Import Mesa 19.2.8

1 files changed, 12 insertions, 93 deletions

diff --git a/lib/mesa/src/compiler/spirv/vtn_opencl.c b/lib/mesa/src/compiler/spirv/vtn_opencl.c
index 55980fd72..e26b5bc04 100644
--- a/lib/mesa/src/compiler/spirv/vtn_opencl.c
+++ b/lib/mesa/src/compiler/spirv/vtn_opencl.c
@@ -25,6 +25,7 @@
  */
 
 #include "math.h"
+
 #include "nir/nir_builtin_builder.h"
 
 #include "vtn_private.h"
@@ -128,18 +129,6 @@ handle_special(struct vtn_builder *b, enum OpenCLstd_Entrypoints opcode,
       return nir_uabs_diff(nb, srcs[0], srcs[1]);
    case OpenCLstd_Bitselect:
       return nir_bitselect(nb, srcs[0], srcs[1], srcs[2]);
-   case OpenCLstd_SMad_hi:
-      return nir_imad_hi(nb, srcs[0], srcs[1], srcs[2]);
-   case OpenCLstd_UMad_hi:
-      return nir_umad_hi(nb, srcs[0], srcs[1], srcs[2]);
-   case OpenCLstd_SMul24:
-      return nir_imul24(nb, srcs[0], srcs[1]);
-   case OpenCLstd_UMul24:
-      return nir_umul24(nb, srcs[0], srcs[1]);
-   case OpenCLstd_SMad24:
-      return nir_imad24(nb, srcs[0], srcs[1], srcs[2]);
-   case OpenCLstd_UMad24:
-      return nir_umad24(nb, srcs[0], srcs[1], srcs[2]);
    case OpenCLstd_FClamp:
       return nir_fclamp(nb, srcs[0], srcs[1], srcs[2]);
    case OpenCLstd_SClamp:
@@ -184,8 +173,6 @@ handle_special(struct vtn_builder *b, enum OpenCLstd_Entrypoints opcode,
       return nir_rotate(nb, srcs[0], srcs[1]);
    case OpenCLstd_Smoothstep:
       return nir_smoothstep(nb, srcs[0], srcs[1], srcs[2]);
-   case OpenCLstd_Clz:
-      return nir_clz_u(nb, srcs[0]);
    case OpenCLstd_Select:
       return nir_select(nb, srcs[0], srcs[1], srcs[2]);
    case OpenCLstd_Step:
@@ -212,34 +199,25 @@ _handle_v_load_store(struct vtn_builder *b, enum OpenCLstd_Entrypoints opcode,
 
    const struct glsl_type *dest_type = type->type;
    unsigned components = glsl_get_vector_elements(dest_type);
+   unsigned stride = components * glsl_get_bit_size(dest_type) / 8;
 
    nir_ssa_def *offset = vtn_ssa_value(b, w[5 + a])->def;
    struct vtn_value *p = vtn_value(b, w[6 + a], vtn_value_type_pointer);
 
-   struct vtn_ssa_value *comps[NIR_MAX_VEC_COMPONENTS];
-   nir_ssa_def *ncomps[NIR_MAX_VEC_COMPONENTS];
-
-   nir_ssa_def *moffset = nir_imul_imm(&b->nb, offset, components);
    nir_deref_instr *deref = vtn_pointer_to_deref(b, p->pointer);
 
-   for (int i = 0; i < components; i++) {
-      nir_ssa_def *coffset = nir_iadd_imm(&b->nb, moffset, i);
-      nir_deref_instr *arr_deref = nir_build_deref_ptr_as_array(&b->nb, deref, coffset);
+   /* 1. cast to vec type with adjusted stride */
+   deref = nir_build_deref_cast(&b->nb, &deref->dest.ssa, deref->mode,
+                                dest_type, stride);
+   /* 2. deref ptr_as_array */
+   deref = nir_build_deref_ptr_as_array(&b->nb, deref, offset);
 
-      if (load) {
-         comps[i] = vtn_local_load(b, arr_deref, p->type->access);
-         ncomps[i] = comps[i]->def;
-      } else {
-         struct vtn_ssa_value *ssa = vtn_create_ssa_value(b, glsl_scalar_type(glsl_get_base_type(dest_type)));
-         struct vtn_ssa_value *val = vtn_ssa_value(b, w[5]);
-         ssa->def = vtn_vector_extract(b, val->def, i);
-         vtn_local_store(b, ssa, arr_deref, p->type->access);
-      }
-   }
    if (load) {
-      struct vtn_ssa_value *ssa = vtn_create_ssa_value(b, dest_type);
-      ssa->def = nir_vec(&b->nb, ncomps, components);
-      vtn_push_ssa(b, w[2], type, ssa);
+      struct vtn_ssa_value *val = vtn_local_load(b, deref, p->type->access);
+      vtn_push_ssa(b, w[2], type, val);
+   } else {
+      struct vtn_ssa_value *val = vtn_ssa_value(b, w[5]);
+      vtn_local_store(b, val, deref, p->type->access);
    }
 }
 
