intel/compiler: Allow MESA_SHADER_KERNEL

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6280>

diff --git a/src/intel/compiler/brw_compiler.c b/src/intel/compiler/brw_compiler.c
index e58ed67..0dc3d90 100644 (file)
--- a/src/intel/compiler/brw_compiler.c
+++ b/src/intel/compiler/brw_compiler.c
@@ -112,7 +112,7 @@ brw_compiler_create(void *mem_ctx, const struct gen_device_info *devinfo)
 
    if (devinfo->gen >= 10) {
       /* We don't support vec4 mode on Cannonlake. */
-      for (int i = MESA_SHADER_VERTEX; i < MESA_SHADER_STAGES; i++)
+      for (int i = MESA_SHADER_VERTEX; i < MESA_ALL_SHADER_STAGES; i++)
          compiler->scalar_stage[i] = true;
    } else {
       compiler->scalar_stage[MESA_SHADER_VERTEX] =
@@ -158,7 +158,7 @@ brw_compiler_create(void *mem_ctx, const struct gen_device_info *devinfo)
       int64_options |= nir_lower_imul_2x32_64;
 
    /* We want the GLSL compiler to emit code that uses condition codes */
-   for (int i = 0; i < MESA_SHADER_STAGES; i++) {
+   for (int i = 0; i < MESA_ALL_SHADER_STAGES; i++) {
       compiler->glsl_compiler_options[i].MaxUnrollIterations = 0;
       compiler->glsl_compiler_options[i].MaxIfDepth =
          devinfo->gen < 6 ? 16 : UINT_MAX;
@@ -247,6 +247,7 @@ brw_prog_data_size(gl_shader_stage stage)
       [MESA_SHADER_GEOMETRY]  = sizeof(struct brw_gs_prog_data),
       [MESA_SHADER_FRAGMENT]  = sizeof(struct brw_wm_prog_data),
       [MESA_SHADER_COMPUTE]   = sizeof(struct brw_cs_prog_data),
+      [MESA_SHADER_KERNEL]    = sizeof(struct brw_cs_prog_data),
    };
    assert((int)stage >= 0 && stage < ARRAY_SIZE(stage_sizes));
    return stage_sizes[stage];
@@ -262,6 +263,7 @@ brw_prog_key_size(gl_shader_stage stage)
       [MESA_SHADER_GEOMETRY]  = sizeof(struct brw_gs_prog_key),
       [MESA_SHADER_FRAGMENT]  = sizeof(struct brw_wm_prog_key),
       [MESA_SHADER_COMPUTE]   = sizeof(struct brw_cs_prog_key),
+      [MESA_SHADER_KERNEL]    = sizeof(struct brw_cs_prog_key),
    };
    assert((int)stage >= 0 && stage < ARRAY_SIZE(stage_sizes));
    return stage_sizes[stage];

diff --git a/src/intel/compiler/brw_compiler.h b/src/intel/compiler/brw_compiler.h
index ccbb5cc..8df2ee5 100644 (file)
--- a/src/intel/compiler/brw_compiler.h
+++ b/src/intel/compiler/brw_compiler.h
@@ -92,9 +92,9 @@ struct brw_compiler {
    void (*shader_debug_log)(void *, const char *str, ...) PRINTFLIKE(2, 3);
    void (*shader_perf_log)(void *, const char *str, ...) PRINTFLIKE(2, 3);
 
-   bool scalar_stage[MESA_SHADER_STAGES];
+   bool scalar_stage[MESA_ALL_SHADER_STAGES];
    bool use_tcs_8_patch;
-   struct gl_shader_compiler_options glsl_compiler_options[MESA_SHADER_STAGES];
+   struct gl_shader_compiler_options glsl_compiler_options[MESA_ALL_SHADER_STAGES];
 
    /**
     * Apply workarounds for SIN and COS output range problems.

diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp
index dbf7391..ec08d84 100644 (file)
--- a/src/intel/compiler/brw_fs.cpp
+++ b/src/intel/compiler/brw_fs.cpp
@@ -7841,7 +7841,7 @@ fs_visitor::allocate_registers(bool allow_spilling)
 
       prog_data->total_scratch = brw_get_scratch_size(last_scratch);
 
-      if (stage == MESA_SHADER_COMPUTE) {
+      if (stage == MESA_SHADER_COMPUTE || stage == MESA_SHADER_KERNEL) {
          if (devinfo->is_haswell) {
             /* According to the MEDIA_VFE_STATE's "Per Thread Scratch Space"
              * field documentation, Haswell supports a minimum of 2kB of
@@ -8229,7 +8229,7 @@ fs_visitor::run_fs(bool allow_spilling, bool do_rep_send)
 bool
 fs_visitor::run_cs(bool allow_spilling)
 {
-   assert(stage == MESA_SHADER_COMPUTE);
+   assert(stage == MESA_SHADER_COMPUTE || stage == MESA_SHADER_KERNEL);
 
    setup_cs_payload();
 
@@ -8796,7 +8796,7 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
 fs_reg *
 fs_visitor::emit_cs_work_group_id_setup()
 {
-   assert(stage == MESA_SHADER_COMPUTE);
+   assert(stage == MESA_SHADER_COMPUTE || stage == MESA_SHADER_KERNEL);
 
    fs_reg *reg = new(this->mem_ctx) fs_reg(vgrf(glsl_type::uvec3_type));
 

diff --git a/src/intel/compiler/brw_fs_nir.cpp b/src/intel/compiler/brw_fs_nir.cpp
index afbb436..f105257 100644 (file)
--- a/src/intel/compiler/brw_fs_nir.cpp
+++ b/src/intel/compiler/brw_fs_nir.cpp
@@ -100,7 +100,7 @@ fs_visitor::nir_setup_uniforms()
 
    uniforms = nir->num_uniforms / 4;
 
