aco/isel: Remove loop_context* parameter from begin_loop() / end_loop() helper functions

We can transparently create the context inside the functions, now.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40364>

src/amd/compiler/instruction_selection/aco_instruction_selection.h

@@ -248,8 +248,8 @@ isel_context setup_isel_context(Program* program, unsigned shader_count,
 /* aco_isel_cfg.cpp */
 void emit_loop_break(isel_context* ctx);
 void emit_loop_continue(isel_context* ctx);
-void begin_loop(isel_context* ctx, loop_context* lc);
-void end_loop(isel_context* ctx, loop_context* lc);
+void begin_loop(isel_context* ctx);
+void end_loop(isel_context* ctx);
 void begin_uniform_if_then(isel_context* ctx, if_context* ic, Temp cond);
 void begin_uniform_if_else(isel_context* ctx, if_context* ic, bool logical_else = true);
 void end_uniform_if(isel_context* ctx, if_context* ic, bool logical_else = true);

src/amd/compiler/instruction_selection/aco_isel_cfg.cpp

@@ -116,15 +116,16 @@ update_exec_info(isel_context* ctx)
 }
 
 void
-begin_loop(isel_context* ctx, loop_context* lc)
+begin_loop(isel_context* ctx)
 {
+   loop_context lc;
    append_logical_end(ctx);
    ctx->block->kind |= block_kind_loop_preheader | block_kind_uniform;
    Builder bld(ctx->program, ctx->block);
    bld.branch(aco_opcode::p_branch);
    unsigned loop_preheader_idx = ctx->block->index;
 
-   lc->loop_exit.kind |= (block_kind_loop_exit | (ctx->block->kind & block_kind_top_level));
+   lc.loop_exit.kind |= (block_kind_loop_exit | (ctx->block->kind & block_kind_top_level));
 
    ctx->program->next_loop_depth++;
 
@@ -135,17 +136,18 @@ begin_loop(isel_context* ctx, loop_context* lc)
 
    append_logical_start(ctx->block);
 
-   lc->cf_info_old = ctx->cf_info;
-   lc->header_idx = loop_header->index;
+   lc.cf_info_old = ctx->cf_info;
+   lc.header_idx = loop_header->index;
    ctx->cf_info.parent_loop = {false, false};
    ctx->cf_info.parent_if.is_divergent = false;
+   ctx->loop_stack.push_back(std::move(lc));
 
    /* Never enter a loop with empty exec mask. */
    assert(!ctx->cf_info.exec.empty());
 }
 
 void
-end_loop(isel_context* ctx, loop_context* lc)
+end_loop(isel_context* ctx)
 {
    /* No need to check exec.potentially_empty_break/continue originating inside the loop. In the
     * only case where it's possible at this point (divergent break after divergent continue), we
@@ -153,7 +155,8 @@ end_loop(isel_context* ctx, loop_context* lc)
     * divergent control flow requires WQM.
     */
    assert(!ctx->cf_info.exec.potentially_empty_discard);
-   Block& header = ctx->program->blocks[lc->header_idx];
+   loop_context& lc = ctx->loop_stack.back();
+   Block& header = ctx->program->blocks[lc.header_idx];
 
    /* Add the trivial continue. */
    if (!ctx->cf_info.has_branch) {
@@ -176,14 +179,16 @@ end_loop(isel_context* ctx, loop_context* lc)
 
    /* emit loop successor block */
    ctx->program->next_loop_depth--;
-   ctx->block = ctx->program->insert_block(std::move(lc->loop_exit));
+   ctx->block = ctx->program->insert_block(std::move(lc.loop_exit));
    append_logical_start(ctx->block);
 
