254 lines
10 KiB
Plaintext
254 lines
10 KiB
Plaintext
/**
|
|
* TMEM Layout D for PV MMA N=64: zero TMEM before PV, then dump.
|
|
* This tells us exactly which columns the PV MMA writes to.
|
|
*/
|
|
|
|
#include <cuda_runtime.h>
|
|
#include <cstdio>
|
|
#include <cstring>
|
|
#include <cmath>
|
|
|
|
#include "dsv4/kernels/attention/fmha_common.cuh"
|
|
#include "dsv4/kernels/attention/fmha_umma_desc.cuh"
|
|
|
|
using namespace dsv4::kernels::attention;
|
|
|
|
static bf16_t f32_to_bf16_host(float f) { uint32_t u; memcpy(&u,&f,4); return (uint16_t)(u>>16); }
|
|
static float bf16_to_f32_host(bf16_t h) { uint32_t u=(uint32_t)h<<16; float f; memcpy(&f,&u,4); return f; }
|
|
|
|
constexpr int HD = 64, SK = 128, BLOCK_MN = 128;
|
|
constexpr int LOCAL_MMA_K = 16;
|
|
constexpr int TILE_SZ = BLOCK_MN * LOCAL_MMA_K;
|
|
constexpr int V_TILE_SZ = (HD / 8) * 2 * 64;
|
|
|
|
__global__ void __launch_bounds__(128)
|
|
test_tmem_zero_before_pv(const bf16_t* q, const bf16_t* k, const bf16_t* v,
|
|
float* tmem_dump, float scale)
|
|
{
|
|
const int tid = threadIdx.x, wid = tid / 32, lane = tid % 32;
|
|
|
|
extern __shared__ char sbuf[];
|
|
uint32_t* sTmemBase = (uint32_t*)sbuf;
|
|
bf16_t* sQ0 = (bf16_t*)(((uintptr_t)(sbuf + 4) + 15) & ~(uintptr_t)15);
|
|
bf16_t* sK0 = sQ0 + 4 * TILE_SZ;
|
|
bf16_t* sPk = (bf16_t*)(((uintptr_t)(sK0 + 4 * TILE_SZ) + 127) & ~(uintptr_t)127);
|
|
bf16_t* sV = (bf16_t*)(((uintptr_t)(sPk + TILE_SZ) + 127) & ~(uintptr_t)127);
|
|
float* s_p_vals = (float*)(sV + 8 * V_TILE_SZ);
|
|
|
|
for (int kt = 0; kt < 4; kt++) {
|
|
bf16_t* sq = sQ0 + kt * TILE_SZ;
|
|
for (int i = tid; i < TILE_SZ; i += 128) sq[i] = 0;
|
|
for (int d = tid; d < LOCAL_MMA_K; d += 128) {
|
|
int ck = d / 8, lc = d % 8;
|
|
sq[ck * 16 * 64 + lc] = q[kt * LOCAL_MMA_K + d];
|
|
}
|
|
}
|
|
for (int kt = 0; kt < 4; kt++) {
|
|
bf16_t* sk = sK0 + kt * TILE_SZ;
|
|
for (int i = tid; i < TILE_SZ; i += 128) sk[i] = 0;
|
|
for (int r = 0; r < SK; r++) {
|
|
for (int d = tid; d < LOCAL_MMA_K; d += 128) {
|
|
int ck = d / 8, lc = d % 8;
|
|
int tmn = r / 8, lr = r % 8;
|
|
sk[ck * 16 * 64 + tmn * 64 + lr * 8 + lc] = k[r * HD + kt * LOCAL_MMA_K + d];
