https://gcc.gnu.org/g:511fc0531e61ea3641191ff811c2c6ea61ad8560

commit 511fc0531e61ea3641191ff811c2c6ea61ad8560
Author: Michael Meissner <[email protected]>
Date:   Thu Sep 10 14:55:31 2026 -0400

    Update ChangeLog.*

Diff:
---
 gcc/ChangeLog.float | 36 ++++++++++++++++++++++++++++++++++++
 1 file changed, 36 insertions(+)

diff --git a/gcc/ChangeLog.float b/gcc/ChangeLog.float
index 8d867bc7ed69..f054a93c4d02 100644
--- a/gcc/ChangeLog.float
+++ b/gcc/ChangeLog.float
@@ -1,3 +1,39 @@
+==================== Branch work262-float, patch #210 ====================
+
+Convert __bf16 to float or double on power8 and power9.
+
+2026-09-10  Michael Meissner  <[email protected]>
+
+gcc/
+
+       * config/rs6000/float16.md (extendbf<mode>): Convert to define_expand.
+       (extendbf<mode>2_p10): Rename from extendbf<mode>2.  Allow HFmode/BFmode
+       to be in all VSX registers, not just Altivec registers.
+       (extendbf<mode>2_not_p10): Implement BFmode to SFmode/DFmode conversions
+       just using shifts.
+       (extendbfsf2_memory): Rename from convert_bf_to_sf_store.
+       (fix_trunchf<mode>2): Split HFmode/BFmode fix truncate insns into
+       separate HFmode and BFmode insns.  On BFmode conversions, do the
+       conversion on power8/power9 using shifts and xscvspdpn.
+       (fixuns_trunchf<mode>2): Likewise.
+       (fix_truncbf<mode>2): Likewise.
+       (fixuns_truncbf<GPR:mode>2): Likewise.
+       (bfloat16_binary_op_internal1): Don't do optimization if
+       -mno-bfloat16-p10-convert.
+       (bfloat16_binary_op_internal2): Likewise.
+       (bfloat16_fma_internal1): Likewise.
+       (bfloat16_fma_internal2): Likewise.
+       (bfloat16_fms_internal1): Likewise.
+       (bfloat16_fms_internal2): Likewise.
+       (bfloat16_nfma_internal1): Likewise.
+       (bfloat16_nfma_internal2): Likewise.
+       (bfloat16_nfma_internal3): Likewise.
+       (bfloat16_nfms_internal1): Likewise.
+       (bfloat16_nfms_internal2): Likewise.
+       (bfloat16_nfms_internal3): Likewise.
+       * config/rs6000/rs6000.opt (-mbfloat16-convert): New option.
+       (-mbfloat16-p10-convert): Likewise.
+
 ==================== Branch work262-float, patch #209 ====================
 
 Add _Float16 and __bf16 tests.

Reply via email to