From e16b4f5b16a3d6b3753ce1a45d40113c8cb6bc23 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Mon, 19 Dec 2022 18:27:50 +0000
Subject: [PATCH] [X86] Fix SLM uops/resources counts for CMPXCHG instructions

LOCK + CMPXCHG8/CMPXCHG16 variants still need overriding as they are not completely correct - already much better though

Based off llvm-exegesis captures, confirmed with Agner + uops.info
---
 llvm/lib/Target/X86/X86ScheduleSLM.td              |  4 +-
 .../tools/llvm-mca/X86/SLM/resources-cmpxchg.s     | 18 ++++----
 .../test/tools/llvm-mca/X86/SLM/resources-x86_64.s | 50 +++++++++++-----------
 3 files changed, 36 insertions(+), 36 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ScheduleSLM.td b/llvm/lib/Target/X86/X86ScheduleSLM.td
index ea6f74e..c59853b 100644
--- a/llvm/lib/Target/X86/X86ScheduleSLM.td
+++ b/llvm/lib/Target/X86/X86ScheduleSLM.td
@@ -117,8 +117,8 @@ defm : X86WriteResPairUnsupported<WriteMULX64>;
 
 defm : X86WriteRes<WriteBSWAP32, [SLM_IEC_RSV01], 1, [1], 1>;
 defm : X86WriteRes<WriteBSWAP64, [SLM_IEC_RSV01], 1, [1], 1>;
-defm : X86WriteRes<WriteCMPXCHG, [SLM_IEC_RSV01], 1, [1], 1>;
-defm : X86WriteRes<WriteCMPXCHGRMW, [SLM_IEC_RSV01, SLM_MEC_RSV], 4, [1, 2], 2>;
+defm : X86WriteRes<WriteCMPXCHG, [SLM_IEC_RSV01], 6, [6], 5>;
+defm : X86WriteRes<WriteCMPXCHGRMW, [SLM_IEC_RSV01, SLM_MEC_RSV], 10, [6, 2], 8>;
 defm : X86WriteRes<WriteXCHG,      [SLM_IEC_RSV01], 1, [1], 1>;
 
 defm : SLMWriteResPair<WriteShift,    [SLM_IEC_RSV0],  1>;
diff --git a/llvm/test/tools/llvm-mca/X86/SLM/resources-cmpxchg.s b/llvm/test/tools/llvm-mca/X86/SLM/resources-cmpxchg.s
index f6636d3..86143c5 100644
--- a/llvm/test/tools/llvm-mca/X86/SLM/resources-cmpxchg.s
+++ b/llvm/test/tools/llvm-mca/X86/SLM/resources-cmpxchg.s
@@ -15,10 +15,10 @@ lock cmpxchg16b (%rax)
 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
-# CHECK-NEXT:  2      4     2.00    *      *            cmpxchg8b	(%rax)
-# CHECK-NEXT:  2      4     2.00    *      *            cmpxchg16b	(%rax)
-# CHECK-NEXT:  2      4     2.00    *      *            lock		cmpxchg8b	(%rax)
-# CHECK-NEXT:  2      4     2.00    *      *            lock		cmpxchg16b	(%rax)
+# CHECK-NEXT:  8      10    3.00    *      *            cmpxchg8b	(%rax)
+# CHECK-NEXT:  8      10    3.00    *      *            cmpxchg16b	(%rax)
+# CHECK-NEXT:  8      10    3.00    *      *            lock		cmpxchg8b	(%rax)
+# CHECK-NEXT:  8      10    3.00    *      *            lock		cmpxchg16b	(%rax)
 