    /* Propagate information about discards and restore previous CF info. */
-   lc->cf_info_old.exec.potentially_empty_discard |= ctx->cf_info.exec.potentially_empty_discard;
-   lc->cf_info_old.had_divergent_discard |= ctx->cf_info.had_divergent_discard;
-   ctx->cf_info = lc->cf_info_old;
+   lc.cf_info_old.exec.potentially_empty_discard |= ctx->cf_info.exec.potentially_empty_discard;
+   lc.cf_info_old.had_divergent_discard |= ctx->cf_info.had_divergent_discard;
+   ctx->cf_info = lc.cf_info_old;
    update_exec_info(ctx);
+
+   ctx->loop_stack.pop_back();
 }
 
 void

src/amd/compiler/instruction_selection/aco_select_nir.cpp

@@ -942,16 +942,14 @@ visit_loop(isel_context* ctx, nir_loop* loop)
 {
    assert(!nir_loop_has_continue_construct(loop));
 
-   ctx->loop_stack.push_back(loop_context());
-   begin_loop(ctx, &ctx->loop_stack.back());
+   begin_loop(ctx);
    ctx->cf_info.parent_loop.has_divergent_break =
       loop->divergent_break && nir_block_num_preds(nir_loop_first_block(loop)) > 1;
    ctx->cf_info.in_divergent_cf |= ctx->cf_info.parent_loop.has_divergent_break;
 
    visit_cf_list(ctx, &loop->body);
 
-   end_loop(ctx, &ctx->loop_stack.back());
-   ctx->loop_stack.pop_back();
+   end_loop(ctx);
 }
 
 void

src/amd/compiler/instruction_selection/aco_select_nir_intrinsics.cpp

@@ -3773,8 +3773,7 @@ pops_await_overlapped_waves(isel_context* ctx)
 
    /* Await the overlapped waves. */
 
-   ctx->loop_stack.push_back(loop_context());
-   begin_loop(ctx, &ctx->loop_stack.back());
+   begin_loop(ctx);
    bld.reset(ctx->block);
 
    const Temp exiting_wave_id = bld.pseudo(aco_opcode::p_pops_gfx9_add_exiting_wave_id, bld.def(s1),
@@ -3795,7 +3794,7 @@ pops_await_overlapped_waves(isel_context* ctx)
    /* Sleep before rechecking to let overlapped waves run for some time. */
    bld.sopp(aco_opcode::s_sleep, ctx->program->gfx_level >= GFX10 ? UINT16_MAX : 3);
 
-   end_loop(ctx, &ctx->loop_stack.back());
+   end_loop(ctx);
    bld.reset(ctx->block);
 
    /* Indicate the wait has been done to subsequent compilation stages. */

src/amd/compiler/instruction_selection/aco_select_trap_handler.cpp

@@ -143,8 +143,7 @@ dump_vgprs_to_mem(isel_context* ctx, Builder& bld, Operand rsrc)
                Operand(PhysReg{1}, s1) /* SRC0 mode */);
    }
 
-   ctx->loop_stack.push_back(loop_context());
-   begin_loop(ctx, &ctx->loop_stack.back());
+   begin_loop(ctx);
    {
       bld.reset(ctx->block);
 
@@ -177,7 +176,7 @@ dump_vgprs_to_mem(isel_context* ctx, Builder& bld, Operand rsrc)
       }
       end_uniform_if(ctx, &loop_break);
    }
-   end_loop(ctx, &ctx->loop_stack.back());
+   end_loop(ctx);
    bld.reset(ctx->block);
 
    if (ctx->program->gfx_level < GFX10) {
@@ -240,8 +239,7 @@ dump_lds_to_mem(isel_context* ctx, Builder& bld, Operand rsrc)
 
       Operand m = load_lds_size_m0(bld);
 
-      ctx->loop_stack.push_back(loop_context());
-      begin_loop(ctx, &ctx->loop_stack.back());
+      begin_loop(ctx);
       {
          bld.reset(ctx->block);
 
@@ -274,7 +272,7 @@ dump_lds_to_mem(isel_context* ctx, Builder& bld, Operand rsrc)
          }
          end_uniform_if(ctx, &loop_break);
       }
-      end_loop(ctx, &ctx->loop_stack.back());
+      end_loop(ctx);
       bld.reset(ctx->block);
    }
    begin_uniform_if_else(ctx, &ic);