@@ -266,52 +244,6 @@ handle_printf(struct vtn_builder *b, enum OpenCLstd_Entrypoints opcode,
    return nir_imm_int(&b->nb, -1);
 }
 
-static nir_ssa_def *
-handle_shuffle(struct vtn_builder *b, enum OpenCLstd_Entrypoints opcode, unsigned num_srcs,
-               nir_ssa_def **srcs, const struct glsl_type *dest_type)
-{
-   struct nir_ssa_def *input = srcs[0];
-   struct nir_ssa_def *mask = srcs[1];
-
-   unsigned out_elems = glsl_get_vector_elements(dest_type);
-   nir_ssa_def *outres[NIR_MAX_VEC_COMPONENTS];
-   unsigned in_elems = input->num_components;
-   if (mask->bit_size != 32)
-      mask = nir_u2u32(&b->nb, mask);
-   mask = nir_iand(&b->nb, mask, nir_imm_intN_t(&b->nb, in_elems - 1, mask->bit_size));
-   for (unsigned i = 0; i < out_elems; i++)
-      outres[i] = nir_vector_extract(&b->nb, input, nir_channel(&b->nb, mask, i));
-
-   return nir_vec(&b->nb, outres, out_elems);
-}
-
-static nir_ssa_def *
-handle_shuffle2(struct vtn_builder *b, enum OpenCLstd_Entrypoints opcode, unsigned num_srcs,
-                nir_ssa_def **srcs, const struct glsl_type *dest_type)
-{
-   struct nir_ssa_def *input0 = srcs[0];
-   struct nir_ssa_def *input1 = srcs[1];
-   struct nir_ssa_def *mask = srcs[2];
-
-   unsigned out_elems = glsl_get_vector_elements(dest_type);
-   nir_ssa_def *outres[NIR_MAX_VEC_COMPONENTS];
-   unsigned in_elems = input0->num_components;
-   unsigned total_mask = 2 * in_elems - 1;
-   unsigned half_mask = in_elems - 1;
-   if (mask->bit_size != 32)
-      mask = nir_u2u32(&b->nb, mask);
-   mask = nir_iand(&b->nb, mask, nir_imm_intN_t(&b->nb, total_mask, mask->bit_size));
-   for (unsigned i = 0; i < out_elems; i++) {
-      nir_ssa_def *this_mask = nir_channel(&b->nb, mask, i);
-      nir_ssa_def *vmask = nir_iand(&b->nb, this_mask, nir_imm_intN_t(&b->nb, half_mask, mask->bit_size));
-      nir_ssa_def *val0 = nir_vector_extract(&b->nb, input0, vmask);
-      nir_ssa_def *val1 = nir_vector_extract(&b->nb, input1, vmask);
-      nir_ssa_def *sel = nir_ilt(&b->nb, this_mask, nir_imm_intN_t(&b->nb, in_elems, mask->bit_size));
-      outres[i] = nir_bcsel(&b->nb, sel, val0, val1);
-   }
-   return nir_vec(&b->nb, outres, out_elems);
-}
-
 bool
 vtn_handle_opencl_instruction(struct vtn_builder *b, SpvOp ext_opcode,
                               const uint32_t *w, unsigned count)
@@ -356,12 +288,6 @@ vtn_handle_opencl_instruction(struct vtn_builder *b, SpvOp ext_opcode,
       return true;
    case OpenCLstd_SAbs_diff:
    case OpenCLstd_UAbs_diff:
-   case OpenCLstd_SMad_hi:
-   case OpenCLstd_UMad_hi:
-   case OpenCLstd_SMad24:
-   case OpenCLstd_UMad24:
-   case OpenCLstd_SMul24:
-   case OpenCLstd_UMul24:
    case OpenCLstd_Bitselect:
    case OpenCLstd_FClamp:
    case OpenCLstd_SClamp:
@@ -388,7 +314,6 @@ vtn_handle_opencl_instruction(struct vtn_builder *b, SpvOp ext_opcode,
    case OpenCLstd_Smoothstep:
    case OpenCLstd_S_Upsample:
    case OpenCLstd_U_Upsample:
-   case OpenCLstd_Clz:
       handle_instr(b, ext_opcode, w, count, handle_special);
       return true;
    case OpenCLstd_Vloadn:
@@ -397,12 +322,6 @@ vtn_handle_opencl_instruction(struct vtn_builder *b, SpvOp ext_opcode,
    case OpenCLstd_Vstoren:
       vtn_handle_opencl_vstore(b, ext_opcode, w, count);
       return true;
-   case OpenCLstd_Shuffle:
-      handle_instr(b, ext_opcode, w, count, handle_shuffle);
-      return true;
-   case OpenCLstd_Shuffle2:
-      handle_instr(b, ext_opcode, w, count, handle_shuffle2);
-      return true;
    case OpenCLstd_Printf:
       handle_instr(b, ext_opcode, w, count, handle_printf);
       return true;