zink: unify ntv code for storing shared/scratch memory

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37358>
This commit is contained in:
Mike Blumenkrantz 2025-09-05 08:27:36 -04:00
parent b6f0826ce5
commit d74eff651b

View file

@ -2351,7 +2351,7 @@ emit_load_shared(struct ntv_context *ctx, nir_intrinsic_instr *intr)
}
static void
emit_store_shared(struct ntv_context *ctx, nir_intrinsic_instr *intr)
emit_store_special(struct ntv_context *ctx, nir_intrinsic_instr *intr, SpvId block, SpvStorageClass storage_class)
{
nir_alu_type atype;
SpvId src = get_src(ctx, &intr->src[0], &atype);
@ -2360,27 +2360,34 @@ emit_store_shared(struct ntv_context *ctx, nir_intrinsic_instr *intr)
unsigned bit_size = nir_src_bit_size(intr->src[0]);
SpvId uint_type = get_uvec_type(ctx, bit_size, 1);
SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder,
SpvStorageClassWorkgroup,
storage_class,
uint_type);
nir_alu_type otype;
SpvId offset = get_src(ctx, &intr->src[1], &otype);
if (otype == nir_type_float)
offset = bitcast_to_uvec(ctx, offset, nir_src_bit_size(intr->src[0]), 1);
SpvId shared_block = get_shared_block(ctx, bit_size);
if (otype != nir_type_uint)
offset = bitcast_to_uvec(ctx, offset, nir_src_bit_size(intr->src[1]), 1);
/* this is a partial write, so we have to loop and do a per-component write */
u_foreach_bit(i, wrmask) {
SpvId shared_offset = emit_binop(ctx, SpvOpIAdd, spirv_builder_type_uint(&ctx->builder, 32), offset, emit_uint_const(ctx, 32, i));
SpvId mask_offset = emit_binop(ctx, SpvOpIAdd, spirv_builder_type_uint(&ctx->builder, 32), offset, emit_uint_const(ctx, 32, i));
SpvId val = src;
if (nir_src_num_components(intr->src[0]) != 1)
val = spirv_builder_emit_composite_extract(&ctx->builder, uint_type, src, &i, 1);
val = spirv_builder_emit_composite_extract(&ctx->builder, get_alu_type(ctx, atype, 1, bit_size), src, &i, 1);
if (atype != nir_type_uint)
val = emit_bitcast(ctx, get_alu_type(ctx, nir_type_uint, 1, bit_size), val);
SpvId member = spirv_builder_emit_access_chain(&ctx->builder, ptr_type,
shared_block, &shared_offset, 1);
block, &mask_offset, 1);
spirv_builder_emit_store(&ctx->builder, member, val);
}
}
static void
emit_store_shared(struct ntv_context *ctx, nir_intrinsic_instr *intr)
{
unsigned bit_size = nir_src_bit_size(intr->src[0]);
SpvId shared_block = get_shared_block(ctx, bit_size);
emit_store_special(ctx, intr, shared_block, SpvStorageClassWorkgroup);
}
static void
emit_load_scratch(struct ntv_context *ctx, nir_intrinsic_instr *intr)
{
@ -2415,32 +2422,9 @@ emit_load_scratch(struct ntv_context *ctx, nir_intrinsic_instr *intr)
static void
emit_store_scratch(struct ntv_context *ctx, nir_intrinsic_instr *intr)
{
nir_alu_type atype;
SpvId src = get_src(ctx, &intr->src[0], &atype);
unsigned wrmask = nir_intrinsic_write_mask(intr);
unsigned bit_size = nir_src_bit_size(intr->src[0]);
SpvId uint_type = get_uvec_type(ctx, bit_size, 1);
SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder,
SpvStorageClassPrivate,
uint_type);
nir_alu_type otype;
SpvId offset = get_src(ctx, &intr->src[1], &otype);
if (otype != nir_type_uint)
offset = bitcast_to_uvec(ctx, offset, nir_src_bit_size(intr->src[1]), 1);
SpvId scratch_block = get_scratch_block(ctx, bit_size);
/* this is a partial write, so we have to loop and do a per-component write */
u_foreach_bit(i, wrmask) {
SpvId scratch_offset = emit_binop(ctx, SpvOpIAdd, spirv_builder_type_uint(&ctx->builder, 32), offset, emit_uint_const(ctx, 32, i));
SpvId val = src;
if (nir_src_num_components(intr->src[0]) != 1)
val = spirv_builder_emit_composite_extract(&ctx->builder, uint_type, src, &i, 1);
if (atype != nir_type_uint)
val = emit_bitcast(ctx, get_alu_type(ctx, nir_type_uint, 1, bit_size), val);
SpvId member = spirv_builder_emit_access_chain(&ctx->builder, ptr_type,
scratch_block, &scratch_offset, 1);
spirv_builder_emit_store(&ctx->builder, member, val);
}
emit_store_special(ctx, intr, scratch_block, SpvStorageClassPrivate);
}
static void