The old vote_eq implementation supported only booleans, but now
we have to support arbitrary values, so use the read_first_invocation
intrinsic + ballot.
I took this as an opportunity to figure out how easy it was to do this
in nir instead of in the nir_to_llvm pass, and it actually turned out
pretty okay IMO. Only creating the pass is some extra code.
Reviewed-by: Dave Airlie <airl...@redhat.com>
---
src/amd/Makefile.sources | 1 +
src/amd/common/ac_lower_subgroups.c | 92 +++++++++++++++++++++++++++++++++++++
src/amd/common/ac_nir_to_llvm.c | 8 ++--
src/amd/common/ac_nir_to_llvm.h | 2 +
src/amd/common/meson.build | 1 +
5 files changed, 99 insertions(+), 5 deletions(-)
create mode 100644 src/amd/common/ac_lower_subgroups.c
diff --git a/src/amd/Makefile.sources b/src/amd/Makefile.sources
index 10c4827e19..f57dd403c5 100644
--- a/src/amd/Makefile.sources
+++ b/src/amd/Makefile.sources
@@ -44,6 +44,7 @@ AMD_COMPILER_FILES = \
common/ac_llvm_helper.cpp \
common/ac_llvm_util.c \
common/ac_llvm_util.h \
+ common/ac_lower_subgroups.c \
common/ac_shader_abi.h \
common/ac_shader_info.c \
common/ac_shader_info.h \
diff --git a/src/amd/common/ac_lower_subgroups.c
b/src/amd/common/ac_lower_subgroups.c
new file mode 100644
index 0000000000..d0782b481b
--- /dev/null
+++ b/src/amd/common/ac_lower_subgroups.c
@@ -0,0 +1,92 @@
+/*
+ * Copyright © 2018 Google Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#include "nir/nir.h"
+#include "nir/nir_builder.h"
+
+#include "ac_nir_to_llvm.h"
+
+static nir_ssa_def *ac_lower_subgroups_intrin(nir_builder *b,
nir_intrinsic_instr *intrin)
+{
+ switch(intrin->intrinsic) {
+ case nir_intrinsic_vote_ieq:
+ case nir_intrinsic_vote_feq: {
+ nir_intrinsic_instr *rfi =
+ nir_intrinsic_instr_create(b->shader,
nir_intrinsic_read_first_invocation);
+ nir_ssa_dest_init(&rfi->instr, &rfi->dest,
+ 1, intrin->src[0].ssa->bit_size, NULL);
+ nir_src_copy(&rfi->src[0], &intrin->src[0], rfi);
+ rfi->num_components = 1;
+
+ nir_ssa_def *is_ne;
+ if (intrin->intrinsic == nir_intrinsic_vote_feq)
+ is_ne = nir_fne(b, &rfi->dest.ssa, intrin->src[0].ssa);
+ else
+ is_ne = nir_ine(b, &rfi->dest.ssa, intrin->src[0].ssa);
+
+ nir_intrinsic_instr *ballot =
+ nir_intrinsic_instr_create(b->shader,
nir_intrinsic_ballot);
+ nir_ssa_dest_init(&ballot->instr, &ballot->dest,
+ 1, 64, NULL);
+ ballot->src[0] = nir_src_for_ssa(is_ne);
+ ballot->num_components = 1;
+
+ return nir_ieq(b, &ballot->dest.ssa, nir_imm_int64(b, 0));
+ }
+ default:
+ return NULL;
+ }
+}
+
+bool ac_lower_subgroups(struct nir_shader *shader)
+{
+ bool progress = false;
+
+ nir_foreach_function(function, shader) {
+ if (!function->impl)
+ continue;
+
+ nir_builder b;
+ nir_builder_init(&b, function->impl);
+
+ nir_foreach_block(block, function->impl) {
+ nir_foreach_instr_safe(instr, block) {
+ if (instr->type != nir_instr_type_intrinsic)
+ continue;
+
+ nir_intrinsic_instr *intrin =
nir_instr_as_intrinsic(instr);
+ b.cursor = nir_before_instr(instr);
+
+ nir_ssa_def *lower =
ac_lower_subgroups_intrin(&b, intrin);
+ if (!lower)
+ continue;
+
+ nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
nir_src_for_ssa(lower));
+ nir_instr_remove(instr);
+ progress = true;
+ }
+ }
+ }
+
+ return progress;
+}
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index d2df2837c8..3bf1a4ba3c 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -4563,11 +4563,6 @@ static void visit_intrinsic(struct ac_nir_context *ctx,
result = LLVMBuildSExt(ctx->ac.builder, tmp, ctx->ac.i32, "");
break;
}
- case nir_intrinsic_vote_eq: {
- LLVMValueRef tmp = ac_build_vote_eq(&ctx->ac, get_src(ctx,
instr->src[0]));
- result = LLVMBuildSExt(ctx->ac.builder, tmp, ctx->ac.i32, "");
- break;
- }
default:
fprintf(stderr, "Unknown intrinsic: ");
nir_print_instr(&instr->instr, stderr);
@@ -6737,6 +6732,9 @@ void ac_nir_translate(struct ac_llvm_context *ac, struct
ac_shader_abi *abi,
struct ac_nir_context ctx = {};
struct nir_function *func;
+ /* Last minute passes for both radv & radeonsi */
+ ac_lower_subgroups(nir);
+
ctx.ac = *ac;
ctx.abi = abi;
diff --git a/src/amd/common/ac_nir_to_llvm.h b/src/amd/common/ac_nir_to_llvm.h
index eea393a9c2..306039aecf 100644
--- a/src/amd/common/ac_nir_to_llvm.h
+++ b/src/amd/common/ac_nir_to_llvm.h
@@ -234,4 +234,6 @@ void ac_lower_indirect_derefs(struct nir_shader *nir, enum
chip_class);
void ac_nir_translate(struct ac_llvm_context *ac, struct ac_shader_abi *abi,
struct nir_shader *nir);
+bool ac_lower_subgroups(struct nir_shader *shader);
+
#endif /* AC_NIR_TO_LLVM_H */
diff --git a/src/amd/common/meson.build b/src/amd/common/meson.build
index 22c13b955f..6c5c2f45ac 100644
--- a/src/amd/common/meson.build
+++ b/src/amd/common/meson.build
@@ -35,6 +35,7 @@ amd_common_files = files(
'ac_llvm_helper.cpp',
'ac_llvm_util.c',
'ac_llvm_util.h',
+ 'ac_lower_subgroups.c',
'ac_shader_abi.h',
'ac_shader_info.c',
'ac_shader_info.h',
