astc: Various robustness enhancements for the gpu decoder

These changes should help in reducing crashes/drivers panics that may
occur due to synchronization issues between the shader completion and
later access of the decoded texture.

1 files changed, 6 insertions, 9 deletions

diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp
index eaba1b103..71327e233 100644
--- a/src/video_core/host_shaders/astc_decoder.comp
+++ b/src/video_core/host_shaders/astc_decoder.comp
@@ -14,9 +14,8 @@
 #define BINDING_6_TO_8_BUFFER 2
 #define BINDING_7_TO_8_BUFFER 3
 #define BINDING_8_TO_8_BUFFER 4
-#define BINDING_BYTE_TO_16_BUFFER 5
-#define BINDING_SWIZZLE_BUFFER 6
-#define BINDING_OUTPUT_IMAGE 7
+#define BINDING_SWIZZLE_BUFFER 5
+#define BINDING_OUTPUT_IMAGE 6
 
 #else // ^^^ Vulkan ^^^ // vvv OpenGL vvv
 
@@ -29,7 +28,6 @@
 #define BINDING_6_TO_8_BUFFER 3
 #define BINDING_7_TO_8_BUFFER 4
 #define BINDING_8_TO_8_BUFFER 5
-#define BINDING_BYTE_TO_16_BUFFER 6
 #define BINDING_OUTPUT_IMAGE 0
 
 #endif
@@ -86,9 +84,6 @@ layout(binding = BINDING_7_TO_8_BUFFER, std430) readonly buffer REPLICATE_7_BIT_
 layout(binding = BINDING_8_TO_8_BUFFER, std430) readonly buffer REPLICATE_8_BIT_TO_8 {
     uint REPLICATE_8_BIT_TO_8_TABLE[];
 };
-layout(binding = BINDING_BYTE_TO_16_BUFFER, std430) readonly buffer REPLICATE_BYTE_TO_16 {
-    uint REPLICATE_BYTE_TO_16_TABLE[];
-};
 
 layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly image2DArray dest_image;
 
@@ -207,8 +202,7 @@ uint Replicate(uint val, uint num_bits, uint to_bit) {
 }
 
 uvec4 ReplicateByteTo16(uvec4 value) {
-    return uvec4(REPLICATE_BYTE_TO_16_TABLE[value.x], REPLICATE_BYTE_TO_16_TABLE[value.y],
-                 REPLICATE_BYTE_TO_16_TABLE[value.z], REPLICATE_BYTE_TO_16_TABLE[value.w]);
+    return value * 0x101;
 }
 
 uint ReplicateBitTo7(uint value) {
@@ -1327,6 +1321,9 @@ void main() {
     offset += swizzle;
 
     const ivec3 coord = ivec3(gl_GlobalInvocationID * uvec3(block_dims, 1));
+    if (any(greaterThanEqual(coord, imageSize(dest_image)))) {
+        return;
+    }
     uint block_index =
         pos.z * gl_WorkGroupSize.x * gl_WorkGroupSize.y + pos.y * gl_WorkGroupSize.x + pos.x;