60 lines · plain
1; RUN: llc < %s -O0 -mtriple=nvptx64 -mcpu=sm_20 | FileCheck %s2; RUN: %if ptxas %{ llc < %s -O0 -mtriple=nvptx64 -mcpu=sm_20 | %ptxas-verify %}3 4; CHECK-LABEL: .visible .func callee(5; CHECK-NEXT: .param .align 16 .b8 callee_param_0[16],6; CHECK-NEXT: .param .align 16 .b8 callee_param_1[16],7define void @callee(i128, i128, ptr) {8 ; CHECK-DAG: ld.param.v2.b64 {%[[REG0:rd[0-9]+]], %[[REG1:rd[0-9]+]]}, [callee_param_0];9 ; CHECK-DAG: ld.param.v2.b64 {%[[REG2:rd[0-9]+]], %[[REG3:rd[0-9]+]]}, [callee_param_1];10 11 ; CHECK: mul.lo.s64 %[[REG4:rd[0-9]+]], %[[REG0]], %[[REG3]];12 ; CHECK-NEXT: mul.hi.u64 %[[REG5:rd[0-9]+]], %[[REG0]], %[[REG2]];13 ; CHECK-NEXT: add.s64 %[[REG6:rd[0-9]+]], %[[REG5]], %[[REG4]];14 ; CHECK-NEXT: mul.lo.s64 %[[REG7:rd[0-9]+]], %[[REG1]], %[[REG2]];15 ; CHECK-NEXT: add.s64 %[[REG8:rd[0-9]+]], %[[REG6]], %[[REG7]];16 ; CHECK-NEXT: mul.lo.s64 %[[REG9:rd[0-9]+]], %[[REG0]], %[[REG2]];17 %a = mul i128 %0, %118 19 store i128 %a, ptr %220 ret void21}22 23; CHECK-LABEL: .visible .entry caller_kernel(24; CHECK-NEXT: .param .align 16 .b8 caller_kernel_param_0[16],25; CHECK-NEXT: .param .align 16 .b8 caller_kernel_param_1[16],26define ptx_kernel void @caller_kernel(i128, i128, ptr) {27start:28 ; CHECK-DAG: ld.param.v2.b64 {%[[REG0:rd[0-9]+]], %[[REG1:rd[0-9]+]]}, [caller_kernel_param_0];29 ; CHECK-DAG: ld.param.v2.b64 {%[[REG2:rd[0-9]+]], %[[REG3:rd[0-9]+]]}, [caller_kernel_param_1];30 31 ; CHECK: { // callseq [[CALLSEQ_ID:[0-9]]], 032 ; CHECK-DAG: .param .align 16 .b8 param0[16];33 ; CHECK-DAG: .param .align 16 .b8 param1[16];34 ; CHECK-DAG: st.param.v2.b64 [param0], {%[[REG0]], %[[REG1]]}35 ; CHECK-DAG: st.param.v2.b64 [param1], {%[[REG2]], %[[REG3]]}36 ; CHECK: } // callseq [[CALLSEQ_ID]]37 call void @callee(i128 %0, i128 %1, ptr %2)38 39 ret void40}41 42; CHECK-LABEL: .visible .func caller_func(43; CHECK-NEXT: .param .align 16 .b8 caller_func_param_0[16],44; CHECK-NEXT: .param .align 16 .b8 caller_func_param_1[16],45define void @caller_func(i128, i128, ptr) {46start:47 ; CHECK-DAG: ld.param.v2.b64 {%[[REG0:rd[0-9]+]], %[[REG1:rd[0-9]+]]}, [caller_func_param_0]48 ; CHECK-DAG: ld.param.v2.b64 {%[[REG2:rd[0-9]+]], %[[REG3:rd[0-9]+]]}, [caller_func_param_1]49 50 ; CHECK: { // callseq [[CALLSEQ_ID:[0-9]]], 051 ; CHECK-DAG: .param .align 16 .b8 param0[16];52 ; CHECK-DAG: .param .align 16 .b8 param1[16];53 ; CHECK-DAG: st.param.v2.b64 [param0], {%[[REG0]], %[[REG1]]}54 ; CHECK-DAG: st.param.v2.b64 [param1], {%[[REG2]], %[[REG3]]}55 ; CHECK: } // callseq [[CALLSEQ_ID]]56 call void @callee(i128 %0, i128 %1, ptr %2)57 58 ret void59}60