vulkan: add missing clamps in new mul_mat_id paths (ggml-org#15702)

jeffbolznv · web-flow · commit 35a42edac846 · 2025-09-01T21:01:10.000+02:00
This is a missing interaction between ggml-org#15546 and ggml-org#15652
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp
@@ -494,6 +494,9 @@ void main() {
                     sum = coopMatMulAdd(mat_a, mat_b, sum);
                 }
             }
+#if defined(ACC_TYPE_MAX)
+            [[unroll]] for (uint i = 0; i < sum.length(); ++i) { sum[i] = clamp(sum[i], -ACC_TYPE_MAX, ACC_TYPE_MAX); }
+#endif
 
             // Convert from ACC_TYPE to D_TYPE
             coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BNover4, gl_MatrixUseAccumulator> mat_d;
@@ -535,6 +538,9 @@ void main() {
                     sum = coopMatMulAdd(mat_a, mat_b, sum);
                 }
             }
+#if defined(ACC_TYPE_MAX)
+            [[unroll]] for (uint i = 0; i < sum.length(); ++i) { sum[i] = clamp(sum[i], -ACC_TYPE_MAX, ACC_TYPE_MAX); }
+#endif
 
             // Convert from ACC_TYPE to D_TYPE
             coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BNover2, gl_MatrixUseAccumulator> mat_d;

Original file line number	Diff line number	Diff line change
`@@ -494,6 +494,9 @@ void main() {`
`494`	`494`	`sum = coopMatMulAdd(mat_a, mat_b, sum);`
`495`	`495`	`}`
`496`	`496`	`}`
	`497`	`+#if defined(ACC_TYPE_MAX)`
	`498`	`+ [[unroll]] for (uint i = 0; i < sum.length(); ++i) { sum[i] = clamp(sum[i], -ACC_TYPE_MAX, ACC_TYPE_MAX); }`
	`499`	`+#endif`
`497`	`500`
`498`	`501`	`// Convert from ACC_TYPE to D_TYPE`
`499`	`502`	`coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BNover4, gl_MatrixUseAccumulator> mat_d;`
`@@ -535,6 +538,9 @@ void main() {`
`535`	`538`	`sum = coopMatMulAdd(mat_a, mat_b, sum);`
`536`	`539`	`}`
`537`	`540`	`}`
	`541`	`+#if defined(ACC_TYPE_MAX)`
	`542`	`+ [[unroll]] for (uint i = 0; i < sum.length(); ++i) { sum[i] = clamp(sum[i], -ACC_TYPE_MAX, ACC_TYPE_MAX); }`
	`543`	`+#endif`
`538`	`544`
`539`	`545`	`// Convert from ACC_TYPE to D_TYPE`
`540`	`546`	`coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BNover2, gl_MatrixUseAccumulator> mat_d;`