nir/spirv: Add real support for outer products

src/glsl/nir/spirv/vtn_alu.c

@@ -187,10 +187,6 @@ vtn_handle_matrix_alu(struct vtn_builder *b, SpvOp opcode,
       dest->ssa = vtn_ssa_transpose(b, src0);
       break;
 
-   case SpvOpOuterProduct:
-      dest->ssa = matrix_multiply(b, src0, vtn_ssa_transpose(b, src1));
-      break;
-
    case SpvOpMatrixTimesScalar:
       if (src0->transposed) {
          dest->ssa = vtn_ssa_transpose(b, mat_times_scalar(b, src0->transposed,
@@ -292,6 +288,14 @@ vtn_handle_alu(struct vtn_builder *b, SpvOp opcode,
    case SpvOpSMod:                  op = nir_op_umod;    break; /* FIXME? */
    case SpvOpFMod:                  op = nir_op_fmod;    break;
 
+   case SpvOpOuterProduct: {
+      for (unsigned i = 0; i < src[1]->num_components; i++) {
+         val->ssa->elems[i]->def =
+            nir_fmul(&b->nb, src[0], nir_channel(&b->nb, src[1], i));
+      }
+      return;
+   }
+
    case SpvOpDot:
       assert(src[0]->num_components == src[1]->num_components);
       switch (src[0]->num_components) {