hacky KHR cooperative matrix prototype

ggml/src/vulkan-shaders/mul_mm.comp

@@ -2,6 +2,9 @@
 
 #extension GL_EXT_control_flow_attributes : enable
 #extension GL_EXT_shader_16bit_storage : require
+#extension GL_KHR_cooperative_matrix : require
+#extension GL_KHR_memory_scope_semantics : require
+#extension GL_EXT_shader_explicit_arithmetic_types : require
 
 #ifdef FLOAT16
 #extension GL_EXT_shader_explicit_arithmetic_types_float16 : require
@@ -152,12 +155,10 @@ void main() {
     uint pos_b = (batch_idx * p.batch_stride_b + ic * BN * p.stride_b + start_k) / LOAD_VEC_B;
 #endif
 
-    float sums[WMITER * TM * WNITER * TN];
-    FLOAT_TYPE cache_a[WMITER * TM];
-    FLOAT_TYPE cache_b[WNITER * TN];
+    coopmat<float, gl_ScopeSubgroup, 16, 16, gl_MatrixUseAccumulator> sums[WM * WN / 16 / 16];
 
-    [[unroll]] for (uint i = 0; i < WMITER*TM*WNITER*TN; i++) {
-        sums[i] = 0.0f;
+    [[unroll]] for (uint i = 0; i < WM * WN / 16 / 16; i++) {
+        sums[i] = coopmat<float, gl_ScopeSubgroup, 16, 16, gl_MatrixUseAccumulator>(0.0);
     }
 
     [[unroll]] for (uint block = start_k; block < end_k; block += BK) {
@@ -446,27 +447,14 @@ void main() {
         pos_a += BK / LOAD_VEC_A;
         pos_b += BK / LOAD_VEC_B;
 
-        for (uint i = 0; i < BK; i++) {
-            // Load from shared into cache
-            [[unroll]] for (uint wsir = 0; wsir < WMITER; wsir++) {
-                [[unroll]] for (uint j = 0; j < TM; j++) {
-                    cache_a[wsir * TM + j] = buf_a[(warp_r * WM + wsir * WSUBM + tiwr * TM + j) * (BK+1) + i];
-                }
-            }
-            [[unroll]] for (uint wsic = 0; wsic < WNITER; wsic++) {
-                [[unroll]] for (uint j = 0; j < TN; j++) {
-                    cache_b[wsic * TN + j] = buf_b[(warp_c * WN + wsic * WSUBN + tiwc * TN + j) * (BK+1) + i];
-                }
-            }
-
-            [[unroll]] for (uint wsic = 0; wsic < WNITER; wsic++) {
-                [[unroll]] for (uint wsir = 0; wsir < WMITER; wsir++) {
-                    [[unroll]] for (uint cc = 0; cc < TN; cc++) {
-                        [[unroll]] for (uint cr = 0; cr < TM; cr++) {
-                            const uint sums_idx = (wsic * TN + cc) * (WMITER * TM) + wsir * TM + cr;
-                            sums[sums_idx] = fma(float(cache_a[wsir * TM + cr]), float(cache_b[wsic * TN + cc]), sums[sums_idx]);
-                        }
-                    }
+        [[unroll]] for (uint i = 0; i < WM; i += 16) {
+            [[unroll]] for (uint j = 0; j < WN; j += 16) {
+                [[unroll]] for (uint k = 0; k < BK; k += 16) {
+                    coopmat<float16_t, gl_ScopeSubgroup, 16, 16, gl_MatrixUseA> matA;
+                    coopmat<float16_t, gl_ScopeSubgroup, 16, 16, gl_MatrixUseB> matB;
+                    coopMatLoad(matA, buf_a, (warp_r * WM + i) * (BK+1) + k, BK+1, gl_CooperativeMatrixLayoutRowMajor);
+                    coopMatLoad(matB, buf_b, (warp_c * WN + j) * (BK+1) + k, BK+1, gl_CooperativeMatrixLayoutColumnMajor);
+                    sums[(i / 16) * (WN / 16) + (j / 16)] = coopMatMulAdd(matA, matB, sums[(i / 16) * (WN / 16) + (j / 16)]);
                 }
             }
         }
@@ -481,6 +469,19 @@ void main() {
     const uint offsets = batch_idx * p.batch_stride_d + ik * p.batch_stride_d * gl_NumWorkGroups.z;
 #endif
 
+#if 1
+#ifndef MUL_MAT_ID
+    // XXX TODO this is missing bounds checking against p.M and p.N,
+    // which probably requires spilling to shared memory and doing scalar stores.
+    // But sums[] may not all fit in shared memory...
+    [[unroll]] for (uint i = 0; i < WM; i += 16) {
+        [[unroll]] for (uint j = 0; j < WN; j += 16) {
+            coopMatStore(sums[(i / 16) * (WN / 16) + (j / 16)], data_d, offsets + (dc + j) * p.stride_d + dr + i, p.stride_d, gl_CooperativeMatrixLayoutColumnMajor);
+        }
+    }
+#endif
+#else
+
     [[unroll]] for (uint wsic = 0; wsic < WNITER; wsic++) {
         [[unroll]] for (uint wsir = 0; wsir < WMITER; wsir++) {
 
@@ -505,4 +506,5 @@ void main() {
             }
         }
     }
+#endif
 }