[llvm-branch-commits] [llvm] [AMDGPU] Error out on too large lit() constants (PR #197803)

[Stanislav Mekhanoshin via llvm-branch-commits]

================
@@ -2008,16 +2008,16 @@ v_add_nc_u64 v[0:1], v[0:1], 0x123456789
 // NOSI: :[[@LINE-6]]:1: error: instruction not supported on this GPU (tahiti): v_add_nc_u64
 // NOVI: :[[@LINE-7]]:1: error: instruction not supported on this GPU (tonga): v_add_nc_u64
 
-// FIXME: Literal is truncated but shall be rejected
 v_add_nc_u64 v[0:1], v[0:1], lit(0x123456789)
 // GFX1250-ASM: v_add_nc_u64_e64 v[0:1], v[0:1], lit(0x23456789) ; encoding: [0x00,0x00,0x28,0xd5,0x00,0xff,0x01,0x02,0x89,0x67,0x45,0x23]
 // GFX1250-DIS: v_add_nc_u64_e64 v[0:1], v[0:1], 0x23456789 ; encoding: [0x00,0x00,0x28,0xd5,0x00,0xff,0x01,0x02,0x89,0x67,0x45,0x23]
 // NOCI: :[[@LINE-3]]:1: error: instruction not supported on this GPU (bonaire): v_add_nc_u64
 // NOGFX11: :[[@LINE-4]]:1: error: instruction not supported on this GPU (gfx1100): v_add_nc_u64
 // NOGFX12: :[[@LINE-5]]:1: error: instruction not supported on this GPU (gfx1200): v_add_nc_u64
-// NOGFX9: :[[@LINE-6]]:1: error: instruction not supported on this GPU (gfx900): v_add_nc_u64
-// NOSI: :[[@LINE-7]]:1: error: instruction not supported on this GPU (tahiti): v_add_nc_u64
-// NOVI: :[[@LINE-8]]:1: error: instruction not supported on this GPU (tonga): v_add_nc_u64
+// NOGFX1250: :[[@LINE-6]]:1: error: invalid literal value
----------------
rampitec wrote:

Probably the validation here has to happen earlier, when we are parsing the immediate. There is a chance of proper bail there.

https://github.com/llvm/llvm-project/pull/197803