|
|
}
|
|
}
|
|
}
|
|
for (int kt = 0; kt < 8; kt++) {
|
|
bf16_t* sv = sV + kt * V_TILE_SZ;
|
|
for (int i = tid; i < V_TILE_SZ; i += 128) sv[i] = 0;
|
|
for (int d = tid; d < HD; d += 128) {
|
|
for (int lr = 0; lr < LOCAL_MMA_K; lr++) {
|
|
int r = kt * LOCAL_MMA_K + lr;
|
|
int g_mn = d / 8, g_k = lr / 8;
|
|
int llr = d % 8, lc = lr % 8;
|
|
sv[g_k * 8 * 64 + g_mn * 64 + llr * 8 + lc] = v[d * SK + r];
|
|
}
|
|
}
|
|
}
|
|
__syncthreads();
|
|
|
|
if (wid == 1) tmem_alloc(__cvta_generic_to_shared(sTmemBase), 128);
|
|
__syncthreads();
|
|
uint32_t tb = *sTmemBase;
|
|
|
|
// QK GEMM
|
|
{
|
|
uint32_t idesc = make_idesc(BLOCK_MN, BLOCK_MN);
|
|
for (int kt = 0; kt < 4; kt++) {
|
|
bf16_t* sq = sQ0 + kt * TILE_SZ;
|
|
bf16_t* sk = sK0 + kt * TILE_SZ;
|
|
uint64_t dq = make_umma_desc_kmajor_none(__cvta_generic_to_shared(sq), BLOCK_MN);
|
|
uint64_t dk = make_umma_desc_kmajor_none(__cvta_generic_to_shared(sk), BLOCK_MN);
|
|
if (tid == 0) umma_ss_f16(tb, dq, dk, idesc, kt > 0);
|
|
asm volatile("tcgen05.fence::after_thread_sync;" ::: "memory");
|
|
__syncthreads();
|
|
}
|
|
}
|
|
|
|
// Softmax
|
|
if (wid == 0) {
|
|
float s_vals[SK], row_max = -INFINITY;
|
|
for (int n = 0; n < SK / 8; n++) {
|
|
float tmp[8];
|
|
asm volatile("tcgen05.ld.sync.aligned.32x32b.x8.b32 {%0,%1,%2,%3,%4,%5,%6,%7},[%8];"
|
|
: "=f"(tmp[0]),"=f"(tmp[1]),"=f"(tmp[2]),"=f"(tmp[3]),
|
|
"=f"(tmp[4]),"=f"(tmp[5]),"=f"(tmp[6]),"=f"(tmp[7])
|
|
: "r"(tb + n*8));
|
|
asm volatile("tcgen05.wait::ld.sync.aligned;");
|
|
if (lane == 0) for (int c=0;c<8;c++) {
|
|
s_vals[n*8+c] = tmp[c] * scale;
|
|
row_max = fmaxf(row_max, tmp[c] * scale);
|
|
}
|
|
}
|
|
row_max = wmax(row_max);
|
|
float row_sum = 0.0f;
|
|
if (lane == 0) for (int j=0;j<SK;j++) {
|
|
s_vals[j] = expf(s_vals[j] - row_max);
|
|
row_sum += s_vals[j];
|
|
}
|
|
row_sum = wsum(row_sum);
|
|
if (lane == 0) for (int j=0;j<SK;j++) s_vals[j] /= row_sum;
|
|
if (lane == 0) for (int j=0;j<SK;j++) s_p_vals[j] = s_vals[j];
|
|
}
|
|
__syncthreads();
|
|
|
|
// ===== ZERO ALL 128 TMEM COLUMNS using 32x32b.x8 stores =====
|
|
// All 32 lanes must call. Each lane writes 8 zero FP32 values.
