panfrost: Add MSAA mode selection field

This field enables MSAA, either writing samples to separate surfaces, to
a single large-bpp surface, or implicitly resolved and to a single
surface.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6161>

diff --git a/src/gallium/drivers/panfrost/pan_mfbd.c b/src/gallium/drivers/panfrost/pan_mfbd.c
index d253b897b98417dfd981b31063fd8ac59df4e7ad..43a26a81e0338d4f4610544888340ef183535050 100644 (file)
--- a/src/gallium/drivers/panfrost/pan_mfbd.c
+++ b/src/gallium/drivers/panfrost/pan_mfbd.c
@@ -48,7 +48,7 @@ panfrost_mfbd_format(struct pipe_surface *surf)
                 .unk2 = 0x1,
                 .nr_channels = MALI_POSITIVE(desc->nr_channels),
                 .unk3 = 0x4,
-                .flags = 0x8,
+                .flags = 0x2,
                 .swizzle = panfrost_translate_swizzle_4(swizzle),
                 .no_preload = true
         };
@@ -225,7 +225,11 @@ panfrost_mfbd_set_cbuf(
         rt->format = panfrost_mfbd_format(surf);
 
         if (layer_stride)
-                rt->format.flags |= MALI_MFBD_FORMAT_MSAA | MALI_MFBD_FORMAT_LAYERED;
+                rt->format.msaa = MALI_MSAA_LAYERED;
+        else if (surf->nr_samples)
+                rt->format.msaa = MALI_MSAA_AVERAGE;
+        else
+                rt->format.msaa = MALI_MSAA_SINGLE;
 
         /* Now, we set the layout specific pieces */
 
@@ -580,7 +584,7 @@ panfrost_mfbd_fragment(struct panfrost_batch *batch, bool has_draws)
                         };
 
                         if (is_bifrost) {
-                                null_rt.flags = 0x8;
+                                null_rt.flags = 0x2;
                                 null_rt.unk3 = 0x8;
                         }
 
@@ -613,8 +617,6 @@ panfrost_mfbd_fragment(struct panfrost_batch *batch, bool has_draws)
         /* Actualize the requirements */
 
         if (batch->requirements & PAN_REQ_MSAA) {
-                rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
-
                 /* XXX */
                 fb.unk1 |= (1 << 4) | (1 << 1);
                 fb.rt_count_2 = 4;

diff --git a/src/panfrost/include/panfrost-job.h b/src/panfrost/include/panfrost-job.h
index 8ac799b81f2c2004c04625bdc35ffe0acbb6bf34..ab0f8f7048fc3855b01e22b94b14d4edfae75733 100644 (file)
--- a/src/panfrost/include/panfrost-job.h
+++ b/src/panfrost/include/panfrost-job.h
@@ -1693,14 +1693,22 @@ struct mali_single_framebuffer {
         /* More below this, maybe */
 } __attribute__((packed));
 
-/* Format bits for the render target flags. Setting MSAA alone works for on
- * chip MSAA. Setting MSAA with the LAYERED flag works for MSAA where each
- * sample is its own image (implements the ES3 spec directly but inefficient on
- * mobile). */
 
-#define MALI_MFBD_FORMAT_LAYERED  (1 << 0)
-#define MALI_MFBD_FORMAT_MSAA    (1 << 1)
-#define MALI_MFBD_FORMAT_SRGB    (1 << 2)
+/* SINGLE to disable multisampling, AVERAGE for
+ * EXT_multisampled_render_to_texture operation where multiple tilebuffer
+ * samples are implicitly resolved before writeout, MULTIPLE to write multiple
+ * samples inline, and LAYERED for ES3-style multisampling with each sample in
+ * a different buffer.
+ */
+
+enum mali_msaa_mode {
+        MALI_MSAA_SINGLE = 0,
+        MALI_MSAA_AVERAGE = 1,
+        MALI_MSAA_MULTIPLE = 2,
+        MALI_MSAA_LAYERED = 3,
+};
+
+#define MALI_MFBD_FORMAT_SRGB    (1 << 0)
 
 struct mali_rt_format {
         unsigned unk1 : 32;
@@ -1711,7 +1719,8 @@ struct mali_rt_format {
         unsigned unk3 : 4;
         unsigned unk4 : 1;
         enum mali_block_format block : 2;
-        unsigned flags : 4;
+        enum mali_msaa_mode msaa : 2;
+        unsigned flags : 2;
 
         unsigned swizzle : 12;
 

diff --git a/src/panfrost/pandecode/decode.c b/src/panfrost/pandecode/decode.c
index 119fcc0287ff64f093c8fdc57b33e44c8d276914..04add67b82002b41de279ce82d230a769fe172b2 100644 (file)
--- a/src/panfrost/pandecode/decode.c
+++ b/src/panfrost/pandecode/decode.c
@@ -245,8 +245,6 @@ static const struct pandecode_flag_info u4_flag_info[] = {
 
 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
 static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
-        FLAG_INFO(LAYERED),
-        FLAG_INFO(MSAA),
         FLAG_INFO(SRGB),
         {}
 };
@@ -411,6 +409,22 @@ pandecode_func(enum mali_func mode)
 }
 #undef DEFINE_CASE
 
+#define DEFINE_CASE(name) case MALI_MSAA_ ## name: return "MALI_MSAA_" #name
+static char *
+pandecode_msaa_mode(enum mali_msaa_mode mode)
+{
+        switch (mode) {
+                DEFINE_CASE(SINGLE);
+                DEFINE_CASE(AVERAGE);
+                DEFINE_CASE(MULTIPLE);
+                DEFINE_CASE(LAYERED);
+        default:
+                unreachable("Impossible");
+                return "";
+        }
+}
+#undef DEFINE_CASE
+
 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
 static char *
 pandecode_stencil_op(enum mali_stencil_op op)
@@ -1028,6 +1042,8 @@ pandecode_rt_format(struct mali_rt_format format)
         pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
         pandecode_log_cont(",\n");
 
+        pandecode_prop("msaa = %s", pandecode_msaa_mode(format.msaa));
+
         /* In theory, the no_preload bit can be cleared to enable MFBD preload,
          * which is a faster hardware-based alternative to the wallpaper method
          * to preserve framebuffer contents across frames. In practice, MFBD