i965/fs: Add an allow_spilling flag to brw_compile_fs

This allows us to disable spilling for blorp shaders since blorp state
setup doesn't handle spilling.  Without this, blorp fails hard if you run
with INTEL_DEBUG=spill.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Tested-by: Francisco Jerez <currojerez@riseup.net>

diff --git a/src/intel/vulkan/anv_pipeline.c b/src/intel/vulkan/anv_pipeline.c
index a8e31b13cf11931c3f66065e100e98358e145df3..7d265d8d2158e3c3033179a7ee972883b6a99dfd 100644 (file)
--- a/src/intel/vulkan/anv_pipeline.c
+++ b/src/intel/vulkan/anv_pipeline.c
@@ -675,7 +675,8 @@ anv_pipeline_compile_fs(struct anv_pipeline *pipeline,
       unsigned code_size;
       const unsigned *shader_code =
          brw_compile_fs(compiler, NULL, mem_ctx, &key, &prog_data, nir,
-                        NULL, -1, -1, pipeline->use_repclear, &code_size, NULL);
+                        NULL, -1, -1, true, pipeline->use_repclear,
+                        &code_size, NULL);
       if (shader_code == NULL) {
          ralloc_free(mem_ctx);
          return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);

diff --git a/src/mesa/drivers/dri/i965/brw_blorp.c b/src/mesa/drivers/dri/i965/brw_blorp.c
index 09a0fd1a632114ca4497bbe76126285f832f0a1a..9590968eef263535c1452cc182ba5f7d87d0820e 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_blorp.c
+++ b/src/mesa/drivers/dri/i965/brw_blorp.c
@@ -223,7 +223,7 @@ brw_blorp_compile_nir_shader(struct brw_context *brw, struct nir_shader *nir,
 
    const unsigned *program =
       brw_compile_fs(compiler, brw, mem_ctx, wm_key, &wm_prog_data, nir,
-                     NULL, -1, -1, use_repclear, program_size, NULL);
+                     NULL, -1, -1, false, use_repclear, program_size, NULL);
 
    /* Copy the relavent bits of wm_prog_data over into the blorp prog data */
    prog_data->dispatch_8 = wm_prog_data.dispatch_8;

diff --git a/src/mesa/drivers/dri/i965/brw_compiler.h b/src/mesa/drivers/dri/i965/brw_compiler.h
index 0db1d0d3561d991dadf9a773b945835aa1a5778b..f4b9d3dddc6aa91dc4be4b79d44c8ee14ef2a70b 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_compiler.h
+++ b/src/mesa/drivers/dri/i965/brw_compiler.h
@@ -790,6 +790,7 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
                struct gl_program *prog,
                int shader_time_index8,
                int shader_time_index16,
+               bool allow_spilling,
                bool use_rep_send,
                unsigned *final_assembly_size,
                char **error_str);

diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
index e62f2fe578df2c07b7d381acd2db711e56f6bbcd..65b64b6d3f641da3e8735c068e9ad71edd6c3f91 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
@@ -5486,7 +5486,7 @@ fs_visitor::fixup_3src_null_dest()
 }
 
 void
-fs_visitor::allocate_registers()
+fs_visitor::allocate_registers(bool allow_spilling)
 {
    bool allocated_without_spills;
 
@@ -5496,6 +5496,8 @@ fs_visitor::allocate_registers()
       SCHEDULE_PRE_LIFO,
    };
 
+   bool spill_all = allow_spilling && (INTEL_DEBUG & DEBUG_SPILL_FS);
+
    /* Try each scheduling heuristic to see if it can successfully register
     * allocate without spilling.  They should be ordered by decreasing
     * performance but increasing likelihood of allocating.
@@ -5507,7 +5509,7 @@ fs_visitor::allocate_registers()
          assign_regs_trivial();
          allocated_without_spills = true;
       } else {
-         allocated_without_spills = assign_regs(false);
+         allocated_without_spills = assign_regs(false, spill_all);
       }
       if (allocated_without_spills)
          break;
@@ -5532,12 +5534,14 @@ fs_visitor::allocate_registers()
       /* Since we're out of heuristics, just go spill registers until we
        * get an allocation.
        */
-      while (!assign_regs(true)) {
+      while (!assign_regs(true, spill_all)) {
          if (failed)
             break;
       }
    }
 
+   assert(last_scratch == 0 || allow_spilling);
+
    /* This must come after all optimization and register allocation, since
     * it inserts dead code that happens to have side effects, and it does
     * so based on the actual physical registers in use.
@@ -5583,7 +5587,7 @@ fs_visitor::run_vs(gl_clip_plane *clip_planes)
    assign_vs_urb_setup();
 
    fixup_3src_null_dest();
-   allocate_registers();
+   allocate_registers(true);
 
    return !failed;
 }
@@ -5665,7 +5669,7 @@ fs_visitor::run_tcs_single_patch()
    assign_tcs_single_patch_urb_setup();
 
    fixup_3src_null_dest();
-   allocate_registers();
+   allocate_registers(true);
 
    return !failed;
 }
@@ -5699,7 +5703,7 @@ fs_visitor::run_tes()
    assign_tes_urb_setup();
 
    fixup_3src_null_dest();
-   allocate_registers();
+   allocate_registers(true);
 
    return !failed;
 }
@@ -5748,13 +5752,13 @@ fs_visitor::run_gs()
    assign_gs_urb_setup();
 
