nx2: lower TEX.B with an explicit LOD

Kirby Air Riders samples a cube texture at an explicit level through a
bindless handle (TEX.B.LL R20, R16, R12, R16, CUBE): the coordinates stay
in Ra (R12-R14) and Rb holds the handle, then the LOD (R16, R17), since
five values do not split evenly. Place the values that way and take .LL
and .LB (the bias in the LOD's place).

1 files changed, 5 insertions, 4 deletions

diff --git a/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h b/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h
index e1c3095abe..fb3ec893d0 100644
--- a/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h
+++ b/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h
@@ -1658,12 +1658,12 @@ inline LoweredProgram LowerToMaxwell(std::span<const Instruction> code) {
         case 0xb60: case 0xf60: case 0x361: { // TEX; 0xf60: handle in a uniform register; 0x361: TEX.B
             // Bindless forms become Maxwell TEX.B, the handle first among the meta registers (the
             // texture pass tracks it back to its constant buffer words). 0x361 lists the handle
-            // right after the coordinates (see general_tex); forms with a LOD value or offsets,
+            // right after the coordinates, then the LOD (see general_tex); forms with offsets,
             // whose place next to the handle is unverified, are not taken.
             const bool bindless = opcode != 0xb60;
             const u32 handle = opcode == 0xf60 ? uniform_register(insn.Bits(40, 6)) : 255;
             if (bindless && ((opcode == 0xf60 && (handle == ~0U || handle >= 246 || insn.Bits(46, 13))) ||
-                             (opcode == 0x361 && (insn.Bits(40, 19) || insn.Bits(87, 3) > 1 || insn.Bits(76, 1)))))
+                             (opcode == 0x361 && (insn.Bits(40, 19) || insn.Bits(87, 3) > 3 || insn.Bits(76, 1)))))
                 return fail(index, "native bindless texture mode");
             const u32 mask = insn.Bits(72, 4);
             const u32 second_dest = insn.Bits(64, 8);
@@ -1708,11 +1708,12 @@ inline LoweredProgram LowerToMaxwell(std::span<const Instruction> code) {
                 const u32 coordinates = coordinate_count[dimension];
                 const bool has_lod = lod == 2 || lod == 3 || lod == 6 || lod == 7;
                 // 0x361's handle is one more value after the coordinates: 2D.DC is (x, y | handle,
-                // reference), ARRAY_2D is (layer, x | y, handle).
+                // reference), ARRAY_2D is (layer, x | y, handle); with five or more values the
+                // coordinates keep Ra (CUBE.LL is (x, y, z | handle, LOD)).
                 const u32 handle_slot = opcode == 0x361 ? 1 : 0;
                 const u32 count = coordinates + handle_slot + (has_lod ? 1 : 0) + (offsets ? 1 : 0) +
                                   (depth_compare ? 1 : 0);
-                const u32 in_a = count <= 4 ? (count + 1) / 2 : 4;
+                const u32 in_a = count <= 4 ? (count + 1) / 2 : opcode == 0x361 ? coordinates : 4;
                 const auto value = [&](u32 i) -> u32 {
                     if (i < in_a) return source_a == 255 ? 255U : source_a + i;
                     return source_b == 255 ? 255U : source_b + (i - in_a);