|
|
if (wid == 0) {
|
|
float z0 = 0.0f, z1 = 0.0f, z2 = 0.0f, z3 = 0.0f;
|
|
float z4 = 0.0f, z5 = 0.0f, z6 = 0.0f, z7 = 0.0f;
|
|
for (int n = 0; n < 16; n++) {
|
|
asm volatile("tcgen05.st.sync.aligned.32x32b.x8.b32 [%0], {%1, %2, %3, %4, %5, %6, %7, %8};"
|
|
:: "r"(tb + n*8), "f"(z0), "f"(z1), "f"(z2), "f"(z3), "f"(z4), "f"(z5), "f"(z6), "f"(z7));
|
|
}
|
|
tmem_fence_store();
|
|
}
|
|
__syncthreads();
|
|
|
|
// PV MMA (N=64) with accumulate=false for first K-tile
|
|
{
|
|
uint32_t idesc_pv = make_idesc(BLOCK_MN, HD);
|
|
for (int kt = 0; kt < 8; kt++) {
|
|
for (int i = tid; i < TILE_SZ; i += 128) sPk[i] = 0;
|
|
if (tid < 16) {
|
|
int c = tid;
|
|
int ck = c / 8, lc = c % 8;
|
|
sPk[ck * 16 * 64 + 0 * 64 + 0 * 8 + lc] = f32_to_bf16(s_p_vals[kt * LOCAL_MMA_K + c]);
|
|
}
|
|
__syncthreads();
|
|
bf16_t* sv = sV + kt * V_TILE_SZ;
|
|
uint64_t dp = make_umma_desc_kmajor_none(__cvta_generic_to_shared(sPk), BLOCK_MN);
|
|
uint64_t dv = make_umma_desc_kmajor_none(__cvta_generic_to_shared(sv), HD);
|
|
if (tid == 0) umma_ss_f16(tb, dp, dv, idesc_pv, kt > 0);
|
|
asm volatile("tcgen05.fence::after_thread_sync;" ::: "memory");
|
|
__syncthreads();
|
|
}
|
|
}
|
|
|
|
// ===== Dump ALL 128 TMEM columns (lane 0 only) =====
|
|
if (wid == 0) {
|
|
for (int n = 0; n < 16; n++) {
|
|
float tmp[8];
|
|
asm volatile("tcgen05.ld.sync.aligned.32x32b.x8.b32 {%0,%1,%2,%3,%4,%5,%6,%7},[%8];"
|
|
: "=f"(tmp[0]),"=f"(tmp[1]),"=f"(tmp[2]),"=f"(tmp[3]),
|
|
"=f"(tmp[4]),"=f"(tmp[5]),"=f"(tmp[6]),"=f"(tmp[7])
|
|
: "r"(tb + n*8));
|
|
asm volatile("tcgen05.wait::ld.sync.aligned;");
|
|
for (int c = 0; c < 8; c++) {
|
|
int col = n * 8 + c;
|
|
tmem_dump[col] = tmp[c]; // lane 0's value for this column
|
|
}
|
|
}
|
|
}
|
|
__syncthreads();
|
|
|
|
if (wid == 0) tmem_dealloc(tb, 128);
|
|
}
|
|
|
|
int main() {
|
|
printf("=== TMEM dump after zeroing + PV MMA N=64 ===\n");
|
|
const float SCALE = 1.0f / sqrtf((float)HD);
|
|
|
|
bf16_t* h_q = (bf16_t*)malloc(HD*sizeof(bf16_t));
|
|
bf16_t* h_k = (bf16_t*)malloc(SK*HD*sizeof(bf16_t));
|
|
bf16_t* h_v = (bf16_t*)malloc(HD*SK*sizeof(bf16_t));
|
|
|
|
srand(42);
|
|
for (int d=0;d<HD;d++) h_q[d] = f32_to_bf16_host((float)(rand()%100)/100.0f-0.5f);
|
|
for (int i=0;i<SK*HD;i++) h_k[i] = f32_to_bf16_host((float)(rand()%100)/100.0f-0.5f);
|
|
for (int i=0;i<HD*SK;i++) h_v[i] = f32_to_bf16_host((float)(rand()%100)/100.0f-0.5f);
|
|
|
|
bf16_t *d_q,*d_k,*d_v;
|
|
float* d_tmem_dump;
|
|
cudaMalloc(&d_q, HD*sizeof(bf16_t));
|
|
cudaMalloc(&d_k, SK*HD*sizeof(bf16_t));
|
|
cudaMalloc(&d_v, HD*SK*sizeof(bf16_t));
|
|
cudaMalloc(&d_tmem_dump, 128 * sizeof(float));
|
|
cudaMemset(d_tmem_dump, 0, 128 * sizeof(float));
