].x = x2; quad[1].y = y1; quad[1].z = 0.5f;
+         quad[1].u = 1.0f; quad[1].v = 0.0f; quad[1].color = col[1];
+
+         quad[2].x = x1; quad[2].y = y2; quad[2].z = 0.5f;
+         quad[2].u = 0.0f; quad[2].v = 1.0f; quad[2].color = col[2];
+
+         quad[3].x = x2; quad[3].y = y2; quad[3].z = 0.5f;
+         quad[3].u = 1.0f; quad[3].v = 1.0f; quad[3].color = col[3];
+      }
 
       /* Top-down ortho: maps X [0,1]->[-1,1], Y [0,1]->[1,-1] (Y=0 at top).
        * Column-major (transposed) layout for the Cg stock shader which

gfx/drivers/d3d9hlsl.c

@@ -42,6 +42,7 @@
 #include <string.h>
 #include <retro_inline.h>
 #include <retro_math.h>
+#include <math.h>
 
 #ifndef _XBOX
 #define WIN32_LEAN_AND_MEAN
@@ -944,17 +945,66 @@ static void gfx_display_d3d9_hlsl_draw(gfx_display_ctx_draw_t *draw,
          y2 = cy + hh;
       }
 
-      quad[0].x = x1; quad[0].y = y1; quad[0].z = 0.5f;
-      quad[0].u = 0.0f; quad[0].v = 0.0f; quad[0].color = col[0];
+      /* Apply draw->rotation around the quad center.  Rotation is
+       * computed in pixel space (using draw->width / draw->height as
+       * the icon's true square extents) and then converted back to
+       * normalised [0,1], so a non-square viewport does not skew the
+       * rotated icon. */
+      if (draw->rotation != 0.0f)
+      {
+         float cx     = (x1 + x2) * 0.5f;
+         float cy     = (y1 + y2) * 0.5f;
+         float half_w = draw->width  * 0.5f;
+         float half_h = draw->height * 0.5f;
+         if (draw->scale_factor && draw->scale_factor != 1.0f)
+         {
+            half_w *= draw->scale_factor;
+            half_h *= draw->scale_factor;
+         }
+         {
+            float c            = cosf(draw->rotation);
+            float s            = sinf(draw->rotation);
+            float inv_vw       = 1.0f / (float)video_width;
+            float inv_vh       = 1.0f / (float)video_height;
+            /* Pixel-space corner offsets (dx, dy) for each of the
+             * four quad vertices.  Order matches the (x1,y1)..
+             * (x2,y2) layout used immediately below. */
+            float ox[4]        = { -half_w,  half_w, -half_w,  half_w };
+            float oy[4]        = { -half_h, -half_h,  half_h,  half_h };
+            float rx[4], ry[4];
+            int   k;
+            for (k = 0; k < 4; k++)
+            {
+               rx[k] = (ox[k] * c - oy[k] * s) * inv_vw;
+               ry[k] = (ox[k] * s + oy[k] * c) * inv_vh;
+            }
+            quad[0].x = cx + rx[0]; quad[0].y = cy + ry[0]; quad[0].z = 0.5f;
+            quad[0].u = 0.0f; quad[0].v = 0.0f; quad[0].color = col[0];
 
-      quad[1].x = x2; quad[1].y = y1; quad[1].z = 0.5f;
-      quad[1].u = 1.0f; quad[1].v = 0.0f; quad[1].color = col[1];
+            quad[1].x = cx + rx[1]; quad[1].y = cy + ry[1]; quad[1].z = 0.5f;
+            quad[1].u = 1.0f; quad[1].v = 0.0f; quad[1].color = col[1];
 
-      quad[2].x = x1; quad[2].y = y2; quad[2].z = 0.5f;
-      quad[2].u = 0.0f; quad[2].v = 1.0f; quad[2].color = col[2];
+            quad[2].x = cx + rx[2]; quad[2].y = cy + ry[2]; quad[2].z = 0.5f;
+            quad[2].u = 0.0f; quad[2].v = 1.0f; quad[2].color = col[2];
 
-      quad[3].x = x2; quad[3].y = y2; quad[3].z = 0.5f;
-      quad[3].u = 1.0f; quad[3].v = 1.0f; quad[3].color = col[3];
+            quad[3].x = cx + rx[3]; quad[3].y = cy + ry[3]; quad[3].z = 0.5f;
+            quad[3].u = 1.0f; quad[3].v = 1.0f; quad[3].color = col[3];
+         }
+      }
+      else
+      {
+         quad[0].x = x1; quad[0].y = y1; quad[0].z = 0.5f;
+         quad[0].u = 0.0f; quad[0].v = 0.0f; quad[0].color = col[0];
+
+         quad[1].x = x2; quad[1].y = y1; quad[1].z = 0.5f;
+         quad[1].u = 1.0f; quad[1].v = 0.0f; quad[1].color = col[1];
+
+         quad[2].x = x1; quad[2].y = y2; quad[2].z = 0.5f;
+         quad[2].u = 0.0f; quad[2].v = 1.0f; quad[2].color = col[2];
+
+         quad[3].x = x2; quad[3].y = y2; quad[3].z = 0.5f;
+         quad[3].u = 1.0f; quad[3].v = 1.0f; quad[3].color = col[3];
+      }
 
       /* Top-down ortho: maps X [0,1]→[-1,1], Y [0,1]→[1,-1] (Y=0 at top).
        * Row-major layout for mul(vector, matrix) in the stock HLSL vertex shader.