 # CHECK:      Resources:
 # CHECK-NEXT: [0]   - SLMDivider
@@ -32,11 +32,11 @@ lock cmpxchg16b (%rax)
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]
-# CHECK-NEXT:  -      -      -      -      -     2.00   2.00   8.00
+# CHECK-NEXT:  -      -      -      -      -     12.00  12.00  8.00
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    Instructions:
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   cmpxchg8b	(%rax)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   cmpxchg16b	(%rax)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   lock		cmpxchg8b	(%rax)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   lock		cmpxchg16b	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   cmpxchg8b	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   cmpxchg16b	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   lock		cmpxchg8b	(%rax)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   lock		cmpxchg16b	(%rax)
diff --git a/llvm/test/tools/llvm-mca/X86/SLM/resources-x86_64.s b/llvm/test/tools/llvm-mca/X86/SLM/resources-x86_64.s
index 7b27f1b..15441af 100644
--- a/llvm/test/tools/llvm-mca/X86/SLM/resources-x86_64.s
+++ b/llvm/test/tools/llvm-mca/X86/SLM/resources-x86_64.s
@@ -1296,18 +1296,18 @@ xorq (%rax), %rdi
 # CHECK-NEXT:  1      100   1.00                  U     cmpsw	%es:(%rdi), (%rsi)
 # CHECK-NEXT:  1      100   1.00                  U     cmpsl	%es:(%rdi), (%rsi)
 # CHECK-NEXT:  1      100   1.00                  U     cmpsq	%es:(%rdi), (%rsi)
-# CHECK-NEXT:  1      1     0.50                        cmpxchgb	%cl, %bl
-# CHECK-NEXT:  2      4     2.00    *      *            cmpxchgb	%cl, (%rbx)
-# CHECK-NEXT:  2      4     2.00    *      *            lock		cmpxchgb	%cl, (%rbx)
-# CHECK-NEXT:  1      1     0.50                        cmpxchgw	%cx, %bx
-# CHECK-NEXT:  2      4     2.00    *      *            cmpxchgw	%cx, (%rbx)
-# CHECK-NEXT:  2      4     2.00    *      *            lock		cmpxchgw	%cx, (%rbx)
-# CHECK-NEXT:  1      1     0.50                        cmpxchgl	%ecx, %ebx
-# CHECK-NEXT:  2      4     2.00    *      *            cmpxchgl	%ecx, (%rbx)
-# CHECK-NEXT:  2      4     2.00    *      *            lock		cmpxchgl	%ecx, (%rbx)
-# CHECK-NEXT:  1      1     0.50                        cmpxchgq	%rcx, %rbx
-# CHECK-NEXT:  2      4     2.00    *      *            cmpxchgq	%rcx, (%rbx)
-# CHECK-NEXT:  2      4     2.00    *      *            lock		cmpxchgq	%rcx, (%rbx)
+# CHECK-NEXT:  5      6     3.00                        cmpxchgb	%cl, %bl
+# CHECK-NEXT:  8      10    3.00    *      *            cmpxchgb	%cl, (%rbx)
+# CHECK-NEXT:  8      10    3.00    *      *            lock		cmpxchgb	%cl, (%rbx)
+# CHECK-NEXT:  5      6     3.00                        cmpxchgw	%cx, %bx
+# CHECK-NEXT:  8      10    3.00    *      *            cmpxchgw	%cx, (%rbx)
+# CHECK-NEXT:  8      10    3.00    *      *            lock		cmpxchgw	%cx, (%rbx)
+# CHECK-NEXT:  5      6     3.00                        cmpxchgl	%ecx, %ebx
+# CHECK-NEXT:  8      10    3.00    *      *            cmpxchgl	%ecx, (%rbx)
+# CHECK-NEXT:  8      10    3.00    *      *            lock		cmpxchgl	%ecx, (%rbx)
+# CHECK-NEXT:  5      6     3.00                        cmpxchgq	%rcx, %rbx
+# CHECK-NEXT:  8      10    3.00    *      *            cmpxchgq	%rcx, (%rbx)
+# CHECK-NEXT:  8      10    3.00    *      *            lock		cmpxchgq	%rcx, (%rbx)
 # CHECK-NEXT:  1      100   1.00                  U     cpuid
 # CHECK-NEXT:  1      1     0.50                        decb	%dil
 # CHECK-NEXT:  1      5     2.00    *      *            decb	(%rax)
@@ -1960,7 +1960,7 @@ xorq (%rax), %rdi
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]
-# CHECK-NEXT: 400.00  -      -     49.00   -     654.50 595.50 835.00
+# CHECK-NEXT: 400.00  -      -     49.00   -     684.50 625.50 835.00
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    Instructions:
@@ -2226,18 +2226,18 @@ xorq (%rax), %rdi
 # CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cmpsw	%es:(%rdi), (%rsi)
 # CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cmpsl	%es:(%rdi), (%rsi)
 # CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cmpsq	%es:(%rdi), (%rsi)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -     cmpxchgb	%cl, %bl
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   cmpxchgb	%cl, (%rbx)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   lock		cmpxchgb	%cl, (%rbx)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -     cmpxchgw	%cx, %bx
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   cmpxchgw	%cx, (%rbx)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   lock		cmpxchgw	%cx, (%rbx)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -     cmpxchgl	%ecx, %ebx
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   cmpxchgl	%ecx, (%rbx)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   lock		cmpxchgl	%ecx, (%rbx)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -     cmpxchgq	%rcx, %rbx
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   cmpxchgq	%rcx, (%rbx)
-# CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   lock		cmpxchgq	%rcx, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -     cmpxchgb	%cl, %bl
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   cmpxchgb	%cl, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   lock		cmpxchgb	%cl, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -     cmpxchgw	%cx, %bx
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   cmpxchgw	%cx, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   lock		cmpxchgw	%cx, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -     cmpxchgl	%ecx, %ebx
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   cmpxchgl	%ecx, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   lock		cmpxchgl	%ecx, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -     cmpxchgq	%rcx, %rbx
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   cmpxchgq	%rcx, (%rbx)
+# CHECK-NEXT:  -      -      -      -      -     3.00   3.00   2.00   lock		cmpxchgq	%rcx, (%rbx)
 # CHECK-NEXT:  -      -      -     1.00    -      -      -      -     cpuid
 # CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -     decb	%dil
 # CHECK-NEXT:  -      -      -      -      -     0.50   0.50   2.00   decb	(%rax)
-- 
2.7.4