|
|
cudaMemcpy(d_q, h_q, HD*sizeof(bf16_t), cudaMemcpyHostToDevice);
|
|
cudaMemcpy(d_k, h_k, SK*HD*sizeof(bf16_t), cudaMemcpyHostToDevice);
|
|
cudaMemcpy(d_v, h_v, HD*SK*sizeof(bf16_t), cudaMemcpyHostToDevice);
|
|
|
|
int smem = (4+16 + 4*TILE_SZ*2 + 4*TILE_SZ*2 + TILE_SZ*2 + 8*V_TILE_SZ*2 + SK*4 + 256 + 127) & ~127;
|
|
cudaFuncSetAttribute(test_tmem_zero_before_pv, cudaFuncAttributeMaxDynamicSharedMemorySize, smem);
|
|
test_tmem_zero_before_pv<<<1, 128, smem>>>(d_q, d_k, d_v, d_tmem_dump, SCALE);
|
|
|
|
cudaError_t err = cudaDeviceSynchronize();
|
|
if (err != cudaSuccess) { printf("CUDA ERROR: %s\n", cudaGetErrorString(err)); return 1; }
|
|
|
|
float* h_dump = (float*)malloc(128 * sizeof(float));
|
|
cudaMemcpy(h_dump, d_tmem_dump, 128 * sizeof(float), cudaMemcpyDeviceToHost);
|
|
|
|
// Reference
|
|
float s[SK];
|
|
for (int j=0;j<SK;j++) {
|
|
float dot = 0.0f;
|
|
for (int d=0;d<HD;d++) dot += bf16_to_f32_host(h_q[d]) * bf16_to_f32_host(h_k[j*HD+d]);
|
|
s[j] = dot * SCALE;
|
|
}
|
|
float mx = -INFINITY;
|
|
for (int j=0;j<SK;j++) mx = fmaxf(mx, s[j]);
|
|
float sm = 0.0f;
|
|
for (int j=0;j<SK;j++) { s[j] = expf(s[j]-mx); sm += s[j]; }
|
|
for (int j=0;j<SK;j++) s[j] /= sm;
|
|
float o_ref[HD];
|
|
for (int d=0;d<HD;d++) {
|
|
float ov = 0.0f;
|
|
for (int j=0;j<SK;j++) ov += s[j] * bf16_to_f32_host(h_v[d*SK+j]);
|
|
o_ref[d] = ov;
|
|
}
|
|
|
|
printf("\n=== TMEM dump (lane 0, columns 0-127) after zero + PV ===\n");
|
|
for (int col = 0; col < 128; col++) {
|
|
if (fabsf(h_dump[col]) > 1e-6f) {
|
|
// Try to find matching reference output
|
|
int best_d = -1; float best_diff = 1e10f;
|
|
for (int d = 0; d < HD; d++) {
|
|
float diff = fabsf(h_dump[col] - o_ref[d]);
|
|
if (diff < best_diff) { best_diff = diff; best_d = d; }
|
|
}
|
|
printf(" col %3d: %10.6f", col, h_dump[col]);
|
|
if (best_diff < 1e-3f) printf(" ≈ ref[%d]=%.6f (diff=%.2e)", best_d, o_ref[best_d], best_diff);
|
|
printf("\n");
|
|
}
|
|
}
|
|
|
|
// Count non-zero columns in 0-63 range
|
|
int nz = 0;
|
|
for (int col = 0; col < 64; col++) if (fabsf(h_dump[col]) > 1e-6f) nz++;
|
|
printf("\nNon-zero columns 0-63: %d (expected 64)\n", nz);
|
|
nz = 0;
|
|
for (int col = 64; col < 128; col++) if (fabsf(h_dump[col]) > 1e-6f) nz++;
|
|
printf("Non-zero columns 64-127: %d (expected 0)\n", nz);
|
|
|
|
// Which columns 0-63 are ZERO?
|
|
printf("\nMissing columns (0 in 0-63): ");
|
|
for (int col = 0; col < 64; col++) if (fabsf(h_dump[col]) < 1e-6f) printf("%d ", col);
|
|
printf("\n");
|
|
|
|
cudaFree(d_q); cudaFree(d_k); cudaFree(d_v); cudaFree(d_tmem_dump);
|
|
free(h_q); free(h_k); free(h_v); free(h_dump);
|
|
return 0;
|
|
}
|