A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 05:45 UTC
[Build] Fix for the Wswitch-bool clang warning (#10060)
Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
Gregory Shtrasberg committed
Dec 7, 2024 at 04:03 UTC
f13cf9ad5049e386f766014877dee78d2f438799
2 files changed
+8
-14
csrc/attention/paged_attention_v1.cu
+4
-7
@@ -140,13 +140,10 @@ void paged_attention_v1_launcher(
140
blocksparse_block_size, blocksparse_head_sliding_step);
141
142
#define CALL_V1_LAUNCHER_SPARSITY(T, CACHE_T, BLOCK_SIZE, IS_FP8_KV_CACHE) \
143
- switch (is_block_sparse) { \
144
- case true: \
145
- CALL_V1_LAUNCHER(T, CACHE_T, BLOCK_SIZE, IS_FP8_KV_CACHE, true); \
146
- break; \
147
- case false: \
148
- CALL_V1_LAUNCHER(T, CACHE_T, BLOCK_SIZE, IS_FP8_KV_CACHE, false); \
149
- break; \
143
+ if (is_block_sparse) { \
144
+ CALL_V1_LAUNCHER(T, CACHE_T, BLOCK_SIZE, IS_FP8_KV_CACHE, true); \
145
+ } else { \
146
+ CALL_V1_LAUNCHER(T, CACHE_T, BLOCK_SIZE, IS_FP8_KV_CACHE, false); \
147
}
148
149
// NOTE(woosuk): To reduce the compilation time, we omitted block sizes
csrc/attention/paged_attention_v2.cu
+4
-7
@@ -147,13 +147,10 @@ void paged_attention_v2_launcher(
147
blocksparse_head_sliding_step);
148
149
#define CALL_V2_LAUNCHER_SPARSITY(T, CACHE_T, BLOCK_SIZE, IS_FP8_KV_CACHE) \
150
- switch (is_block_sparse) { \
151
- case true: \
152
- CALL_V2_LAUNCHER(T, CACHE_T, BLOCK_SIZE, IS_FP8_KV_CACHE, true); \
153
- break; \
154
- case false: \
155
- CALL_V2_LAUNCHER(T, CACHE_T, BLOCK_SIZE, IS_FP8_KV_CACHE, false); \
156
- break; \
150
+ if (is_block_sparse) { \
151
+ CALL_V2_LAUNCHER(T, CACHE_T, BLOCK_SIZE, IS_FP8_KV_CACHE, true); \
152
+ } else { \
153
+ CALL_V2_LAUNCHER(T, CACHE_T, BLOCK_SIZE, IS_FP8_KV_CACHE, false); \
154
}
155
156
// NOTE(woosuk): To reduce the compilation time, we omitted block sizes