[PowerPC][NFC] Define new alias for mma accumulate builtins #147382
Merged
+64
−105
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
llvmbot
added
clang
|
@llvm/pr-subscribers-backend-powerpc @llvm/pr-subscribers-clang Author: Lei Huang (lei137) ChangesInitial clean up in prep for adding more accumulate builtins to clang. Full diff: https://github.com/llvm/llvm-project/pull/147382.diff 1 Files Affected:
diff --git a/clang/include/clang/Basic/BuiltinsPPC.def b/clang/include/clang/Basic/BuiltinsPPC.def
index 7c278d6841c74..3fa9961a9602c 100644
--- a/clang/include/clang/Basic/BuiltinsPPC.def
+++ b/clang/include/clang/Basic/BuiltinsPPC.def
@@ -35,6 +35,13 @@
#define UNALIASED_CUSTOM_BUILTIN(ID, TYPES, ACCUMULATE, FEATURE) \
CUSTOM_BUILTIN(ID, ID, TYPES, ACCUMULATE, FEATURE)
+#define UNALIASED_CUSTOM_MMA_BUILTIN(ID, TYPES, FEATURE) \
+ UNALIASED_CUSTOM_BUILTIN(ID, TYPES, false, FEATURE) \
+ UNALIASED_CUSTOM_BUILTIN(ID##nn, TYPES, true, FEATURE) \
+ UNALIASED_CUSTOM_BUILTIN(ID##np, TYPES, true, FEATURE) \
+ UNALIASED_CUSTOM_BUILTIN(ID##pn, TYPES, true, FEATURE) \
+ UNALIASED_CUSTOM_BUILTIN(ID##pp, TYPES, true, FEATURE)
+
// GCC predefined macros to rename builtins, undef them to keep original names.
#if defined(__GNUC__) && !defined(__clang__)
#undef __builtin_vsx_xvnmaddadp
@@ -1032,12 +1039,6 @@ UNALIASED_CUSTOM_BUILTIN(mma_xvi16ger2, "vW512*VV", false,
"mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_xvi16ger2s, "vW512*VV", false,
"mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf16ger2, "vW512*VV", false,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf32ger, "vW512*VV", false,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf64ger, "vW512*W256V", false,
- "mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_pmxvi4ger8, "vW512*VVi15i15i255", false,
"mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_pmxvi8ger4, "vW512*VVi15i15i15", false,
@@ -1046,12 +1047,6 @@ UNALIASED_CUSTOM_BUILTIN(mma_pmxvi16ger2, "vW512*VVi15i15i3", false,
"mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_pmxvi16ger2s, "vW512*VVi15i15i3", false,
"mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf16ger2, "vW512*VVi15i15i3", false,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf32ger, "vW512*VVi15i15", false,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf64ger, "vW512*W256Vi15i3", false,
- "mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_xvi4ger8pp, "vW512*VV", true,
"mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_xvi8ger4pp, "vW512*VV", true,
@@ -1072,74 +1067,6 @@ UNALIASED_CUSTOM_BUILTIN(mma_pmxvi16ger2pp, "vW512*VVi15i15i3", true,
"mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_pmxvi16ger2spp, "vW512*VVi15i15i3", true,
"mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf16ger2pp, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf16ger2pn, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf16ger2np, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf16ger2nn, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf16ger2pp, "vW512*VVi15i15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf16ger2pn, "vW512*VVi15i15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf16ger2np, "vW512*VVi15i15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf16ger2nn, "vW512*VVi15i15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf32gerpp, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf32gerpn, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf32gernp, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf32gernn, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf32gerpp, "vW512*VVi15i15", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf32gerpn, "vW512*VVi15i15", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf32gernp, "vW512*VVi15i15", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf32gernn, "vW512*VVi15i15", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf64gerpp, "vW512*W256V", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf64gerpn, "vW512*W256V", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf64gernp, "vW512*W256V", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvf64gernn, "vW512*W256V", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf64gerpp, "vW512*W256Vi15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf64gerpn, "vW512*W256Vi15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf64gernp, "vW512*W256Vi15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvf64gernn, "vW512*W256Vi15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvbf16ger2, "vW512*VV", false,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvbf16ger2, "vW512*VVi15i15i3", false,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvbf16ger2pp, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvbf16ger2pn, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvbf16ger2np, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_xvbf16ger2nn, "vW512*VV", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvbf16ger2pp, "vW512*VVi15i15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvbf16ger2pn, "vW512*VVi15i15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvbf16ger2np, "vW512*VVi15i15i3", true,
- "mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_pmxvbf16ger2nn, "vW512*VVi15i15i3", true,
- "mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_dmxvi8gerx4, "vW1024*W256V", false,
"mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_pmdmxvi8gerx4, "vW1024*W256Vi255i15i15", false,
@@ -1148,10 +1075,26 @@ UNALIASED_CUSTOM_BUILTIN(mma_dmxvi8gerx4pp, "vW1024*W256V", true,
"mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_pmdmxvi8gerx4pp, "vW1024*W256Vi255i15i15", true,
"mma,paired-vector-memops")
-UNALIASED_CUSTOM_BUILTIN(mma_dmxvi8gerx4spp, "vW1024*W256V", true,
+UNALIASED_CUSTOM_BUILTIN(mma_dmxvi8gerx4spp, "vW1024*W256V", true,
"mma,paired-vector-memops")
UNALIASED_CUSTOM_BUILTIN(mma_pmdmxvi8gerx4spp, "vW1024*W256Vi255i15i15", true,
"mma,paired-vector-memops")
+UNALIASED_CUSTOM_MMA_BUILTIN(mma_xvf16ger2, "vW512*VV",
+ "mma,paired-vector-memops")
+UNALIASED_CUSTOM_MMA_BUILTIN(mma_xvf32ger, "vW512*VV",
+ "mma,paired-vector-memops")
+UNALIASED_CUSTOM_MMA_BUILTIN(mma_xvf64ger, "vW512*W256V",
+ "mma,paired-vector-memops")
+UNALIASED_CUSTOM_MMA_BUILTIN(mma_pmxvf16ger2, "vW512*VVi15i15i3",
+ "mma,paired-vector-memops")
+UNALIASED_CUSTOM_MMA_BUILTIN(mma_pmxvf32ger, "vW512*VVi15i15",
+ "mma,paired-vector-memops")
+UNALIASED_CUSTOM_MMA_BUILTIN(mma_pmxvf64ger, "vW512*W256Vi15i3",
+ "mma,paired-vector-memops")
+UNALIASED_CUSTOM_MMA_BUILTIN(mma_xvbf16ger2, "vW512*VV",
+ "mma,paired-vector-memops")
+UNALIASED_CUSTOM_MMA_BUILTIN(mma_pmxvbf16ger2, "vW512*VVi15i15i3",
+ "mma,paired-vector-memops")
// FIXME: Obviously incomplete.
|
amy-kwan
reviewed
Jul 8, 2025
the macros are defined vs down below where the macro is first used.
amy-kwan
approved these changes
Jul 9, 2025
apply nit Co-authored-by: Amy Kwan <amy.kwan1@ibm.com>
maryammo
reviewed
Jul 9, 2025
| "mma,paired-vector-memops") | ||
| UNALIASED_CUSTOM_BUILTIN(mma_pmdmxvi8gerx4spp, "vW1024*W256Vi255i15i15", true, | ||
| "mma,paired-vector-memops") | ||
|
|
||
| // MMA builtins with positive/negative multiply/accumulate. |
There was a problem hiding this comment.
please add a list of builtin's name as a comment here.
There was a problem hiding this comment.
I feel that would be a bit of an overkill. I have added an example at the top to further document the new macro usage.
maryammo
approved these changes
Jul 9, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Initial clean up in prep for adding more accumulate builtins to clang.