ac/llvm: handle write mask for nir_intrinsic_store_buffer_amd

tess lowering may generate buffer store with partial write mask.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16705>

diff --git a/src/amd/llvm/ac_nir_to_llvm.c b/src/amd/llvm/ac_nir_to_llvm.c
index 920fd5a..7fc61f4 100644 (file)
--- a/src/amd/llvm/ac_nir_to_llvm.c
+++ b/src/amd/llvm/ac_nir_to_llvm.c
@@ -4241,10 +4241,19 @@ static void visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
       if (slc)
          cache_policy |= ac_slc;
 
-      LLVMValueRef voffset = LLVMBuildAdd(ctx->ac.builder, addr_voffset,
-                                          LLVMConstInt(ctx->ac.i32, const_offset, 0), "");
-      ac_build_buffer_store_dword(&ctx->ac, descriptor, store_data, NULL, voffset, addr_soffset,
-                                  cache_policy);
+      unsigned writemask = nir_intrinsic_write_mask(instr);
+      while (writemask) {
+         int start, count;
+         u_bit_scan_consecutive_range(&writemask, &start, &count);
+
+         LLVMValueRef voffset = LLVMBuildAdd(
+            ctx->ac.builder, addr_voffset,
+            LLVMConstInt(ctx->ac.i32, const_offset + start * 4, 0), "");
+
+         LLVMValueRef data = extract_vector_range(&ctx->ac, store_data, start, count);
+         ac_build_buffer_store_dword(&ctx->ac, descriptor, data, NULL, voffset, addr_soffset,
+                                     cache_policy);
+      }
       break;
    }
    case nir_intrinsic_has_input_vertex_amd: {