On January 2, 2018 15:22:51 Ian Romanick <[email protected]> wrote:

On 01/01/2018 08:09 PM, Jason Ekstrand wrote:
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104424
---
 src/compiler/spirv/vtn_variables.c | 31 +++++++++++++++++++++++++------
 1 file changed, 25 insertions(+), 6 deletions(-)

diff --git a/src/compiler/spirv/vtn_variables.c b/src/compiler/spirv/vtn_variables.c
index d69b056..48797f6 100644
--- a/src/compiler/spirv/vtn_variables.c
+++ b/src/compiler/spirv/vtn_variables.c
@@ -1899,6 +1899,28 @@ vtn_create_variable(struct vtn_builder *b, struct vtn_value *val,
    }
 }

+static void
+vtn_assert_types_equal(struct vtn_builder *b, SpvOp opcode,
+                       struct vtn_type *dst_type, struct vtn_type *src_type)
+{
+   if (dst_type->val == src_type->val)
+      return;
+
+   if (dst_type->type == src_type->type) {
+      /* Early versions of GLSLang would re-emit types unnecessarily and you
+       * would end up with OpLoad, OpStore, or OpCopyMemory opcodes which have
+       * mismatched source and destination types.
+       */
+      vtn_warn("Source and destination types of %s do not match",
+               spirv_op_to_string(opcode));

This is deep compare vs. shallow compare, right?  Looking at the SPIR-V
spec, it's not clear to me which kind of "equality" is necessary.  What
does the validator do?  I'm just wondering of we should even bother
emitting a warning since this may just be a sub-optimal SPIR-V binary.
Emitting this particular warning here is a bit misleading.  The real
problem is that there are multiple identical types with different names,
right?

That's an interesting question. The SPIR-V spec is definitely unclear on point and, if you wind back the clock a bit, I think you'll find that the working group didn't really have consensus for quite some time on what "the same type" really means. At this point in time, I believe the consensus is that it means the same SPIR-V id (in this patch we compare value pointers but that's equivalent). However, this consensus was reached long after the initial SPIR-V spec was released.

We certainly could ditch the warning and keep the vtn_fail_if only using a looser condition. That said, if the consensus is going to be that they must have the same id then I'm a fan of enforcing the rules even if it's just a warning.

Also, fyi, I sent a new version of this patch today which uses a new vtn_types_compatible helper which is a bit more obvious in it's behavior than comparing ->type.

+   } else {
+      vtn_fail("Source and destination types of %s do not match: %s vs. %s",
+               spirv_op_to_string(opcode),
+               glsl_get_type_name(dst_type->type),
+               glsl_get_type_name(src_type->type));
+   }
+}
+
 void
 vtn_handle_variables(struct vtn_builder *b, SpvOp opcode,
                      const uint32_t *w, unsigned count)
@@ -1975,8 +1997,7 @@ vtn_handle_variables(struct vtn_builder *b, SpvOp opcode,
       struct vtn_value *dest = vtn_value(b, w[1], vtn_value_type_pointer);
       struct vtn_value *src = vtn_value(b, w[2], vtn_value_type_pointer);

-      vtn_fail_if(dest->type->deref != src->type->deref,
-                  "Dereferenced pointer types to OpCopyMemory do not match");
+      vtn_assert_types_equal(b, opcode, dest->type->deref, src->type->deref);

       vtn_variable_copy(b, dest->pointer, src->pointer);
       break;
@@ -1988,8 +2009,7 @@ vtn_handle_variables(struct vtn_builder *b, SpvOp opcode,
       struct vtn_value *src_val = vtn_value(b, w[3], vtn_value_type_pointer);
       struct vtn_pointer *src = src_val->pointer;

-      vtn_fail_if(res_type != src_val->type->deref,
-                  "Result and pointer types of OpLoad do not match");
+      vtn_assert_types_equal(b, opcode, res_type, src_val->type->deref);

       if (src->mode == vtn_variable_mode_image ||
           src->mode == vtn_variable_mode_sampler) {
@@ -2006,8 +2026,7 @@ vtn_handle_variables(struct vtn_builder *b, SpvOp opcode,
       struct vtn_pointer *dest = dest_val->pointer;
       struct vtn_value *src_val = vtn_untyped_value(b, w[2]);

-      vtn_fail_if(dest_val->type->deref != src_val->type,
-                  "Value and pointer types of OpStore do not match");
+      vtn_assert_types_equal(b, opcode, dest_val->type->deref, src_val->type);

       if (glsl_type_is_sampler(dest->type->type)) {
          vtn_warn("OpStore of a sampler detected.  Doing on-the-fly copy "




_______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Reply via email to