[llvm] f34a195 - [AMDGPU] Fix AddedComplexity for s_buffer_load patterns. NFCI.

[Jay Foad via llvm-commits]

Author: Jay Foad
Date: 2023-04-10T17:26:50+01:00
New Revision: f34a1953ce334b87cfe25cdf72677d31ebe1d50b

URL: https://github.com/llvm/llvm-project/commit/f34a1953ce334b87cfe25cdf72677d31ebe1d50b
DIFF: https://github.com/llvm/llvm-project/commit/f34a1953ce334b87cfe25cdf72677d31ebe1d50b.diff

LOG: [AMDGPU] Fix AddedComplexity for s_buffer_load patterns. NFCI.

We set AddedComplexity = 100 for s_load patterns to prefer them over
global loads, but for s_buffer_load patterns there is no need to do
this and it was quietly overriding the AddedComplexity of each
individual GCNPat that is defined inside SMLoad_Pattern (but in practice
that did not appear to make any difference).

Differential Revision: https://reviews.llvm.org/D145396

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/SMInstructions.td

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/SMInstructions.td b/llvm/lib/Target/AMDGPU/SMInstructions.td
index 0bc608d3a9d1..c8b6cbee099b 100644
--- a/llvm/lib/Target/AMDGPU/SMInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SMInstructions.td
@@ -890,6 +890,8 @@ foreach vt = SReg_512.RegTypes in {
 defm : SMRD_Pattern <"S_LOAD_DWORDX16", vt>;
 }
 
+} // End let AddedComplexity = 100
+
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORD",     i32>;
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX2",   v2i32>;
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX4",   v4i32>;
@@ -901,7 +903,6 @@ defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX2",   v2f32>;
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX4",   v4f32>;
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX8",   v8f32>;
 defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX16",  v16f32>;
-} // End let AddedComplexity = 100
 
 let OtherPredicates = [HasSMemTimeInst] in {
 def : GCNPat <