[Mesa-dev] [PATCH 4/8] nir/spirv: Implement OpPtrAccessChain for buffers

Iago Toral itoral at igalia.com
Fri Jul 14 08:53:02 UTC 2017


On Thu, 2017-07-13 at 12:41 -0700, Jason Ekstrand wrote:
> ---
>  src/compiler/spirv/spirv_to_nir.c  |  4 +++-
>  src/compiler/spirv/vtn_private.h   | 11 ++++++++---
>  src/compiler/spirv/vtn_variables.c | 23 +++++++++++++++++++++++
>  3 files changed, 34 insertions(+), 4 deletions(-)
> 
> diff --git a/src/compiler/spirv/spirv_to_nir.c
> b/src/compiler/spirv/spirv_to_nir.c
> index 89ebc5f..7038bd9 100644
> --- a/src/compiler/spirv/spirv_to_nir.c
> +++ b/src/compiler/spirv/spirv_to_nir.c
> @@ -600,7 +600,8 @@ type_decoration_cb(struct vtn_builder *b,
>     switch (dec->decoration) {
>     case SpvDecorationArrayStride:
>        assert(type->base_type == vtn_base_type_matrix ||
> -             type->base_type == vtn_base_type_array);
> +             type->base_type == vtn_base_type_array ||
> +             type->base_type == vtn_base_type_pointer);
>        type->stride = dec->literals[0];
>        break;
>     case SpvDecorationBlock:
> @@ -3067,6 +3068,7 @@ vtn_handle_body_instruction(struct vtn_builder
> *b, SpvOp opcode,
>     case SpvOpCopyMemory:
>     case SpvOpCopyMemorySized:
>     case SpvOpAccessChain:
> +   case SpvOpPtrAccessChain:
>     case SpvOpInBoundsAccessChain:
>     case SpvOpArrayLength:
>        vtn_handle_variables(b, opcode, w, count);
> diff --git a/src/compiler/spirv/vtn_private.h
> b/src/compiler/spirv/vtn_private.h
> index 7cb5035..2f96c09 100644
> --- a/src/compiler/spirv/vtn_private.h
> +++ b/src/compiler/spirv/vtn_private.h
> @@ -220,15 +220,15 @@ struct vtn_type {
>     /* Specifies the length of complex types. */
>     unsigned length;
>  
> +   /* for arrays, matrices and pointers, the array stride */
> +   unsigned stride;
> +
>     union {
>        /* Members for scalar, vector, and array-like types */
>        struct {
>           /* for arrays, the vtn_type for the elements of the array
> */
>           struct vtn_type *array_element;
>  
> -         /* for arrays and matrices, the array stride */
> -         unsigned stride;
> -
>           /* for matrices, whether the matrix is stored row-major */
>           bool row_major:1;
>  
> @@ -308,6 +308,11 @@ struct vtn_access_link {
>  struct vtn_access_chain {
>     uint32_t length;
>  
> +   /** Whether or not to treat the base pointer as an array.  This
> is only
> +    * true if this access chain came from an OpPtrAccessChain.
> +    */
> +   bool ptr_as_array;
> +
>     /** Struct elements and array offsets.
>      *
>      * This is an array of 1 so that it can conveniently be created
> on the
> diff --git a/src/compiler/spirv/vtn_variables.c
> b/src/compiler/spirv/vtn_variables.c
> index 4f21fdd..a9ba392 100644
> --- a/src/compiler/spirv/vtn_variables.c
> +++ b/src/compiler/spirv/vtn_variables.c
> @@ -67,6 +67,12 @@ vtn_access_chain_pointer_dereference(struct
> vtn_builder *b,
>        vtn_access_chain_extend(b, base->chain, deref_chain->length);
>     struct vtn_type *type = base->type;
>  
> +   /* OpPtrAccessChain is only allowed on things which support
> variable
> +    * pointers.  For everything else, the client is expected to just
> pass us
> +    * the right access chain.
> +    */
> +   assert(!deref_chain->ptr_as_array);
> +
>     unsigned start = base->chain ? base->chain->length : 0;
>     for (unsigned i = 0; i < deref_chain->length; i++) {
>        chain->link[start + i] = deref_chain->link[i];
> @@ -135,6 +141,21 @@ vtn_ssa_offset_pointer_dereference(struct
> vtn_builder *b,
>     struct vtn_type *type = base->type;
>  
>     unsigned idx = 0;
> +   if (deref_chain->ptr_as_array) {
> +      /* We need ptr_type for the stride */
> +      assert(base->ptr_type);
> +      /* This must be a pointer to an actual element somewhere */
> +      assert(block_index && offset);
> +      /* We need at least one element in the chain */
> +      assert(deref_chain->length >= 1);

I was surprised that this wasn't '>= 2' but looking at the
implementation in vtn_handle_variables() it seems that we support an
empty list of indices with SpvOpAccessChain too. The specs do not state
clearly whether this is valid or not though. Is it correct?

> +
> +      nir_ssa_def *elem_offset =
> +         vtn_access_link_as_ssa(b, deref_chain->link[idx],
> +                                base->ptr_type->stride);
> +      offset = nir_iadd(&b->nb, offset, elem_offset);
> +      idx++;
> +   }
> +
>     if (!block_index) {
>        assert(base->var);
>        if (glsl_type_is_array(type->type)) {
> @@ -1699,8 +1720,10 @@ vtn_handle_variables(struct vtn_builder *b,
> SpvOp opcode,
>     }
>  
>     case SpvOpAccessChain:
> +   case SpvOpPtrAccessChain:
>     case SpvOpInBoundsAccessChain: {
>        struct vtn_access_chain *chain = vtn_access_chain_create(b,
> count - 4);
> +      chain->ptr_as_array = (opcode == SpvOpPtrAccessChain);
>  
>        unsigned idx = 0;
>        for (int i = 4; i < count; i++) {


More information about the mesa-dev mailing list