-   if (stage == MESA_SHADER_COMPUTE) {
+   if (stage == MESA_SHADER_COMPUTE || stage == MESA_SHADER_KERNEL) {
       /* Add uniforms for builtins after regular NIR uniforms. */
       assert(uniforms == prog_data->nr_params);
 
@@ -184,7 +184,8 @@ emit_system_values_block(nir_block *block, fs_visitor *v)
          break;
 
       case nir_intrinsic_load_work_group_id:
-         assert(v->stage == MESA_SHADER_COMPUTE);
+         assert(v->stage == MESA_SHADER_COMPUTE ||
+                v->stage == MESA_SHADER_KERNEL);
          reg = &v->nir_system_values[SYSTEM_VALUE_WORK_GROUP_ID];
          if (reg->file == BAD_FILE)
             *reg = *v->emit_cs_work_group_id_setup();
@@ -489,6 +490,7 @@ fs_visitor::nir_emit_instr(nir_instr *instr)
          nir_emit_fs_intrinsic(abld, nir_instr_as_intrinsic(instr));
          break;
       case MESA_SHADER_COMPUTE:
+      case MESA_SHADER_KERNEL:
          nir_emit_cs_intrinsic(abld, nir_instr_as_intrinsic(instr));
          break;
       default:
@@ -3714,7 +3716,7 @@ void
 fs_visitor::nir_emit_cs_intrinsic(const fs_builder &bld,
                                   nir_intrinsic_instr *instr)
 {
-   assert(stage == MESA_SHADER_COMPUTE);
+   assert(stage == MESA_SHADER_COMPUTE || stage == MESA_SHADER_KERNEL);
    struct brw_cs_prog_data *cs_prog_data = brw_cs_prog_data(prog_data);
 
    fs_reg dest;
@@ -3792,7 +3794,7 @@ fs_visitor::nir_emit_cs_intrinsic(const fs_builder &bld,
 
    case nir_intrinsic_load_shared: {
       assert(devinfo->gen >= 7);
-      assert(stage == MESA_SHADER_COMPUTE);
+      assert(stage == MESA_SHADER_COMPUTE || stage == MESA_SHADER_KERNEL);
 
       const unsigned bit_size = nir_dest_bit_size(instr->dest);
       fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
@@ -3828,7 +3830,7 @@ fs_visitor::nir_emit_cs_intrinsic(const fs_builder &bld,
 
    case nir_intrinsic_store_shared: {
       assert(devinfo->gen >= 7);
-      assert(stage == MESA_SHADER_COMPUTE);
+      assert(stage == MESA_SHADER_COMPUTE || stage == MESA_SHADER_KERNEL);
 
       const unsigned bit_size = nir_src_bit_size(instr->src[0]);
       fs_reg srcs[SURFACE_LOGICAL_NUM_SRCS];
@@ -4280,7 +4282,7 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
          break;
       }
 
-      if (stage != MESA_SHADER_COMPUTE)
+      if (stage != MESA_SHADER_COMPUTE && stage != MESA_SHADER_KERNEL)
          slm_fence = false;
 
       /* If the workgroup fits in a single HW thread, the messages for SLM are

diff --git a/src/intel/compiler/brw_fs_visitor.cpp b/src/intel/compiler/brw_fs_visitor.cpp
index 5e06f10..d891885 100644 (file)
--- a/src/intel/compiler/brw_fs_visitor.cpp
+++ b/src/intel/compiler/brw_fs_visitor.cpp
@@ -845,7 +845,7 @@ fs_visitor::emit_cs_terminate()
    assert(devinfo->gen >= 7);
 
    /* We are getting the thread ID from the compute shader header */
-   assert(stage == MESA_SHADER_COMPUTE);
+   assert(stage == MESA_SHADER_COMPUTE || stage == MESA_SHADER_KERNEL);
 
    /* We can't directly send from g0, since sends with EOT have to use
     * g112-127. So, copy it to a virtual register, The register allocator will
@@ -880,7 +880,7 @@ fs_visitor::emit_barrier()
    }
 
    /* We are getting the barrier ID from the compute shader header */
-   assert(stage == MESA_SHADER_COMPUTE);
+   assert(stage == MESA_SHADER_COMPUTE || stage == MESA_SHADER_KERNEL);
 
    fs_reg payload = fs_reg(VGRF, alloc.allocate(1), BRW_REGISTER_TYPE_UD);
 

diff --git a/src/intel/compiler/brw_nir_lower_cs_intrinsics.c b/src/intel/compiler/brw_nir_lower_cs_intrinsics.c
index 102bd29..401fbdd 100644 (file)
--- a/src/intel/compiler/brw_nir_lower_cs_intrinsics.c
+++ b/src/intel/compiler/brw_nir_lower_cs_intrinsics.c
@@ -211,7 +211,8 @@ lower_cs_intrinsics_convert_impl(struct lower_intrinsics_state *state)
 bool
 brw_nir_lower_cs_intrinsics(nir_shader *nir)
 {
-   assert(nir->info.stage == MESA_SHADER_COMPUTE);
+   assert(nir->info.stage == MESA_SHADER_COMPUTE ||
+          nir->info.stage == MESA_SHADER_KERNEL);
 
    struct lower_intrinsics_state state = {
       .nir = nir,