MirrorHub/construct
0
0
Fork 0
mirror of https://github.com/matrix-construct/construct synced 2025-02-18 01:30:12 +01:00
Code Issues Releases Wiki Activity

ircd::simt: Simplify keywording; tease out OpenCL version by target.

Browse source
This commit is contained in:
Jason Volk 2022-10-19 00:41:35 +00:00
parent e89703aa97
commit ac163274f1
3 changed files with 16 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
include/ircd/gpt/vector.h
View file

@ -11,14 +11,14 @@
#pragma once
#define HAVE_IRCD_GPT_VECTOR_H
static __constant const uint
__constant const uint
ircd_gpt_context_tokens = 512, // 1024,
ircd_gpt_vector_elems = 768,
ircd_gpt_attn_rank = 12,
ircd_gpt_attn_segs = 3,
ircd_gpt_ffnn_segs = 4;
static __constant const uint
__constant const uint
ircd_gpt_vector_attn_elems = ircd_gpt_vector_elems / ircd_gpt_attn_rank,
ircd_gpt_attn_fcon_elems = ircd_gpt_vector_elems * ircd_gpt_attn_segs,
ircd_gpt_ffnn_fcon_elems = ircd_gpt_vector_elems * ircd_gpt_ffnn_segs;

4
include/ircd/simt/portable.h
View file

@ -30,8 +30,6 @@
#if defined(__OPENCL_VERSION__)
#if __OPENCL_VERSION__ < 120
#define static __attribute__((internal_linkage))
#elif __OPENCL_VERSION__ < 200
#define static __constant static
#endif
#endif
@ -42,7 +40,7 @@
//
#if !defined(__OPENCL_VERSION__)
#define __constant
#define __constant static
#endif
//

17
ircd/Makefile.am
View file

@ -388,7 +388,6 @@ if CLANG
if OPENCL
GPU_CPPFLAGS =#
GPU_CPPFLAGS += -D__OPENCL_VERSION__=110
GPU_CPPFLAGS += -Dcl_clang_storage_class_specifiers
GPU_CPPFLAGS += -DNOFP64
GPU_CPPFLAGS += -I$(top_srcdir)/include
@ -398,7 +397,6 @@ GPU_CPPFLAGS += $(CWARNS)
GPU_CPPFLAGS += $(DEFS)
GPU_CFLAGS =#
GPU_CFLAGS += -std=cl1.1
GPU_CFLAGS += -fident
GPU_CFLAGS += -fno-builtin
GPU_CFLAGS += -fverbose-asm
@ -440,9 +438,16 @@ GPU_ASFLAGS += -Rpass-analysis=asm-printer
# SPV
#
SPV_CPPFLAGS = $(GPU_CPPFLAGS)
SPV_CPPFLAGS += -D__OPENCL_VERSION__=120
SPV_CFLAGS = $(GPU_CFLAGS)
SPV_CFLAGS += -std=cl2.0
SPV_CFLAGS += -target spir--
BUILT_SOURCES += gpt_gpu.spv.bc
gpt_gpu.spv.bc: gpt_gpu.cl
$(CC) -target spir-- $(GPU_CPPFLAGS) -O0 -emit-llvm -o $@ -x cl -c $^
$(CC) $(SPV_CFLAGS) $(SPV_CPPFLAGS) -O0 -emit-llvm -o $@ -x cl -c $^
BUILT_SOURCES += gpt_gpu.spv
gpt_gpu.spv: gpt_gpu.spv.bc
@ -463,14 +468,18 @@ R600_TARGET = r600--
# R600 Saint Barthélemy
#
R600_BARTS_CPPFLAGS = $(GPU_CPPFLAGS)
R600_BARTS_CPPFLAGS += -D__OPENCL_VERSION__=110
R600_BARTS_CFLAGS = $(GPU_CFLAGS)
R600_BARTS_CFLAGS += -std=cl1.1
R600_BARTS_CFLAGS += -target $(R600_TARGET)
R600_BARTS_CFLAGS += -mcpu=barts
R600_BARTS_CFLAGS += -Xclang -mlink-bitcode-file -Xclang /usr/lib/clc/barts-r600--.bc
BUILT_SOURCES += gpt_gpu.r600_barts.bc
gpt_gpu.r600_barts.bc: gpt_gpu.cl
$(CC) $(R600_BARTS_CFLAGS) $(GPU_CPPFLAGS) -emit-llvm -o $@ -x cl -c $^
$(CC) $(R600_BARTS_CFLAGS) $(R600_BARTS_CPPFLAGS) -emit-llvm -o $@ -x cl -c $^
#BUILT_SOURCES += gpt_gpu.r600_barts.link.bc
gpt_gpu.r600_barts.link.bc: gpt_gpu.r600_barts.bc