[Mesa-dev] [PATCH 15/19] radeonsi: emit TGSI_OPCODE_BALLOT

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle at amd.com>

---
 src/gallium/drivers/radeonsi/si_shader.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 12eca9f..37346c6 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -5238,20 +5238,37 @@ static void vote_eq_emit(
 	vote_set = si_emit_ballot(ctx, emit_data->args[0]);
 
 	all = LLVMBuildICmp(gallivm->builder, LLVMIntEQ, vote_set, active_set, "");
 	none = LLVMBuildICmp(gallivm->builder, LLVMIntEQ,
 			     vote_set, LLVMConstInt(ctx->i64, 0, 0), "");
 	tmp = LLVMBuildOr(gallivm->builder, all, none, "");
 	emit_data->output[emit_data->chan] =
 		LLVMBuildSExt(gallivm->builder, tmp, ctx->i32, "");
 }
 
+static void ballot_emit(
+	const struct lp_build_tgsi_action *action,
+	struct lp_build_tgsi_context *bld_base,
+	struct lp_build_emit_data *emit_data)
+{
+	struct si_shader_context *ctx = si_shader_context(bld_base);
+	LLVMBuilderRef builder = ctx->gallivm.builder;
+	LLVMValueRef tmp;
+
+	tmp = lp_build_emit_fetch(bld_base, emit_data->inst, 0, TGSI_CHAN_X);
+	tmp = si_emit_ballot(ctx, tmp);
+	tmp = LLVMBuildBitCast(builder, tmp, ctx->v2i32, "");
+
+	emit_data->output[0] = LLVMBuildExtractElement(builder, tmp, ctx->i32_0, "");
+	emit_data->output[1] = LLVMBuildExtractElement(builder, tmp, ctx->i32_1, "");
+}
+
 static unsigned si_llvm_get_stream(struct lp_build_tgsi_context *bld_base,
 				       struct lp_build_emit_data *emit_data)
 {
 	struct si_shader_context *ctx = si_shader_context(bld_base);
 	struct tgsi_src_register src0 = emit_data->inst->Src[0].Register;
 	LLVMValueRef imm;
 	unsigned stream;
 
 	assert(src0.File == TGSI_FILE_IMMEDIATE);
 
@@ -6690,20 +6707,21 @@ static void si_init_shader_ctx(struct si_shader_context *ctx,
 	bld_base->op_actions[TGSI_OPCODE_CLOCK].emit = clock_emit;
 
 	bld_base->op_actions[TGSI_OPCODE_DDX].emit = si_llvm_emit_ddxy;
 	bld_base->op_actions[TGSI_OPCODE_DDY].emit = si_llvm_emit_ddxy;
 	bld_base->op_actions[TGSI_OPCODE_DDX_FINE].emit = si_llvm_emit_ddxy;
 	bld_base->op_actions[TGSI_OPCODE_DDY_FINE].emit = si_llvm_emit_ddxy;
 
 	bld_base->op_actions[TGSI_OPCODE_VOTE_ALL].emit = vote_all_emit;
 	bld_base->op_actions[TGSI_OPCODE_VOTE_ANY].emit = vote_any_emit;
 	bld_base->op_actions[TGSI_OPCODE_VOTE_EQ].emit = vote_eq_emit;
+	bld_base->op_actions[TGSI_OPCODE_BALLOT].emit = ballot_emit;
 
 	bld_base->op_actions[TGSI_OPCODE_EMIT].emit = si_llvm_emit_vertex;
 	bld_base->op_actions[TGSI_OPCODE_ENDPRIM].emit = si_llvm_emit_primitive;
 	bld_base->op_actions[TGSI_OPCODE_BARRIER].emit = si_llvm_emit_barrier;
 }
 
 #define EXP_TARGET (HAVE_LLVM >= 0x0500 ? 0 : 3)
 #define EXP_OUT0 (HAVE_LLVM >= 0x0500 ? 2 : 5)
 
 /* Return true if the PARAM export has been eliminated. */
-- 
2.9.3