gfx/drivers/d3d_shaders/sprite_sm4.hlsl.h

@@ -20,6 +20,7 @@ SRC(
          float4 color1 : COLOR1;
          float4 color2 : COLOR2;
          float4 color3 : COLOR3;
+         float2 params : PARAMS;
       };
       struct PSInput
       {
@@ -42,34 +43,75 @@ SRC(
          output.color1 = input.color1;
          output.color2 = input.color2;
          output.color3 = input.color3;
+         /* params.x is consumed in the VS (scaling, above);
+          * forward params.y (rotation, in radians) to the GS. */
+         output.params = input.params;
          return output;
       }
 
       [maxvertexcount(4)]
       void GSMain(point GSInput input[1], inout TriangleStream<PSInput> triStream)
       {
+         /* Apply rotation around the sprite centre.
+          *
+          * input[0].position.xy = top-left corner of the quad (clip space)
+          * input[0].position.zw = quad size               (clip space)
+          *
+          * The four corners are at `position.xy + position.zw * c`
+          * with `c` in {(0,0), (1,0), (0,1), (1,1)}.  We rewrite as
+          * `centre + position.zw * (c - 0.5)` and rotate the
+          * `(c - 0.5)`-scaled offset by `params.y` radians around
+          * the centre.  The rotation is computed in pixel space and
+          * then mapped back to clip space via the per-axis `.zw`
+          * sizes; this keeps a square pixel-space icon square on
+          * screen even on a non-square viewport, because position.z
+          * and position.w both carry the same pixel-space length W
+          * (z = 2W/vw, w = 2W/vh).  Non-square icons would skew, but
+          * the only caller of rotation today (the task-message
+          * hourglass) draws a square. */
          PSInput output;
+         float  cosT = cos(input[0].params.y);
+         float  sinT = sin(input[0].params.y);
+         float2 centre = input[0].position.xy + input[0].position.zw * 0.5;
+
+         /* Corner sign ordering matches the four triStream.Append
+          * calls below; (sx, sy) ∈ {±1} are the corner-relative-to-
+          * centre signs in pixel space. */
+         float2 corner_signs[4] = {
+            float2( 1.0, -1.0),  /* top-right    -> color0 */
+            float2(-1.0, -1.0),  /* top-left     -> color1 */
+            float2( 1.0,  1.0),  /* bottom-right -> color2 */
+            float2(-1.0,  1.0),  /* bottom-left  -> color3 */
+         };
+         float2 corner_uv[4] = {
+            float2(1.0, 0.0),
+            float2(0.0, 0.0),
+            float2(1.0, 1.0),
+            float2(0.0, 1.0),
+         };
+         float4 corner_colour[4] = {
+            input[0].color0, input[0].color1,
+            input[0].color2, input[0].color3,
+         };
+
          output.position.zw = float2(0.0f, 1.0f);
 
-         output.position.xy = input[0].position.xy +  input[0].position.zw * float2(1.0, 0.0);
-         output.texcoord = input[0].texcoord.xy + input[0].texcoord.zw * float2(1.0, 0.0);
-         output.color = input[0].color0;
-         triStream.Append(output);
-
-         output.position.xy = input[0].position.xy +  input[0].position.zw * float2(0.0, 0.0);
-         output.texcoord = input[0].texcoord.xy + input[0].texcoord.zw * float2(0.0, 0.0);
-         output.color = input[0].color1;
-         triStream.Append(output);
-
-         output.position.xy = input[0].position.xy +  input[0].position.zw * float2(1.0, 1.0);
-         output.texcoord = input[0].texcoord.xy + input[0].texcoord.zw * float2(1.0, 1.0);
-         output.color = input[0].color2;
-         triStream.Append(output);
-
-         output.position.xy = input[0].position.xy +  input[0].position.zw * float2(0.0, 1.0);
-         output.texcoord = input[0].texcoord.xy + input[0].texcoord.zw * float2(0.0, 1.0);
-         output.color = input[0].color3;
-         triStream.Append(output);
+         [unroll]
+         for (int i = 0; i < 4; i++)
+         {
+            float sx = corner_signs[i].x;
+            float sy = corner_signs[i].y;
+            /* Rotate the half-extent corner offset, then scale back
+             * by the per-axis clip-space half-size.  See block
+             * comment above for why this preserves square icons. */
+            float dx = (sx * cosT - sy * sinT) * input[0].position.z * 0.5;
+            float dy = (sx * sinT + sy * cosT) * input[0].position.w * 0.5;
+            output.position.xy = centre + float2(dx, dy);
+            output.texcoord    = input[0].texcoord.xy
+               + input[0].texcoord.zw * corner_uv[i];
+            output.color       = corner_colour[i];
+            triStream.Append(output);
+         }
       }
 
       uniform sampler s0;

gfx/gfx_widgets.c

@@ -1278,6 +1278,13 @@ static void gfx_widgets_draw_task_msg(
          texture    = MENU_WIDGETS_ICON_HOURGLASS;
          color      = msg_queue_bar;
          radians    = msg->hourglass_rotation;
+         /* The display drivers that consume this rotation via the
+          * cosine/sine pair (gl, gl1, glcore, vulkan) need the
+          * trig values recomputed to match the live rotation -
+          * leaving them at the zero-rotation defaults above pins
+          * the hourglass icon to its starting orientation. */
+         cosine     = cosf(radians);
+         sine       = sinf(radians);
       }
       else if (msg->flags & DISPWIDG_FLAG_POSITIVE)
       {