github-mirrors/vortex
1
0
Fork 0
mirror of https://github.com/vortexgpgpu/vortex.git synced 2025-04-24 13:57:17 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Added extra test cases

Browse source
This commit is contained in:
Rishabh Ravi 2025-01-14 13:40:02 -05:00
parent b293397c01
commit a51fb03d5c
3 changed files with 97 additions and 27 deletions
Download patch file Download diff file Expand all files Collapse all files

62
kernel/include/vx_intrinsics.h
View file

@ -263,12 +263,72 @@ inline void vx_shfl() {
".insn i %0, 3, x14, x13, 1067" :: "i"(RISCV_CUSTOM2)); //(c(01)+b(00001)+membermask(address(01011)))
//".insn i opcode6, func3, rd, rs1, simm12"
}
inline void vx_tile(unsigned int tile_mask, int thread_count) {
__asm__ volatile (
".insn r %0, 1, 0, x0, %1, %2" :: "i"(RISCV_CUSTOM3),"r"(tile_mask),"r"(thread_count));
}
// mode: 0:ALL, 1:ANY, 2:UNI, 3:BALLOT
inline int vx_vote_sync(int pred,
int neg,
int mode,
int threadMask)
{
int func3 = ((neg & 0x1) << 2) | (mode & 0x3);
int rs1 = pred;
int rd;
__asm__ volatile (
"addi a2, %[tm], 0\n\t"
".insn i %[opcode], %[f3], %[rd], %[rs1], 12\n\t"
: [rd] "=r" (rd)
: [tm] "r" (threadMask),
[opcode] "i" (RISCV_CUSTOM1),
[f3] "i" (func3),
[rs1] "r" (rs1)
: "a2"
);
return rd;
}
inline int vx_shfl_sync(int offset, int mode, int val, int threadMask)
{
int func3 = mode & 0x3;
int rs1 = val;
int rd;
int imm12 = (11) | ((offset & 0x1F) << 5) | (1 << 10);
int c = -1;
__asm__ volatile (
"addi a1, %[tm], 0\n\t" // Load immediate value 15 into a1(x11) register (membermask)
"addi a2, %[c], 0\n\t" // Load immediate value 15 into a2(x12) register (c)
".insn i %[opcode], %[f3], %[rd], %[rs], %[imm]\n\t"
: "=r"(rd)
: [tm] "r"(threadMask),
[c] "r"(c),
[opcode] "i"(RISCV_CUSTOM2),
[f3] "i"(func3),
[rd] "r"(rd),
[rs] "r"(rs1),
[imm] "i"(imm12)
:
"a1", "a2"
);
return rd;
}
#ifdef __cplusplus
}
#endif

12
tests/regression/Makefile
View file

@ -18,6 +18,8 @@ all:
$(MAKE) -C vote
$(MAKE) -C shfl
$(MAKE) -C coop
$(MAKE) -C cg_mse_cm_hw
$(MAKE) -C cg_mse_tm_hw
run-simx:
$(MAKE) -C basic run-simx
@ -36,6 +38,8 @@ run-simx:
$(MAKE) -C vote run-simx
$(MAKE) -C shfl run-simx
$(MAKE) -C coop run-simx
$(MAKE) -C cg_mse_cm_hw run-simx
$(MAKE) -C cg_mse_tm_hw run-simx
run-rtlsim:
$(MAKE) -C basic run-rtlsim
@ -54,6 +58,8 @@ run-rtlsim:
$(MAKE) -C vote run-rtlsim
$(MAKE) -C shfl run-rtlsim
$(MAKE) -C coop run-rtlsim
$(MAKE) -C cg_mse_cm_hw run-rtlsim
$(MAKE) -C cg_mse_tm_hw run-rtlsim
run-opae:
$(MAKE) -C basic run-opae
@ -72,6 +78,8 @@ run-opae:
$(MAKE) -C vote run-opae
$(MAKE) -C shfl run-opae
$(MAKE) -C coop run-opae
$(MAKE) -C cg_mse_cm_hw run-opae
$(MAKE) -C cg_mse_tm_hw run-opae
clean:
$(MAKE) -C basic clean
@ -89,4 +97,6 @@ clean:
$(MAKE) -C sgemm2x clean
$(MAKE) -C vote clean
$(MAKE) -C shfl clean
$(MAKE) -C coop clean
$(MAKE) -C coop clean
$(MAKE) -C cg_mse_cm_hw clean
$(MAKE) -C cg_mse_tm_hw clean

50
tests/regression/coop/kernel.cpp
View file

@ -27,30 +27,30 @@ void kernel_body(kernel_arg_t* __UNIFORM__ arg) {
// vx_tile(0b10000000,32); // default config
// //CASE 2
// int i = vx_thread_id();
// vx_printf("A (thread %d)\n", i);
int i = vx_thread_id();
vx_printf("A (thread %d)\n", i);
// vx_tile(0b10001000,16);
vx_tile(0b10001000,16);
// vx_tile(0b00001000,16);
// {
// vx_printf("B (thread %d)\n", i);
// vx_barrier(0,0b00001000);
// vx_printf("C (thread %d)\n", i);
// }
// vx_tile(0b10000000,16);
// {
// vx_printf("D (thread %d)\n", i);
// }
vx_tile(0b00001000,16);
{
vx_printf("B (thread %d)\n", i);
vx_barrier(0,0b00001000);
vx_printf("C (thread %d)\n", i);
}
vx_tile(0b10000000,16);
{
vx_printf("D (thread %d)\n", i);
}
// vx_tile(0b10001000,16);
// if(i%16 == 0)
// {
// vx_printf("E (thread %d)\n", i);
// }
vx_tile(0b10001000,16);
if(i%16 == 0)
{
vx_printf("E (thread %d)\n", i);
}
// vx_tile(0b10000000,32); //default config
vx_tile(0b10000000,32); //default config
// //CASE 3
@ -164,17 +164,17 @@ void kernel_body(kernel_arg_t* __UNIFORM__ arg) {
// //CASE 9
// vx_printf("A (thread %d)\n", vx_thread_id());
vx_printf("A (thread %d)\n", vx_thread_id());
// vx_tile(0b10101010,8);
vx_tile(0b10101010,8);
// vx_printf("B (thread %d)\n", vx_thread_id());
vx_printf("B (thread %d)\n", vx_thread_id());
// vx_barrier(0,0b10101010);
vx_barrier(0,0b10101010);
// vx_printf("C (thread %d)\n", vx_thread_id());
vx_printf("C (thread %d)\n", vx_thread_id());
// vx_tile(0b10000000,32); //default config
vx_tile(0b10000000,32); //default config
// //CASE 10