    fixup_3src_null_dest();
-   allocate_registers();
+   allocate_registers(true);
 
    return !failed;
 }
 
 bool
-fs_visitor::run_fs(bool do_rep_send)
+fs_visitor::run_fs(bool allow_spilling, bool do_rep_send)
 {
    brw_wm_prog_data *wm_prog_data = (brw_wm_prog_data *) this->prog_data;
    brw_wm_prog_key *wm_key = (brw_wm_prog_key *) this->key;
@@ -5818,7 +5822,7 @@ fs_visitor::run_fs(bool do_rep_send)
       assign_urb_setup();
 
       fixup_3src_null_dest();
-      allocate_registers();
+      allocate_registers(allow_spilling);
 
       if (failed)
          return false;
@@ -5861,7 +5865,7 @@ fs_visitor::run_cs()
    assign_curb_setup();
 
    fixup_3src_null_dest();
-   allocate_registers();
+   allocate_registers(true);
 
    if (failed)
       return false;
@@ -5986,6 +5990,7 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
                const nir_shader *src_shader,
                struct gl_program *prog,
                int shader_time_index8, int shader_time_index16,
+               bool allow_spilling,
                bool use_rep_send,
                unsigned *final_assembly_size,
                char **error_str)
@@ -6029,7 +6034,7 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
    fs_visitor v8(compiler, log_data, mem_ctx, key,
                  &prog_data->base, prog, shader, 8,
                  shader_time_index8);
-   if (!v8.run_fs(false /* do_rep_send */)) {
+   if (!v8.run_fs(allow_spilling, false /* do_rep_send */)) {
       if (error_str)
          *error_str = ralloc_strdup(mem_ctx, v8.fail_msg);
 
@@ -6047,7 +6052,7 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
                      &prog_data->base, prog, shader, 16,
                      shader_time_index16);
       v16.import_uniforms(&v8);
-      if (!v16.run_fs(use_rep_send)) {
+      if (!v16.run_fs(allow_spilling, use_rep_send)) {
          compiler->shader_perf_log(log_data,
                                    "SIMD16 shader failed to compile: %s",
                                    v16.fail_msg);

diff --git a/src/mesa/drivers/dri/i965/brw_fs.h b/src/mesa/drivers/dri/i965/brw_fs.h
index f9e6792e8d643f26eb653906ed22dfa032aeba3d..ac270cdefab0a5abb386732a1145f7d9190a5768 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs.h
+++ b/src/mesa/drivers/dri/i965/brw_fs.h
@@ -105,14 +105,14 @@ public:
                                    uint32_t const_offset);
    void DEP_RESOLVE_MOV(const brw::fs_builder &bld, int grf);
 
-   bool run_fs(bool do_rep_send);
+   bool run_fs(bool allow_spilling, bool do_rep_send);
    bool run_vs(gl_clip_plane *clip_planes);
    bool run_tcs_single_patch();
    bool run_tes();
    bool run_gs();
    bool run_cs();
    void optimize();
-   void allocate_registers();
+   void allocate_registers(bool allow_spilling);
    void setup_fs_payload_gen4();
    void setup_fs_payload_gen6();
    void setup_vs_payload();
@@ -127,7 +127,7 @@ public:
    void assign_tcs_single_patch_urb_setup();
    void assign_tes_urb_setup();
    void assign_gs_urb_setup();
-   bool assign_regs(bool allow_spilling);
+   bool assign_regs(bool allow_spilling, bool spill_all);
    void assign_regs_trivial();
    void calculate_payload_ranges(int payload_node_count,
                                  int *payload_last_use_ip);

diff --git a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
index 2347cd5d33f726233b1f04d29c337623584594c0..e65f73bc0c729ff23d0a13f58dd180cefdfb76fe 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
@@ -542,7 +542,7 @@ setup_mrf_hack_interference(fs_visitor *v, struct ra_graph *g,
 }
 
 bool
-fs_visitor::assign_regs(bool allow_spilling)
+fs_visitor::assign_regs(bool allow_spilling, bool spill_all)
 {
    /* Most of this allocation was written for a reg_width of 1
     * (dispatch_width == 8).  In extending to SIMD16, the code was
@@ -668,7 +668,7 @@ fs_visitor::assign_regs(bool allow_spilling)
    }
 
    /* Debug of register spilling: Go spill everything. */
-   if (unlikely(INTEL_DEBUG & DEBUG_SPILL_FS)) {
+   if (unlikely(spill_all)) {
       int reg = choose_spill_reg(g);
 
       if (reg != -1) {

diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c
index 6473b021245e10860eca41ced8b92bcbe314c7e8..d5841f38b2eeba16039a954e320aab84386248f4 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_wm.c
+++ b/src/mesa/drivers/dri/i965/brw_wm.c
@@ -137,7 +137,8 @@ brw_codegen_wm_prog(struct brw_context *brw,
    program = brw_compile_fs(brw->intelScreen->compiler, brw, mem_ctx,
                             key, &prog_data, fp->program.Base.nir,
                             &fp->program.Base, st_index8, st_index16,
-                            brw->use_rep_send, &program_size, &error_str);
+                            true, brw->use_rep_send,
+                            &program_size, &error_str);
    if (program == NULL) {
       if (prog) {
          prog->LinkStatus = false;