radeonsi: keep serialized NIR instead of nir_shader in si_shader_selector

This decreases memory usage, because serialized NIR is more compact.

The main shader part is compiled from nir_shader.
Monolithic shader variants are compiled from nir_binary.

Reviewed-by: Timothy Arceri <[email protected]>

5 files changed, 65 insertions, 13 deletions

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 446b3e393ee..fba7187204d 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -42,6 +42,7 @@
 #include "sid.h"
 
 #include "compiler/nir/nir.h"
+#include "compiler/nir/nir_serialize.h"
 
 static const char scratch_rsrc_dword0_symbol[] =
 	"SCRATCH_RSRC_DWORD0";
@@ -6069,7 +6070,8 @@ static bool si_vs_needs_prolog(const struct si_shader_selector *sel,
 	return sel->vs_needs_prolog || key->ls_vgpr_fix;
 }
 
-static bool si_compile_tgsi_main(struct si_shader_context *ctx)
+static bool si_compile_tgsi_main(struct si_shader_context *ctx,
+				 struct nir_shader *nir, bool free_nir)
 {
 	struct si_shader *shader = ctx->shader;
 	struct si_shader_selector *sel = shader->selector;
@@ -6304,7 +6306,10 @@ static bool si_compile_tgsi_main(struct si_shader_context *ctx)
 			return false;
 		}
 	} else {
-		if (!si_nir_build_llvm(ctx, sel->nir)) {
+		bool success = si_nir_build_llvm(ctx, nir);
+		if (free_nir)
+			ralloc_free(nir);
+		if (!success) {
 			fprintf(stderr, "Failed to translate shader from NIR to LLVM\n");
 			return false;
 		}
@@ -6924,6 +6929,27 @@ static bool si_should_optimize_less(struct ac_llvm_compiler *compiler,
 	       sel->info.num_memory_instructions > 1000;
 }
 
+static struct nir_shader *get_nir_shader(struct si_shader_selector *sel,
+					 bool *free_nir)
+{
+	*free_nir = false;
+
+	if (sel->nir) {
+		return sel->nir;
+	} else if (sel->nir_binary) {
+		struct pipe_screen *screen = &sel->screen->b;
+		const void *options =
+			screen->get_compiler_options(screen, PIPE_SHADER_IR_NIR,
+						     sel->type);
+
+		struct blob_reader blob_reader;
+		blob_reader_init(&blob_reader, sel->nir_binary, sel->nir_size);
+		*free_nir = true;
+		return nir_deserialize(NULL, options, &blob_reader);
+	}
+	return NULL;
+}
+
 int si_compile_tgsi_shader(struct si_screen *sscreen,
 			   struct ac_llvm_compiler *compiler,
 			   struct si_shader *shader,
@@ -6931,6 +6957,8 @@ int si_compile_tgsi_shader(struct si_screen *sscreen,
 {
 	struct si_shader_selector *sel = shader->selector;
 	struct si_shader_context ctx;
+	bool free_nir;
+	struct nir_shader *nir = get_nir_shader(sel, &free_nir);
 	int r = -1;
 
 	/* Dump TGSI code before doing TGSI->LLVM conversion in case the
@@ -6940,20 +6968,20 @@ int si_compile_tgsi_shader(struct si_screen *sscreen,
 		if (sel->tokens)
 			tgsi_dump(sel->tokens, 0);
 		else
-			nir_print_shader(sel->nir, stderr);
+			nir_print_shader(nir, stderr);
 		si_dump_streamout(&sel->so);
 	}
 
 	si_init_shader_ctx(&ctx, sscreen, compiler, si_get_shader_wave_size(shader),
-			   sel->nir != NULL);
-	si_llvm_context_set_ir(&ctx, shader);
+			   nir != NULL);
+	si_llvm_context_set_ir(&ctx, shader, nir);
 
 	memset(shader->info.vs_output_param_offset, AC_EXP_PARAM_UNDEFINED,
 	       sizeof(shader->info.vs_output_param_offset));
 
 	shader->info.uses_instanceid = sel->info.uses_instanceid;
 
-	if (!si_compile_tgsi_main(&ctx)) {
+	if (!si_compile_tgsi_main(&ctx, nir, free_nir)) {
 		si_llvm_dispose(&ctx);
 		return -1;
 	}
@@ -6997,15 +7025,16 @@ int si_compile_tgsi_shader(struct si_screen *sscreen,
 			parts[3] = ctx.main_fn;
 
 			/* VS as LS main part */
+			nir = get_nir_shader(ls, &free_nir);
 			struct si_shader shader_ls = {};
 			shader_ls.selector = ls;
 			shader_ls.key.as_ls = 1;
 			shader_ls.key.mono = shader->key.mono;
 			shader_ls.key.opt = shader->key.opt;
 			shader_ls.is_monolithic = true;
-			si_llvm_context_set_ir(&ctx, &shader_ls);
+			si_llvm_context_set_ir(&ctx, &shader_ls, nir);
 
