[AMDGPU] Restrict operands of ld_scale_paired to low 256 vgprs. NFCI

[rampitec]

This is NFC because these instructions are never selected just
by itself.

[rampitec]

• [AMDGPU] Restrict operands of ld_scale_paired to low 256 vgprs. NFCI #157935 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Stanislav Mekhanoshin (rampitec)

Changes

This is NFC because these instructions are never selected just
by itself.

---

Full diff: https://github.com/llvm/llvm-project/pull/157935.diff

1 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/VOP3PInstructions.td (+2-2)

diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index c8a59efc106fe..6f778a0d262af 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -1857,8 +1857,8 @@ defm V_WMMA_SCALE_F32_32X16X128_F4_w32   : WMMAInstGFX12<"v_wmma_scale_f32_32x16
 defm V_WMMA_SCALE16_F32_32X16X128_F4_w32 : WMMAInstGFX12<"v_wmma_scale16_f32_32x16x128_f4", F32_32X16X128_F4_SCALE16_w32, "_w32">;
 } // End is_wmma_xdl = 1.
 
-defm V_WMMA_LD_SCALE_PAIRED_B32   : VOP3PInst<"v_wmma_ld_scale_paired_b32",   VOP_WMMA_LD_SCALE<i32, VCSrc_b32>>;
-defm V_WMMA_LD_SCALE16_PAIRED_B64 : VOP3PInst<"v_wmma_ld_scale16_paired_b64", VOP_WMMA_LD_SCALE<i64, VCSrc_b64>>;
+defm V_WMMA_LD_SCALE_PAIRED_B32   : VOP3PInst<"v_wmma_ld_scale_paired_b32",   VOP_WMMA_LD_SCALE<i32, VCSrc_b32_Lo256>>;
+defm V_WMMA_LD_SCALE16_PAIRED_B64 : VOP3PInst<"v_wmma_ld_scale16_paired_b64", VOP_WMMA_LD_SCALE<i64, VCSrc_b64_Lo256>>;
 } // End SubtargetPredicate = isGFX125xOnly
 } // End WaveSizePredicate = isWave32
 

[rampitec]

I.e, these 2 instructions are a glue to support disasm on partial 8 dword encoding, it is never used by itself, but always combined into wmma_scale. The wmma_scale was already updated to this and tests too.