nx2: place TEX.B's handle after the coordinates; take .DC and .LZ

Ampere TEX.B lists its values as the coordinates, then the handle, then
the depth reference, split over Ra and Rb as for bound TEX: 2D.DC is
(x, y | handle, reference) and ARRAY_2D.LZ is (layer, x | y, handle), as
Kirby Air Riders' shaders build them (the handle is the register they
load from the texture constant buffer). The lowering read the handle from
Rb and all coordinates from Ra, which holds only without a depth
reference for up to two coordinates; place the handle by the layout and
take .DC and .LZ (101 TEX.B in 21 programs). Forms with a LOD value or
offsets stay unlowered: their place next to the handle is unverified.

1 files changed, 19 insertions, 14 deletions

diff --git a/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h b/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h
index 9cbd45ab88..5677215b67 100644
--- a/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h
+++ b/src/shader_recompiler/frontend/ampere/lower_to_maxwell.h
@@ -1587,14 +1587,13 @@ 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 keeps the
-            // coordinates together from Ra and the handle in Rb; only plain samples are taken.
+            // 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,
+            // 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)) : insn.Bits(32, 8);
-            if (bindless && (handle == ~0U || handle >= 246 ||
-                             (opcode == 0xf60 && insn.Bits(46, 13)) ||
-                             (opcode == 0x361 && (insn.Bits(40, 19) || insn.Bits(87, 3) || insn.Bits(76, 1) ||
-                                                  insn.Bits(78, 1)))))
+            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)))))
                 return fail(index, "native bindless texture mode");
             const u32 mask = insn.Bits(72, 4);
             const u32 second_dest = insn.Bits(64, 8);
@@ -1605,7 +1604,7 @@ inline LoweredProgram LowerToMaxwell(std::span<const Instruction> code) {
             const bool depth_compare = insn.Bits(78, 1);
             const u32 dimension = insn.Bits(61, 3), lod = insn.Bits(87, 3);
             const u32 source_a = insn.Bits(24, 8), source_b = insn.Bits(32, 8);
-            if (dimension == 5 && depth_compare && lod == 0 && !split && mask == 1) {
+            if (!bindless && dimension == 5 && depth_compare && lod == 0 && !split && mask == 1) {
                 if (!insn.Bits(60, 1) || insn.Bits(76, 2) || insn.Bits(81, 3) != 7 ||
                     insn.Bits(59, 1) || insn.Bits(40, 14) > 0x1fff || insn.Bits(54, 5) != 2 ||
                     (source_a & 1) || (source_b & 1) || source_a >= 248 || source_b >= 248 || dest >= 248)
@@ -1638,15 +1637,21 @@ inline LoweredProgram LowerToMaxwell(std::span<const Instruction> code) {
                     return false;
                 const u32 coordinates = coordinate_count[dimension];
                 const bool has_lod = lod == 2 || lod == 3 || lod == 6 || lod == 7;
-                const u32 count = coordinates + (has_lod ? 1 : 0) + (offsets ? 1 : 0) + (depth_compare ? 1 : 0);
+                // 0x361's handle is one more value after the coordinates: 2D.DC is (x, y | handle,
+                // reference), ARRAY_2D is (layer, x | y, handle).
+                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 auto value = [&](u32 i) -> u32 {
-                    if (opcode == 0x361 || i < in_a) return source_a == 255 ? 255U : source_a + i;
+                    if (i < in_a) return source_a == 255 ? 255U : source_a + i;
                     return source_b == 255 ? 255U : source_b + (i - in_a);
                 };
-                if ((source_a != 255 && source_a + (opcode == 0x361 ? count : in_a) > 246) ||
-                    (opcode != 0x361 && count > in_a && source_b != 255 && source_b + (count - in_a) > 246))
+                if ((source_a != 255 && source_a + in_a > 246) ||
+                    (count > in_a && source_b != 255 && source_b + (count - in_a) > 246))
                     return false;
+                const u32 texture_handle = opcode == 0x361 ? value(coordinates) : handle;
+                if (bindless && texture_handle >= 246) return false;
                 if (!bindless) {
                     const u32 binding = insn.Bits(54, 5);
                     if (result.required_texture_buffer != ~0U && result.required_texture_buffer != binding)
@@ -1659,8 +1664,8 @@ inline LoweredProgram LowerToMaxwell(std::span<const Instruction> code) {
                 // Coordinates to R250.., then LOD/bias, offsets and reference to R246...
                 for (u32 i = 0; i < coordinates; ++i) copy(250 + i, value(i));
                 u32 meta = 246;
-                u32 next = coordinates;
-                if (bindless) copy(meta++, handle);
+                u32 next = coordinates + handle_slot;
+                if (bindless) copy(meta++, texture_handle);
                 if (has_lod) copy(meta++, value(next++));
                 if (offsets) copy(meta++, value(next++));
                 if (depth_compare) copy(meta++, value(next++));