-			if (!si_compile_tgsi_main(&ctx)) {
+			if (!si_compile_tgsi_main(&ctx, nir, free_nir)) {
 				si_llvm_dispose(&ctx);
 				return -1;
 			}
@@ -7063,6 +7092,7 @@ int si_compile_tgsi_shader(struct si_screen *sscreen,
 			gs_prolog = ctx.main_fn;
 
 			/* ES main part */
+			nir = get_nir_shader(es, &free_nir);
 			struct si_shader shader_es = {};
 			shader_es.selector = es;
 			shader_es.key.as_es = 1;
@@ -7070,9 +7100,9 @@ int si_compile_tgsi_shader(struct si_screen *sscreen,
 			shader_es.key.mono = shader->key.mono;
 			shader_es.key.opt = shader->key.opt;
 			shader_es.is_monolithic = true;
-			si_llvm_context_set_ir(&ctx, &shader_es);
+			si_llvm_context_set_ir(&ctx, &shader_es, nir);
 
-			if (!si_compile_tgsi_main(&ctx)) {
+			if (!si_compile_tgsi_main(&ctx, nir, free_nir)) {
 				si_llvm_dispose(&ctx);
 				return -1;
 			}
diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 1d41b7aa042..20f2c812c21 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -328,6 +328,9 @@ struct si_shader_selector {
 
 	struct tgsi_token       *tokens;
 	struct nir_shader       *nir;
+	void			*nir_binary;
+	unsigned		nir_size;
+
 	struct pipe_stream_output_info  so;
 	struct tgsi_shader_info		info;
 	struct tgsi_tessctrl_info	tcs_info;
diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h
index 98abbdfc693..8f904471d1f 100644
--- a/src/gallium/drivers/radeonsi/si_shader_internal.h
+++ b/src/gallium/drivers/radeonsi/si_shader_internal.h
@@ -282,7 +282,8 @@ void si_llvm_context_init(struct si_shader_context *ctx,
 			  unsigned wave_size,
 			  unsigned ballot_mask_bits);
 void si_llvm_context_set_ir(struct si_shader_context *ctx,
-			    struct si_shader *shader);
+			    struct si_shader *shader,
+			    struct nir_shader *nir);
 
 void si_llvm_create_func(struct si_shader_context *ctx,
 			 const char *name,
diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
index 91e9bd3dd68..cb965cc8eb3 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
@@ -1029,7 +1029,8 @@ void si_llvm_context_init(struct si_shader_context *ctx,
 /* Set the context to a certain TGSI shader. Can be called repeatedly
  * to change the shader. */
 void si_llvm_context_set_ir(struct si_shader_context *ctx,
-			    struct si_shader *shader)
+			    struct si_shader *shader,
+			    struct nir_shader *nir)
 {
 	struct si_shader_selector *sel = shader->selector;
 	const struct tgsi_shader_info *info = &sel->info;
@@ -1058,7 +1059,7 @@ void si_llvm_context_set_ir(struct si_shader_context *ctx,
 	ctx->num_samplers = util_last_bit(info->samplers_declared);
 	ctx->num_images = util_last_bit(info->images_declared);
 
-	if (sel->nir)
+	if (nir)
 		return;
 
 	if (info->array_max[TGSI_FILE_TEMPORARY] > 0) {
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index 10b95ff0e78..ddd27fb8176 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -2581,6 +2581,22 @@ static void si_init_shader_selector_async(void *job, int thread_index)
 
 		si_shader_vs(sscreen, sel->gs_copy_shader, sel);
 	}
+
+	if (sel->nir) {
+		/* Serialize NIR to save memory. Monolithic shader variants
+		 * have to deserialize NIR before compilation.
+		 */
+		struct blob blob;
+		blob_init(&blob);
+		nir_serialize(&blob, sel->nir, false);
+		sel->nir_binary = malloc(blob.size);
+		memcpy(sel->nir_binary, blob.data, blob.size);
+		sel->nir_size = blob.size;
+		blob_finish(&blob);
+
+		ralloc_free(sel->nir);
+		sel->nir = NULL;
+	}
 }
 
 void si_schedule_initial_compile(struct si_context *sctx, unsigned processor,
@@ -3281,6 +3297,7 @@ void si_destroy_shader_selector(struct si_context *sctx,
 	simple_mtx_destroy(&sel->mutex);
 	free(sel->tokens);
 	ralloc_free(sel->nir);
+	free(sel->nir_binary);
 	free(sel);
 }