================
@@ -34,6 +37,87 @@ enum GPUKind : uint8_t {
#include "llvm/TargetParser/IntelGPUTargetParser.def"
};
+/// Represents an IGCA feature set, including whether or not it is exact. The
+/// values are bitfields where the last bit represents exact. See IGCA Target
+/// representation below for more info.
+enum class IGCAFeatureSet : uint8_t {
+ Core = 0b000,
+ Compute = 0b010,
+ ComputeExact = 0b011,
+ Render = 0b100,
+ RenderExact = 0b101,
+};
+
+/// Wrapper around an IGCA target's uint32_t representation, packed using the
+/// following format:
+///
+/// 31 16 15 3 2 1 0 0
+/// +------------------+--------------+-------------+-----------+
+/// | Target | Reserved | Feature set | Is Exact? |
+/// +------------------+--------------+-------------+-----------+
+/// 16 bits 13 bits 2 bits 1 bit
----------------
srividya-sundaram wrote:
nit: the diagram labels bits `2–1` as `"Feature set"`, but in the code
`IGCAFeatureSet, FeatureSetMask (0x7) and getFeatureSet()` all cover bits
`2–0`, exact bit included.
Could the diagram label bits` 2–0` as the feature set, with bit 0 as the exact
flag, so the two match?
https://github.com/llvm/llvm-project/pull/227468
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
