nil: Add GOBType::TegraColor

Reviewed-by: Mohamed Ahmed <mohamedahmedegypt2001@gmail.com>
Reviewed-by: Mary Guillemard <mary@mary.zone>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37824>

src/nouveau/nil/copy.rs

@@ -181,6 +181,54 @@ impl<C: CopyBytes> CopyGOBLines for CopyGOBFermi<C> {
     }
 }
 
+/// Implements copies for [`GOBType::TegraColor`]
+struct CopyGOBTegra<C: CopyBytes> {
+    phantom: std::marker::PhantomData<C>,
+}
+
+impl<C: CopyBytes> CopyGOBLines for CopyGOBTegra<C> {
+    const GOB_EXTENT_B: Extent4D<units::Bytes> = Extent4D::new(64, 8, 1, 1);
+    const LINE_WIDTH_B: u32 = 16;
+    const X_DIVISOR: u32 = C::X_DIVISOR;
+
+    unsafe fn copy(tiled: *mut u8, linear: *mut u8, bytes: usize) {
+        unsafe {
+            C::copy(tiled, linear, bytes);
+        }
+    }
+
+    unsafe fn copy_whole_line(tiled: *mut u8, linear: *mut u8) {
+        unsafe {
+            C::copy_16b(tiled as *mut _, linear as *mut _);
+        }
+    }
+
+    #[inline(always)]
+    fn for_each_gob_line(mut f: impl FnMut(u32, u32, u32, u32)) {
+        for i in 0..2 {
+            f(i * 0x100 + 0x00, i * 32 + 0, 0, 0);
+            f(i * 0x100 + 0x10, i * 32 + 0, 1, 0);
+            f(i * 0x100 + 0x20, i * 32 + 16, 0, 0);
+            f(i * 0x100 + 0x30, i * 32 + 16, 1, 0);
+
+            f(i * 0x100 + 0x40, i * 32 + 0, 2, 0);
+            f(i * 0x100 + 0x50, i * 32 + 0, 3, 0);
+            f(i * 0x100 + 0x60, i * 32 + 16, 2, 0);
+            f(i * 0x100 + 0x70, i * 32 + 16, 3, 0);
+
+            f(i * 0x100 + 0x80, i * 32 + 0, 4, 0);
+            f(i * 0x100 + 0x90, i * 32 + 0, 5, 0);
+            f(i * 0x100 + 0xa0, i * 32 + 16, 4, 0);
+            f(i * 0x100 + 0xb0, i * 32 + 16, 5, 0);
+
+            f(i * 0x100 + 0xc0, i * 32 + 0, 6, 0);
+            f(i * 0x100 + 0xd0, i * 32 + 0, 7, 0);
+            f(i * 0x100 + 0xe0, i * 32 + 16, 6, 0);
+            f(i * 0x100 + 0xf0, i * 32 + 16, 7, 0);
+        }
+    }
+}
+
 /// Implements copies for [`GOBType::TuringColor2D`]
 struct CopyGOBTuring2D<C: CopyBytes> {
     phantom: std::marker::PhantomData<C>,
@@ -676,6 +724,16 @@ pub unsafe extern "C" fn nil_copy_linear_to_tiled(
                     end_B,
                 );
             }
+            GOBType::TegraColor => {
+                copy_tiled::<CopyGOBTegra<RawCopyToTiled>>(
+                    *tiling,
+                    level_extent_B,
+                    tiled_dst,
+                    linear_pointer,
+                    offset_B,
+                    end_B,
+                );
+            }
             GOBType::FermiColor => {
                 copy_tiled::<CopyGOBFermi<RawCopyToTiled>>(
                     *tiling,
@@ -745,6 +803,16 @@ pub unsafe extern "C" fn nil_copy_tiled_to_linear(
                     end_B,
                 );
             }
+            GOBType::TegraColor => {
+                copy_tiled::<CopyGOBTegra<RawCopyToLinear>>(
+                    *tiling,
+                    level_extent_B,
+                    tiled_src,
+                    linear_pointer,
+                    offset_B,
+                    end_B,
+                );
+            }
             GOBType::FermiColor => {
                 copy_tiled::<CopyGOBFermi<RawCopyToLinear>>(
                     *tiling,

src/nouveau/nil/modifiers.rs

@@ -55,12 +55,17 @@ struct GOBTypeModifierInfo {
     sector_layout: SectorLayout,
 }
 
-const GOB_TYPE_MODIFIER_INFOS: [GOBTypeModifierInfo; 2] = [
+const GOB_TYPE_MODIFIER_INFOS: [GOBTypeModifierInfo; 3] = [
     GOBTypeModifierInfo {
         gob_type: GOBType::FermiColor,
         gob_kind_version: GOBKindVersion::Fermi,
         sector_layout: SectorLayout::Desktop,
     },
+    GOBTypeModifierInfo {
+        gob_type: GOBType::TegraColor,
+        gob_kind_version: GOBKindVersion::Fermi,
+        sector_layout: SectorLayout::TegraK1,
+    },
     GOBTypeModifierInfo {
         gob_type: GOBType::TuringColor2D,
         gob_kind_version: GOBKindVersion::Turing,

src/nouveau/nil/tiling.rs

@@ -40,6 +40,24 @@ pub enum GOBType {
     /// `CopyGOBFermi` implements CPU copies for Fermi color GOBs.
     FermiColor,
 
+    /// The Tegra GOB format for color images
+    ///
+    /// A `TegraColor` GOB is 512 bytes, arranged in a 64x8 layout and split
+    /// into Sectors. Each Sector is 32 Bytes, and the Sectors in a GOB are
+    /// arranged in a 16x2 layout (i.e., two 16B lines on top of each other).
+    /// It's then arranged into two columns that are 2 sectors by 4, leading to
+    /// a 4x4 grid of sectors:
+    ///
+    /// |           |           |           |           |
+    /// |-----------|-----------|-----------|-----------|
+    /// | Sector  0 | Sector  1 | Sector  8 | Sector  9 |
+    /// | Sector  2 | Sector  3 | Sector 10 | Sector 11 |
+    /// | Sector  4 | Sector  5 | Sector 12 | Sector 13 |
+    /// | Sector  6 | Sector  7 | Sector 14 | Sector 15 |
+    ///
+    /// `CopyGOBTegra` implements CPU copies for Tegra color GOBs.
+    TegraColor,
+
     /// The Turing 2D GOB format for color images
     ///
     /// A `TuringColor2D` GOB is 512 bytes, arranged in a 64x8 layout and split
@@ -126,6 +144,8 @@ impl GOBType {
         } else if dev.cls_eng3d >= cl9097::FERMI_A {
             if format.is_depth_or_stencil() {
                 GOBType::FermiZS
+            } else if dev.type_ == NV_DEVICE_TYPE_SOC {
+                GOBType::TegraColor
             } else {
                 GOBType::FermiColor
             }
@@ -139,6 +159,7 @@ impl GOBType {
             GOBType::Linear => Extent4D::new(1, 1, 1, 1),
             GOBType::FermiZS
             | GOBType::FermiColor
+            | GOBType::TegraColor
             | GOBType::TuringColor2D
             | GOBType::Blackwell8Bit
             | GOBType::Blackwell16Bit