agx: simplify block image store offset

just make 32-bit offset.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36257>

src/asahi/compiler/agx_compile.c

@@ -897,11 +897,6 @@ agx_emit_block_image_store(agx_builder *b, nir_intrinsic_instr *instr)
    enum agx_dim dim = agx_tex_dim(nir_intrinsic_image_dim(instr), array);
    bool explicit = nir_intrinsic_explicit_coord(instr);
 
-   /* 32-bit source physically, 16-bit in NIR, top half ignored but needed
-    * logically to ensure alignment.
-    */
-   offset = agx_pad_to_32(b, offset);
-
    /* Modified coordinate descriptor */
    if (!explicit) {
       if (array) {

src/asahi/lib/agx_bg_eot.c

@@ -172,8 +172,8 @@ agx_build_end_of_tile_shader(struct agx_bg_eot_cache *cache,
          layer = nir_u2u16(&b, nir_load_layer_id(&b));
 
       nir_image_store_block_agx(
-         &b, nir_imm_intN_t(&b, rt, 16), nir_imm_intN_t(&b, offset_B, 16),
-         layer, .format = agx_tilebuffer_physical_format(&key->tib, rt),
+         &b, nir_imm_intN_t(&b, rt, 16), nir_imm_int(&b, offset_B), layer,
+         .format = agx_tilebuffer_physical_format(&key->tib, rt),
          .image_dim = dim, .image_array = key->tib.layered);
    }
 

src/asahi/vulkan/hk_cmd_meta.c

@@ -490,7 +490,7 @@ build_image_copy_shader(const struct vk_meta_image_copy_key *key)
       assert(isa_format != PIPE_FORMAT_NONE);
    }
 
-   nir_def *local_offset = nir_imm_intN_t(b, 0, 16);
+   nir_def *local_offset = nir_imm_int(b, 0);
    nir_def *lid = nir_trim_vector(b, nir_load_local_invocation_id(b), 2);
    lid = nir_u2u16(b, lid);
 

src/gallium/drivers/asahi/agx_blit.c

@@ -166,7 +166,7 @@ asahi_blit_compute_shader(struct pipe_context *ctx, struct asahi_blit_key *key)
    else if (clamp == ASAHI_BLIT_CLAMP_UINT_TO_SINT)
       color = nir_umin(b, color, nir_imm_int(b, INT32_MAX));
 
-   nir_def *local_offset = nir_imm_intN_t(b, 0, 16);
+   nir_def *local_offset = nir_imm_int(b, 0);
    nir_def *lid = nir_trim_vector(b, nir_load_local_invocation_id(b), 2);
    lid = nir_u2u16(b, lid);