diff options
author | Rob Clark <robdclark@chromium.org> | 2022-12-21 20:17:13 -0800 |
---|---|---|
committer | Marge Bot <emma+marge@anholt.net> | 2022-12-29 17:42:42 +0000 |
commit | 75eb0d2891c2eb7574bd943a6a40493484047b48 (patch) | |
tree | 74ceeb7b6511cb9a39601d9805bd746ec16ae693 /src/freedreno/ir3 | |
parent | 7cf7bf8b9ee87691816cf7d33d522b6417099f4b (diff) | |
download | mesa-75eb0d2891c2eb7574bd943a6a40493484047b48.tar.gz |
freedreno/ir3: Allow isam for non-bindless ssbo loads
We already had the infrastructure for this, dating back to a5xx, so this
is low hanging fruit to hit the faster isam path.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20429>
Diffstat (limited to 'src/freedreno/ir3')
-rw-r--r-- | src/freedreno/ir3/ir3_compiler_nir.c | 15 |
1 files changed, 9 insertions, 6 deletions
diff --git a/src/freedreno/ir3/ir3_compiler_nir.c b/src/freedreno/ir3/ir3_compiler_nir.c index 3d2109f8ee7..45cd8647814 100644 --- a/src/freedreno/ir3/ir3_compiler_nir.c +++ b/src/freedreno/ir3/ir3_compiler_nir.c @@ -1337,7 +1337,7 @@ struct tex_src_info { * to handle with the image_mapping table.. */ static struct tex_src_info -get_image_ssbo_samp_tex_src(struct ir3_context *ctx, nir_src *src) +get_image_ssbo_samp_tex_src(struct ir3_context *ctx, nir_src *src, bool image) { struct ir3_block *b = ctx->block; struct tex_src_info info = {0}; @@ -1382,9 +1382,13 @@ get_image_ssbo_samp_tex_src(struct ir3_context *ctx, nir_src *src) } else { info.flags |= IR3_INSTR_S2EN; unsigned slot = nir_src_as_uint(*src); - unsigned tex_idx = ir3_image_to_tex(&ctx->so->image_mapping, slot); + unsigned tex_idx = image ? + ir3_image_to_tex(&ctx->so->image_mapping, slot) : + ir3_ssbo_to_tex(&ctx->so->image_mapping, slot); struct ir3_instruction *texture, *sampler; + ctx->so->num_samp = MAX2(ctx->so->num_samp, tex_idx + 1); + texture = create_immed_typed(ctx->block, tex_idx, TYPE_U16); sampler = create_immed_typed(ctx->block, tex_idx, TYPE_U16); @@ -1440,7 +1444,7 @@ emit_intrinsic_load_image(struct ir3_context *ctx, nir_intrinsic_instr *intr, } struct ir3_block *b = ctx->block; - struct tex_src_info info = get_image_ssbo_samp_tex_src(ctx, &intr->src[0]); + struct tex_src_info info = get_image_ssbo_samp_tex_src(ctx, &intr->src[0], true); struct ir3_instruction *sam; struct ir3_instruction *const *src0 = ir3_get_src(ctx, &intr->src[1]); struct ir3_instruction *coords[4]; @@ -1482,7 +1486,7 @@ emit_intrinsic_image_size_tex(struct ir3_context *ctx, struct ir3_instruction **dst) { struct ir3_block *b = ctx->block; - struct tex_src_info info = get_image_ssbo_samp_tex_src(ctx, &intr->src[0]); + struct tex_src_info info = get_image_ssbo_samp_tex_src(ctx, &intr->src[0], true); struct ir3_instruction *sam, *lod; unsigned flags, ncoords = ir3_get_image_coords(intr, &flags); type_t dst_type = nir_dest_bit_size(intr->dest) == 16 ? TYPE_U16 : TYPE_U32; @@ -1526,7 +1530,6 @@ emit_intrinsic_load_ssbo(struct ir3_context *ctx, { /* Note: isam currently can't handle vectorized loads/stores */ if (!(nir_intrinsic_access(intr) & ACCESS_CAN_REORDER) || - !ir3_bindless_resource(intr->src[0]) || intr->dest.ssa.num_components > 1) { ctx->funcs->emit_intrinsic_load_ssbo(ctx, intr, dst); return; @@ -1535,7 +1538,7 @@ emit_intrinsic_load_ssbo(struct ir3_context *ctx, struct ir3_block *b = ctx->block; struct ir3_instruction *offset = ir3_get_src(ctx, &intr->src[2])[0]; struct ir3_instruction *coords = ir3_collect(b, offset, create_immed(b, 0)); - struct tex_src_info info = get_image_ssbo_samp_tex_src(ctx, &intr->src[0]); + struct tex_src_info info = get_image_ssbo_samp_tex_src(ctx, &intr->src[0], false); unsigned num_components = intr->dest.ssa.num_components; struct ir3_instruction *sam = |