r600/sfn: use a per stream index register in GS

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7714>

diff --git a/src/gallium/drivers/r600/sfn/sfn_instruction_export.cpp b/src/gallium/drivers/r600/sfn/sfn_instruction_export.cpp
index cf40f2a..d5796d2 100644 (file)
--- a/src/gallium/drivers/r600/sfn/sfn_instruction_export.cpp
+++ b/src/gallium/drivers/r600/sfn/sfn_instruction_export.cpp
@@ -329,12 +329,13 @@ void MemRingOutIntruction::remap_registers_child(std::vector<rename_reg_pair>& m
    map[m_index->sel()].used = true;
 }
 
-void MemRingOutIntruction::patch_ring(int stream)
+void MemRingOutIntruction::patch_ring(int stream, PValue index)
 {
    const ECFOpCode ring_op[4] = {cf_mem_ring, cf_mem_ring1, cf_mem_ring2, cf_mem_ring3};
 
    assert(stream < 4);
    m_ring_op = ring_op[stream];
+   m_index = index;
 }
 
 }

diff --git a/src/gallium/drivers/r600/sfn/sfn_instruction_export.h b/src/gallium/drivers/r600/sfn/sfn_instruction_export.h
index a2c74e8..1bc8308 100644 (file)
--- a/src/gallium/drivers/r600/sfn/sfn_instruction_export.h
+++ b/src/gallium/drivers/r600/sfn/sfn_instruction_export.h
@@ -152,7 +152,7 @@ public:
    void replace_values_child(const ValueSet& candiates, PValue new_value) override;
    void remap_registers_child(std::vector<rename_reg_pair>& map,
                         ValueMap& values) override;
-   void patch_ring(int stream);
+   void patch_ring(int stream, PValue index);
 private:
    bool is_equal_to(const Instruction& lhs) const override;
    void do_print(std::ostream& os) const override;

diff --git a/src/gallium/drivers/r600/sfn/sfn_shader_geometry.cpp b/src/gallium/drivers/r600/sfn/sfn_shader_geometry.cpp
index 3862986..58ac270 100644 (file)
--- a/src/gallium/drivers/r600/sfn/sfn_shader_geometry.cpp
+++ b/src/gallium/drivers/r600/sfn/sfn_shader_geometry.cpp
@@ -65,7 +65,7 @@ bool GeometryShaderFromNir::do_emit_store_deref(const nir_variable *out_var, nir
 
    auto ir = new MemRingOutIntruction(cf_mem_ring, mem_write_ind, out_value,
                                       4 * out_var->data.driver_location,
-                                      instr->num_components, m_export_base);
+                                      instr->num_components, m_export_base[0]);
 
    streamout_data[out_var->data.location] = ir;
 
@@ -180,8 +180,14 @@ bool GeometryShaderFromNir::do_allocate_reserved_registers()
    m_invocation_id.reset(reg);
    inject_register(1, 3, m_invocation_id, false);
 
-   m_export_base = get_temp_register();
-   emit_instruction(new AluInstruction(op1_mov, m_export_base, Value::zero, {alu_write, alu_last_instr}));
+   m_export_base[0] = get_temp_register(0);
+   m_export_base[1] = get_temp_register(0);
+   m_export_base[2] = get_temp_register(0);
+   m_export_base[3] = get_temp_register(0);
+   emit_instruction(new AluInstruction(op1_mov, m_export_base[0], Value::zero, {alu_write, alu_last_instr}));
+   emit_instruction(new AluInstruction(op1_mov, m_export_base[1], Value::zero, {alu_write, alu_last_instr}));
+   emit_instruction(new AluInstruction(op1_mov, m_export_base[2], Value::zero, {alu_write, alu_last_instr}));
+   emit_instruction(new AluInstruction(op1_mov, m_export_base[3], Value::zero, {alu_write, alu_last_instr}));
 
    sh_info().ring_item_sizes[0] = m_next_input_ring_offset;
 
@@ -193,7 +199,7 @@ bool GeometryShaderFromNir::do_allocate_reserved_registers()
 
 void GeometryShaderFromNir::emit_adj_fix()
 {
-   PValue adjhelp0(new  GPRValue(m_export_base->sel(), 1));
+   PValue adjhelp0(new  GPRValue(m_export_base[0]->sel(), 1));
    emit_instruction(op2_and_int, adjhelp0, {m_primitive_id, Value::one_i}, {alu_write, alu_last_instr});
 
    int help2 = allocate_temp_register();
@@ -203,7 +209,7 @@ void GeometryShaderFromNir::emit_adj_fix()
    int rotate_indices[6] = {4, 5, 0, 1, 2, 3};
 
    reg_indices[0] = reg_indices[1] = reg_indices[2] = reg_indices[3] = help2;
-   reg_indices[4] = reg_indices[5] = m_export_base->sel();
+   reg_indices[4] = reg_indices[5] = m_export_base[0]->sel();
 
    std::array<PValue, 6> adjhelp;
 
@@ -270,7 +276,7 @@ bool GeometryShaderFromNir::emit_vertex(nir_intrinsic_instr* instr, bool cut)
 
    for(auto v: streamout_data) {
       if (stream == 0 || v.first != VARYING_SLOT_POS) {
-         v.second->patch_ring(stream);
+         v.second->patch_ring(stream, m_export_base[stream]);
          emit_instruction(v.second);
       } else
          delete v.second;
@@ -279,7 +285,7 @@ bool GeometryShaderFromNir::emit_vertex(nir_intrinsic_instr* instr, bool cut)
    emit_instruction(new EmitVertex(stream, cut));
 
    if (!cut)
-      emit_instruction(new AluInstruction(op2_add_int, m_export_base, m_export_base,
+      emit_instruction(new AluInstruction(op2_add_int, m_export_base[stream], m_export_base[stream],
                                           PValue(new LiteralValue(sh_info().noutput)),
                                           {alu_write, alu_last_instr}));
 

diff --git a/src/gallium/drivers/r600/sfn/sfn_shader_geometry.h b/src/gallium/drivers/r600/sfn/sfn_shader_geometry.h
index 3a66b9d..1e07769 100644 (file)
--- a/src/gallium/drivers/r600/sfn/sfn_shader_geometry.h
+++ b/src/gallium/drivers/r600/sfn/sfn_shader_geometry.h
@@ -64,7 +64,7 @@ private:
    std::array<PValue, 6> m_per_vertex_offsets;
    PValue m_primitive_id;
    PValue m_invocation_id;
-   PValue m_export_base;
+   PValue m_export_base[4];
    bool m_first_vertex_emitted;
 
    std::map<unsigned, ArrayDeref> m_in_array_deref;