From d315f61414ef06126ad019607e7c577b59a831fa Mon Sep 17 00:00:00 2001 From: Jens Steube Date: Sun, 24 Nov 2019 08:50:31 +0100 Subject: [PATCH] Fix -m 1800 speed on ROCM --- src/modules/module_01800.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff --git a/src/modules/module_01800.c b/src/modules/module_01800.c index 9b6d0c196..4b1540dd7 100644 --- a/src/modules/module_01800.c +++ b/src/modules/module_01800.c @@ -434,7 +434,14 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY { char *jit_build_options = NULL; - hc_asprintf (&jit_build_options, "-D NO_UNROLL"); + if ((device_param->opencl_device_vendor_id == VENDOR_ID_AMD) && (device_param->has_vperm == true)) + { + // unroll is faster on rocm in this kernel + } + else + { + hc_asprintf (&jit_build_options, "-D NO_UNROLL"); + } return jit_build_options; }