agx: Pass mask into ld/st_tile instructions

Properly handle render target formats with <4 components.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18687>

diff --git a/src/asahi/compiler/agx_compile.c b/src/asahi/compiler/agx_compile.c
index 74f32a5..32cec3a 100644 (file)
--- a/src/asahi/compiler/agx_compile.c
+++ b/src/asahi/compiler/agx_compile.c
@@ -475,7 +475,8 @@ agx_emit_fragment_out(agx_builder *b, nir_intrinsic_instr *instr)
 
    b->shader->did_writeout = true;
    return agx_st_tile(b, agx_src_index(&instr->src[0]),
-             b->shader->key->fs.tib_formats[rt]);
+             b->shader->key->fs.tib_formats[rt],
+             nir_intrinsic_write_mask(instr));
 }
 
 static void
@@ -494,8 +495,10 @@ agx_emit_load_tile(agx_builder *b, agx_index dest, nir_intrinsic_instr *instr)
    b->shader->did_writeout = true;
    b->shader->out->reads_tib = true;
 
-   agx_ld_tile_to(b, dest, b->shader->key->fs.tib_formats[rt]);
-   agx_emit_cached_split(b, dest, 4);
+   unsigned nr_comps = nir_dest_num_components(instr->dest);
+   agx_ld_tile_to(b, dest, b->shader->key->fs.tib_formats[rt],
+                  BITFIELD_MASK(nr_comps));
+   agx_emit_cached_split(b, dest, nr_comps);
 }
 
 static enum agx_format

diff --git a/src/asahi/compiler/agx_opcodes.py b/src/asahi/compiler/agx_opcodes.py
index df0a98d..42dd753 100644 (file)
--- a/src/asahi/compiler/agx_opcodes.py
+++ b/src/asahi/compiler/agx_opcodes.py
@@ -221,11 +221,10 @@ op("get_sr", (0x72, 0x7F | L, 4, _), dests = 1, imms = [SR])
 op("sample_mask", (0x7fc1, 0xffff, 6, _), dests = 0, srcs = 1, can_eliminate = False)
 
 # Essentially same encoding
-op("ld_tile", (0x49, 0x7F, 8, _), dests = 1, srcs = 0,
-      can_eliminate = False, imms = [FORMAT])
+op("ld_tile", (0x49, 0x7F, 8, _), dests = 1, srcs = 0, imms = [FORMAT, MASK])
 
 op("st_tile", (0x09, 0x7F, 8, _), dests = 0, srcs = 1,
-      can_eliminate = False, imms = [FORMAT])
+      can_eliminate = False, imms = [FORMAT, MASK])
 
 for (name, exact) in [("any", 0xC000), ("none", 0xC200)]:
    op("jmp_exec_" + name, (exact, (1 << 16) - 1, 6, _), dests = 0, srcs = 0,

diff --git a/src/asahi/compiler/agx_pack.c b/src/asahi/compiler/agx_pack.c
index b55c908..3e304b4 100644 (file)
--- a/src/asahi/compiler/agx_pack.c
+++ b/src/asahi/compiler/agx_pack.c
@@ -426,8 +426,7 @@ agx_pack_instr(struct util_dynarray *emission, struct util_dynarray *fixups, agx
       bool load = (I->op == AGX_OPCODE_LD_TILE);
       unsigned D = agx_pack_alu_dst(load ? I->dest[0] : I->src[0]);
       unsigned rt = 0; /* TODO */
-      unsigned mask = I->mask ?: 0xF;
-      assert(mask < 0x10);
+      assert(I->mask < 0x10);
 
       uint64_t raw =
          0x09 |
@@ -436,7 +435,7 @@ agx_pack_instr(struct util_dynarray *emission, struct util_dynarray *fixups, agx
          ((uint64_t) (I->format) << 24) |
          ((uint64_t) (rt) << 32) |
          (load ? (1ull << 35) : 0) |
-         ((uint64_t) (mask) << 36) |
+         ((uint64_t) (I->mask) << 36) |
          ((uint64_t) 0x0380FC << 40) |
          (((uint64_t) (D >> 8)) << 60);