nv50,nvc0: provide debug messages with shader compilation stats

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h

@@ -96,6 +96,7 @@ struct nv50_ir_prog_info
       uint32_t tlsSpace;  /* required local memory per thread */
       uint32_t *code;
       uint32_t codeSize;
+      uint32_t instructions;
       uint8_t sourceRep;  /* NV50_PROGRAM_IR */
       const void *source;
       void *relocData;

src/gallium/drivers/nouveau/codegen/nv50_ir_target.cpp

@@ -373,6 +373,7 @@ Program::emitBinary(struct nv50_ir_prog_info *info)
    if (!code)
       return false;
    emit->setCodeLocation(code, binSize);
+   info->bin.instructions = 0;
 
    for (ArrayList::Iterator fi = allFuncs.iterator(); !fi.end(); fi.next()) {
       Function *fn = reinterpret_cast<Function *>(fi.get());
@@ -382,6 +383,7 @@ Program::emitBinary(struct nv50_ir_prog_info *info)
       for (int b = 0; b < fn->bbCount; ++b) {
          for (Instruction *i = fn->bbArray[b]->getEntry(); i; i = i->next) {
             emit->emitInstruction(i);
+            info->bin.instructions++;
             if (i->sType == TYPE_F64 || i->dType == TYPE_F64)
                info->io.fp64 = true;
          }

src/gallium/drivers/nouveau/nv50/nv50_program.c

@@ -318,7 +318,8 @@ nv50_program_create_strmout_state(const struct nv50_ir_prog_info *info,
 }
 
 bool
-nv50_program_translate(struct nv50_program *prog, uint16_t chipset)
+nv50_program_translate(struct nv50_program *prog, uint16_t chipset,
+                       struct pipe_debug_callback *debug)
 {
    struct nv50_ir_prog_info *info;
    int ret;
@@ -406,6 +407,11 @@ nv50_program_translate(struct nv50_program *prog, uint16_t chipset)
       prog->so = nv50_program_create_strmout_state(info,
                                                    &prog->pipe.stream_output);
 
+   pipe_debug_message(debug, SHADER_INFO,
+                      "type: %d, local: %d, gpr: %d, inst: %d, bytes: %d",
+                      prog->type, info->bin.tlsSpace, prog->max_gpr,
+                      info->bin.instructions, info->bin.codeSize);
+
 out:
    FREE(info);
    return !ret;

src/gallium/drivers/nouveau/nv50/nv50_program.h

@@ -106,7 +106,8 @@ struct nv50_program {
    struct nv50_stream_output_state *so;
 };
 
-bool nv50_program_translate(struct nv50_program *, uint16_t chipset);
+bool nv50_program_translate(struct nv50_program *, uint16_t chipset,
+                            struct pipe_debug_callback *);
 bool nv50_program_upload_code(struct nv50_context *, struct nv50_program *);
 void nv50_program_destroy(struct nv50_context *, struct nv50_program *);
 

src/gallium/drivers/nouveau/nv50/nv50_shader_state.c

@@ -113,7 +113,7 @@ nv50_program_validate(struct nv50_context *nv50, struct nv50_program *prog)
 {
    if (!prog->translated) {
       prog->translated = nv50_program_translate(
-         prog, nv50->screen->base.device->chipset);
+         prog, nv50->screen->base.device->chipset, &nv50->base.debug);
       if (!prog->translated)
          return false;
    } else

src/gallium/drivers/nouveau/nv50/nv50_state.c

@@ -727,7 +727,8 @@ nv50_sp_state_create(struct pipe_context *pipe,
       prog->pipe.stream_output = cso->stream_output;
 
    prog->translated = nv50_program_translate(
-         prog, nv50_context(pipe)->screen->base.device->chipset);
+         prog, nv50_context(pipe)->screen->base.device->chipset,
+         &nouveau_context(pipe)->debug);
 
    return (void *)prog;
 }

src/gallium/drivers/nouveau/nvc0/nvc0_compute.c

@@ -120,7 +120,7 @@ nvc0_compute_validate_program(struct nvc0_context *nvc0)
 
    if (!prog->translated) {
       prog->translated = nvc0_program_translate(
-         prog, nvc0->screen->base.device->chipset);
+         prog, nvc0->screen->base.device->chipset, &nvc0->base.debug);
       if (!prog->translated)
          return false;
    }

src/gallium/drivers/nouveau/nvc0/nvc0_context.h

@@ -224,7 +224,8 @@ void nvc0_default_kick_notify(struct nouveau_pushbuf *);
 extern struct draw_stage *nvc0_draw_render_stage(struct nvc0_context *);
 
 /* nvc0_program.c */
-bool nvc0_program_translate(struct nvc0_program *, uint16_t chipset);
+bool nvc0_program_translate(struct nvc0_program *, uint16_t chipset,
+                            struct pipe_debug_callback *);
 bool nvc0_program_upload_code(struct nvc0_context *, struct nvc0_program *);
 void nvc0_program_destroy(struct nvc0_context *, struct nvc0_program *);
 void nvc0_program_library_upload(struct nvc0_context *);

src/gallium/drivers/nouveau/nvc0/nvc0_program.c

@@ -517,7 +517,8 @@ nvc0_program_dump(struct nvc0_program *prog)
 #endif
 
 bool
-nvc0_program_translate(struct nvc0_program *prog, uint16_t chipset)
+nvc0_program_translate(struct nvc0_program *prog, uint16_t chipset,
+                       struct pipe_debug_callback *debug)
 {
    struct nv50_ir_prog_info *info;
    int ret;
@@ -639,6 +640,11 @@ nvc0_program_translate(struct nvc0_program *prog, uint16_t chipset)
       prog->tfb = nvc0_program_create_tfb_state(info,
                                                 &prog->pipe.stream_output);
 
+   pipe_debug_message(debug, SHADER_INFO,
+                      "type: %d, local: %d, gpr: %d, inst: %d, bytes: %d",
+                      prog->type, info->bin.tlsSpace, prog->num_gprs,
+                      info->bin.instructions, info->bin.codeSize);
+
 out:
    FREE(info);
    return !ret;

src/gallium/drivers/nouveau/nvc0/nvc0_shader_state.c

@@ -72,7 +72,7 @@ nvc0_program_validate(struct nvc0_context *nvc0, struct nvc0_program *prog)
 
    if (!prog->translated) {
       prog->translated = nvc0_program_translate(
-         prog, nvc0->screen->base.device->chipset);
+         prog, nvc0->screen->base.device->chipset, &nvc0->base.debug);
       if (!prog->translated)
          return false;
    }

src/gallium/drivers/nouveau/nvc0/nvc0_state.c

@@ -681,7 +681,8 @@ nvc0_sp_state_create(struct pipe_context *pipe,
       prog->pipe.stream_output = cso->stream_output;
 
    prog->translated = nvc0_program_translate(
-      prog, nvc0_context(pipe)->screen->base.device->chipset);
+      prog, nvc0_context(pipe)->screen->base.device->chipset,
+      &nouveau_context(pipe)->debug);
 
    return (void *)prog;
 }