[clang-tools-extra] [llvm] [RISCV][Zba] Optimize mul with SH*ADD (PR #68144)

2023-11-23 Thread Alexandr Lekomtsev via cfe-commits 

https://github.com/vacma updated 
https://github.com/llvm/llvm-project/pull/68144

>From 64e5deade92206cd04f86e0665a482ea78d0f3bf Mon Sep 17 00:00:00 2001
From: Alexandr 
Date: Sun, 1 Oct 2023 19:57:01 +0300
Subject: [PATCH 1/3] [RISCV][Zba] Optimize mul with SH*ADD

Optimized multiplication by constants 23, 29, 35, 39, 43, 47,
49, 51, 53, 55, 57, 59, 61, 69, 75, 77, 83, 85, 87, 89, 91,
93, 95, 97, 99, 101, 103, 105, 107, 109, 111, 113, 117, 121,
123, 125, 135, 137, 145, 147, 149, 153, 155, 157, 161, 163,
165, 169, 171, 173, 181, 185, 189, 201, 203, 205, 217, 219,
225, 243, 293, 297, 301, 305, 325, 329, 333, 361, 365, 369,
405, 585, 593, 649, 657, 729
---
 llvm/lib/Target/RISCV/RISCVInstrInfoZb.td |  161 +++
 llvm/test/CodeGen/RISCV/rv32zba.ll| 1292 +
 llvm/test/CodeGen/RISCV/rv64zba.ll| 1292 +
 3 files changed, 2745 insertions(+)

diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td 
b/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
index a21c3d132636bea..cbd8fad30cd1a20 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
@@ -718,12 +718,173 @@ def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 41)),
   (SH3ADD (SH2ADD GPR:$r, GPR:$r), GPR:$r)>;
 def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 73)),
   (SH3ADD (SH3ADD GPR:$r, GPR:$r), GPR:$r)>;
+
 def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 27)),
   (SH1ADD (SH3ADD GPR:$r, GPR:$r), (SH3ADD GPR:$r, GPR:$r))>;
 def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 45)),
   (SH2ADD (SH3ADD GPR:$r, GPR:$r), (SH3ADD GPR:$r, GPR:$r))>;
 def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 81)),
   (SH3ADD (SH3ADD GPR:$r, GPR:$r), (SH3ADD GPR:$r, GPR:$r))>;
+
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 61)),
+  (SH2ADD (SH2ADD (SH1ADD GPR:$r, GPR:$r), (SH1ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 121)),
+  (SH3ADD (SH2ADD (SH1ADD GPR:$r, GPR:$r), (SH1ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 55)),
+  (SH1ADD (SH3ADD (SH1ADD GPR:$r, GPR:$r), (SH1ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 109)),
+  (SH2ADD (SH3ADD (SH1ADD GPR:$r, GPR:$r), (SH1ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 217)),
+  (SH3ADD (SH3ADD (SH1ADD GPR:$r, GPR:$r), (SH1ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 51)),
+  (SH1ADD (SH2ADD (SH2ADD GPR:$r, GPR:$r), (SH2ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 101)),
+  (SH2ADD (SH2ADD (SH2ADD GPR:$r, GPR:$r), (SH2ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 201)),
+  (SH3ADD (SH2ADD (SH2ADD GPR:$r, GPR:$r), (SH2ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 91)),
+  (SH1ADD (SH3ADD (SH2ADD GPR:$r, GPR:$r), (SH2ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 181)),
+  (SH2ADD (SH3ADD (SH2ADD GPR:$r, GPR:$r), (SH2ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 361)),
+  (SH3ADD (SH3ADD (SH2ADD GPR:$r, GPR:$r), (SH2ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 163)),
+  (SH1ADD (SH3ADD (SH3ADD GPR:$r, GPR:$r), (SH3ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 325)),
+  (SH2ADD (SH3ADD (SH3ADD GPR:$r, GPR:$r), (SH3ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 649)),
+  (SH3ADD (SH3ADD (SH3ADD GPR:$r, GPR:$r), (SH3ADD GPR:$r, GPR:$r)), 
GPR:$r)>;
+
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 29)),
+  (SH2ADD (SH1ADD (SH1ADD GPR:$r, GPR:$r), GPR:$r), GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 57)),
+  (SH3ADD (SH1ADD (SH1ADD GPR:$r, GPR:$r), GPR:$r), GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 53)),
+  (SH2ADD (SH2ADD (SH1ADD GPR:$r, GPR:$r), GPR:$r), GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 105)),
+  (SH3ADD (SH2ADD (SH1ADD GPR:$r, GPR:$r), GPR:$r), GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 23)),
+  (SH1ADD (SH1ADD (SH2ADD GPR:$r, GPR:$r), GPR:$r), GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 89)),
+  (SH3ADD (SH1ADD (SH2ADD GPR:$r, GPR:$r), GPR:$r), GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 43)),
+  (SH1ADD (SH2ADD (SH2ADD GPR:$r, GPR:$r), GPR:$r), GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 85)),
+  (SH2ADD (SH2ADD (SH2ADD GPR:$r, GPR:$r), GPR:$r), GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 169)),
+  (SH3ADD (SH2ADD (SH2ADD GPR:$r, GPR:$r), GPR:$r), GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 83)),
+  (SH1ADD (SH3ADD (SH2ADD GPR:$r, GPR:$r), GPR:$r), GPR:$r)>;
+def : Pat<(mul_const_oneuse GPR:$r, (XLenVT 165)),
+

[clang-tools-extra] [llvm] [RISCV][Zba] Optimize mul with SH*ADD (PR #68144)

2023-11-24 Thread Alexandr Lekomtsev via cfe-commits 


@@ -983,6 +983,1299 @@ define i64 @mul81(i64 %a) {
   ret i64 %c
 }
 
+
+define i64 @mul153(i64 %a) {
+; RV64I-LABEL: mul153:

vacma wrote:

Thanks for your feedback,
tests show that 3 shXadds is _a little_ slower than mul, but for some reason 
GCC uses 3 shXadds - https://godbolt.org/z/oez5Pddde
And even 4 - https://godbolt.org/z/czK3Kd7bs
I'll try to find out why they do this and write later.

https://github.com/llvm/llvm-project/pull/68144
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits