[Mesa-dev] [PATCH] radeon/llvm: Dynamically allocate branch/loop stack arrays

[Michel Dänzer]

From: Michel Dänzer <michel.daenzer at amd.com>

This prevents us from silently overflowing the stack arrays, and allows
arbitrary stack depths.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=85454

Reported-and-Tested-by: Nick Sarnie <commendsarnex at gmail.com>
Signed-off-by: Michel Dänzer <michel.daenzer at amd.com>
---
 src/gallium/drivers/radeon/radeon_llvm.h           | 10 ++++---
 .../drivers/radeon/radeon_setup_tgsi_llvm.c        | 33 ++++++++++++++++++++--
 2 files changed, 37 insertions(+), 6 deletions(-)

diff --git a/src/gallium/drivers/radeon/radeon_llvm.h b/src/gallium/drivers/radeon/radeon_llvm.h
index 00714fb..8612ef8 100644
--- a/src/gallium/drivers/radeon/radeon_llvm.h
+++ b/src/gallium/drivers/radeon/radeon_llvm.h
@@ -33,10 +33,10 @@
 
 #define RADEON_LLVM_MAX_INPUTS 32 * 4
 #define RADEON_LLVM_MAX_OUTPUTS 32 * 4
-#define RADEON_LLVM_MAX_BRANCH_DEPTH 16
-#define RADEON_LLVM_MAX_LOOP_DEPTH 16
 #define RADEON_LLVM_MAX_ARRAYS 16
 
+#define RADEON_LLVM_INITIAL_CF_DEPTH 4
+
 #define RADEON_LLVM_MAX_SYSTEM_VALUES 4
 
 struct radeon_llvm_branch {
@@ -122,11 +122,13 @@ struct radeon_llvm_context {
 
 	/*=== Private Members ===*/
 
-	struct radeon_llvm_branch branch[RADEON_LLVM_MAX_BRANCH_DEPTH];
-	struct radeon_llvm_loop loop[RADEON_LLVM_MAX_LOOP_DEPTH];
+	struct radeon_llvm_branch *branch;
+	struct radeon_llvm_loop *loop;
 
 	unsigned branch_depth;
+	unsigned branch_depth_max;
 	unsigned loop_depth;
+	unsigned loop_depth_max;
 
 	struct tgsi_declaration_range arrays[RADEON_LLVM_MAX_ARRAYS];
 	unsigned num_arrays;
diff --git a/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c b/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c
index 2fa23ed..c30a9d0 100644
--- a/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c
+++ b/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c
@@ -446,7 +446,19 @@ static void bgnloop_emit(
 						endloop_block, "LOOP");
 	LLVMBuildBr(gallivm->builder, loop_block);
 	LLVMPositionBuilderAtEnd(gallivm->builder, loop_block);
-	ctx->loop_depth++;
+
+	if (++ctx->loop_depth > ctx->loop_depth_max) {
+		unsigned new_max = ctx->loop_depth_max << 1;
+
+		if (!new_max)
+			new_max = RADEON_LLVM_INITIAL_CF_DEPTH;
+
+		ctx->loop = REALLOC(ctx->loop, ctx->loop_depth_max *
+				    sizeof(ctx->loop[0]),
+				    new_max * sizeof(ctx->loop[0]));
+		ctx->loop_depth_max = new_max;
+	}
+
 	ctx->loop[ctx->loop_depth - 1].loop_block = loop_block;
 	ctx->loop[ctx->loop_depth - 1].endloop_block = endloop_block;
 }
@@ -577,7 +589,18 @@ static void if_cond_emit(
 	LLVMBuildCondBr(gallivm->builder, cond, if_block, else_block);
 	LLVMPositionBuilderAtEnd(gallivm->builder, if_block);
 
-	ctx->branch_depth++;
+	if (++ctx->branch_depth > ctx->branch_depth_max) {
+		unsigned new_max = ctx->branch_depth_max << 1;
+
+		if (!new_max)
+			new_max = RADEON_LLVM_INITIAL_CF_DEPTH;
+
+		ctx->branch = REALLOC(ctx->branch, ctx->branch_depth_max *
+				      sizeof(ctx->branch[0]),
+				      new_max * sizeof(ctx->branch[0]));
+		ctx->branch_depth_max = new_max;
+	}
+
 	ctx->branch[ctx->branch_depth - 1].endif_block = endif_block;
 	ctx->branch[ctx->branch_depth - 1].if_block = if_block;
 	ctx->branch[ctx->branch_depth - 1].else_block = else_block;
@@ -1440,4 +1463,10 @@ void radeon_llvm_dispose(struct radeon_llvm_context * ctx)
 	LLVMContextDispose(ctx->soa.bld_base.base.gallivm->context);
 	FREE(ctx->temps);
 	ctx->temps = NULL;
+	FREE(ctx->loop);
+	ctx->loop = NULL;
+	ctx->loop_depth_max = 0;
+	FREE(ctx->branch);
+	ctx->branch = NULL;
+	ctx->branch_depth_max = 0;
 }
-- 
2.1.1