From b57e63b07a7b70ebfb5f794648e2102b7c1bd3a3 Mon Sep 17 00:00:00 2001 From: Matt Arsenault Date: Tue, 25 Feb 2025 18:23:04 +0700 Subject: [PATCH] libclc: Stop using asm declarations for r600 on amdgcn for get_global_size (#128692) Comparing the case where each dimension is used alone, the only codegen difference is a missed addressing mode fold for the constant offset in the old version due to an ancient bug. --- libclc/amdgcn/lib/workitem/get_global_size.cl | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-) diff --git a/libclc/amdgcn/lib/workitem/get_global_size.cl b/libclc/amdgcn/lib/workitem/get_global_size.cl index 2f28ca606665..0fec7e24966f 100644 --- a/libclc/amdgcn/lib/workitem/get_global_size.cl +++ b/libclc/amdgcn/lib/workitem/get_global_size.cl @@ -1,17 +1,13 @@ #include -uint __clc_amdgcn_get_global_size_x(void) __asm("llvm.r600.read.global.size.x"); -uint __clc_amdgcn_get_global_size_y(void) __asm("llvm.r600.read.global.size.y"); -uint __clc_amdgcn_get_global_size_z(void) __asm("llvm.r600.read.global.size.z"); - _CLC_DEF _CLC_OVERLOAD size_t get_global_size(uint dim) { switch (dim) { case 0: - return __clc_amdgcn_get_global_size_x(); + return __builtin_amdgcn_grid_size_x(); case 1: - return __clc_amdgcn_get_global_size_y(); + return __builtin_amdgcn_grid_size_y(); case 2: - return __clc_amdgcn_get_global_size_z(); + return __builtin_amdgcn_grid_size_z(); default: return 1; }